Best Analogous Situations Information System. User's guide for
Contents
1. 4 9 Saving and loading your model GLOSSARY zemen sne vere e 39 REFERENCES ona eae Vande Trees ve den 43 Tuus nisl Se IRL se OPENT ol Kee vie E EK TD COR gelen sneed lets 45 Appendix 1 Start up file 45 Appendix 2 How to new dala in dir E 46 47 LIST OF FIGURES How to use vertical scroll bars a Schedule of the method of predictions b UOTE 51 a Schedule of the method of cross validation b Example To get help on any topic press twice F1 and choose a topic in the help index After selection of the variables the screen looks like this The menu A logical expression can be entered to select part of the database Click to examine individual points Two dimensional plot with the WAP arde Three dimensional plot of two conditional variables and one response variable T Test predictions and calculate the percentage explained variance adjusted R Optimization of weights LIST Of TABLES Logical operators are used to combine to relational equations that can be TRUE or FALSE How these operators work is explained in this table RARRSSRAREKRSES s 28 SUMMARY BASIS Best Analogous Situations Information System is a graphical oriented computer program that runs on Personal Computers It aims to be an aid for ecologists f2. transformed and standardized value of the variable k in lake i multiplied with the weight of the variable default weights are 1 n number of variables The properties of the Euclidean Distance and the Cord Distance are discussed by Jongman ef al 1987 3 2 How to predict a response variable After calculation of the dissimilarities of all lakes with the question lake the lakes in the database are ranked according to the obtained values The N nearest lakes are used to make a prediction of the selected response variables see Fig 2b we call N the number of nearest points The default number for N is 25 The response variables of these cases are weighted so that the influence of the cases declines with the dissimilarity from the lake being estimated N r Yo a EE ED iel 18 In which P Prediction of the transformed response variable needs to be transformed back by using the inverse of the transformation N Number of nearest points Ya transformed not standardized response variable of lake i D Dissimilarity of lake i with the question lake p Distance weighting power is always negative The more negative the distance weighting power the faster the decline in influence and the less the effect of points further out will have on the interpolation 3 3 How accurate is the prediction To answer the question how accurate the prediction is a special technique was used called cross validation also cal
3. be entered to select part of the database 29 4 4 Finding analogies In this exercise we shown how analogous lakes can be found It is possible to browse through the database starting with the best analogous lake To find analogies 1 Select the bar Analogies in the main menu and press Enter or click Analogies The variables of best analogous case are shown on your screen Fig 9 Only the selected condi tional and response variables are shown Best analogy distance 1 0706 Schoonrevoerdse Wiel 1985 Conditions your lake relative distance 2 0 051 mg l 0 106 mg l MEANDEPT 6 000 m 8 900 n Response Chlorophull a Jul Aug R 22 566 yor Secchi depth Jul Aug 0 973 n p tF2 Full info fPgDnl PgUpl Next Previous Esc Stop Fig 9 The conditional and response variables of the best analogous lake 2 To view more variables press or click on F2 toggles between Summary and Full info mode Now all available information is shown Use cursor keys Up Down to scroll if not all variables are visible or click on the scroll bar 3 Press or click on F2 again to return to the summary mode 4 Press or click on PgDn to view other analogies in descending order of similarity 5 Press or click on Esc to return to the main menu 30 4 5 Predicting a response In this exercise we predict the response of Loch Neach Further we learn how to zoom in scatter plots how to highlight
4. Fig 3 Schedule of the method of cross validation b Example 20 3 4 Optimization of the prediction To help the user to set the values of parameters like the weights of the conditional variables an optimization routine is included in the program The following parameters used by the prediction method can be optimized automatically Weights of conditional variables Distance weighting power Number of nearest points The parameters are optimized iteratively by use of the Controlled Random Search CRS algorithm de Hoop et aL 1992 Klepper et al subm This algorithm is an improvement of pure random search an algorithm searching the best set of parameters by trying at random After each iteration the goodness of fit adjusted R is calculated by cross validation see 3 3 The CRS algorithm first selects N sets of model parameters uniformly distributed over prior parameter ranges calculates the goodness of fit for each and puts them in a vase It then selects m 1 points at random from the vase and mirrors the last point over the average centroid of the first m The mirrored point is the new trial point The goodness of fit is calculated If the R better than the worst set of parameters in the vase the worst element of the vase is replaced by this new guess This process continues until a convergence criterion is reached An option is to add all variables stepwise Each step is followed by an iterative optimiz
5. lt equal not equal greater than less than greater than or equal to less than or equal to a value or class name Class names e g Sand must be written between delimiters 27 An example of simple relational expression SEDIMENT lt gt Sand To select all data from the database except the lakes with sandy sediment gt 0 1 To select all lakes with tot P of more than 0 1 Relational expressions can be combined by logical operators Parentheses must be used in combina tion with the logical operators The logical operators are see also Table 1 negation logical and to combine two conditions logical or to combine two conditions SRE logical exclusive or to combine two conditions Examples Sediment Sand and PTOT1 lt 0 2 or PTOT1 gt 0 1 not sediment Sand the same as Sediment lt gt Sand Table 1 Logical operators are used to combine to relational equations that can be TRUE or FALSE How these operators work is explained in this table suppose Ais TRUE FALSE TRUE FALSE and B is TRUE TRUE FALSE FALSE than A and B TRUE FALSE FALSE FALSE AorB TRUE TRUE TRUE FALSE A xor B FALSE TRUE TRUE FALSE not A FALSE TRUE FALSE TRUE To select all lakes with a maximal depth of less or equal than 10 m except manipulated lakes Select the bar Select if in the main menu and press Enter or click Select if Now a dialog screen is displayed You ca
