Home

southern lakes, yukon territory earth cover classification user's guide

1. 57 Table 2 Error Matrix at 5 Level of Variation Classified Data tease aes ee se eee ae ee eee ee error of commission ClosedNeedieleaf_ 4 s5 So Joad T T T S S T 8 Open Needleleaf Lichen 3f o loll Jo T S S S f S f o Open Spruce Open Pine B r R Open Low Shrub Hero 15 Jo o Too To ST S S S f Ssa S fS o o To T T S S S S f S day fa o T S S T e S S S Saa ed Dwarf Sirub Herb E E Se ee ee ee ee ee eee 9 po IH IH ITH IIH I IIT S S 5 d ID IH ee ID IT ee ID ID ee IHE d oo S sm 7 0o lez lo lalm O T E L H o E C E E l 9 ET IT error of omission 86 1 Total Number of Sites 115 Diagonal Sum 99 Denotes mapped classes that accuracy sites from other mapped classes fell into due to error of commission Off diagonal Sum 16 These classes were therefore included in the error matrix but no accuracy assessment was done for these two classes Overall Accurac 86 1 58 Appendix F DUFF Data and Photo Viewer Extension Installation and How to Run the Extension Version 1 0 Edited By Ruth Spell Ducks Unlimited Inc Western Regional Office By LEMA Center Utah State University John Lowry Project Coordinator jlofwry gis usu edu Bill Wagner LEMA Manager bwag cnr usu edu Nathan Bentley LEMA Coordinator Nathan cnr usu edu Pat Terletzky GIS Technician patt cnr usu edu July 2000 Installation and Requirements Requirements 1 ArcView 3
2. 275 Needleleaf Single Species Closed Spruce Fir Shrub Closed Spruce Fir Moss Open Spruce Fir Shrub Open Spruce Fir Moss Open Spruce Fir Lichen Open Spruce Fir Herb NO me Closed Spruce Fir Closed Spruce Fir Other YES Closed Pine Closed Pine Shrub Closed Pine Moss Closed Pine Other Closed Mixed NL Shrub Closed Mixed NL gt 25 Understory YES Closed Mixed NL Moss Closed Mixed NL Other Open Needleleaf Open YES i YES Spruce Fir pen Needleleaf Liche 7 p 25 59 cc Open Needleleaf Moss Sin A E aeS gt on Understory YES Open Needleleaf Herb ERE 2 20 Lichen Open Pine Open Needleleaf Shrub NO Open Pine Shrub Open Pine Moss Open Pine Lichen Open Pine Herb Open Pine Other Open Spruce Fir Other Open Mixed NL Shrub 2 gt 25 Understory YES M NL F Open Mixed gt 20 Lichen NO Open Mixed NL Moss Open Mixed NL Lichen Open Mixed NL Herb Open Mixed NL Other 43 YES ae Closed N YES gt 75 Deciduous 275 Single Species NO NO NO Closed Birch Poplar YES gt 25 Understory Closed Aspen NO Closed BP Shrub Closed BP Herb Closed Aspen Shrub Closed Aspen Herb Closed BP Other Closed Aspen Other Closed Mixed Deciduous c YE 25 59 CC 275 Single Species 2 NO YES Closed Mixed YES YES 0 0 gt 60 ce NL Deciduous gt 25 Understory Open Birch Poplar
3. HEIGHT Year of field data collection 4 letter code for project area ID number of crew that collected data Field site number relates to SITE NUM of field site polygon coverage in a one to many relationship Each site may have multiple species records in this table Percent coverage of the species as observed at site by vegetation caller Height of tree or shrub species at site as observed by vegetation caller 63
4. gt 25 Understory 15 Open Aspen NO Open BP Shrub Open BP Herb Open Aspen Shrub Open Aspen Herb Open Aspen Other Open BP Other Open Mixed Deciduous Closed Mixed NL Deciduous Shrub Closed Mixed NL Deciduous Moss Closed Mixed NL Deciduous Other Open Mixed NL Deciduous 44 l YES YES Trees 10 24 1 M Height Woodland NO YES Highest Shrub YES Shrubs 25 100 is gt 1 5M tall OR NO highest is Tall Shrub NO Highest Shrub is gt 0 25 lt 1 5 tall OR YES highest is Low Shrub Highest Shrub is lt 0 25 tall OR highest is Dwarf Shrub YES gt 30 Herbaceous AND gt 50 Byroi lt 10 Water eRe NO NO YE gt 5 and lt 10 Water gt 50 Graminoid Sedge Grass Tussock NO gt 50 Grass amp or Tussock gt 25 Understory YES 220 Lichen gt 60 canopy cover 1E NO gt 25 Understory YES 220 Lichen NO YES gt 60 canopy cover NO YES Open Low Shrub Dwarf Shrub gt 50 Lichen ZES NO gt 35 Tussock NO gt 50 Sedge gt 50 Grass gt 25 Understory 220 Lichen Woodland Lichen Woodland Shrub Woodland Herbaceous Woodland Other Closed Tall Shrub Open TS Herbaceous Open TS LS Open TS Lichen Open TS Moss Open TS Other Closed Low Shrub gt 25 Understory YES 220 Lichen DS Other NO gt 20 Lichen NO Y
5. rock water ice cloud and shadow classes were added Other classes that could not reliably be discerned from satellite imagery had to be collapsed such as open and closed tall shrub classes ericaceous willow and dwarf shrub classes Because of the importance of lichen for site characterization and wildlife and because the presence of lichen can be detected by satellite imagery shrub and forested classes with and without a component of lichen were distinguished A few classes from Level IV of the Viereck et al 1992 classification were also mapped because of their identifiable satellite signature and their importance for wildlife management These Level IV classes included tussock tundra low shrub tussock tundra and low shrub willow alder Table 1 Classification scheme developed at the BLM Earth Cover Workshop Level II Level III Level IV 1 0 Forest 1 1 Closed Needleleaf 1 2 Open Needleleaf 1 21Open Needleleaf Lichen 1 3 Woodland Needleleaf 1 31 Woodland Needleleaf Lichen 1 4 Closed Deciduous 1 41 Closed Paper Birch 1 42 Closed Aspen 1 43 Closed Balsam Poplar Cottonwood 1 44 Closed Mixed Deciduous 1 5 Open Deciduous 1 51 Open Paper Birch 1 52 Open Aspen 1 53 Open Balsam Poplar Cottonwood 1 54 Open Mixed Deciduous 1 6 Closed Mixed Needleleaf Deciduous 1 7 Open Mixed Needleleaf Deciduous 2 0 Shrub 2 1 Tall Shrub 3 0 Herbaceous 4 0 Aquatic Vegetation 5 0 Water 6 0 Barren 7 0 Urban 2 2 Low Shrub 2 3 Dwarf Shrub 3
6. Moss 1 133 Closed Mixed NL Other 1 2111 Open Spruce Fir Lichen 1 2112 Open Spruce Fir Shrub 1 2113 Open Spruce Fir Moss 1 2114 Open Spruce Fir Herb 1 2115 Open Spruce Fir Other 1 261 Open Pine Lichen 1 262 Open Pine Shrub 1 263 Open Pine Moss 1 264 Open Pine Herb 1 265 Open Pine Other 1 271 Open Nixed NL Lichen 1 272 Open Nixed NL Shrub 1 273 Open Nixed NL Moss 1 274 Open Nixed NL Herb 1 275 Open Nixed NL Other 11 2 0 Shrub 3 0 Herbaceous 4 0 Aquatic Veg 5 0 Water 6 0 Barren 7 0 Urban 8 0 Agriculture 9 0 Cloud Shadow 10 0 Other 1 4 Closed Deciduous 1 5 Open Deciduous 1 6 Closed Mixed NL Deciduous 1 7 Open Mixed NL Deciduous 2 1 Tall Shrub 2 2 Low Shrub 2 3 Dwarf Shrub 3 1 Bryoid 3 2 Wet Herbaceous 3 3 Mesic Dry Herbaceous 4 1 Aquatic Bed 4 2 Emergent Vegetation 5 1 Snow 5 2 Ice 5 3 Clear Water 5 4 Turbid Water 6 1 Sparsely Vegetated 6 2 Rock Gravel 6 3 Non Veg Soil 9 1 Cloud 9 2 Shadow 1 34 Woodland Needleleaf Other 1 41 Closed Paper Birch Poplar 1 42 Closed Aspen 1 44 Closed Mixed Deciduous 1 51 Open Paper Birch Poplar 1 52 Open Aspen 1 53 Open Mixed Deciduous 1 61 Closed Mixed NL Deciduous Shrub 1 62 Closed Mixed NL Deciduous Moss 1 63 Closed Mixed NL Deciduous Other 2 11 Closed Tall Shrub 2 12 Open Tall Shrub 2 21 Closed Low Shrub 2 22 Open Low Shrub 2 31 Dwarf Shrub Lichen 2 32 Dwarf Shrub
7. it is certain that both the user s and producer s accuracy for these classes would be at or very near 100 thus only acting to improve the overall accuracy for the final earth cover map One site was discarded from the accuracy assessment where the error was not able to be reconciled e the site was labeled as closed needeleleaf when it occurred in a dwarf shrub lichen area In a few other cases the accuracy sites were reclassed from one class to another class where they were interpreted to be incorrect One example was a tall shrub site classed as dwarf shrub The actual shrub height of over 4 meters was not recorded on the field form and the site was calculated as dwarf shrub This site was reclassed to closed tall shrub from dwarf shrub other 7 4 1 Overall and Class Accuracy The difference in classification accuracy between the 0 variation in interpretation level 76 5 and the 5 variation in interpretation level 86 1 indicated that characteristics of many of the reference data sites were close to the threshold of two or more mapping classes It is generally accepted that variation in interpretation of 10 is common for human interpreters either from aerial photography or on the ground When this natural and accepted variation is measured and accounted for as in the case of the error matrix in Table 2 Appendix E a more reliable and informative measure of accuracy 1s presented The acceptance or unacceptance of each a
8. ocsiccescsacsecssvaidesteeecasseccaaesiusraaniescndenstssuceadesusesncesseveatecestsseseereirtess 1 ECOL EIO UTOS oruc E E E L T 4 Leco TADIS car E E meee L 4 ADS AGE aerosani aE EEE O AE EEEE EE E E 5 LO RTS S UC CIO Dessiner sesati aaa EEEE IA EEEa 6 20 Projec G R LO A 6 3 0 Project Participants Funding secessvcsssessesavenssessesiacestccasavsseesacevanessdessnsennesiccansensses 6 AW 2 N c E E E O E E E O O EO EE E A E E OE 7 30 Data ACOSO icsse a EEN aes 8 OO Methods srar riaa EE EAEE EEEE EAA 9 OT Samphino 1 T ne EE OEN EENES 9 Odet ebeL SO NENE onna E E E E 9 CE DPR I T 13 OL SDE o T 13 GLA Sampie SCICCUON sasn n E 14 6 2 Image PVCPVOCESSING T 14 6 3 Field dE Ta T a A R 15 64 Field Data ANGIV SAS a I OY CUUSSIPIC AMON wis con scoscasiesesncccacsnst tases E E EE saeeseeaseu sucess 19 6 5 1 Generation of New Bands sese 19 O52 SC CAINS PrOCE S S aaea E E E 19 6 5 3 Generation of Unsupervised Signatures cccccccccsssseeeeeeeeeeeeeeeees 19 6 5 4 Modified Supervised Unsupervised Classification cccccccccceeeees 19 OS Sra ET IN eN Ode T 21 6 6 Change R rT a a i 21 s TIE 6 201 CX TTT 21 6 6 2 Change Detection Methods acc sescsccavecceseceesocadversdasaceasccslassdsssdaweeasnteess 22 O ACCUTAC ASSESSMENT oaa ERRE TEE E E AEE 23 See ee Enor E eT 23 PAW PROS UIE S ocsicccesescscnsescescanccdacsesevdusanescecssceusawscanscnawsionwswsansncceeceavesusacescvarsiarssaceacdsceseane 24 Tel Prelad VCViUC
9. the value of a pixel might be the result of the mud water trees and wetland plants all at once For this reason wetland areas often confused with closed needleleaf open needleleaf rock gravel and terrain shadow cover types All of the sites containing wetland pixels that were selected during the site selection process classified as other in the classification scheme These sites were a mix of several different classes and were not useful as training sites for wetland types Four emergent wetland sites were added in the field and notes were made if emergent or aquatic bed areas were encountered during the field work This information was used to generate the classification for these types 29 Figure 8 One of the many shallow depressional lakes in the study area 7 2 3 Shrub Vegetation Cover Types Shrub dominated cover types were very common throughout the image They were common in burn areas areas with poorer soils not able to support forested vegetation and the areas above the treeline and below the bare rock mountaintops The difference in shrub height was not always easily distinguishable in the imagery as there was a wide variation in greenness in the shrub classes throughout the image The coastal influenced valleys in the southwestern portion of the image were much more lush than many of the interior influenced shrub sites Since these coastally influenced shrub areas exhibited a different signature than the other parts of t
10. 31 earth cover categories The classification scheme was based on Viereck et al 1992 in Alaska but was modified to represent the earth cover communities found in the Yukon Mapped classes included open and closed canopy forests tall low and dwarf shrubs herbaceous lichen and wetlands as well as unvegetated and anthropogenic types The overall accuracy of the mapped categories was 86 at the 5 level of variation A change detection was performed using an August 24 1987 Landsat TM 5 image Changes in earth cover included those due to fires urban expansion conversion of arable land to agriculture receding snow packs and lowering of water levels Standardized procedures were used to conduct the earth cover classification An unsupervised clustering technique was used to determine the location of the field sites to ground truth the satellite imagery Global positioning system GPS technology was used to navigate to the pre selected sites and to record new sites selected in the field A total of 479 sites were visited by Bell Jet Ranger helicopter during July 15 26 2000 Field information was recorded on a custom field data form and digital database Approximately 30 of the field sites were set aside for accuracy assessment A modified supervised unsupervised classification technique was performed to classify the satellite imagery 1 0 Introduction Ducks Unlimited Inc and Ducks Unlimited Canada are in the second year of a multi year pr
11. Moss 2 33 Dwarf Shrub Herb 2 34 Dwarf Shrub Other 3 11 Lichen 3 12 Moss 3 21 Wet Graminoid 3 22 Wet Forb 3 31 Tussock Tundra 3 32 Tussock Tundra Lichen 3 33 Mesic Dry Sedge 3 34 Mesic Dry Grass 3 35 Mesic Dry Graminoid 3 36 Mesic Dry Forb 1 411 Closed Birch Poplar Shrub 1 412 Closed Birch Poplar Herb 1 413 Closed Birch Poplar Other 1 421 Closed Aspen Shrub 1 422 Closed Aspen Herb 1 423 Closed Aspen Other 1 511 Open Paper Birch Poplar Shrub 1 512 Open Paper Birch Poplar Herb 1 513 Open Paper Birch Poplar Other 1 521 Open Aspen Shrub 1 522 Open Aspen Herb 1 523 Open Aspen Other 2 121 Open Tall Shrub Lichen 2 122 Open Tall Shrub Moss 2 123 Open Tall Shrub Herb 2 124 Open Tall Shrub Low Shrub 2 125 Open Tall Shrub Other 2 221 Open Low Shrub Lichen 2 222 Open Low Shrub Moss 2 223 Open Low Shrub Herb 2 224 Open Low Shrub Other 12 6 1 2 Sampling Unit The sampling unit for a project refers to the size of the field site area that will be sampled for the training of the classification and accuracy assessment The sampling unit most useful for both techniques was determined to have the following criteria 1 Larger than the minimum mapping unit to minimize errors of omission due to misregistration 2 Could be located easily within the image using GPS technology 3 Was not an arbitrary designation 1 e formed a representative area of a cover class be
12. and the TM Band 7 difference image was used to detect changes in snow cover urban expansion fires and agriculture changes Image differencing techniques provided the difference in spectral reflectance between 1987 and 1999 for each of the bands This final difference image exhibits a range of values from negative to positive depending on the brightness values of each image For this specific change detection modeling or statistical classification of change signatures were not possible due to the overlap of difference values among change classes For example a fire and a clearcut might exhibit the same difference in spectral signatures between the two dates but would belong to different change classes In addition some changes were nearly impossible to separate using unsupervised classification techniques but were readily classified by an analyst using visual references Pattern recognition context and other visual cues were all used to separate classes For these reasons a decision was made to manually classify the difference image into the different change categories This process was done using field data aerial photographs and the interpretive ability of the analyst Because no data was collected at the time of the 1987 image overpass no accuracies were provided for the change detection analysis Sources of error were minimized where possible One source of error that was present was misregistration Although the overall pixel error for
