User guide to the STatistical Analogue Resampling Scheme Version
Contents
1. generating the date to date mapping If everything is prepared the simulations can be started by typing gt stars_2 4 exe number 12 where number is an optional integer number which serves as seed for the random number generator This is useful if several runs of the program are started simultaneously as it is thereby possible to make sure that every single run starts with a different random seed For each simulation a directory is created labelled with integer numbers starting from 0 All output for these simulations is stored in the respective directories The output of the model is written to the directory of the current simu lation It consists of four files e order _years dat The order of the rearrangement of the yearwise segments from the observation period which constitutes the first ap proximation see Section 1 1 Its elements range from 1 to the number of years of the observation period e final _order_dates dat As the ultimate goal of the simulation con sists of the generation of a date to date mapping of calendar dates of the observation period to dates of the simulation period this mapping is written to file final order _dates dat which looks like 11 3 1976 lt 21 2 1961 12 3 1976 lt 22 2 1961 13 3 1976 lt 13 3 1974 14 3 1976 lt 14 3 1974 15 3 1976 lt 15 3 1974 16 3 1976 lt 16 3 1974 17 3 1976 lt 17 3 1974 18 3 1976 lt 18 3 1974 19 3 1976 lt 19 3 1974 202. lations for post processing and further analysis Note that this statistics can be carried out reasonable only if the variables in the input files are placed like in the example file in Section 3 2 1 Also given is the rank of each simulation on a scale from wet to dry This rank is determined as follows Mean and trend of the climatological water balance are calculated for each representative station and each simulation E g if 4 representative stations are used an 8 tuple 8 4 x 2 2 because of 14 mean and trend characterises each simulation These tuples are written into the rows of a matrix with as many rows as simulations Each column of the resulting matrix is then sorted individually such that the first row contains the minimal means and trends and the last row the corresponding maxima The rank of a given simulation can then be determined by looking for the matrix row which is most similar to the 8 tuple of the simulation similarity measured by the Euclidean distance At the end of this file the most prominent quantiles and their simulations are given They give a reasonable starting point if one is interested in analysing the whole spectrum of the ensemble without analysing every single simulation which eventually can be quite a lot of data 3 5 Post processing generating the final simulation files To generate the final simulation files from the date to date mapping files the program post process exe is used It
3. 2 prepare blocks cpp The blocks which are to be replaced are identified using a heuristic cri terion which decides loosely speaking whether any given block in the first approximation series contributes too much to the mismatch between the pre scribed and the first approximation regression line For this a cluster analy sis of the blocks is required which classifies blocks into groups consisting of similar blocks where similar refers to the temperature observations of the blocks See in Sec 4 4 identify cpp The replacing blocks are selected from the blocks of the observation period according to several heuristic rules Besides temperature observations these rules make use of the calendar dates of the blocks in order to select blocks from appropriate seasons They define a set of potential replacements which firstly bring the regression line of the resulting series closer to the prescribed one and secondly make sure that the inserted replacement fits well into the parts of the series which have been set already From this set of potential replacements a block is chosen randomly After the replacement of all bad blocks the date to date mapping is defined for the current iteration See in Sec 4 4 replace cpp used_ red cpp date_ red cpp bridge red cpp If despite these replacements the regression line of the simulated an nual temperature means does not match the prescribed parameters within the given tolerance