6. You can use this scroll bars with a mouse to scroll the contents of the menu 1 To scroll one line at a time click the arrow at either end of the bar 2 To scroll continuously keep the mouse button pressed 3 To scroll on page at a time click the bar to either side of the scroll box 4 To move quickly drag the scroll box to any spot on the scroll bar 11 Click here to scroll one line up Click here to scroll one page up Drag scroll box to any spot Click here to scroll one page down Click here to scroll one line down Fig 1 How to use vertical scroll bars Special use of the mouse in scatter plots In some scatter plots it is possible to get more information about the points that are displayed Just a mouse click on the point gives the name of the lake that is displayed More information about that lake can be obtained by pressing F2 afterwards see also 4 5 3 METHODS The main purpose of the program BASIS is to find analogous cases based on available information In chapter 3 1 is explained how this is done Based on these analogous cases is possible to predict the response of the question lake In chapter 3 2 the averaging method is explained The next question is how good this prediction is In chapter 3 3 is explained how the percentage of variance explained can be calculated The methods and parameters used by BASIS can be optimized automatically Chapter 3 4 explains b
7. range from xlo to xhi 47 Appendix 4 Error messages Error 1 Out of memory Not enough memory to run BASIS Error 2 A nominal variable may not be selected as response variable Select another variable Error 3 Not a valid value entered Acceptable numbers are 123 232 12E3 1223 2 Error 4 Negative weight The weight of a variable should be a positive value otherwise analogies are down weighted Error 5 Value of question lake too small The value of the question lake is out of the range of valid input values Please enter a larger value or deselect the conditional variable Check the units Error 6 Value of question lake too large The value of the question lake is out of the range of valid input values Please enter a smaller value or deselect the conditional variable Check the units Error 7 No response variables selected None of the variables is selected as response variable Select the variables that should be predicted Error 8 No conditional variables selected None of the variables is selected as conditional variable Select the variables that should be used to make a prediction Error 9 No value entered of conditional variable No value of one of the conditional variables is entered therefore no predictions can be made Please enter a value or deselect the conditional variable Error 10 Not enough lakes selected Make a new selection Select if Or select another binary file Start up file Error 11 Not enough lak
8. a part of a plot and how individual points of the plot can be examined These options also can be used for two dimensional WAP relations 4 6 To make a prediction 1 Select the bar Predictions in the main menu to make a prediction A scatter plot is drawn on your screen Fig 10 goShlerephull a Jul Aug 8 Weighted Euclidean Distance Expected CHLA2 jn question lake 69 777 yq l rz0 268 Valid n go F1 Help F2I Zoom f3 Highlight part Other key Continue Fig 10 Plot of the prediction The predicted chlorophyll is 39 ug l The prediction is visualized by plotting of each lake in the database the response variable against the dissimilarity with the question lake When the prediction makes sense a converging cloud of points is expected with the lowest dispersion near the y axis Under the figure the prediction the question lake is displayed The r that is displayed is a measure of how well the cloud converges It is the correlation between dissimilarity and difference between response variable and prediction So near 1 perfect converg ing cloud near 1 diverging 2 Press Enter to view the prediction of the next response variable SECCHI2 31 Zooming 1 Press F2 to enter the zoom mode The selected area is indicated by a rectangle with the blinking red cross in the right upper corner You will find this rectangle in the lower left corner of the scatter plot 2 To change the Size of the
9. analysis of Poisson distributed variables Xu Joa 15 4 Inverse transformation Used e g to make the relation between Secchi depth and extinction approximately linear Ni p 5 No transformation Use untransformed data Symbols value of the variable in lake i n number of variables Standardization The different variables need to be standardized to give equal weight to different scaled variables 1 Normalization By this method the relative position of an observation within a distribution is described The normalized value also called standard or Z score SPSS 1987 shows how many times the standard deviation an observation deviates from the mean of the population The mean of the normalized values is O and the standard deviation is l It is obtained by subtracting the mean from a value and dividing this difference by the standard deviation Lu sd 2 MinMax standardization By this method the variables are scaled between the minimum and maximum values in the database The minimum gets value 0 and the maximum gets value 1 __ MIN VT MAX MIN O 5 No Use unstandardized data Use this option only if the data are already standardized 16 Symbols Yu transformed value of the variable k in lake i value of the variable k in lake i MIN minimal value of the variable k in the database MAX Yu maximal value of the variable in the database P mean of variable k in the database sd s
10. cursor keys or the scroll bar to highlight the right item Alternative press the first character of the item Select the highlighted item by pressing Enter 4 Go to the previous screen Esc If you press Esc you go to the previous screen The help system of BASIS recalls the last 10 topics that are examined as long as you don t leave the help system 2 10 2 5 Special keys and how to use the mouse The following keys can be used F1 Help F1 F1 Help index F2 Zoom scatter plot this key has also other functions see status bar F3 Highlight part of a scatter plot F10 Go from the input screen to the main menu Esc Cancel calculation or go to preyious menu All calculation can be cancelled by pressing Esc It may however take a few seconds before the program reacts Up Down Move cursor or scroll PgUp PgDn Scroll up or down one page Home End Go to the first or last page Enter Select or change an item Alt F Save to or load selection from file How to use a mouse The use of a mouse is fully supported The highlighted keys e g F1 of the status line below can be activated by a mouse click on the key name or on the text after the key name In menu s a single mouse click selects an item In the input screen a single mouse click moves only the cursor A double click is required to select a value or to enter a new value or choice In some menu s vertical scroll bars are displayed Fig 1