13. in Appendix B Although this classification scheme provided very detailed information for all of the sites visited in the field the final classification did not have all of the classes listed Since there were nearly 100 classes about 1500 or more sites would have to be recorded in the field in order to get a statistically valid sample for each class and to accurately classify all of the classes in this scheme This level of sampling was beyond the scope of this project The classification scheme was designed so that the higher level classes could be rolled up into more general classes if the inherent limits of the TM sensor to discriminate that level of vegetation were exceeded Table 2 Southern Lakes Yukon Classification Scheme Level II Level HI 1 0 Forest 1 1 Closed Needleleaf NL 1 2 Open Needleleaf 1 3 Woodland Needleleaf Level IV 1 11 Closed Spruce Fir 1 12 Closed Pine 1 13 Closed Mixed Needleleaf 1 21 Open Needleleaf Lichen 1 22 Open Needleleaf Moss 1 23 Open Needleleaf Herb 1 24 Open Needleleaf Shrub 1 25 Open Spruce Fir 1 26 Open Pine 1 27 Open Mixed Needleleaf 1 31 Woodland Needleleaf Lichen 1 32 Woodland Needleleaf Shrub 1 33 Woodland Needleleaf Herb Level V 1 111 Closed Spruce Fir Shrub 1 112 Closed Spruce Fir Moss 1 113 Closed Spruce Fir Other 1 121 Closed Pine Shrub 1 122 Closed Pine Moss 1 123 Closed Pine Other 1 131 Closed Mixed NL Shrub 1 132 Closed Mixed NL
14. in this project was not a robust test of the classification it gives the user some confidence in using the classification It provided enough detail for the end user to determine where discrepancies in the classification may cause a problem while using the data 38 7 5 Final Products The final products include a digital earth cover classification a 1987 1999 change detection coverage a hard copy map of the entire project area and a digital database of field data collected for the 479 field sites visited during this project The digital map was delivered in Erdas Imagine format Digital photographs of the field sites are stored in jpg format and can be viewed using the DUFF Data and Photo Viewer ArcView extension Appendix F The site photos field site information and vegetative species lists are stored as digital tables in Dbase IV format described in Appendix G Hardcopy maps of the entire project area at the 1 250 000 scale as well as 1 50 000 scale quadrangles were also produced of the classified area All of the delivered datasets were loaded into Arcview projects for display purposes 8 0 Summary The total area mapped in this project was 3 8 million hectares Classification was performed using Landsat 7 ETM satellite scenes Path 59 Rows 17 and 18 acquired on a cloudless day on August 1 1999 The project area was classified into 31 earth cover categories with an overall accuracy of 86 at the 5 level of variation in interpr
15. plant groups were present including low and tall shrubs forbs grasses sedges horsetails mosses and lichens The deciduous tree species generally found were aspen Populus tremuloides and balsam poplar P balsamifera and less commonly paper birch Betula papyfera Under some conditions willow Salix spp and alder Alnus incana formed a significant part of the tree canopy Deciduous stands were found in major river valleys on alluvial flats surrounding lakes or most commonly on the steep south facing slopes Mixed deciduous coniferous stands were present in the same areas as pure deciduous stands 1 1 Closed Needleleaf At least 60 of the cover was trees and gt 75 of the trees were needleleaf trees Common distribution throughout image 1 2 Open Needleleaf From 25 59 of the cover was trees and gt 75 of the trees were needleleaf Distributed throughout study area in transitional areas 1 21 Open Needleleaf Lichen From 25 59 of the cover was trees gt 75 of the trees were needleleaf and gt 20 of the understory was lichen Common throughout study area 1 3 Woodland Needleleaf From 10 24 of the cover was trees and gt 75 of the trees were needleleaf Very common highly variable 1 31 Woodland Needleleaf Lichen From 10 24 of the cover was trees gt 75 of the trees were needleleaf and gt 20 of the understory was lichen 1 4 Closed Deciduous Mixed Deciduous Species 1 45 At least 60 of the cover was t
16. programmed in Visual Basic The user interface was organized similarly to the field form to facilitate data entry Figure 3 The application utilized pull down menus to minimize keystrokes and checked for data integrity to minimize data entry errors The database program also calculated an overall class name for each site based on the recorded species and its cover percentage Digital images from each site were stored in the database and accessible from within the user interface The number of field sites per earth cover class was tracked daily to ensure that adequate samples were being obtained within each class 17 Field Site Closed Mixed Needleleaf Deciduous LN As WA oat j High site photo T T Low site photo DUFF INTERFACE jets Ducks Unlimited File Tools Help 2000 SLAK 4 4 76 Observation Crews Check Flag military EEE K Tear Project Crew Ste Nay Veg Rec Obsery Date Obs Level Obs Time click to search Delete New JKS JKS Hw 21 Jul 00 2 03 44 Ie Session Photo Lat degrees decimal min Long Z Slope m Aspect Avg Dist Bhan Stem l o0ao0 OOoo00 oooao0 o000 5 o ME 1 gt 1080 1 gt 1079 1 1072 All Species Latin Common Show All Species Observed Classes Add Delete Edit a Observed Species C FOREST CLOSED SPRUCE FIR MOS C FOREST CLOSED SPRUCE FIR OTH oe oa S 2 _ c Hent C FOREST CLOSED PINE SHALIB C FOREST C
17. 1 Bryoid 3 2 Wet Herbaceous 3 3 Mesic Dry Herbaceous 4 1 Aquatic Bed 4 2 Emergent Vegetation 5 1 Snow 5 2 Ice 5 3 Clear Water 5 4 Turbid Water 6 1 Sparsely Vegetated 6 2 Rock Gravel 6 3 Mud Silt Sand 2 21 Low Shrub Willow Alder 2 22 Low Shrub Tussock Tundra 2 23 Low Shrub Lichen 2 24 Low Shrub Other 2 31 Dwarf Shrub Lichen 2 32 Dwarf Shrub Other 3 11 Lichen 3 12 Moss 3 21 Wet Graminoid 3 22 Wet Forb 3 31 Tussock Tundra 3 32 Mesic Dry Sedge Meadow 3 33 Mesic Dry Grass Meadow 3 34 Mesic Dry Graminoid 3 35 Mesic Dry Forb 10 8 0 Agriculture 9 0 Cloud Shadow 10 0 Other 9 1 Cloud 9 2 Shadow 6 1 1 3 Derivation of Southern Lakes Classification Scheme As stated earlier the final classification scheme for the Southern Lakes was based on the Alaskan classification scheme but was adapted on several levels in an attempt to better classify the earth cover types found in the Yukon Territory Also the classification scheme was designed to record in the field as much variation in the vegetation as possible For instance in the Southern Lakes classification scheme the open needleleaf class was broken into pine and spruce classes and then each one of those had several understory classes The understory classes were moss lichen herb shrub and other This procedure greatly expanded the number of classes Table 2 The classification decision tree is shown in Appendix A and written descriptions of the classes are
18. 1 or 3 2 with the Dialog Designer extension NOTE The Dialog Designer extension comes with ArcView 3 1 and 3 2 2 Itis assumed that the machine on which the extension is to be run has the environmental variable AVHOME set to the directory where ArcView is located The environmental variable AVEXT should also be set to the Ext32 directory where ArcView is located 3 The structure of the field data is important The following directories must be at the same directory level arccovs photos where the photos are set up under the photos directory as shown in the following example photos 2000 SLAK 1 0806 sess1 0806_ 001 jpg where SLAK is the 4 letter project code 0806 is the month day and 001 is the photo number 59 Installation l Copy the duff avx file to the ArcView extension directory Ext32 Generally this directory is located at c esri3 1 AV_gis30 Arcview Ext32 Ifthe user has set the USEREXT variable to another location the extension can be placed there since ArcView will look in that directory for all extensions Open a new ArcView session go to File gt Extensions and turn on the DUFF Data and Photo Viewer extension Using the DUFF Data and Photo Extension k 2 As described under Installation make sure the DUFF Data and Photo Viewer extension is turned on in ArcView Load the slak_fld_ sts coverage from the arccovs directory into a view where slak is the project abbreviation Ex SLAK f
19. 3 93047 4 307574 19 8733 87 2831 58 12476 79 3817308 51 9 9 1 6 2 8 8 4 4 5 1 5 5 4 2 0 3 4 7 9 2 3 4 2 4 8 5 9 2 7 4 3 4 6 4 1 1 5 0 2 0 3 0 1 0 1 3 0 1 2 2 3 2 4 8 1 0 2 0 1 0 3 In the lower elevations with better soils and drainage forested cover types were typically the dominant earth cover classes Closed and open canopy needleleaf stands could be found throughout many of the river valleys and well drained hillsides along the valleys Lodgepole Pine Pinus contorta var latifolia was extensive in old burn areas in the interior influenced valleys such as the Yukon and Teslin river valleys While deciduous trees often dominate the fire scar regeneration areas pine regeneration in pure stands can be found along the river valleys Figure 7 27 Figure 7 Pure Lodgepole pine stands in an old burn area White Spruce Picea glauca was common throughout the image whereas Black Spruce Picea mariana was more common in the eastern portion of the image Since their growth forms were very similar in many areas they were grouped together in the field data collection and the classification Spruce trees were among the most widely distributed trees in the Southern Lakes project area ranging from just below tree line to the interior and coastal influenced river bottoms Subalpine Fir Abies lasiocarpa was often associated with the spruce trees and also shared a wide range in
20. ES Open LS Herbaceous Open LS Moss Open LS Lichen Open LS Other DS Herbaceous DS Lichen Lichen Moss NO YES gt 50 Graminoid YES 0 ae cee Sedge Grass Tussock Wet Graminoid Wet Forb YES Tussock Tundra Lichen Tussock Tundra Mesic Dry Sedge Mesic Dry Grass Mesic Dry Gramin Mesic Dry Forh 45 gt 30 Aquatic Vegetation lt NO 7 gt 75 Water NO YES NO ss YES YES YES YES YES YES gt 30 Aquatic Bed Open Water i O Z O gt 20 Vegetation O gt 50 Rock Gravel NO YES YES YES Urban Agriculture Snow Ice Cloud Shadow Other Aquatic Bed Emergent Vegetation Clear Water Turbid Water Sparse Vegetation Rock Gravel Non Veg Soil 46 Appendix B Southern Lakes Earth Cover Class Descriptions 1 0 Forest Needleleaf and Deciduous Trees The needleleaf species generally found were white spruce Picea glauca and black spruce P mariana White spruce tended to occur on warmer sites with better drainage and deeper soils while black spruce dominated poorly drained sites with poorer soils Since the two spruce types were difficult to separate visually from a helicopter the needleleaf classes included both white and black spruce Lodgepole pine Pinus contorta var latifolia was common throughout the interior influenced valleys throughout the project A wide variety of understory
21. LOSED PINE MOSS Moss MOSS MOSS 10 L C FOREST CLOSED PINE OTHER FUR FORE SPP FORE SPP 5 2 FOREST CLOSED Mis NEEDLELEAF ALTRE 4LNUS SPP TREE ALDER TREE 25 L C FOREST CLOSED MIS NEEDLELEAF aA SALI SPP WILLOW 5 l C FOREST CLOSED Mix NEEDLELEAF PICO PINUS CONTORTA PINE LOCDGE POLE 15 1 C FOREST OPEN SPRUCE EIR SHALIB BEPA pE TULA PAPYRIFERA BIRCH PAPER 5 O FOREST OPEN SPRUCE FIR HERB PATRICI PAPII IS T DEMM DINF S SOPPR lak b m C FOREST OPEN SPRUCE EIR MOSS i L C FOREST OPEN SPRUCE EIR LICHEN Comments Sum of Covers 100 C FOREST OPEN SPRUCE EIR OTHER C FOREST OPEN PINE SHALIB C FOREST OPEN PINE HERB 2 _ FOREST OPEN PINE MOSS T FOREST OPEN PINE LICHEN Calculated Class S CLOSED MISED NEEDLELEAF DECIDUOUS OTHER Aerial Photos Quad B B Satellite Image aig TRS AE Flight Line Photo Date Source Township Range Section Figure 3 The customized database and user interface for field data entry DUFF 18 6 5 Classification Every image is unique and presents special problems in the classification process but we use the same basic approach in all projects Figure 4 as it has proven successful It is particularly valuable that the image processor was actively involved in the field data collection and had first hand knowledge of every training site Erdas Imagine v 8 4 was used to perform the classification Arc Info v 7 2 1 was utilized to manage the field
22. NW Ffit Average Distance Open or Woodland Needieaf Only me L OS Hr Min t i r HHI Figure 2 Custom field data collection form These procedures for collecting field data have evolved into a very efficient and effective means of data collection The navigator located the site with a GPS and verified the location on the field map As the helicopter approached the site at about 300 meters above ground level the navigator described the site to the biologist and the pilot and took a high overhead picture with a digital camera The pilot then descended to approximately 5 10 meters above the vegetation and laterally moved across the site while the biologist called out the vegetation to the recorder The navigator took another picture with the digital 16 camera for a close up view of the site The pilot then ascended to approximately 100 meters so that the biologist could estimate the percentages of each species to the recorder The navigator then snapped an oblique angle photo and directed the pilot to the next site On average it took 5 8 minutes to collect all of the information for one site 6 4 Field Data Analysis The collected field information was entered into a digital database using a custom data entry application DUFF designed jointly by the BLM and DU Inc and programmed by GeoNorth The structure of the relational database is described in Appendix C It was powered by SQL Anywhere while the user interface was
23. OOM T 24 Tle CASSI CANON T 25 TV StS e CONSE TYDE Surp i E E EE E 27 T22 Wetland OVC S KT 29 Tez OUO Vedeta C Over TYPES sireno e eT 30 P Mion PIC VAUON Cover Type rennararnir e O 30 T29 Iniuences on EnC Over K s 31 6 Mod WINS oeste ANE EE TE 32 PEO E EE E T E AE 32 GS Change DCW CHON assan E E 34 E AACCULUACY ASSESSMENT soevexcoussasiaveseusseacccessnwsssvaasaeseessanavessentonesseostessesousevsstaess 35 A1 Overalland ae T 36 TAZ Accuracy R T 38 h Final VOOUCIS rnan E E OEE O 39 S0 SUMMATV Aesse seoser enner ue EEE OEE AE EEE SR 39 90 Ey D aT nE ET EEE a 39 BIS ACUTE CIC eorerinsiroriensienrno rrenetan esk TNE RE EE EELNES ETNE EEN TEER 41 Appendix A Classification Decision Tree eesessssscccosssscccocsssssecocosssseceoossseeeosoo 43 Appendix B Southern Lakes Earth Cover Class Descriptions 0000 47 Appendix C Structure of the DUFF Field Databagp ssssssssssssssssssssssssssssssssssssss 52 Appendix D Summary of Final Map Classes ssssseecoccossssseccccocssssssescssoossssseees 55 Appendix E Accuracy Assessment sseseccssccecsssececcssecessssecocsssecessssecossssecesssseeesso 57 Appendix F DUFF Data and Photo Viewer Extension sssssssssssssssssssssssssssssssssss 59 Appendix G Field Site Attribute Tables and Field Description 62 List of Figures Figure 1 Southern Lakes project area lOCAtION Lc sss 8 Figure 2 Custom field data colleccion Te 16 Figure 3 The customize