4. 3 1976 lt 20 3 1974 21 3 1976 lt 21 3 1974 22 3 1976 lt 22 3 1974 23 3 1976 lt 23 3 1974 24 3 1976 lt 24 3 1974 25 3 1976 lt 30 3 1972 26 3 1976 lt 31 3 1972 27 3 1976 lt 1 4 1972 28 3 1976 lt 2 4 1972 where e g the first row means that March 11th 1976 from the simula tion period is assigned February 21st 1961 from the observation period and so on 13 e final order days dat The output in file final_ order_ days dat is a single list of the row indices corresponding to the assigned dates This will in general be the preferred starting point for building and analysing the corresponding simulation Row indices start at 1 For example if the first few lines in final_ order days dat look like 4 o N DH ot 9 10 11 12 13 14 15 382 383 384 then the first day of the simulation period is assigned the fourth day of the observation period the second the fifth the 13th the 382nd and so On e statistics dat Several statistics concerning the current simulation 3 4 Characterising the simulations Once all simulations are run a file called sim characteristics dat is writ ten to the working directory It contains a summarising line for each simu lation indicating mean and trend with respect to annual means of all mete orological variables and the climatological water balance averaged over all stations and grid points This information is intended to help choosing simu
5. not be negative the format of the observation files as described above is mandatory for a correct post processing If there are more columns than in the list above the last column is expected to contain categorical information which is not altered Any additional columns up to this categorical column are modified without any con siderations for meaningful ranges 2 Rearrange correct for interannual variance and shift the simulated series such that there is no break between the end of the observation series and the begin of the simulated series This can be useful if a period directly following the observation period is to be simulated No break in this setting means that the regression lines of the annual means from the observation and from the simulation period connect smoothly For additional columns in the list above applies the same as for level 1 3 As level 2 but with modified treatment of sunshine duration sunsh and global radiation rad For post processing level 0 and 1 the abso lute values of these variables are rearranged This can cause a break of the physical limitations due to the obliquity of the ecliptic For example a day with maximum attainable sunshine duration in summer can be reallocated to a day in autumn with a lower possible sunshine duration To avoid this effect post processing level 3 applies the reordering to the relative values of sunshine duration and global radiation Therefore in a first
6. see Section 3 2 3 this second step is iteratively re peated Therefore in order to compensate for a potential bias from the previous iteration the following iteration is prescribed with exaggerated re gression parameters see Orlowsky et al 2008 for details See in Sec 4 5 check regression parms cpp Both at step 1 and at the end of step 2b years or blocks are drawn randomly This makes any simulation a stochastic realization of the popula tion of possible simulations given the prescribed regression parameters and the set of heuristic rules Its range can be estimated by generating large ensembles of simulations 1 2 Extension to several locations The approach works pretty much along the same lines for multi location simulations except for that it takes place in a parameter space of higher dimensionality The ultimate goal however a date to date mapping is the same As the simulated series are generated by rearranging observations from the observation period according to this mapping and the order of the rearrangement is the same at all stations or grid points the series of simu lated fields consist of spatial fields that were observed during the observation period In order to limit the dimensionality of the task in a preparatory step climatologically similar locations are classified using cluster analysis The stations are characterised by selected climate statistics from the observation period e g mean temperature an
7. step the absolute observations of each day are transformed to values relative to the multi year average of that day Then the rear rangement is applied to the transformed values Finally the time series is transformed back into absolute values using the multi year average of the observations Example gt post_process exe data elbe_data 23 2 17 4 The program files Files are organised in several directories which correspond to different units in the program Each directory contains a hpp header file which defines one or several classes or just functions see below main h and main cpp The files which bundle it all There are contri butions from the following directories organised along the boxes in Figure 2 4 1 Input This directory corresponds to the boxes on the left side of Figure 2 Input hpp The input level consists of three classes e Simparm cpp The Simparm class where the parameters from sim parm dat are stored e Data cpp The Data class where the observational data from the representative stations and the respective regression line parameters are stored e Dates cpp The Dates class where calendar dates of simulation and observation period are stored read methods cpp Methods for reading matrices and vectors from files 4 2 Preparation This directory corresponds to the lower half of the left top box in Figure 2 namely the preparatory organisation of the input data