11. on a point to see the name of the lake and the year of the data see Fig 11 2 Click on the lake name or press F2 to get a full list of information about the lake 3 Scroll the text if necessary 3 Press Other key to return to the plot 32 Secchi depth Jul Aug mJ ben UT MUN Heerlo 986 B NE EL ET EE et 2 4 6 Usighted Euclidean Distance Expected SECCHI2 in question lake 0 440 m r 0 288 Valid n 75 F1 Help F2 Mor info Other key Continue Fig 11 Click to examine individual points 4 Click between points Now the mouse cursor jumps to the nearest point 5 Click again to remove the text from the plot 6 Press Space until you are returned to the main menu 4 6 Viewing relations between variables In this exercise we learn about the option for exploratory analysis of the data set Two dimensional scatter plots and three dimensional surface plots show the relationships between the conditional and the response variables The relationships between the variables are shown by making WAP predictions The technique is than comparable to the running averages method The way of averaging of WAP 3 2 however differs and can be optimized Note Only the visible conditional variables are used in the calculations 33 To make two dimensional plots 1 Select the bar WAP relations in the main menu 2 Select the bar 2D relations from the next menu Note only if there ar
12. same with SECCHI2 24 To enter the conditional variables 1 Move to the column Cond conditional variable and position the cursor on the right row MEANDEPT 2 Press Enter or double click with mouse The cursor moves to the column Tran sform We don t change the default transformation so press to the Right arrow key 3 Press the Right arrow key To change the default weight 1 0 to be given to the variable press Enter else move to the right 4 Press Enter to enter the value of the question lake in our case 8 9 followed by Enter 5 Now enter the second conditional variable in the same way PTOT2 transformation Logarithmic weight 1 0 and value 0 106 6 After selection of the conditional variable the screen looks like Fig 6 Variable Cond Resp Transiorm eight RETENTIO SEDIMENT XCOORO YCooRD CHLAL CHLAZ CLL CL2 CONOUCL CONDUC2 NTOTL NTOT2 2 SECCHIL SECCHI2 o Logarithaic 1 0000 3O0Oso0o00000000000 Enter input value of question lake CFL Help EF10J Menu Fig 6 After selection of the variables the screen looks like this 25 4 2 The main menu After selection of the desired variables different kinds of analysis can be selected in the main menu To open the menu from the input screen 1 Press F10 Predictions HAP relations Test predictions Optimize Methods Select if Input screen Esc F
13. the database or to highlight points in plots that are entered and the used methods are saved To save your optimized model 1 Select the bar File in the main menu 2 Select the bar Save to file in the next menu 3 Enter the name of the file or press Enter to select the default name BASIS VAR You go back to the main menu exit BASIS and restart BASIS with the same selection l Select the bar Exit in the main menu 2 Restart BASIS typing BASIS followed by Enter 3 Press Alt F both alt key and F 4 Enter the same name of the file or press Enter to select the default name BASIS VAR You enter the main menu 38 5 GLOSSARY Adjusted R The statistic adjusted R is the proportion of variance explained by a model It attempts to correct R to reflect more closely the goodness of fit of the model in the population Analogies Analogies are lakes that resemble the question lake most closely BASIS Best Analogous Information System a computer system to search for information about analogous cases Classes Nominal or ordinal variables have classes These variables can only have some discrete values Examples Sediment type country manager Conditional variables Conditional variables are called independent variables in regression analysis They explain the state of the ecosystem of a lake Examples of conditional variables in limnology Mean depth Maximum depth Area Nutrients tot P tot N Chlorid
14. use of logical operators 39 Logical operators Logical operators are used to combine two or more relational expressions The following operators are supported not negation and logical and or logical or xor logical exclusive or Nominal variables Nominal variables are class variables Values assigned to these classes do not imply any order Differences between them have no function Number of nearest points The number of cases on which the prediction is based If you want to select all available points enter 0 Missing values If one or more variables are not known no dissimilarity can be calculated These lakes are discarded in the calculated In the data set these values are marked with value 9999 9999 Ordinal variables Ordinal variables are class variables Values assigned to the classes imply an order Differences have no absolute meaning Question lake lake with known conditions but with an unknown response Protected mode The protected mode is a mode of operation of a 80286 or later processor This mode is not compatible with earlier processors because it extends the addressing up to 16 Mb Protected mode programs that are developed in Borland Pascal use a special run time manager RTM EXE that manages the memory usage A 80286 or later processor and at least 2Mb extended memory is required Real mode The real mode is a mode of operation of a processor of the PC This mode is compatible with the original 8086 processor Only
15. vine 4 1 n wiles deed ank AG vi A24 vong tr M pd dio IE 1 Ea A E I rae Fiera 5 REFERENCES de Hoop B J P M J Herman Scholten Soutaert 1992 Seneca 1 5 A Simulation ENviron ment for ECological Application Manual Netherlands Institute of Ecology Yerseke The Netherlands ISBN 90 9004976 2 Jongman R H C J F ter Braak amp O F R van Tongeren 1987 Data analysis in community and landscape ecology Pudoc Wageningen Klepper amp E M T Hendrix submitted A comparison of algorithms for global characterization of confidence regions for nonlinear models J Env Toxicol Chemistry Sas ed 1989 Lake restoration by reduction of nutrient loading expectations experiences extrapolations St Augustin Academia Verl Richarz Scheffer M 1991 On the predictability of aquatic vegetation in shallow lakes Mem Ist ital Idrobiol 48 207 217 SPSS 1989 SPSS PC V2 0 Base Manual for the IBM PX XT AT and PS 2 SPSS Inc Chicago Sokal R R amp F J Rolff 1981 Biometry The Principles and Practice of Statistics in Biological Research 2nd edition W H Freeman and Company New York Surfer Surfer Reference manual Trimbee A M amp E E Prepas 1987 Evaluation of total phosphorus as a predictor of the relative biomass of blue green algae with emphasis on Alberta Lakes Can J Fish Aqu
16. 1 Mb of memory can be addressed in this mode Relational expression A relational expression is a simple logical expression It consist of l a variable name 2 a relational operator e g gt lt lt gt and 3 a value or class name Class names e g Sand must be written between delimiters Relational operators The following operators can be used in relational expressions equal lt gt not equal gt greater than lt less than gt greater than or equal to lt less than or equal to Response variables Response variables are variables that express the state of the ecosystem of a lake In regression these variables are called dependent variables Examples of response variables in limnology Secchi depth Chlorophyll Algae composition Fish species 40 Scroll bars If not all information can be displayed vertical scroll bars are shown You can use this scroll bars with a mouse to scroll the contents of the screen Start up file default extension INI In this ASCII file the colors of BASIS can be changed and the names of the input and output files Use an editor to make changes The commands are not case sensitive see Appendix 1 Status line The status line is a bar at the bottom of the screen where information and hot keys are displayed The hot keys can also be activated by a mouse click on the key name or the text after the key name Standardization Give the same weight to different scaled variables Transfo