24. SOUTHERN LAKES YUKON TERRITORY EARTH COVER CLASSIFICATION SER S GUIDE August 2002 Final Report Ducks Unlimited Canada Canadian Wildlife Service Environment Canada YUKON FISH ANID WILDLIFE EXMHANMCEMENT TRUST Alsek Renewable Resources Council Citation A Collaborative Project by Ducks Unlimited Canada Edmonton Alberta Ducks Unlimited Inc Rancho Cordova California Yukon Territorial Government Whitehorse Yukon Canadian Wildlife Service Whitehorse Yukon August 2002 Southern Lakes Yukon Territory Earth Cover Classification User s Guide 63pp Acknowledgements A lot of work and help from various sources went into this project Funding was provided by Ducks Unlimited Canada DUC Ducks Unlimited Inc DU Inc the Yukon Territorial Government YTG Environment Canada Canadian Wildlife Service CWS the Yukon Energy Corporation and the Yukon Fish and Wildlife Enhancement Trust Cooperation and support was received from the Teslin Tlingit Council the Champagne and Aishihik First Nation and the Alsek Renewable Resources Council Gary Stewart DUC and Dr Fritz Reid DU Inc were instrumental in the development and overall management of the project Al Richard DUC and Debbie van de Wetering DUC managed the field work and was responsible for much of the project planning Staff of Yukon Department of Renewable Resources Y TG and the Canadian Wildlife Service CWS aided in all aspect
25. apply but could result in sampling from only limited areas of the map A stratified approach uses prior knowledge about the study area in order to divide the image into earth cover strata and then each stratum is randomly sampled The sampling procedure used in this project was a modified stratified approach The image was separated into strata using an unsupervised classification which provided an unbiased grouping of strata that collectively represent the entire spectral variation found in the imagery For this project 30 distinct strata were determined and a set number of samples 30 for each class were randomly selected from each of the spectral groups Field sites are hand digitized polygons selected from pixel clusters of sufficient size for sampling greater than 10 pixels of each spectral stratum The standard deviation of the spectral variation found within each field site polygon was calculated and typically sites over three standard deviations from the mean in any of the bands were thrown out In a few cases classes with higher standard deviations were kept due to the wide range of spectral variation found within a spectral class In these cases there tend to be few or no areas within the class that exhibit less than three standard deviations from the spectral mean Thus in order to provide samples within spectral classes with a wide range of spectral variation such as in rocky dwarf shrub areas higher standard deviations were allow
26. as also used in this project Use of the ancillary data was limited and therefore minimally influenced the final earth cover classification The path images were mosaicked using the path correction information in the image header file and then georeferenced and orthorectified to within one pixel root mean squared error 6 0 Methods 6 1 Sampling Design The sampling design followed the following steps 1 Determination of a Classification Scheme of classes 2 Determination of the Sampling Unit 3 Calculation of the Number of Samples 4 Sample selection The sampling design for this project was ultimately influenced by the detail of the classification number of classes and the overall project budget 6 1 1 Classification Scheme The classification system categorized the features to be mapped The system was derived from the anticipated uses of the map information and the features of the earth that could be discerned by TM data The classification system had two critical components 1 a set of labels e g forest shrub and water and 2 a set of rules or a system for assigning labels The set of rules for assigning labels was mutually exclusive and totally exhaustive Congalton 1991 Any given area fell into only one category and every area was to be included in the classification The classification system used to classify the earth cover types for the Southern Lakes project was based on an Alaskan earth cover classification
27. cal Low shrub areas at a height of 0 3 meters were confused with dwarf shrub areas with a height of 0 2 meters Also low shrub and tall shrub were confused as well for the same reason at heights near 1 5 meters The aquatic bed and emergent classes cover types commonly required extra editing because they were generally both limited in extent and highly variable Emergent vegetation typically occurred in narrow strips often only a few pixels wide making it very difficult to obtain reliable ground samples Small differences in soil moisture content density of vegetation and the proportion of senescent plants drastically affected the reflectance values Standing water created a very dark signature while senescent plants created a very bright signature Emergent vegetation signatures were confused with a wide variety of other cover types including open needleleaf open and closed mixed needleleat deciduous low shrub dwarf shrub and even open and closed deciduous The pixel mixing of water mud emergent vegetation and upland vegetation caused much of the confusion in these small areas Due to the number of ponds lakes and wetlands found throughout the image a considerable effort went into editing these errors where possible Undoubtedly some of the areas around the lakes and ponds in the final classification will contain an erroneous scattering of open needleleaf pixels 33 Manual editing of some classes can be a more effective classification
28. ccuracy assessment site with an off diagonal map class provides insight into the vegetation composition of that reference site Similarly since the number of misclassified sites is still indicated in the matrix a user can determine in which classes the map is least reliable and with which mapping classes the unreliable classes are confused If lumping of classes 1s still desired this can easily be accomplished through application of the techniques utilized in previous projects The computed accuracies for the needleleaf forested classes were the same at both the 0 and 5 variation levels 1 e Closed Needleleaf 71 4 open spruce 100 and open pine 71 4 The user s accuracy for the same classes was in the 83 85 range These measures were encouraging since a large portion of the study area was mapped as one of these forested needleleaf classes When an area was classified as one of the forested needleleaf classes the user can have confidence in the accuracy of that classification The open and woodland needleleaf classes were difficult to map due to their high diversity of possible components In some cases cover types other than trees dominated the signature of woodland sites For example a woodland site could include 40 graminoid cover and just 10 trees or it could contain 36 20 trees and 50 shrubs A great deal of effort was expended in separating these two classes from one another as well as from other similar non forest
29. ce lichen and open mixed needleleaf lichen earth cover classes Since there were not enough combined field sites in these sub classes they were grouped into the more general open needleleaf lichen mapped earth cover class Despite the lack of an accuracy assessment on this class this class was mapped because the spectral signature was distinct and separable from the other open needleleaf classes 28 The mixed needleleaf deciduous classes occurred as mature stands shorter stature stands in successional areas and as sapling regrowth in recent burn scars which resulted in a wide range of cover types being grouped into one classification class Mixed Needleleaf Deciduous classes are very difficult to separate A wide variation of mixtures are possible including stands where the conifers and deciduous trees are mixed tree by tree or in clumps of trees within the site Open and closed deciduous stands were found in association with the conifer forests throughout the image in regenerating burn areas along well drained south facing slopes and in the coastal influenced valleys Aspen Populus tremuloides was the most common although tree willows Salix sp and Balsam Poplar Populus balsamifera were also found in abundance Relatively few stands of Paper Birch Betula papyrifera were found in the study area Due to a limited number of balsam poplar and paper birch field sites as well as to their spectral similarity to aspen attempts to distinguis
30. ctral classes were adequately matched and labeled or until the remaining confused classes were spectrally inseparable Throughout this iterative process interim checks of classification accuracy were performed by intersecting the classified image with a coverage of the training sites to determine if the training sites were being accurately labeled by the classification Areas with incorrectly classified training sites were run through further iterations of the supervised unsupervised classification and further refined The iterative process of interim accuracy assessments and refining classifications was terminated when the accuracy assessments indicated no improvements between iterations 20 6 5 5 Editing and Modeling The Southern Lakes project area exhibited considerable topographic effects on the remotely sensed imagery To minimize the effect of this on the final classification some modeling was done to enhance the classification Modeling was performed using shaded relief slope and elevation images derived from DEMs All of the relief slope and elevation images were created using Erdas IMAGINE The modeling process was used primarily to identify potentially misclassified cover types due to the influence of topography throughout the study area The final steps of the classification process were to model the confused classes remaining after the iterative supervised unsupervised classification process and to make final edits in areas that st
31. d database and user interface for field data entry DUE 18 Figure 4 The image processing flow Cid Qram icccccccccccccccccccsccceceeeeeeesccceeeeeeeeee eee e ea aaa aan nee e escent es 20 Figure 5 Fuel site locations and field site distributions for the Southern Lakes project 23 r Southern AGAR CS COMIN cover IED wisi sie se cessed e E a each acstee Aaes 26 Figure 7 Pure Lodgepole pine stands in an old burn dred sese 28 Figure 8 One of the many shallow depressional lakes in the study area sese sss 30 Figure 9 Surface profile of the Southern Lakes project area near the Teslin River 000664 32 Figure 10 Change areas around Whitehorse F 35 F ieureTI SNOrCOFOwWN TIMO TTT 37 List of Tables Table 1 Classification scheme developed at the BLM Earth Cover Workshop sss 10 Table 2 Southern Lakes Yukon Classification scheme eee 1l Table 3 Area and percent cover of earth cover classes within the study area sss sese 27 Table 4 1987 1999 Southern Lakes project area change sss 34 Abstract Ducks Unlimited Inc and Ducks Unlimited Canada have been mapping wetlands in the boreal regions of Alaska since 1988 and in Canada for the past two years The project area chosen was a 3 8 million hectare 9 4 million acre area in the Southern Lakes region centered on Whitehorse Yukon Territory Portions of two cloud free Landsat TM scenes Path 59 Rows 17 and 18 imaged on August 1 1999 were used to classify the project area into
32. d spectral range in brightness values distribution in stable areas or areas throughout the image that did not change between the two dates The sites encompassed all of the major earth cover types from water to forest to shrub to rock Once the training sites were selected a simple linear regression was used to determine the radiometric normalization for each band Then the intercept and coefficient of the regression equation were applied to each band providing the final normalized 1987 image which was used in the change detection 6 6 2 Change Detection Methods After the images were correctly orthorectified and normalized change detection procedures were run on the images For this analysis a simple band differencing change detection was used All of the bands were put into a simple model and the difference either positive or negative was displayed as an image Difference images were determined for each band in the image A threshold was set where everything above the threshold was labeled as change and everything below was labeled no change The threshold is an arbitrary value set by the analyst after reviewing the image The threshold level is highly variable from image to image Since land cover changes showed up in all of the bands there was a lot of redundancy in the band difference images After a review of all of the difference images the TM Band 4 difference image was chosen to determine the vegetation and water level changes
33. d water and snow classes These classes were easily recognizable on the imagery and aerial photography The output of the seeding process in Imagine was a signature file that contained all of the statistics for the training areas The signature file was then used in the modified supervised unsupervised classification 6 5 3 Generation of Unsupervised Signatures An unsupervised classification was generated using the six raw bands and the NDVI ratio One hundred and fifty signatures were derived from the unsupervised classification using the ISODATA program in Imagine The output of this process was a signature file similar to that of the seeding process but containing the 150 unsupervised signatures A maximum likelihood classification of the 150 unsupervised signatures was generated using the supervised classification program in Imagine 6 5 4 Modified Supervised Unsupervised Classification A modified supervised unsupervised classification approach Chuvieco and Congalton 1988 was used for the classification This approach uses a statistical program to group the spectrally unique signatures 19 QA QC Imagery and Generation of New Clip to Project Area Bands Ratios PCA Classify amp Remove Merge Raw Bands Clouds Shadows With New Bands Unsupervised Clustering Mask Out Digitize Pre selected Fisuds Shadows Field Site Locations Plot and Laminate Field Seer oe marines Maps with Field Sites P 8 From Imagery Combine and Recode Into Dra
34. distribution and growth forms from the stunted bush like Krummholtz form near the treeline to the 10 20 meter tall mature trees in the eastern portion of the study area Because fir and spruce were often mixed and the spectral signatures of each were also indistinguishable in many areas spruce and fir were grouped into one class In general the spruce and fir stands dominated the forested areas of higher elevations and pine occurred at lower elevations Despite the spectral similarities among all of the conifer types the open needleleaf class was separated into open pine and open spruce categories This separation was more difficult than expected since several factors such as slope aspect stand age understory components and other factors caused significant confusion between the spectral classes In areas where the two could not be distinguished those classes were assigned back to the open needleleaf class Where possible every attempt was made to determine the understory component within the major earth cover type class For instance open needleleaf lichen was readily determined but the other types of understory components were not readily separated from each other so they had to be lumped into a larger category open needleleaf If spruce and pine had exhibited different and distinguishable understories they would have been separated based on the understory The open needleleaf lichen class was comprised of open pine lichen open spru