8. Preparation hpp and Preparation cpp The Prepared_ data class which contains the data organised as e prepare _years cpp years and as e prepare _blocks cpp blocks including the cluster analysis 18 4 3 Firstapprox This directory corresponds to the second level grey box the rearrangement of entire years Firstapprox hpp and Firstapprox cpp The Firstapprox class which contains the first approximation i e a rearrangement of entire years generate fa cpp contains the necessary functions for this 4 4 Mending This directory corresponds to the yellow boxes of step 2a and 2b in Figure 2 Mending hpp and Mending cpp The Mending class which contains the mended first approximation i e the first approximation with some se lected blocks replaced in order to obtain series which match the regression parameters identify cpp Functions for identifying blocks which are to be replaced replace cpp Functions which replace the identified blocks These functions call code from e used_red cpp A function which selects unused blocks e date_red cpp A function which selects blocks with an appropriate Julian day e bridge _red cpp A function which selects blocks which match the predecessor and or successor in the simulated series 4 5 check This directory contains the functionality for checking whether the Mending object complies with the prescribed regression parameters and if not up dates the internally
9. The selected characteristic climate variable must be located in one of those columns for the introductory description of the model this variable was mean temperature NA values should be entered as 999 9 It is important to know that all kind of variables in any order can be written in the input files However some processes in the model e g pre processing post processing simulation characterisations demand the mete orological variables in the order shown above otherwise the results are not reasonable It is therefore recommended to use the specific order and to add other variables in additional columns Obviously some of the processes also can not deal with NA values Hence they should be avoided if a proper simulation characterisation is desired 3 2 2 Regression parameter files The regression parameters characterise the development of the selected char acteristic climate variable More specifically the simulation series are gener ated such that the annual means of the selected variable feature a regression line which corresponds to the prescribed parameters If not calculated from a general prescribed trend see Section 1 2 and 3 2 3 the files containing two values each the starting level of the regression line for the simulation period and the final level E g if you want to simulate a climate of which the temperature starts at 8 7 C and arrives at 9 9 C in terms of the linear trend of mean annual temperature then
10. This program is open source you may use and adapt it whenever it appears to be useful However it comes without any warranty Climate simulations obtained by this model should be considered as an estimation of possible future development not as predictions Even though the program has been tested and evaluated carefully no guarantee can be given with respect to its bug freeness So if you use it use it at your own risk And cite its published foundations Orlowsky 2007 and Orlowsky et al 2008 Further model applications can be found in Orlowsky and Fraedrich 2009 Orlowsky et al 2010 and Lutz et al 2013 This manual is organised as follows After a brief description of the steps within the model and their corresponding objects in the implementation the next two sections will describe how to install and use the program Finally the single source files are summarised The model works with station as well as with gridded data e g from GCM or RCM simulations The program consists of two three parts stars 2 4 exe which gener ates a date to date mapping see Section 1 and post process exe which generates the final simulation files from the date to date mapping with op tional corrections for inter annual variability and a smoothing of the series between observation and simulation period Optionally pre process exe can be used in order to derive a meaningful set of representative stations see Section 1 2 1 The model a b