17. Best Analogous Situations Information System User s guide for BASIS Version 1 0 Notanr 93 044 ministerie van verkeer en waterstaat rijkswaterstaat NOTA auteur s datum riza rijksinstituut voor integraal zoetwaterbeheer afvalwaterbehandeling tel 03200 70411 fax 03200 49218 doorkiesnummer 70713 Best Analogous Situations Information System User s guide for BASIS Version 1 0 93 044 E H van Nes amp M Scheffer December 1993 ISBN 90 369 0103 0 SUMMARY irade anan na owls ee ee DIEI INTRODUCTION 1 1 1 What is BASIS 7 1 2 Why WAP 7 INSTALLATION soe venda it eco toan Nu eva Mew Fe le eek 9 2 1 System requirements 9 2 2 Using INSTALL 9 2 3 Running BASIS 10 2 4 Getting help 10 2 5 E n the Reims e ROT oer NOU MONA 5 1 METHODS Ee wx MR ve 13 3 1 How to find analogies e E o i don dot Bees beds 13 3 2 How to predict a response variable 18 3 3 How accurate is the prediction 19 3 4 Optimization of the prediction 21 TUTORIAL AN EXAMPLE OF A PREDICTION 4 1 Selecting your variables 4 2 The main menu 4 3 Filtering the database 44 Finding analogies 4 5 Predicting a response 4 6 Viewing relations between variables eee nnnm 4 7 Calculate percentage explained variance R 4 8 Optimizing your model
18. IR m AIR 56 E a E E 3 Ele R amp B_RUDD i d 8 amp i Bream and White Bream No mer a hmm 78 home pierin Zeeli Coregonus No Coregonus aantal Ov Snoek me san em HEN eem gt Zeelt en kroeskarper aantal Other fish No N PERCH Perch No kere Ruffe No i g ae en kroeskarper aantal ige vissen aan aantal 3 t Smelt No E i i Z E m EJ H E a E a amp 53 For the nominal and ordinal variables the following classes were defined CLASS LABEL comments PRU Provincie Utrecht 54
19. at Sci 44 1335 1342 43 APPENDICES Appendix 1 Start up file Start up file default extension INI In this ASCII file the colors of BASIS can be changed and the names of the input and output files Use an editor to make changes All commands are not case sensitive The following commands are supported Comment line DefColor Normal text color Monochrome Monochrome mode True or False ShadowColor Color of shadows HighColor Highlighted color DefBkColor Background color DefFillColor Color inside box of screen LogoColor Color of the logo LogoFillColor Background of the logo CursorColor Color of the cursor WapColor Color of wap relation DataFile Name of binary file with data DbfVarsFile Name of dBase file with variables DbfClassFile Name of dBase file with data DbfDataFile Name of dBase file with data OutFile Name of output text file WriteOutput Output written to ASCII file Yes No Optimization report test prediction and others can produce ASCII output GridFile Name of output grid file for surfer NLines Number of lines in input screen Condition Any condition e g Sediment Sand to select only lakes with sandy sediment Possible colors Black Blue Green Cyan Red Gray VGA only Magenta not VGA Brown LightGray Darkgray LightBlue LightGreen LightCyan LightRed LightMagenta Yellow White BASIS INI is the default start up file with defaults To use a different start up file f
20. ation 21 4 TUTORIAL EXAMPLE OF A PREDICTION In the following exercises we will make a simple prediction and optimize the results We try to predict the summer chlorophyll a concentrations and the Secchi depth of Lough Neagh Sas 1989 using the mean depth 8 9 m and the total P concentration in the summer 1989 0 106 mg l The summer mean Secchi and chlorophyll values were respectively 0 85 m and 79 ug l Sas 1989 4 1 Selecting your variables In the first exercise we ll start the program from DOS and select the conditional independent and the response dependent variables How to get on line help from the program is explained too To start the program and get some help 1 Enter in DOS BASIS followed by Enter The program displays a message about restrictions in the use of the data set 2 Press any key to enter the input screen see Fig 4 The first column shows codes for variables The full names of the variables are shown below in the status bar only if the cursor is on the first column Variable Cond Resp Transform _ Height MANI PULA o a a o a o a o o a a n Joooooo0000000000 Fig 4 input screen 23 3 To get help about the column Cond press the right arrow key once and press F1 Now a help screen about Conditional variables is displayed The highlighted words are related topics 4 get information about a related topic e g the inpu
21. computer program It runs on any DOS system if one of the following graphical drivers or compatible drivers is available VGA or SVGA recommended EGA Hercules monochrome CGA EGA monochrome During installation of the program approximately 1 Mb of free disk space is required The installed program including a data set with 718 lakes years x 71 variables takes 0 7 to 0 9 Mb of disk space depending on your hardware Since some options in the program require heavy calculations a numerical coprocessor is strongly recommended Three dimensional plots can only be made automatically when the grid program SURFER is present on your disk not necessarily in your DOS path If a mouse is available it can be used There are two versions of BASIS that differ completely in the use of RAM memory During the installation the program INSTALL automatically selects the right version The two modes are 1 Real mode This version runs on any DOS machine with at least 640 Kb memory It uses expanded memory EMS if available 2 Protected mode This version runs an 80286 processor or higher if there is at least 2 extended or expanded memory available It uses all available memory and runs DOS in the protected mode It is faster than the real mode version in particular with large data sets 2 2 Using INSTALL INSTALL detects what hardware you are using and configurates BASIS appropriately Files that are not necessary for your machi
22. e concentration Sediment class Controlled Random Search Algorithm to optimize parameters of a model by trying at random within predefined ranges During the search the ranges are shrunk to speed up the search process Cross validation all but one method An efficient way of testing a model is to predict each lake in the database using all other lakes Dissimilarity The dissimilarity between the question lake and all other lakes is calculated by a dissimilarity index The Euclidean Distance ED is the most frequently used index It is the distance in the n dimensional space constrained by the conditional variables Each variable is one dimension of the space Distance weighting power Data points that are used to make a prediction ofthe response variable of the question lake are weighted such that the influence of the data points declines with the dissimilarity from the point being estimated The more negative the weighting power the faster the decline in influence and the less the effect of points further out will have on the interpolation Input screen Screen to select conditional and response variables LABDA Largely Analogous But Differences Also an approach of predicting the responnse of a lake It involves two steps 1 crude estimation using analogies 2 fine tuning of the prediction using models or knowlegde of processes Logical expressions logical expression consists of one or more relational expressions that are combined by