35. dleleaf Herb 3 Open Needleleaf Open Needleleaf Shrub 2 _ Open Needleleaf Other 1 Sub Total 6 Closed Mixed Needleleaf Deciduous Closed Mixed Needleleaf Deciduous l Open Mixed Needleleaf Deciduous Open Mixed Needleleaf Deciduous Open Aspen Shrub Open Aspen Other Open Birch Poplar Shrub Open Birch Poplar Herb Open Mixed Deciduous Sub Total Closed Deciduous Closed Aspen Closed Mixed Deciduous Sub Total Woodland Needleleaf Shrub lt Woodland Needleleaf Shrub Woodland Needleleaf Lichen Woodland Needleleaf Herb Woodland Needleleaf Other Sub Total Open Deciduous Woodland Needleleaf Other 55 Appendix D cont Final Map roll up Class Tall Shrub Closed Low Shrub Open Low Shrub Open Low Shrub Herb Open Low Shrub Lichen Dwarf Shrub Other Dwarf Shrub Herb Dwarf Shrub Lichen Lichen Mesic Dry Forb Aquatic Bed Emergent Vegetation Clear Water Turbid Water Sparse Vegetation Rock Gravel Agriculture Snow Ice Shadow classes was significant 1 Map classes are often comprised of ie a roll up of more than one field class where confusion between field Closed Tall Shrub _ Open Tall Shrub Herb Open Tall Shrub Low Shrub Open Tall Shrub Lichen Open Tall Shrub Other _ Sub Total Closed Low Shrub Open Low Shrub Other Open Low Shrub Herb Open Low Shrub Lichen _ Dwarf Shrub Other Dwarf Shrub Herb _ Dwarf Shrub Lichen _ Lichen Mesic Dry Forb _ Aquatic Bed Emerg
36. e Since these classes were readily separated in the final classification the new classification 1s more accurate Some manual edits were done on the image for the final classification where errors were indicated by YTG staff These areas include some of the pixels within deciduous forest classes that were mislabeled as closed low shrub mixed pixels that occurred in wetland classes and other minor revisions throughout the image Also some of the classes were combined for simplicity at the request of YTG staff reviewers This included combining snow and ice into one class merging the deciduous classes into two classes closed deciduous and open deciduous classes and merging the closed pine class back into the closed needleleaf class Thus the total number of earth cover classes were reduced from 34 to 31 40 Literature Cited Chuvieco E and R G Congalton 1988 Using cluster analysis to improve the selection for training statistics in classifying remotely sensed data Photogrammetric Engineering and Remote Sensing 54 1275 1281 Congalton R and K Green 1993 A practical look at the sources of confusion in error matrix generation Photogrammetric Engineering amp Remote Sensing 59 641 644 Congalton R G K Green and J Teply 1993 Mapping Old Growth Forest on National Forest and Park Lands in the Pacific Northwest from Remotely Sensed Data Photogrammetric Engineering and Remote Sensing 59 529 535 Congalton R G 1991 A revie
37. e to be examined select yes Next select a field with which the DUFF field site theme will be queried Finally select a value to use as a limiter This inquiry may run a long time due to large numbers of records being chosen 60 Limitations of the DUFF Data and Photo Viewer Extension l 2 gt If there 1s more than one view in the ArcView Project the user will have to identify which view contains the field site theme The field site theme must be named with the same project name as the tables ie for the Southern Lakes project slak fld sts slak species dbf slak site species dbf slak site photo dbf To sort the species box the user needs to click on the column which they want sorted DUFF Photo tool looks specifically for jpg files not jgw files If sites are currently selected when you use the Inquiry Button no sites will be returned in response to your query regardless of the query This is due to a bug in the program To avoid this problem deselect all sites before using the Inquiry Button added by Ducks Unlimited Inc 6l Appendix G Field Site Attribute Tables and Field Descriptions SLAK SITE PHOTO dbf Dbase file containing site photo information YEAR Year of field data collection AREA NAME 4 letter code for project area CREW NUM ID number of crew that collected data SITE NUM Field site number relates to SITE NUM of field site polygon coverage in a one to many relationship ie each site may have m
38. ea Perimeter Utik Td sts Utik fld sts id Site num Year Area name Crew num Obs nav Obs date Percent slp Aspect dir Elevation Latitude Longitude Obs level Stem dist Obs id Maj _obs Obs class Comments Calc class Calc cl id Aa flag Slak_site_photo dbf Year Area_name Crew num Site num Sess num Photo num slak _species dbf slak_site_species dbf Symbol Year Family Area name Species Crew num Author Site num Common Percnt_cov Alternate Height General Symbol Specific Showit SLAK FLD_STS polygon coverage ArcInfo coverage of the field sites Field AREA PERIMETER COVERAGE COVERAGE ID SITE NUM YEAR AREA NAME CREW NUM OBS NAV OBS VEG OBS REC OBS DATE PERCNT_ SLP ASPECT DIR LATITUDE LONGITUDE OBS LEVEL STEM DIST OBS ID MAJ OBS Description ArcInfo internal field area in coverage units ArcInfo internal field ArcInfo internal field ArcInfo internal field Field site number Year of field data collection 4 letter code for project area ID number of crew that collected data Navigator for field data collection Vegetation caller for field data collection Recorder for field data collection Date of field data collection Percent slope of site Aspect of site Latitude of site only entered for non preselected sites Longitude of site only entered for non preselected sites Observation level where 1 Observation made from within site on the ground 2 Observation made fro
39. eaf class included forested areas with 25 59 crown closure trees of varying height and a diverse understory composition The image analyst individually selected training sites from the spectrally unique areas of the unsupervised classification Whenever possible training sites were grouped in clusters to reduce the amount of travel time between sites The image analyst also placed training sites near landmarks that were easily recognizable in the field such as lakes or streams A tally of the estimated number of field sites per class was kept until all of the target map classes were adequately sampled throughout the project area The coordinates of the center points of the field sites were then uploaded into a Garmin II for navigational purposes Training sites were overlain with the satellite imagery and plotted at 2 54cm 1 6km 1 inch 1 mile scale These field maps were used for recording field notes placing additional field sample sites and navigating to field sites 6 3 Field Verification The purpose of field data collection was to assess measure and document the on the ground vegetation variation within the project area This variation was correlated with the spectral variation in the satellite imagery during the image classification process Low level helicopter surveys were a very effective method of field data collection since a much broader area was covered with an orthogonal view from above similar to a satellite sensor In add
40. ed Because earth cover classes are often arbitrary designations of a continuum of ecological variation this sampling approach provides samples that represent the spectral variation of the project area Since the actual ecological variation of each site is collected in the field the sampling scheme allows the classification technique used in this project to relate spectral variation to earth cover types in a non biased manner 6 2 Image Preprocessing Each image was examined for quality and consistency Each band was examined visually and statistically by reviewing histograms Combinations of bands were displayed to check for band to band registration and for clouds shadows and haze Positional accuracy was checked by comparing the image to available ancillary data such as adjacent imagery hydrography and DEM s 14 In order to optimize helicopter efficiency field sites were identified and plotted on field maps before fieldwork began Sufficient samples for each mapped class were selected to span the variation of spectral responses within that class throughout the entire image For example a shrub class in the southern part of an image may have a different spectral response than the same shrub class in the northern part of that image Many factors contribute to such variation including aspect terrain shadow or small differences in soil moisture In addition each earth cover type encompassed a variety of subtypes e g the open needlel
41. ed sites The overall producer s error for the woodland needleleaf shrub class was 47 while the user s error was 100 This means that this class was underclassified in that it kept falling into other classes yet none of the other classes fell into the woodland earth cover class The error matrix indicates that only five of the 17 accuracy 71 woodland needleleaf shrub reference sites were mapped incorrectly when allowing for only 5 variation in interpretation of the reference data This indicates that cover values in several of the reference sites that were not direct matches with the map sites were close to the threshold between two different mapping classes one of which the map presented for the site Similar results were found throughout the error matrix When accounting for those reference sites with cover values close to these class boundaries consistently high user s and producer s accuracy measures were found Appendix E Table 2 Analysis of the off diagonal sites showed several reasons for their misclassification In several cases particularly the dwarf shrub and many of the needleleaf sites terrain shadow caused these classes to fall into a different category in the classification For instance open needleleaf sites on north slopes generally tended to class as closed while dwarf shrub on the north slopes tended to be sparse vegetation This problem contributed nearly 10 of the error in the classification Another p
42. ent Vegetation Clear Water Turbid Water _ Sparse Vegetation Rock Gravel _ Agriculture Snow Ice Shadow Non Vegetated Soil TOTAL _ 364 l5 479 2 Field classes are those identified in the field but also includes those readily mapped through modeling 56 Appendix E Accuracy Assessment Table 1 Error Matrix at 0 Variation Classified Data Ee pesca I TT Closed Needleleaf 5 Jo Too So To Saa T S e S y y S S f e Open Needleleaf Lichen 3 t fs 2 Open Spruce User s Accuracy error of commission Open Pine 6 a E D A A R RRR Open Deciduous g To T Taf Jaf JFJ F So So S S S S y f s Coo o T S S S Saf So So So So S T S y f a o o o T ST S S Sajf z Jaia S S S S S foal Closed Low Shrub 13 S S S CT So Saa Sed J T S S S T Open Low Shrub 14 Too To To T S S oaa Sof Jij 2 Tan Open Low Shrub Herb 5f J Sooo So So S 3 oo ooo To o T ee ee y y S eee aa S PoE T T To T T S S Saa S dJe S y ff y7 Dwarf Shrub Herb oaao o o o o o o oo o o S T S Ja S ff al Coo o To T To o T o S S S S ay S d a f g Rock Gravel Er E M AA AM a Bae e l error of omission 76 5 Total Number of Sites 115 Denotes mapped classes that accuracy sites from other mapped classes fell into due to error of commission Off diagonalSum 27 These classes were therefore included in the error matrix but no accuracy assessment was done for these two classes Overall Accurac 76 5
43. epresented the reference data while the rows indicated the classification Lillesand and Kiefer 1994 In the error matrix numbers along the main diagonal of the matrix indicate an exact match between the reference data site and the map Sites that have been designated as false in the accuracy assessment are shown in the off diagonal cells of the error matrix Thus if an accuracy assessment site was designated in the field as open needleleaf and the majority of the image pixels were calculated as closed needleleaf this accuracy site would be placed in the off diagonal cell that corresponds to closed needleleaf An error matrix was an effective way to represent accuracy in that the individual accuracy of each category was plainly described along with both the errors of inclusion commission errors and errors of exclusion omission errors present in the classification A commission error occurred when an area was included in a category it did not belong An omission error was excluding that area from the category in which it did belong Every error was an omission from the correct category and a commission to a wrong category In addition to clearly showing errors of omission and commission the error matrix was used to compute overall accuracy producer s accuracy and user s accuracy Story and Congalton 1986 Overall accuracy is the sum of the major diagonal 1 e the correctly classified samples divided by the total number of samples in t
44. etation The digital database and map of the classification were the primary products of this project along with hard copy maps of the classification a complete field database including digital site photos and an ArcView project 9 0 Addendum During the review process a software bug was identified which caused a non linear distortion of the original imagery used for the Southern Lakes project The software bug was documented and confirmed by the software vendor A bug report was filed and the software vendor attempted a fix with no success after two months Due to the nature of the non linear distortion of the image rubber sheeting could not be applied to correct the error The remaining solution was to use the original iterative processing files and the original classification to guide a rerun of the classification The methods used to rerun the analysis are the same as those described in Section 6 5 The first step was to correct the image distortion by using the orthorectified imagery This imagery was then processed with a workaround to solve the distortion problem Then the undistorted image was used as the basis for the classification The signature files from the old classification were used to train the new classification as part of the iterative process In cases where the geometric distortion influenced the signature files the old classification was used to supply additional training sites for the classification This involved taking rand