11. User guide to the STatistical Analogue Resampling Scheme Version 2 4 Boris Orlowsky Julia Lutz June 14 2013 Contents 1 The model a brief description 3 1 1 Outline for a single location station or grid point 4 1 2 Extension to several locations 6 2 Installation 7 21 e Sica he og og aye Gag yee Ga lee Gh eA Se ww oh 7 3 Usage 8 3 1 Optional pre processing oo o a a 8 3 2 Preparation of the input data 9 3 2 1 Observation data files o aa 9 3 2 2 Regression parameter files 9 3 2 3 Simulation parameter files 2 0 65 2 2 8 be eed 10 3 2 4 The calendar for the simulation period 12 3 3 Run generating the date to date mapping 12 3 4 Characterising the simulations 14 3 5 Post processing generating the final simulation files 15 4 The program files 18 AN JW c Mets Stee a ote G7 Sei Ge APE a oe Fees de he 18 4 2 Preparation 62 os BK wee Boe wpe eae oS ae he ee 18 Ao Isher Ok y uc wes Serie eed dee ta Vane coud bret dar 4a x 19 AAS Mending cea manata ap RE eed ge SR clon Re A a e 19 a man aoa A E cee 19 As Output 45 ght 4 ht Me AE So ad She BS ee 19 A OGNA rach ia Watt Gee ee Ae A o E 20 ALS MISC foe Aas eh o ee ee e a a ee a eb 20 ADs PLE POEMA AOR oh hg yen ag eae oe ae oe ee Gee eS 20 4 10 post_ processing ca A a a ee ER 20 ATE my E e e peaa PRR epee Pe we geal o moh geal om he 20
12. break adaption number of years which needs multiple sweeps for variance break adap tion maximum number of sweeps for one year maximum amount changed for a single day percentage of variance break adapted days The syntax of the post processing is the following gt post_process exe basz path work directory simulation level where basz path the path of the basis scenarios work directory the path from where the simulations were invoked It must contain a file called stations tab which in its first column has the 5 digit station id and in its second column the latitude This matters to possible corrections of the sunshine duration necessary for level 1 and 2 see below simulation The simulation number which is to be post processed The number corresponds to the directory an integer from 0 to Nsim 1 and can be taken from e g the sim characteristics dat file level One of the following O Just rearrange This works for all kind of input data according to Sec 3 2 1 16 1 Rearrange and correct each variable year by year such that the interannual variance of the simulated series is the same as during the observation period As resampling schemes like STARS sometimes tend to underestimate long term variabilities this may be a welcome post tuning As this means an individual alteration of the meteorological variables and these are subject to differing constraints e g precip itation must
13. can simply rearrange the observa tions according to final order_ days dat or do so plus extra corrections see below If the corrections are to be applied the input data MUST comply with the following file format header line day month year vl v2 v3 v4 vd v6 v7 v8 v9 v10 vil where vl maximum of temperature C v2 mean of temperature C v3 minimum of temperature C v4 precipitation mm v5 relative humidity v6 air pressure hPa v7 vapour pressure hPa v8 sunshine duration h v9 cloud cover 1 8 v10 radiation daily sum J cm vll mean wind speed m s 15 Besides the post processing creates a file called post statistics dat if the level see below is greater than or equal to 1 It is intended to monitor the correction of the inter annual variability and or correction of breaks be tween observation and simulation time series which are likely if level gt 2 The STARS post processing routine uses an iterative adaptation algorithm to achieve these corrections Especially in arid regions this can result into anomalous precipitation values for particular days This affects mostly hu midity and radiation related variables i e precipitation relative humidity vapour pressure sunshine duration cloudiness global radiation and wind speed Therefore for each station the following statistics will be monitored for each affected variable and stored in post statistics dat number of years with false variance
14. cess exe basz path work path where 3See http www boost org more getting_ started unix variants html e basz_ path is the path where the available basis scenarios are stored and e work path denotes the directory where the result is to be stored This result is stored in suggested reference stations dat and contains suggested lists of up to 10 representative stations They can be inserted into simparm dat see below in Section 3 2 3 3 2 Preparation of the input data 3 2 1 Observation data files The files containing the observation data need to be named according to the id of the station or grid point and basz stands for Basis Szenario basis sce nario They need to contain an arbitrary header line and have to consist of columns containing the following elements day month year meteorological variables e g the PIK internal format day mon year tmax tmean tmin hum airp vapor sunsh cloud rad wind lo Be o Cc 1 1 1951 1 5 3 8 12 6 5 83 0 987 7 3 8 1 5 7 3 297 0 4 3 2 1 1951 3 2 5 5 1 0 92 0 983 8 4 9 0 5 0 208 0 4 3 3 1 1951 1 2 2 5 0 0 85 0 991 2 5 2 0 Td 141 0 2 9 4 1 1951 3 4 6 1 0 0 82 0 1003 6 5 4 2 2 4 0 198 0 5 8 5 1 1951 4 1 2 4 1 1 4 8 92 0 999 0 6 6 0 7 0 158 0 5 8 6 a 1951 6 0 4 4 3 0 1 9 99 0 996 7 8 7 0 8 0 35 0 4 8 7 1 1951 6 9 5 4 1 8 1 1 75 0 989 6 7 0 3 7 3 48 0 8 3 8 1 1951 6 1 3 9 2 9 7 85 0 997 9 6 9 4 8 0 93 0 6 6 9 1 1951 5 6 3 5 2 0 5 2 83 0 995 3 6 7 0 Tt 133 0 9 2