23. e more than one conditional variable this menu appears In a two dimensional plot the first conditional variable is plotted against the first response variable The data points crosses and circles and the WAP predictions line are plotted together see Fig 12 1 0 1 5 2 0 Tatal P Jul Ano mol Fig 12 Two dimensional plot with the WAP predictions In this plot it is also possible to Zoom in Highlight part of the plot Examine individual points mouse only In the previous exercise we have learned how to do these things 4 Press Esc to view next plot until all combinations of conditional and response variables are shown Then you return to the mean menu ne 12 08 SES SEES 33 29 3209 EN s We 08 at Pd 3 x ES 3 H v v 0 55 e Fig 13 Three dimensional plot of two conditional variables and one response variable To make three dimensional surface plots 1 2 3 4 Select the bar WAP relations in the main menu Select the bar 3D relations from the next menu If you don t have SURFER on your current drive the following error message is displayed Error 14 SURFER not found If this message appears it is not possible to make 3D plots at run time The ASCII grid files still can be made press Enter and follow the next instructions to do so If you want to cancel the calculations press Enter and Esc Now a two dimensional plot of the conditio
24. e most important options of the program 1 INTRODUCTION 1 1 What is BASIS A central theme in water management research is the problem how to select the optimal measure for restoration Solving the problem requires prognosis of the expected response of the ecosystem Up till now the dominant approach has been to combine all available quantitative information on important processes in computer models for prediction Such a compilation of information however is accompanied by a compilation of errors and uncertainties which is the main reason that this approach is losing popularity Most experts do better than most models We think that this is because experts tend to reach their conclusion in a safer way than models do Experts use analogies the response of a specific lake to removal of fish is expected to be similar to that of comparable lakes where this has been done before Obviously all lakes are different so that some nuance has to be made also We formalized the search for analogous lakes The computer selects the most similar lakes from some relevant variables The characteristics of the selected cases are visualized in such way that differences and similarities are easily recognized This computer information system that we call Best Analogous Situations Information System BASIS aims to be an aid for ecologists for making predictions of the response of aquatic systems to measures Besides the WAP technique the program includ
25. es an option for exploratory analysis of the database Two and three dimensional plots can be made and outliers can easily be examined and selections can be highlighted The computer program is menu driven and includes an extended on line help system with more than 80 entries The use of a mouse is fully supported The computer program is written in Borland Pascal 7 0 some procedures of Turbo Professional 5 0 are used for the real mode version 1 2 Why WAP WAP Weighted Analogies Prediction is a technique to make a prediction based on information of analogous cases The technique involves two steps 1 Find analogous lakes in a database 2 Predict the response of the question lake using the most analogous lakes The advantages of the WAP technique are 1 No prior information or assumptions about the nature of relations between variables are needed 2 It is easy to find and browse through the available information of the most analogous lakes 3 It is the starting point of the LABDA approach Largely Analogous But Differences Also Scheffer 1991 The LABDA approach is a way of predicting the response of a lake It involves two steps A Rough estimations using analogous cases Largely Analogous B Fine tuning of the prediction by quantitative models that predict only the differences of the question lake from the analogous cases But Differences Also 2 INSTALLATION 2 1 System requirements BASIS is a graphical oriented
26. es with valid conditional variables Deselect some conditional variables Error 13 Distance weighting power should be negative Enter negative value Error 14 Number of nearest points should be larger than 0 Enter positive value Error 15 Error in conditional expression Many possible errors can be detected in a logical expression XXXX not found Name of variable not found Class XXXXX not found Name of a class not found Error in number 48 Not valid number Not enough operators More operands than operators No relational operator Two operands but no relational operator Not enough logical operators Not enough logical operators Too many logical operators More logical operators than operands right parentheses missing Not enough right parentheses XXX left parentheses missing Not enough left parentheses Not enough parentheses All logical expressions must be separated by parentheses no message General error in expression Error 16 SURFER not found If SURFER is not available then it is not possible to get 3D plots at run time If you continue an ASCII grid file is made and might be used in another GIS system after necessary transform ations Error 17 File not found Enter name of an existing file or press Esc to cancel Error 18 Wrong type of File Only files that are saved in BASIS can be loaded Reenter name of an existing file or press Esc to cancel 49 Appendix 5 The database The database consists o
27. f a selection of data per lake From each lake one representative year is chosen to ensure statistical independency of the data An exception on this rule is made for lakes that are in a transient phase because they were manipulated recently mostly biomanipulation Although different years are not fully independent the year to year differences are expected to be large From some variables two means are used 1 May pre young of the year young planktivorous fish not yet present Either clear water phase or a spring algal bloom 2 July August post young of the year young planktivorous fish mostly present and high predation on zooplankton Totally there are 145 Dutch lake years 112 lakes in the database The database includes physical and chemical parameters information on morphology abundances of main groups of algae zooplankton and fish and coverage with macrophytes see Table 50 In the following table all variables are listed eee aa mman UIN Institute where the data came no Instituut waar de data vandaan komen m E B 8 E 7 ULA Recently manipulated Yes No Recent ingegrepen Yes No CIR b emm Retention time class Verblijftijd Kl Chlorophyll a May ug l Chlorofyl a mei Chlorophyll a Jul Aug Chlorofyl a juli aug Chloride mei Chloride Jul Aug men Chloride juli aug Conductivity May CEN CNN mg l Conductivi