45. ft Classification Edit and Model Remaining Confusion Final Classification Accuracy Assessment Reprocess if Accuracy Assessment Below Standard Cluster Unsuper Sigs Field Verification with Training Site Sigs amp Label Clusters Data Analysis Separate Training Sites from AA Sites Figure 4 The image processing flow diagram from the unsupervised classification with the signatures of the supervised training areas In this way the spectrally unique areas were labeled according to the supervised training areas This classification approach provided three major benefits 1 it aided in the labeling of the unsupervised classes by grouping them with known supervised training sites 2 it helped to identify classes that possessed no spectral uniqueness 1 e training sites that were spectrally inseparable and 3 it identified areas of spectral reflectance present in the imagery that had not been represented by a training site This approach was an iterative process because all of the supervised signatures do not cluster perfectly with the unsupervised signatures the first time The unsupervised signatures that matched well with the supervised signatures were inspected labeled with the appropriate class label and removed from the classification process The remaining confused clusters were grouped into general categories e g forest shrub non vegetation and the process was repeated This process was continued until all of the spe
46. ge was orthorectified using the same method and control points as for the 1999 imagery Root mean squared error for the orthorectification was less than half of a pixel Once the earlier date image is satisfactorily orthorectified that image must be normalized to the reference image in this case the image associated with the field data Correctly normalized images ensure that changes of small magnitude can be detected and measured and spurious changes will be minimized Selection of anniversary dates where the spectral reflectances of most of the features on the ground are similar in both images is advisable This is because areas that have the same unchanged cover type in both images and have different reflectance values will need to be normalized to each other so images with similar reflectance values minimizes the need for normalization The object of normalization is to correct for all of the differences between the images in the change analysis that are not associated with landscape change in cover or condition These discrepancies may include atmospheric conditions solar illumination phenologic changes in the landscape and sensor calibration 21 The first step in the normalization process was to take training samples of as much spectral variation as possible throughout the 1987 image A total of 35 training sites were selected to normalize the 1987 image to the 1999 image These sites were chosen to maximize spatial image distribution an
47. h birch and balsam poplar from aspen were unsuccessful and therefore not included in the final classification Woodland needleleaf areas were very common throughout the study area particularly in areas influenced by fire or with poorer soils Woodland areas represent over 10 of the project area These classes were among the most difficult earth cover classes to map because 80 95 of the signature was derived from the understory constituents Transitional areas between open needleleaf and tall shrub areas mixed pixel areas at the edge of open closed needleleaf stands and areas regenerating after fires were areas included within this class 7 2 2 Wetland Cover Types Although over 4 of the earth cover in the Southern Lakes project was water there were relatively few large 400ha 1000 acres non forested wetland complexes While there were some areas of black spruce muskegs throughout the image many of these forested wetland types were classified in the woodland other or open spruce classes Emergent vegetation and aquatic bed classes were most common as rings of vegetation in the many small basins and lakes found throughout the image Figure 8 and also as riparian wetlands along the rivers These areas exhibited a wide range of variation in the spectral classes due to the many types of earth cover classes found within a small area The amount of water also has a great influence on the ability of the sensor to detect wetland classes Therefore
48. have dominated a community 3 11 Lichen Composed of gt 30 herbaceous species lt 10 water and gt 50 lichen species Not common to find areas dominated by lichen in study area 3 12 Moss Composed of gt 30 herbaceous species lt 10 water and gt 50 moss species No sites found of this type in study area 3 21 Wet Graminoid Composed of gt 30 herbaceous species gt 5 and lt 10 water and where gt 35 was tussock forming graminoids and where gt 50 of the total was graminoid Moderately well distributed throughout study area areas were typically very small sites 3 34 Mesic Dry Graminoid Composed of gt 30 herbaceous species lt 5 water with gt 50 graminoids but lt 35 tussock forming cotton grass This class was not common and was found generally only at high elevations 49 3 35 Mesic Dry Forb Composed of gt 30 herbaceous species lt 5 water with lt 50 graminoids This class was very rare in the Southern Lakes project area and many of the areas were reclassed into agriculture for example 4 0 Aquatic Vegetation The aquatic vegetation was divided into Aquatic Bed and Emergent classes The Aquatic Bed class was dominated by plants with leaves that float on the water surface generally pond lilies Nuphar polysepalum and some pondweed Potamogeton spp The Emergent Vegetation class was composed of species that were partially submerged in the water and included freshwater herbs such as sedge Carex sp
49. he cover was composed of mud silt or sand and vegetation made up less than 20 of the cover This type was generally along shorelines or rivers 7 0 Urban At least 50 of the area was urban Included areas of human habitation e g Whitehorse 50 8 0 Agriculture At least 50 of the area was agriculture This was an uncommon type 9 0 Cloud Shadow At least 50 of the cover was cloud or shadow The imagery was cloud free 9 1 Cloud At least 50 of the cover was made up of clouds 9 2 Cloud Shadow At least 50 of the cover was made up of clouds shadows 9 3 Terrain Shadow At least 50 of the cover was made up of terrain shadows Terrain shadows were important in the mountainous areas 10 0 Other Sites that did not fall into any other category were assigned to Other For example sites with mixed pixels from several different types such as 25 80 water lt 25 shrub and lt 30 aquatic vegetation were classed as Other Sites classed as Other may have also included extensive areas of vegetative litter such as downed wood 51 Appendix C Structure of the DUFF Field Database The field database is delivered as a field site polygon coverage and three related dbf tables The structure of the related tables is diagrammed below These tables can be linked in MS Access or linked joined in ArcView where they can also be related to the field site polygon coverage Field Site Theme Field Database Tables Slak_fld_ sts Shape Ar
50. he error matrix This value is the most commonly reported accuracy assessment statistic Producer s and user s accuracies are ways of representing individual category accuracy instead 23 of just the overall classification accuracy The producer s accuracy provides an assessment of the error of omission and is calculated by dividing the total number of correctly determined accuracy assessment sites within an earth cover type by the total number of accuracy assessment sites classified into that earth cover type as derived from the reference data The user s accuracy is a measure of the commission error and is calculated by the total number of sites within an earth cover class divided by the total number of sites that were classified into that category An error matrix was calculated for accuracy at 0 and at 5 variation in interpretation within the reference data The 5 refers to adding or subtracting 5 to the cover percentages of the field sites and then recalculating the cover class This technique gives the user an indication of the robustness of the field site vegetation calls For instance say a site 1s called in the field as 55 spruce It would be classed as an open needleleaf when in reality it might typify a closed needleleaf stand By adding 5 to the cover of spruce the value would become 60 and the site would then be correctly calculated as a closed needleleaf site This technique compensates for human error and provide
51. he image this area was processed separately and then merged back into the final classification This area was basically west of Kusawa Lake in the southwestern corner of the image 7 2 4 High Elevation Cover Types The higher elevations in the Southern Lakes project area contained a mix of non forested vegetation communities Rock gravel sparse vegetation snow ice dwarf shrub lichen dwarf shrub low shrub and low shrub lichen were the dominant cover types The imagery in these areas showed a large amount of spectral variation primarily due to the highly variable distribution of cover types within one field site For example a myriad of cover types were often found within one 30 meter area rock gravel sparse vegetation lichen dwarf shrub and snow were commonly associated with each other Within a specific site the relative percentages of each of these components will determine the cover type For instance a site might have dwarf shrub lichen ice rock snow etc The final cover type determination depends on the relative proportions of these components 30 7 2 5 Influences on Earth Cover Type 7 2 5 1 Fires Most of the vegetation in the Southern Lakes project area is based on successional stages of fire regimes Many of the field sites were in old burn areas that were spectrally similar but very different on the ground The more recent burn areas were dominated by standing dead trees and ground litter which caused much of the conf
52. ill had classification errors Editing of classification errors was a process of comparing the classified image to the raw satellite image aerial photography and notes on field maps to identify errors remaining in the classification The pixels that were classified in error were then corrected by manually changing the class value to their correct class value 6 6 Change Detection 6 6 1 Pre processing Like the image classification pre processing change detection images must be accurately georeferenced and orthorectified in areas of significant terrain relief typically within one half pixel within 15 meters root mean squared error Image to image registration quality is critical to a valid change detection analysis Misregistration will lead directly to commission errors and will corrupt the results For example roads that are slightly off show up as a narrow band in change detections due to the error in registration Commission errors are extremely difficult or impossible in many cases to mitigate using post processing techniques In addition since image orthorectification involves pixel resampling a nearest neighbor resampling kernel was used as this technique alters the original pixel values the least The two Landsat 5 images were mosaicked for the change detection analysis for the Southern Lakes study area The 1987 imagery covered a slightly smaller area than the 1999 imagery with less area in the eastern portion of the image The ima
53. ing mapped There are several potential sampling units for a project a pixel a group of pixels or a polygon For this project the sampling unit chosen was a homogeneous polygon since all of the other possible sampling units did not satisfy all of the above criteria desired for this project For instance a 3 X 3 cluster of pixels would be an arbitrary designation and would result in the sampling unit encompassing more than one earth cover type Homogeneous polygons were selected because they directly correspond to areas within a single cover type and also provide a homogeneous spectral signature for the classification 6 1 3 Sample Size An adequate number of samples must be collected in the field for each earth cover class in order to determine the statistical validity of the classification as well as to provide a representative sample of the various spectral signatures of the myriad cover types for the classification Determination and sampling of every distinct spatial area encompassed by all of the cover types within the project area is not feasible particularly for large remote access areas so sample size is in turn influenced by logistic and economic constraints Thus the primary objective of the sample size determination is to balance statistical recommendations with these more practical limitations The first step was to determine the sampling design needed to provide an adequate number of field sites used as training sites for the cla
54. ition aerial surveys are often the only alternative in Canada due to the large amount of roadless areas In order to obtain a reliable and consistent field sample a custom field data collection form Kempka et al 1994 was used to record field information Figure 2 A five person helicopter crew performed the field assessment Each crew consisted of a pilot biologist recorder navigator and alternate The navigator operated the GPS equipment interpreted the satellite image derived field maps to guide the biologist to the pre defined field site and collected photos of the field sites It was valuable for the image processor to gain first hand knowledge of the project area therefore the image processor filled the role of the navigator The biologist identified plant species estimated the percent cover of each and determined the overall earth cover class based on the classification scheme The recorder wrote species percentages and other data on the field form and generally assisted the biologist The alternate was responsible for on ground support data entry and substitution in case of sickness The majority of sites were observed without landing the helicopter Ground verification was performed when identification of dominant plant species was uncertain 15 Kev o ZovUU South Lakes Field Form g 2s a Observers Yr Project Crew Site Number Mo LAT GPS Same LONG GPS C E gl IE a a lope A oe S 7 Tags ON ONE 8 s sw Ww W
55. le to be resolved spectrally some manual editing was needed In some cases a single pixel fell across two cover types for example between a lake and the forested land surrounding it These half water half land signatures were often confused with emergent wetland and closed deciduous signatures Editing was also required to classify areas that fell in the middle of the gradient between one class and another e g between woodland needleleaf and shrub A woodland area of 10 15 trees was easily confused with a shrub area of 5 10 trees The most prevalent example of the confusion within the gradient between classes was found between open and woodland needleleaf classes As evidenced by the field training sites a large number of the open and woodland needleleaf classes exhibited a crown closure between 15 25 A classification decision was made based on the available data for these transitional areas and signatures A similar case of spectral classification confusion involved the misclassification of open mixed needleleaf deciduous pixels in areas of woodland needleleaf that exhibited a dense low and tall shrub understory The mix of the sparse needleleaf trees and the deciduous shrubs mimicked the spectral signatures of two open mixed needleleaf deciduous field training sites This confusion was corrected via manual editing utilizing photo interpretation and review of specific field notes and photos Detecting plant heights also proved problemati