28. iction in the main menu Slowly Fig 14 is build on your screen Press Esc to cancel otherwise wait until the adjusted R is displayed 2 Press any key to view the next relation 3 Repeat this until all response variables are tested and the main menu reappears 36 4 8 Optimizing your model In this exercise we optimize the methods and the weights of the conditional variables The calculations take much time so use a fast computer and let it work for a night or weekend To optimize your model 1 Select Optimize in the main menu 2 Select Enter all variables in the next menu If you have many conditional variables and you want to select the best explaining variables select Stepwise Then the variables are selected stepwise with optionally an iteration after each step 3 Select Optimize full and wait After some time a plot of the weights of the prediction exponent and the adjusted R is drawn on the screen After each iteration a point is drawn in the plot Fig 15 4 If you want to cancel the calculations press Esc The best model sofar is then stored 5 After a convergence criterion is reached an iteration report is written on the screen al Predictinn Fxnanent Fig 15 Optimization of weights 37 4 9 Saving and loading your model In this last exercise is shown how all selections can be saved Not only the selection of the variables will be saved but also the logical expressions to select part of
29. ile CALtI F Exit L1 L1 o L1 El o Fig 7 main menu A short description of the items in the main menu Run Weighted Average Prediction Analogies Find lakes that resemble the question lake Predictions Predict the response variables of the question lake WAP relations Show relations between response variables and conditional variables 26 Optimize methods Test predictions Test the prediction model and calculate the adjusted R for each response variable This is done by cross validation Optimize Optimize weights and or methods automatically by an iterative algorithm Warning this takes some time Methods Change methods of prediction and dissimilarity indices Selection of data or variables Select if Enter or change a condition for the database for example sediment Sand Input screen Retumn to the selection of parameters File Save or load selection to a file Exit Retum to DOS 4 3 Filtering the database In this exercise we will select from the data base only those lakes that are not recently managed and that have a maximum depth of less than 10 meter All other lakes are not used in the analyses To make such selection a logical expression is used A logical expression has one or more relational expressions that are combined by logical operators A relational expression consists of 1 variable name 2 a relational operator lt gt
30. led All but one method This is a data efficient way of testing a model see also Fig 3a Each time one lake from the database is removed and treated as the unknown case The rest ofthe lakes is used to make a prediction using Weighted Analogies Prediction This procedure is repeated until all lakes in the database are predicted The predictions are compared with the observed values The values of observed and predicted are plotted see Fig 3b and the percentage of the variance that is explained by the model is calculated R sometimes called coefficient of determination is the same statistic that is commonly used in linear regression The sample R usually is an optimistic estimate of how well the model fits the reality The statistic adjusted R attempts to correct R to reflect more closely the goodness of fit of the model in the population Formula 21 residual sum of squares N p 1 aa total sum of squares N 1 sample size p number of parameters If the adjusted R equals 1 the model fits perfectly if the adjusted is negative the mean value of the response variables is a better prediction than the model 19 Fig Cross validation select a lake from the database predict the response next lake explained variance compare predicted with observed reponses Fig 3b Goodness of fit of WAP adj R 0 46 regression 0 29 predicted Secchi depth observed Secchi depth
31. n either use the menu or the keyboard to enter an expression We will use the menu 2 Select left parenthesis and press Enter Select Select variable Now a list of variables is displayed Select MANIPULA from the list of variables and press Enter or double click Select the greater than sign gt EN De 12 13 14 15 Fig 8 Oops wrong symbol Press Backspace to clear last symbol gt Select the equal sign Select Select class Select No from the list of class names and press Enter Select the right parenthesis Select AND Select left parenthesis Select Select variable Now a list of variables is displayed Select MAXDEPTH from the list of variables and press Enter or double click Select the less equal sign lt Enter 10 Select the right parenthesis Now the condition is completed see Fig 8 Select Ready to save the expression Note if you press Esc the default expression remains unchanged Select if Conditional use of the database Fondi ti on MANI PULA No MAXDEPTH lt 10 X Ready Select variable Select class equal greater equal not equal less than greater than lam lett parenthesis ptotl lt 0 2 Manipula No and ptotl lt 0 1 or ptotl gt 0 2 Manipu a No and ptotl lt 0 2 LeftJtRight Hove cursor Enter t Backspace Del Escl Can 1 A logical expression
32. nal variables is shown Press F2 if you want to select a certain part of the plot You should do this to exclude parts of the plot with few data to avoid large interpolations You enter the zoom mode of the plot You have learned in 4 5 how to zoom in Press Enter to start the calculations All grid points that are calculated are displayed until the whole plot is filled Then SURFER is started and the three dimensional plot is drawn Fig 13 Press Esc to go on The next question appears Enter rotation of previous plot gt Enter 30 followed by Enter Now you return to the 3D plot which is rotated 30 degrees Press Esc twice to rotate another 30 degrees 35 10 To stop rotation press Esc followed by 0 and Enter 11 Press Y if you want to plot the graph else press N 12 Now a plot of the next combination of conditional variables can be made To cancel press Esc until you retum to the main menu 4 7 Calculate percentage explained variance R The next two exercises show the more advanced options of the program First the cross validation is applied to calculate the percentage explained variance and plot observed against predicted response variables MOL Chlorophyll a Jul Aug 300 400 500 nhserved Fhlaranhuyll a Tul Ang Fig 14 Test predictions and calculate the percentage explained variance adjusted R To test the predictions takes some calculation time 1 Select the bar Test pred