56. m above site ie from helicopter 3 Observation made from a distance outside of site 4 Observation made from aerial photographs Distance between tree stems applies to Open or Woodland Needleleaf only ID of observed class ie classification assigned by vegetation caller Level 1 class of classification hierarchy 1e forest shrub herbaceous etc 53 OBS CLASS COMMENTS CALC_CLASS CALC CL ID AA FLAG Classification label assigned by vegetation caller Notes made by vegetation caller while observing the site Classification of site as calculated using the project decision tree ID number of calculated class Indicates 1f site was used as an accuracy assessment site or training data 0 site used for training L site used for accuracy assessment 54 Appendix D Summary of Final Map Classes Shows Field Classes that comprise them and the number of Training and Accuracy Sites for each shaded map classes are included in accuracy assessment Appendix E Training Accuracy Total Final Map roll up Class Field or Modeled Class _ Closed Spruce Fir Closed Needleleaf Closed Pine Sub Total Open Pine Herb Open Pine Shrub Open Pine Other _Sub Total _ Open Spruce Fir Herb 2 _ Open Spruce Fir Moss Open Spruce Fir Open Spruce Fir Shrub i ists Open Spruce Fir Other _Sub Total Open Needleleaf Lichen Open Spruce Fir Lichen Open Needleleaf Lichen Coe _ Sub Total Open Pine Open Nee
57. mage tended to occur in the valley bottoms Dwarf shrub lichen sites occurred in the highest areas just below the rock gravel class Interspersed between the two were the dwarf shrub low shrub tall shrub deciduous aspen and spruce cover types 31 Figure 9 Surface profile of the Southern Lakes project area near the Teslin River Refer to Figure 6 for color referencing 7 2 6 Modeling Models that incorporated ancillary data sets such as elevation slope aspect shaded relief or hydrography helped to separate confused classes For instance terrain shadow water confusion was easily corrected by creating a model using a shaded relief layer derived from DEMs While this process highlighted some of the areas of confusion it did not eliminate all of the problems associated with shadowing effect These modeling approaches identified 0 30 of the area as terrain shadow in the Southern Lakes area A much larger portion of the image was affected by shadows but not completely blackened by those shadows These in between areas were included in the classification and an earth cover class was determined although the shadows often influenced the signature In some cases the effect of the shadow on a cover type would be that it would cause it to fall into another spectral class Attempts were made to classify any shadowed areas that showed enough spectral reflectance but it was left up to the image processor s discretion whether or not to edit
58. n A total of 31 earth cover classes were mapped in the final earth cover map Figure 6 Appendix D shows the composition of the final map classes in relation to the field classes Table 3 presents the total percent cover and area covered in hectares per class The three most extensive vegetation classes were the closed needleleaf 9 9 of total area open spruce 8 4 and woodland needleleaf shrub 7 9 This agrees with observations made during the field data collection As these summary statistics show none of the earth cover classes were dominant no classes were more than 10 At over 8 the non vegetated rock gravel class was one of the most dominant classes primarily in the mountains in the south central portion of the image The Southern Lakes project area was diverse ranging from glacial icepack in the southern portion of the image to dwarf shrub lowshrub highlands to the forested river valleys The lower elevations were characterized mainly by fire influenced open needleleaf forests which ranged from woodland needleleaf to closed needleleaf with interspersed deciduous forests The 25 mountainous upland areas were characterized by dwarf shrub low shrub tall shrub sparse vegetation rock gravel snow ice and lichen cover types 93 TNT TT Earth Cover Classes Closed Needleleaf Open Mixed Needleleaf E Tall Shrub Open and Closed Open Needleleaf Lichen Closed Low Shrub Open Spruce yyy Open Low Shrub Open Pine Hy O
59. ngit Council the Champagne and Aishihik First Nation and the Alsek Renewable Resources Council 6 This project was administered through Ducks Unlimited Canada Ducks Unlimited Inc in Sacramento California was contracted to produce the earth cover classification Staff of YTG Dept of Renewable Resources now Dept of Environment Parks Canada and the Canadian Wildlife Service participated in various aspects of the project 4 0 Project Area The Southern Lakes Earth Cover Mapping Project consisted of approximately 3 8 million hectares centered roughly on Whitehorse YT 60 73 N 135 098 W and included portions of the traditional territories of the Taan Kwach an Kwanlin Dun Champagne and Aishihik Carcross Tagish and Teslin Tlingit First Nations Figure 1 shows the location of the study area This project area encompassed a wide variety of environments ranging from glaciated mountains to lowland black spruce muskeg Woodland Caribou Rangifer tarandus caribou Dall s sheep Ovis dalli mountain goat Oreamnos americanus moose Alces alces grizzly Ursus arctos horribilis and black bear Ursus americanus are just some of the large mammalian species that can be found within the project area The many small lakes and ponds are important staging molting production habitats for breeding ducks geese trumpeter swans Cygnus buccinator grebes and other waterbirds f T a j PA F 5 U C d al OCA ie 4 dar L a EE So
60. ntifying earth cover types When combined with other GIS data sets e g elevation slope aspect shaded relief and hydrology Landsat TM data produces highly accurate classifications with a moderately detailed classification scheme Earth cover databases allow researchers biologists and managers to define and map crucial areas for wildlife perform analysis of related habitats detect changes in the landscape plot movement patterns for ungulates generate risk assessments for proposed projects and provide baseline data to which wildlife and sociological data can be related 2 0 Project Objective The objective of this project was to develop a baseline earth cover inventory using Landsat TM imagery for a portion of southwestern Yukon and to provide this inventory in an integrated GIS database that can be used for improved natural resources planning A sub objective was to provide coverages of changes in earth cover from an earlier date Change detection information is crucial to long term ecological monitoring of the project area 3 0 Project Participants Funding Funding for this project was a multi agency cooperative effort including the Yukon Territorial Government YTG Department of Renewable Resources the Yukon Fish and Wildlife Enhancement Trust the Yukon Energy Corporation Environment Canada Canadian Wildlife Service CWS Ducks Unlimited Canada and Ducks Unlimited Inc Cooperation and support was received from the Teslin Tli
61. ogram to provide regional earth cover data as part of the Western Boreal Forest Initiative WBFI The WBFI is focussed on understanding how the wetlands of the boreal forest function within the larger landscape and on finding ways to ensure that these wetlands continue to support an abundance of waterfowl and other wildlife Earth cover mapping is an important aspect of the WBFI Some areas that have been mapped include the area around Norman Wells Northwest Territories and Utikuma Lake Alberta This project continued the mapping effort in the Yukon Territory The Yukon portion of the western boreal region is a vast expanse of cordillera and sub arctic woodland covering three distinct ecozones and providing critical habitat for waterfowl and numerous other wildlife species The Yukon Southern Lakes Project area contains highly diverse landscapes and is deemed important for its wildlife cultural and recreational values Landsat Thematic Mapper TM satellite imagery was chosen as the primary source for the DU WBFI s earth cover mapping effort Satellite imagery offers a number of advantages for region wide projects TM data is cost effective processed using automated mapping techniques and collected on a cyclical basis providing a standardized data source for future database updates or change detection studies In addition TM imagery includes a mid infrared band which is sensitive to both vegetation and soil moisture content and is useful in ide
62. om clusters that represented one distinct earth cover class within the old classification These signatures were buffered by at least five pixels to make sure that the geometric distortion would not influence the new classification The signatures from these clusters were then combined with the signature files from the old classification to produce the new classification The iterative process was then rerun and the classification was reassembled The undistorted image allowed the field sites to be more accurately registered to the image producing better separation of spectral clusters and a more accurate classification Differences between the original and final earth cover classifications are primarily due to the improved spatial registration In the new classification there were the increases in open and closed mixed needleleaf deciduous low shrub lichen 39 and aquatic bed vegetation These classes often occurred in narrow bands between vegetation types so the improved spatial registration allowed them to be more accurately determined during the iterative process After the distortion problem was solved comparison between the original and the final classification showed an improved accuracy from 84 to 86 Although the original classification appears to be less pixilated than the final classification this 1s because many of the edge classes like low shrub lichen and needleleaf deciduous were more difficult to isolate using the distorted imag
63. or Southern Lakes Load the three dbf files from arccovs into the ArcView project To do this click on the Tables icon in the Project Menu and then click Add You will be prompted to enter the dbf files to be loaded The three dbf files are slak site species dbf slak_species dbf and slak site photo dbf Once the DUFF Data and Photo Viewer extension is turned on the following buttons tools will appear on the ArcView menu bars the DUFF Info Tool can be used to access the field info from a selected site va Fal the DUFF Photo tool can be used to view the field photos related to a selected field site the DUFF Inquiry button can be used to query the field database according to user specified criteria Steps to run the Inquiry Button NOTE Due to a bug in the program you must deselect all sites before using the Inquiry Button If you have sites selected when you use the Inquiry Button no sites will be returned in response to your inquiry We hope to correct this error in the near future Ducks Unlimited Inc 1 Select a way of identifying up to 3 species 4 letter code Symbol Common Name or Scientific Name 2 Select the type of physical parameter s to examine for a single species either percent cover or height or both 3 Select a mathematical operator for the physical parameter to examine Note this is specific to each species selected 4 Determine if specific sites are to be compared If only certain sites ar
64. ory techniques a case study of the Black River Pages 1 1 in Seventh Annual Symposium on Geographic Information Systems in Forestry Environment and Natural Resources Management GIS World Inc editor GIS 93 Symposium Proceedings Vancouver British Columbia Canada Krebs P 1980 ASVT Project Phase I Denali Study Area Alaska Draft Technical Report U S Department of Interior Bureau of Land Management Anchorage Alaska USA 4 Landis J and G Koch 1977 The measurement of observer agreement for categorical data Biometrics 33 159 174 Lillesand T and R Kiefer 1994 Remote Sensing and Image Interpretation Wiley and Sons Inc New York USA Ritter R A G T Koeln and C Altop 1989 Use of Landsat Thematic Mapper Data for Waterfowl Habitat Inventory in Alaska Story M and R G Congalton 1986 Accuracy Assessment A User s Perspective Photogrammetric Engineering and Remote Sensing 16 529 535 Tortora R 1978 A note on sample size estimation for multinomial populations The American Statistician 43 1135 1137 Viereck L A Dryrness C T Batten and A R Wenzlick 1992 The Alaska Vegetation Classification General Technical Report PNW GTR 286 U S Department of Agriculture Forest Service Pacific Northwest Research Station Portland Oregon USA 42 Appendix A Classification Decision Tree YES YES g 0 Trees 25 100 gt 75 Needleleaf 260 gt Im tall canopy cover YES YES YES Closed
65. p Horsetails Equisetum spp Marestail Hippuris spp and Buckbean Menyanthes trifoliata 4 1 Aquatic Bed There was gt 10 water and aquatic vegetation made up gt 30 of the cover and gt 30 of the vegetation was composed of plants with floating leaves This class was uncommon and spatially limited in distribution Generally dominated by pond lilies 4 2 Emergent Vegetation There was gt 10 water and aquatic vegetation made up gt 30 of the cover and gt 30 of the vegetation was composed of hydrophilic plants Common around ponds but areas usually very small 5 1 Clear Water Composed of gt 75 clear water Common 5 2 Turbid Water Composed of gt 75 turbid water Moderately common 6 0 Barren This class included sparsely vegetated sites e g abandoned gravel pits or riparian gravel bars along with non vegetated sites e g barren mountaintops or glacial till 6 1 Sparse Vegetation At least 50 of the area was barren but vegetation made up gt 20 of the cover This class was often found on riparian gravel bars on rocky or very steep slopes and in abandoned gravel pits The plant species were generally herbs 6 2 Rock Gravel At least 50 of the area was barren gt 50 of the cover was composed of rock and or gravel and vegetation made up less than 20 of the cover This class was most often made up of mountaintops or glaciers Common 6 3 Non vegetated Soil At least 50 of the area was barren gt 50 of t
66. pen Low Shrub Herb Closed Mixed Needleleaf Deciduous SS Open Low Shrub Lichen Open Mixed Needleleaf Deciduous Dwarf Shrub Other Closed Deciduous Dwarf Shrub Lichen Open Deciduous Dwarf Shrub Herb Woodland Shrub Mesic Dry Forb Woodland Other Lichen Figure 6 Southern Lakes earth cover map pas La Aquatic Bed Emergent Vegetation Clear Water Turbid Water Snow Ice Sparse Vegetation Rock Gravel Urban Agriculture Terrain Shadow 26 Table 3 Area and percent cover of earth cover classes within the study area Earth Cover Class Closed Needleleaf Open Mixed Needleleaf Open Needleleaf Lichen Open Spruce Open Pine Closed Mixed Needlelaf Deciduous Open Mixed Needleleaf Deciduous Closed Deciduous Open Deciduous Woodland Needleleaf Shrub Woodland Needleleaf Other Tall Shrub Open and Closed Closed Low Shrub Open Low Shrub Other Open Low Shrub Herb Open Low Shrub Lichen Dwarf Shrub Other Dwarf Shrub Lichen Dwarf Shrub Herb Mesic Dry Forb Lichen Aquatic Bed Emergent Vegetation Clear Water Turbid Water Snow Ice Sparse Vegetation Rock Gravel Urban Agriculture Terrain Shadow Total Area 7 2 1 Forested Cover Types 377862 48 62444 25 105640 29 319683 51 172105 02 57867 12 206270 46 75075 3 130913 19 302427 27 86086 35 158618 61 184310 64 226440 36 102989 52 165563 46 173906 91 154794 15 57326 31 6145 92 13326 12 3587 31 2854 71 113269 23 44384 76 88751 4