33. ne are deleted To install BASIS 1 Insert the installation disk into drive A Type the following command and press Enter A INSTALL 2 Enter the directory of the source usually A and press Enter 3 Enter the directory of the destination usually C BASIS and press Enter 2 3 Running BASIS To start BASIS go to the BASIS directory created with INSTALL Usually this directory is C BASIS To start BASIS type BASIS followed by Enter Now you enter the input screen where you can select variables 2 4 Getting help Wherever you are in the program it is always possible to get case sensitive help Press F1 to enter the on line help system To learn more about BASIS it is possible to get help on related topics and go to an index with 85 entries 1 Case sensitive help Press the F7 key any moment in the program to get case sensitive help Use cursor keys or the vertical scroll bar see 2 5 to scroll the help screen Help on related topics mouse click or 2 In most help screens part of the text is highlighted These are links You can use these highlighted links to display a new help screen that presents information about the related topic Click on the highlighted words to display that screen b If you don t use a mouse press F2 to get a list of related topics Move to the desired bar and press Enter 3 Help index F1 1f you press F1 in a help screen you get the Help Index on your screen Use the
34. or example BASMONO INI start BASIS in DOS the following way BASIS BASMONO 45 Appendix 2 How to enter new data in the database BASIS uses a binary file for its data input It is however possible is to enter the data in three dBASE TII files which will be transformed into a binary file 1 dBASE file with variables The structure of this file is as follows Field Field name 1 VARIABLE NAME UNIT VARMIN VARMAX 6 DEFTRANSFO naun Type Width Dec X Character 10 Character 35 Character 5 w Numeric 10 w Numeric 10 Numeric 1 a d Note the contents of the field VARIABLE must be the exact the same as the field names in the file with lake information 3 2 dBASE file with variable labels The values of class variables can be transformed to labels The unit of these variables see 1 must be CODE The structure of this file is as follows Field Field Name 1 VARIABLE 2 VALUE 3 LABEL Type Width Dec Index Character 8 N Numeric 2 N Character 10 N 3 dBASE file with lake information The structure of this file is as follows Field Field name 1 LAKE 2 NAME 3 YEAR 4 VARI N43 VARN Type Width Dec Index Character 8 N Character 35 N Numeric 4 N Numeric 10 5 N Numeric 10 5 N 1 to VARN contains the values of the N variables Beware the number of variables must be the same as in the file with the variable information and the field names should be the same as the con
35. or making predictions of the response of lakes to measures The user of the program enters relevant properties of the lake in question The program selects in a data set the most similar lakes based on these variables The database includes physical and chemical parameters information on morphology abundances of main groups of algae zooplankton and fish and coverage with macrophytes Information of 112 lakes from Holland is included The selected similar lakes are visualized in a way that differences and similarities are easily recog nized A prediction of the unknown lake is made by averaging the responses of the most similar lakes The program includes a test of how accurate these predictions are and an iterative algorithm to optimize the prediction Besides the predictions the program includes an option for exploratory analysis of the data set Two and three dimensional plots can be made individual points can easily be examined and selections can be highlighted The computer program is menu driven and includes an extended on line help system with more than 80 entries The use of a mouse is fully supported 1 How to install BASIS on your PC 2 How to get started with the program special keys the use of a mouse 3 How precisely the predictions are made and how the predictions can be optimized 4 How to enter new data in the data set 5 The used terminology in a glossary In an extended tutorial you can learn step by step what th
36. rectangle a Without mouse move the upper right corner of the rectangle that indicates the selected area by pressing cursor keys Up Down Left Right b With mouse move the mouse to the upper right corner of the rectangle and click 3 To Move the rectangle press F2 toggles between the move and the Resize Mode Without mouse move the upper right corner of the rectangle that indicates the selected area by pressing cursor keys Up Down Left Right b With mouse move the mouse to the upper right corner of the rectangle and click 4 Press Enter or double click to leave the Zoom mode The zoomed plot is drawn This plot can be slightly different from the indicated rectangle to get nice scales on the axes Highlighting a part of the data In this example we will highlight the lakes with more than 10 coverage of submerged plants SUBMERG gt 10 1 Press F3 2 Enter condition in our case SUBMERG 10 see also 4 3 3 Select Ready and press Enter The selected lakes are plotted as red circles while the not selected lakes plotted as blue crosses If there are cases where there is no information about the vegetation they are plotted as gray crosses Unless this selection is changed this selection is used in all plots predictions and WAP relations during the BASIS run 4 Press F1 to display a legend press Esc to return to the plot Examine individual points mouse only 1 Click
37. rmation non linear Transform variables to change the scale of a variable such that certain parts of the scale are shrunk while other parts are stretched Surfer SURFER isa registered trademark of Golden Software Inc P O Box 281 Golden Colorado 80402 USA BASIS can produce Command files to be used by SURFER and grid files in the format of SURFER If SURFER is present on your PC and there is enough free memory three dimensional Surfer plots can be made automatically Option WAP relations 3D WAP relation These plots can be rotated on your screen without leaving BASIS Otherwise the grid files can be used by other programs Values of the question lake The values of conditional variables of the question lake have to be entered ifa prediction of the response variables is to be made WAP relations The prediction method of the program can also be used as a kind of running avegage method of displaying the relation between two or more variables We call this WAP relations WAP WAP stands for Weighted Analogies Prediction It is a technique to find analogies and to predict the response of an unknown case Weights Less important conditional variables can be down weighted by the user by giving them a lower weight then the other variables By default all weights have the value 1 The weights can also be optimized automatically by a parameter optimization technique Controlled Random Search 41 Go e AY n plees qn Range d