67. raphs were used where they were applicable in determining pre existing agricultural fields An example of some of the urban agricultural and fire change classes around Whitehorse is shown in Figure 10 Some of the natural changes in the earth cover types included snow receding 1n the higher elevations fires channel shift in rivers vegetation change and water table changes Fire and snow receding classes were separated as single classes while the remaining classes were lumped into one class This was done because many of the remaining changes were similar or related in nature for instance if an area dropped in water level often in the change image it was included as vegetation change since the exposed land often was covered by emergent upland herbaceous or shrub cover types Table 4 1987 1999 Southern Lakes project area change Change Class Area Acres Area Hectares Fire 120502 48765 Agriculture Conversion 5175 2094 Urban Expansion 3039 1230 Snow Receding 59214 23963 Water table Vegetation Change 47700 19304 34 Figure 10 Change areas around Whitehorse YT The yellow changes are urban expansion green changes are agriculture conversion and red changes are fire scars The underlying image is a 4 5 3 composite of the 1999 image 7 4 Accuracy Assessment The final tally of training and accuracy sites is shown in Appendix D An assessment of the reliability accuracy of the earth cover classification is found in the error mat
68. rees and gt 75 of the trees were deciduous Common throughout study area 1 41 Closed Birch At least 60 of the cover was trees gt 75 of the trees were deciduous and gt 75 of the deciduous trees were Paper Birch Betula Papyrifera Extremely rare distribution 47 1 42 Closed Aspen At least 60 of the cover was trees gt 75 of the trees were deciduous and gt 75 of the deciduous trees were Aspen Common throughout study area 1 43 Closed Balsam Poplar At least 60 of the cover was trees gt 75 of the trees were deciduous and gt 75 of the deciduous trees were Balsam Poplar Rare in pure stands more common mixed with other deciduous species Open Deciduous Mixed Deciduous Species 1 54 From 25 59 of the cover was trees and gt 75 of the trees were deciduous There was generally a needleleaf component to this class though it was less than 25 This was a relatively uncommon class 1 51 Open Birch From 25 59 of the cover was trees gt 75 of the trees were deciduous and gt 75 of the deciduous trees were Paper Birch No sites were found of this class were found in the study area 1 52 Open Aspen From 25 59 of the cover was trees gt 75 of the trees were deciduous and gt 75 of the deciduous trees were Aspen Relatively common class 1 5 Closed Mixed Needleleaf Deciduous At least 60 of the cover was trees but neither needleleaf nor deciduous trees made up gt 75 of the tree cover This class was uncommon b
69. respond correctly with the classified map 7 4 2 Accuracy Discussion A major assumption of quantitative accuracy assessments 1s that the label from the reference data represents the true label of the site and that all differences between the remotely sensed map classification and the reference data are due to classification and or delineation error Congalton and Green 1993 Unfortunately error matrices can sometimes be inadequate indicators of map error because they are often confused by non map error differences Some of the non map errors that can cause confusion are 1 registration differences between the reference data and classification map 2 digitizing errors 3 data entry errors 4 changes in earth cover between the date of the remotely sensed data and the date of the field collection of reference data 5 mistakes in interpretation of the reference data perhaps the most significant 6 variation in classification and delineation of the reference data due to inconsistencies in human interpretation of vegetation The error matrices included in this report attempt to capture measure and account for the most significant of these sources of inconsistency and error in the development of the reference dataset variation in human interpretation Since the fieldwork occurred in July 2000 only a year after the imagery was acquired on August 1 1999 the field data was assumed to be valid While the accuracy assessment performed
70. rix provided in Appendix E Tables 1 and 2 The overall accuracy of the project was calculated to be 77 86 at the 5 level In addition to the overall map accuracy the error matrix allows the user to evaluate the performance and interrelationships of each of the mapping classes that met the minimum criteria for an accuracy assessment highlighted in Appendix D These classes correspond to most of the major in terms of area extent earth cover classes found within the final earth cover map 35 An accuracy assessment was not done on all of the mapped classes However earth cover classes with no accuracy sites should not be interpreted as being necessarily inaccurately mapped Some earth cover classes were not adequately represented in the field data primarily because of their scarcity within the project area e g open closed birch open closed poplar In some cases these classes were merged together into a higher class e g open closed deciduous Other classes did not meet the minimum requirements for accuracy assessment but had enough training sites to be mapped Earth cover classes were not classified if they were not represented in the field data were not readily determinable from the ancillary data or did not exist in the study area The overall map accuracy did not include classes for which no accuracy assessment was done e g snow ice clear water water turbid water urban agriculture Due to their spectral distinctiveness
71. roblem was the fire regeneration sites which had very different growth forms than the classification scheme could handle As can be seen in Figure 11 the aspen in this picture are small most less than 2 meters in height This site was classed as low shrub in the image classification but came out as open aspen shrub in the decision tree classification scheme Figure 11 Short growth form of Aspen 37 Three classes that are listed in Appendix D do not appear to meet the minimum requirement for accuracy assessment at least 15 sites total 5 for accuracy Dwarf Shrub Herb originally had 15 sites but one site was reclassed into another earth cover class due to errors in the field observations Open Deciduous and Open Low Shrub Herb did have at least 15 sites but several additional sites from each were required for training The remote sensing analyst evaluated each class and determined that the accuracy sites would adequately represent the class and would be sufficient for the analysis In summary based on the quantitative accuracy assessment the earth cover classification map produced for the Southern Lakes project area is very reliable Nearly 75 of the accuracy assessment sites matched the full detailed 31 mapping classes directly even when taking no variation in interpretation into account When as little as 5 variation in interpretation was accounted for more than eight out of ten 85 6 of the reference sites were found to cor
72. s a metric of how many borderline sites were in the classification 7 0 Results 7 1 Field Verification A total of 479 sites were surveyed in the field during the period from July 15 July 26 2000 A total of 54 2 hours of flight time were used to collect the field sites The proportions of sites per class Appendix D largely reflected the proportions of corresponding earth cover types within the project area In some cases more sites were collected for classes that exhibited greater variation in growth form and or spectral response from the satellite Within any cover type there 1s a wide range of phenological variation and therefore spectral variation among cover types caused by a wide range of variables elevation aspect soils moisture regimes etc Although the range of spectral variation within a spectral class is determined before the field work using an unsupervised classification it 1s more difficult to determine pre field the ecological variation within an earth cover class For this reason some new field sites were collected in the field Fuel locations were spaced throughout the image and allowed for a wide distribution of sites Figure 5 24 southern Lakes Earth Cover Mapping Project Field Fuel Site Distribution e Fuel Sites Surveyed Field Sites Scale n Kilometers 50 0 Miles 50 Figure 5 Fuel site locations and field site distributions for the Southern Lakes project 7 2 Classificatio
73. s of the project Val Loewen YTG and Jim Hawkings CWS were instrumental in the development of a Southern Lakes classification scheme Marcus Waterreus YTG georeferenced and orthorectified both the 1987 and 1999 images Val Loewen Bruce Sundbo Parks Canada Marcus Waterreus and Al Richard were the recorders for the field sessions Jennifer Staniforth YTG and Karen McKenna contractor provided expert vegetation classification for the field sites Capital Helicopters of Whitehorse YT provided helicopter support and pilot Mike McPhee flew the helicopter in a manner that allowed for safe accurate and efficient field data collection The staff of the Western Regional Office of Ducks Unlimited Inc put a lot of effort into this project as well Beate Sterrenberg managed the project and Kevin Smith was the navigator in the field season and performed all of the image processing Ruth Spell and Brendan O Hara provided technical support on numerous occasions Roxie Anderson helped to prepare the field gear Brandon Sullivan worked long hours helping to print the final maps Many thanks to Val Loewen Jim Hawkings Marcus Waterreus Debbie van de Wetering and Al Richard for their constructive input to the final report and products A special thanks to Debbie van de Wetering for helping to coordinate all aspects of the field work correspondence status reports and final debriefing Table of Contents CHAON creiere EEE EE A EE 1 Acknowledgements
74. s the project area to reduce bias 13 The final number of samples for each earth cover class often was a reflection of how much area that earth cover class occupied within the project area boundary Thus if one class covered a larger portion of the project area than another class it should have had more samples overall Exceptions to this rule were for classes such as clear water shadow ice snow urban and agriculture which due to their inherent spectral properties did not need to be sampled in the field These classes were readily determined from the imagery so more field emphasis was placed on those earth cover types with more spectral variability 6 1 4 Sample Selection Once the appropriate sample size for the project is determined the samples must be allocated among the earth cover classes in the map The primary purpose of a sampling scheme is to provide an adequate number of samples to represent the earth cover classes within a project area for use in both the classification and the accuracy assessment Thus the sample selection process ultimately determines the distribution of samples throughout the project The three most common methods for selection of field sites are a random approach a systematic approach or a stratified approach All of these approaches have strengths and limitations A purely random approach has excellent statistical properties but 1s difficult and expensive to apply A purely systematic approach is easy to
75. site polygons Various word processing and data analysis software including MS Word Excel and Access was also used during the image classification 6 5 1 Generation of New Bands The Landsat TM imagery contained 7 bands of data 3 visible bands 1 near infrared band 2 mid infrared bands and 1 thermal band One new band the NDVI Normalized Difference Vegetation Index was generated for this project The NDVI was highly correlated with the 4 3 ratio a band ratio that typically reduces the effect of shadows in the image and enhances the differences between vegetation types Kempka et al 1995 Congalton et al 1993 The NDVI has been correlated with various forest and crop canopy characteristics such as biomass and leaf area index This NDVI band replaced thermal band band 6 to retain a 7 band image for classification 6 5 2 Seeding Process Spectral signatures for the field sites to be used as training areas were extracted from the imagery using a seeding process in Erdas Imagine A pixel within each training area was chosen as a seed and adjoining pixels were evaluated for inclusion in each training site using a threshold value based on a spectral Euclidean distance The standard deviations of the seeded areas were kept close to or below 3 and all seeded areas were required to be over 15 pixels approximately 1 5 ha 3 75 acres in size Along with the field training areas additional seeds were generated for clear water turbi
76. ssification as well as enough field sites for the accuracy assessment of the earth cover map For this project the minimum sample size for each earth cover class was set at 15 5 for accuracy assessment 10 for image processing training sites in order to perform an accuracy assessment for that class This number was a balance between the theoretical statistical recommendations and the financial limitations of large scale mapping projects The fieldwork for obtaining the training sites for classifying the imagery and the reference data for the accuracy assessment was accomplished at the same time Special care was taken during the preprocessing stage and in the field to make sure an adequate number of samples were obtained Some earth cover classes were naturally limited in size and distribution so a statistically valid accuracy assessment sample could not be obtained without additional field time For these classes with low sample sizes few if any field sites were withheld for the accuracy assessment This does not indicate that the classification for these types 1s inaccurate but rather that no statistically valid conclusions can be made about the accuracy of these classes However withholding even a small percentage of sites for the accuracy assessment provided some confidence in the classification and guided the image processor and end user in identifying areas of confusion in the classification Accuracy assessment sites were selected randomly acros