38. t screen click on the highlighted word input screen Alternatively press F2 to display a menu with related topics and select input screen 5 Press Esc to go back to the first help screen 6 Press the F1 key Now a full list of topics is displayed see Fig 5 7 Scroll the list of topics cursor keys or scroll bar and press Enter or click on an item Alterna tively press the first character to move faster to the right spot 8 Press space bar to exit the help system Help avialable on the follouing topics Controlled Random Search Cord distance Cross validation DBASE III files Dissimilarities Distance weighting pouer Error 1 Number not valid Error 2 Negative ueight Error 3 Value too small Error 4 Value too large Error 5 No resp variables Error 6 No cond variables IF1J Help on help Enter Select UpllDounl Hove cursor Esc Backuard Fig 5 To get help on any topic press twice F1 and choose a topic in the help index To enter the response variables in our case chlorophyll a and Secchi depth 1 Move with cursor keys right to the column Resp response variable 2 Move down to the column with CHLA2 3 Press Enter alternatively double click with mouse The cursor moves to the right under the column Transform 4 To change the default transformation press Enter and select the desired one In this example the default transformations are used press Right arrow key 5 Do the
39. tandard deviation of variable in the database number of variables Weighting the variables The variables are weighted by multiplying their values with weights that are entered by the user The purpose of this weighting is to give important variables more weight The weights also can be optimized by the computer see 3 4 Calculate dissimilarities The distance between the question lake and all other lakes is calculated by a dissimilarity index Supported indices 1 Euclidean distance default The Euclidean Distance ED is the most frequently used index It is the distance in the n dimensional space constrained by the conditional variables Each variable is one dimension of the space 2 City Block distance also called the Manhattan Distance The City Block Distance CB is the sum of the differences between all variables It weights the variables that are far out stronger than the Euclidean Distance does a CB gt gt bu Yy kel 17 3 Cord Distance The Cord Distance CD is geometrically represented by the distance between points where the sample vectors intersect a unit sphere see Jongman ef al 1987 It gives more weight to qualitative aspects than the other indices of the program 4 Chebychev Distance The Chebychev Distance ChD is the maximum difference between variables It weights one variable that is far out even stronger than the City Block Distance ChD MAX y l Symbols
40. tents of the first field in file 1 To enter these new data change the file names in the start up file see Start up file INI and start BASIS If the new binary file is not found it automatically creates this file by reading the dBASE files 46 Appendix 3 Format of the grid files SURFER isa registered trademark of Golden Software Inc Box 281 Golden Colorado 80402 USA BASIS can produce Command files to be used by SURFER and grid files in the format of SURFER If SURFER is present on your PC and there is enough free memory three dimensional Surfer plots can be made automatically Option WAP relations 3D WAP relation These plots can be rotated on your screen without leaving BASIS Otherwise the grid files can be used by other programs The format of the ASCII grid file GRD is as follows Surfer Reference Manual id id 4 characters always DSAA which means ASCII grid file nx ny nx number of grid lines along X axis columns ny number of grid lines along Y axis rows xLo xHi xLo minimum X coordinate of grid xHi maximum X coordinate of grid yLo yHi yLo minimum Y coordinate of grid yHi maximum Y coordinate of grid zLo zHi zLo minimum Z coordinate of grid zHi maximum Z coordinate of grid grid row 1 grid row 2 grid row ny Z values of the grid organized in row order Each row has a constant Y coordinate with the first row equal to ylo and the last row yhi X coordinates within each row
41. ty Jul Aug EGV juli aug mg l Total N May CIT posee Total P Jul Aug Secchi depth May 818 STE E E AE 8 8 2 2 z Lid 2 i 51 i E emer a ammmon x seam oman NN ee ee diatoms Bio Jul Aug diatome a gewicht Gui aug m Q CHI2 SUSPMATI n BE amp SUSPMAT2 7 g a IHE Els d 2 greens Bio Jul Aug Cyanobacteria May No m Blauwalgen aantal mei om ange deme No QM e stipe a YAN2 blue green No Qul Aug a blauwalgen Gana 8 dione 00 toy s Js diatoms No Gul Aug s diatome a Guliaug IN I LN Cyanobacteria Jul Aug g 8 Z24 i SIE aE gja 5 5 z greens No Jul Aug Coverage emergent plants Coverage submerged macroph a Cladocerans Jul Aug Daphnia cuculata May Daphnia cucculata mei Daphnia sp Jul Aug Daphnia sp juli aug 1 52 Brasem kolblei gewicht z HE ETETETETET JE JE 919 5 Carp kg rper gewicht ARIABLE 3 B_PIKEPE i SI gewicht Ital w o gt a E E SERENE I EPIIT H E IDISIE
42. value for the response variables prediction of these n lakes Fig 2b Weighted Analogies Prediction Lakein Question data base lake total P maximum depth Fig 2 Schedule of the method of predictions b Example 14 In BASIS nominal variables may not be used as response variables because the mean of nominal variable lacks meaning When nominal variables are used as conditional variables the variable is translated into binary dummy variables Each class is one variable that can take two values true 1 or false 0 For ordinary variables these restrictions are not made The results of predictions of these variables still may be not very accurate Transformations Before standardization a non linear transformation can be used to shrunk certain parts of the scale and to stretch other parts The most common used transformation is the logarithmic transformation The following transformations and standardization are available 1 Logarithmic transformation Commonly used transformation to change a log normally distributed variable into a normal distribu tion or to give less weight to large quantities yg 7ln y 1 2 Log x 100 x transformation Special transformation to bring data of percentages of phytoplankton as close as possible to a normal distribution See Trimbee amp Prepas 1987 Ini 100 5 3 Square root transformation Square root transformation used before
43. y which method this is done 3 1 How to find analogies The steps to be taken to find analogies are summarized in Fig 2a Each step is explained below Select variables of question lake The first step in Weighted Analogies Prediction is to select variables that are to be used in the analysis There are two kinds of variables 1 Conditional variables Conditional variables are called independent variables in regression analysis They explain the state of the ecosystem of a lake These variables are known or can be managed in the question lake 2 Response variables Response variables are variables that express the state of the ecosystem of a lake In regression these variables are called dependent variables Class variables Class variables need special attention since these variables are not fully quantitative We distinguish two kinds of class variables nominal and ordinal variables 1 Values assigned to the classes of nominal variables do not imply any order Differences between them have no function Examples sediment type lake manager 2 Values assigned to the classes of ordinal variables imply an order Differences however have no absolute meaning Example Retention time classes 13 Fig 2 How to predict select variables of problem lake transform the conditional variables standardize the conditional variables calculate dissimilarities with all other lakes select the n closest lakes alculate mean
Download Pdf Manuals
Related Search