77. system developed through a cooperative partnership between the Bureau of Land Management BLM and Ducks Unlimited Inc This system was developed over several years and has been field tested on several projects The decision to derive the Southern Lakes classification scheme from the Alaskan classification was based on the similarity of cover types between the two areas 6 1 1 2 Derivation of Alaskan Classification Scheme As projects expanded in size and as other cooperators began mapping and sharing data across Alaska the necessity for a standardized classification system became apparent At the BLM Earth Cover Workshop in Anchorage on 3 6 March 1997 a classification system based on the existing Alaska Vegetation Classification Viereck et al 1992 Table 1 was designed to address this need The goals of this meeting were to 1 develop an earth cover classification system for the state of Alaska that can be used in large regional mapping efforts and 2 build consensus for the system among multiple land management agencies The classification system has been slightly improved since this meeting The classification scheme consisted of 10 major categories and 27 subcategories A classification decision tree and written description was developed in order to clarify the classification Though based largely on Level III of the Viereck et al 1992 classification some classes were modified added or omitted for the earth cover mapping projects e g
78. teaus and may have included some rock 2 1 Tall Shrub Shrubs made up 25 100 of the cover and shrub height was gt 1 5 meters Very common just above treeline 48 2 21 Closed Low Shrub Shrubs made up 60 100 of the cover shrub height was 25 lt 1 5 meters Common throughout study area 2 22 Open Low Shrub Lichen Shrubs made up 25 lt 60 of the cover shrub height was 25 1 3 meters and gt 20 of the cover was made up of lichen Fairly common throughout study area 2 23 Open Low Shrub Other Shrubs made up 25 lt 60 of the cover shrub height was 25 1 3 meters Very common throughout study area 2 24 Open Low Shrub Herb Shrubs made up 25 lt 60 of the cover shrub height was 0 25 1 3 meters and gt 20 of the cover was made up of herbaceous ground cover Uncommon distribution throughout study area mainly in coastally influenced alpine regions 2 31 Dwarf Shrub Lichen Dwarf Shrubs made up gt 25 of the cover shrub height was lt 25 meters and gt 20 of the cover was made up of lichen Very common at higher elevations 2 31 Dwarf Shrub Other Dwarf shrubs made up gt 25 of the cover the shrub height 1s lt 25 meters ground cover is variable Very common at higher elevations 3 0 Herbaceous The classes in this category included bryoids forbs and graminoids Bryoids and forbs were present as a component of most of the other classes but rarely appeared in pure stands Graminoids such as Carex spp may
79. technique in some cases because there are certain features of the image that can be seen by the image processor but are not identified by the computer Classes such as urban and agriculture were manually edited as they are spatially distinct features of the landscape If an analyst tried to separate these classes using the iterative process it would be a long involved attempt that would largely be unsuccessful due to both the wide range of spectral variation and similarity to many other natural non anthropogenic earth cover types Yet these areas can easily be differentiated by use of visual interpretation cues such as spatial context and were manually separated into the appropriate classes 7 3 Change Detection Changes to the Southern Lakes project area and the areas changed are included in Table 4 All of the change classes were merged into one thematic raster coverage for the Southern Lakes project area and coded by change class Most of the anthropogenic change in the landscape occurred near the major transportation arteries including areas around Whitehorse and the Alaska Highway These changes included urban expansion roads houses developments mining sewer etc and conversion of arable land to agriculture The agriculture change class was edited to remove spurious phenologic change crop rotation plowing so that all of the changes in this class were alterations of the natural earth cover to agricultural Aerial photog
80. the accuracy assessment process overall classification accuracies were also generated assuming a 5 variation in estimation of vegetation compositions for each of the accuracy assessment sites In other words if a variation in interpretation of 5 would have resulted in the generation of a different reference site label this new label was also considered an acceptable mapping label for the reference site For this project an accuracy assessment was calculated for each of the designated accuracy assessment sites The total number of earth cover classification pixels that fall within the accuracy assessment site are compared with the field site designation For the site to be assessed as accurate greater than 50 of the earth cover classification pixels must correspond to the earth cover class calculated by the field site Sites with less than 50 coverage in the correct class are labeled as false 6 7 1 Error Matrix The standard method for assessing the accuracy of a map is to build an error matrix also known as a confusion matrix or contingency table The error matrix compares the reference data field site or photo interpreted site with the classification The matrix was designed as a square array of numbers set out in rows and columns that expressed the number of sites assigned to a particular category in the reference data relative to the number of sites assigned to a particular category in the classification The columns r
81. the orthorectification procedure was less than a half a pixel lt 15 meters in some areas the registration was off by more than a pixel This typically occurred in areas where there were few control points available for the registration for example the northwestern portion of the image Other areas such as around Whitehorse and along the major transportation routes were very accurately orthorectified A large portion of the change occurred in these areas while very little occurred in the areas with registration errors In areas with registration error all changes were checked visually and false changes were edited out of the final change coverage 22 6 7 Accuracy Assessment There were two primary motivations for accuracy assessment 1 to understand the errors in the map so they can be corrected and 2 to provide an overall assessment of the reliability of the map Gopal and Woodcock 1992 There is no accuracy assessment method available that would provide a precise representation of the true accuracy of the final earth cover map due to limitations including sampling error logistics project funding and the degree of complexity found within the landscape Because sampling for every spatial unit 1s not economically or logistically feasible particularly for large land areas the sample design for this project provides an estimate of the accuracy of the map In an effort to account for some of the variation in human interpretation in
82. the shadowed area into the terrain class or into the appropriate earth cover class 7 2 7 Editing Despite the extensive iterative process and modeling techniques devised to break up spectral classes into earth cover classes there are some instances where the image analyst s ability to discern cover types visually far exceeds the computer s ability to mathematically or statistically separate classes In some cases different cover types can have exactly the same spectral reflectance values For this reason the final step in the classification is for the image analyst to check all of the assigned classes and manually 32 edit pixels where field data aerial photographs or personal knowledge allows for greater accuracy Only a very small portion of the Southern Lakes image was manually edited but the edits were scattered throughout the image Manual pixel editing was performed on all classes to a minor extent depending on how well the iterative classification and modeling processes separated the spectral classes Some editing centered on ecological differences across the project area For instance the tall and low shrub classes found in the coastal influenced valleys in the southwestern portion were significantly different than the same type of class in the drier interior influenced areas Editing in this case consisted of separating the two areas and re running the classification on each type simultaneously Where the differences were not ab
83. ultiple photos SESS NUM Session number for field data collection PHOTO NUM Photo number Photos are numbered consecutively within each session SLAK SPECIES dbf The data in the following table are based on the PLANTS National Database developed by the National Resource Conservation Service Edits have been made to some species codes to facilitate the use of the data with the DUFF data entry program Also species have been added to the list and a species code created as necessary when compiling field data SYMBOL FAMILY SPECIES AUTHOR COMMON ALTERNATE GENERAL SPECIFIC SHOWIT Species code Most of these are from the PLANTS National Database When new species were added to the field database a code was created to represent it Plant family Plant genus and species or in some cases genus only Author citation for species information Common name Alternate name General plant type used to pipe information correctly through the decision tree Specific plant type used to pipe information correctly through the decision tree Internal field used by the DUFF program 62 SLAK SITE SPECIES dbf Dbase table containing species composition information for each site Each record describes an individual species observed at a site Each site can have multiple records in this table depending on how many different species were observed within the site YEAR AREA NAME CREW NUM SITE NUM PERCNT COV
84. usion between the different field sites Most of the confusion in the old burn areas was among tall shrub low shrub and woodland needleleaf The majority of these old burn areas were classified solely on field site verification Areas that had been burned within a close time frame of the image date left a very bright and homogenous signature because of the abundance of standing dead litter and burned vegetation The standing dead trees litter and exposed soils give a very high reflection value because the features do not absorb any IR The actual live vegetation within the area was not represented in the signature Therefore it is impossible to collect any variation in signatures for classification of these areas For this reason very few field sites were delineated in the burned areas and they were classified based on the spectral signatures from the imagery date Also areas within the fire scar are highly variable year to year For example the 1998 Braeburn fire was mostly bare soil during the imagery date 1999 but during the field season 2000 it had considerably more vegetation 1e regrowth So the image was classified based solely on remote sensing analyst interpretation not based on field determination 7 2 5 2 Elevation Another factor that greatly influenced the earth cover types in this project was elevation As can be seen in Figure 9 vegetation was clearly stratified by elevation Pine areas shown as forest green in the i
85. ut distributed throughout image 1 7 Open Mixed Needleleaf Deciduous From 25 59 of the cover was trees but neither needleleaf nor deciduous trees made up gt 75 of the tree cover Relatively common 2 0 Shrub The tall and low shrub classes were dominated by willow species dwarf birch Betula glandulosa and Vaccinium species with alder being somewhat less common However the proportions of willow to birch and the relative heights of the shrub species varied widely which created difficulties in determining whether a site was made up of tall or low shrub As a result the height of the shrub species making up the largest proportion of the site dictated whether the site was called a low or tall shrub The shrub heights were averaged within a genus as in the case of a site with both tall and low willow shrubs Tall shrubs generally had a major willow component that was mixed with dwarf birch and or alder but could also have been dominated by nearly pure stands of alder It was found most often in wet drainages at the head of streams or on slopes The low shrub class was found at mid high elevations The shrub species in this class were nearly always dwarf birch Dwarf shrub was usually composed of dwarf ericaceous shrubs and Dryas species but often included a variety of forbs and graminoids The species composition of this class varied widely from site to site and included rare plant species It 1s nearly always found on hill tops or mountain pla
86. uthern Lakes Project At a URDD Tertony e ieee ee Pi A L Lc ie D Sc me Wea As Oty ole tor x Figure 1 Southern Lakes project area location 5 0 Data Acquisition The imagery for the earth cover classification was contributed to the project at the systematic correction level by Forest Resources of the Department of Indian and Northern Affairs Canada Two Landsat 7 ETM scenes were used in the project and the final study area used the top 50 of row 18 and the bottom 50 of row 17 row 17 shifted 50 south The acquisition date for the imagery was August 1 1999 and the field data was collected July 15 July 26 2000 both within the peak summer season Ducks Unlimited Canada purchased two historical TM 5 images Path 59 rows 17 amp 18 for the change detection They were acquired on August 24 1987 which was close to the acquisition date of the 1999 image August 1 All imagery was cloud free Ancillary data was limited for the Southern Lakes project The data that was used was provided by YTG and included some 1 20 000 color infrared and color photographs taken in 1978 The areas covered by these photographs were limited to the agricultural areas along the Yukon and Takhini River 8 valleys Due to the long period between the dates of the photographs and the imagery these files were used to help determine land cover in areas where there was no photo interpreted change A forest cover GIS layer w
87. w of assessing the accuracy of classifications of remotely sensed data Remote Sensing of Environment 37 35 46 Ducks Unlimited Inc 1998 Earth Cover Classification System Draft Unpublished Report Rancho Cordova California USA Gopal S and C Woodcock 1992 Accuracy assessment of the Stanislaus vegetation map using fuzzy sets Pages 378 394 Remote Sensing and Natural Resource Management Proceedings of the Fourth Forest Service Remote Sensing Applications Conference American Society for Photogrammetry and Remote Sensing City State USA Kempka R G R D Macleod J Payne F A Reid D A Yokel and G R Balogh 1995 National Petroleum Reserve Alaska Landcover Inventory Exploring Arctic Coastal Plain Using Remote Sensing Pages 788 798 in Ninth Annual Symposium on Geographic Information Systems in Natural Resources Management GIS World Inc editor GIS 95 Symposium Proceedings Vancouver British Columbia Canada Kempka R G B S Maurizi F A Reid R C Altop and J W Denton 1994 Standardizing reference data collection for satellite land cover mapping in Alaska Pages 419 426 in Eighth Annual Symposium on Geographic Information Systems in Forestry Environmental and Natural Resources Management GIS World Inc editor GIS 94 Symposium Proceedings Vancouver British Columbia Canada Kempka R G F A Reid and R C Altop 1993 Developing large regional databases for waterfowl habitat in Alaska using satellite invent

southern lakes, yukon territory earth cover classification user's guide

Contents

Download Pdf Manuals

Related Search

Related Contents