15. d precipitation standard deviation of tem perature and precipitation and the difference between the mean of the sec ond half and the mean of the first half of the observation period They thus roughly describe climatological level variability and temporal development at the stations For each of the clusters a representative station is identified by looking for the station most similar to the cluster centre of mass Linear regres sion parameters for the characteristic variable are prescribed at each of the representative stations thereby allowing for the simulation of spatially dif ferentiated developments Suggestions of sets of representative stations are made by the program pre_process exe which determines useful sets of 1 to 10 out of all available stations see in Section 4 9 pre processing pre processing cpp It uses a simple kmeans clustering algorithm with random initialisation from the SiMath library see below After all simulations are run their most important characteristics are written to a file see Section 4 7 2 Installation The program is implemented in C It makes use of parts of the open source C library SiMath 1 0 which were partly edited and corrected by Boris Orlowsky They are integrated directly into the source code so for this no extra library compilation and linking is necessary However the program requires a part of the boost library namely the filesystem part 2 1 Linux Unix A Makefile fo
16. emes like STARS are prone See in Sec 4 2 prepare years cpp and in Sec 4 3 gener ate fa cpp The second step iteratively alters this first approximation in order to find a series which matches the prescribed regression parameters exactly This step replaces segments of a given number of consecutive days according to INPUT FLOW STEP 1 STEP 2 Daily meteorological data Yearwise Blockwise Jan 1st 1977 Feb 2nd 1977 Dec 31st 1977 Feb 13th 1977 Clusteranalysis lof blocks 2b Replace blocks with appropriate replacement blocks 1 Rearrangement of entire years from the observation series 2a Identify blocks which are to be re placed Clusteranalysis Prescribed regression line for the annual means of a characteristic variable Internal adjustment of the prescribed regression line Figure 2 Summary of the generation of the date to date mapping the parameter blocklength see Section 3 2 3 Replacing blocks instead of single days helps to obtain simulated series with realistic persistences as the weather sequences within the blocks are entirely copied from the observed series Experiments with station data in Central Europe have shown that a block length of 12 days essentially captures the persistence of the observed time series This number might be different for other regions in the world Thus this should be tested and the number should be adjusted if necessary see in Sec 4
17. lt and the simulation is run 10 internal cluster y n If yes the data is clustered by the pro gram internally using a kmeans clustering and a random seed Al ternatively if internal_cluster n the user is expected to provide an external clustering in the file cluster dat which contains a class id for each block see Section 1 cluster dat therefore contains as many entries as there are rows in block dat The cluster ids have to range in 0 no_ clusters 1 no clusters integer number The number of clusters This has to be provided also if the clustering is done externally defaults to 50 no sims integer number in 1 many The number of simulations to be run For each simulation a directory is generated which will keep its results points separated by blanks The program works with any number of stations however convergence to the prescribed regression lines of each station can become very hard to achieve for numbers larger than approximately 6 In that case tolerances need to be very generous see below delta_ temp floating point number General trend of the character istic climate variable for the entire region starting from the level of the end of the observation period This number will be rescaled for each representative station or grid point by assuming constant ratios of the observed trends at these stations for the simulation period calc trends y n Whether regression parameter
18. of representative stations or grid points has to be identified in a preparatory step Suggestions for these stations or grid points are given by the program pre _ process exe but the user is free to identify the stations or grid points which in his expertise optimally represent the climatological variability of the region of interest Additionally to the data prescribed regression parameters of the selected climate variable must be provided This can be done in two ways by pre scribing a general trend for the entire region which is understood as starting from the level of the end of the observation period It is rescaled for each representative station or grid point according to the evolution of the obser vation period see Section 3 2 3 for more details Alternatively a file for each representative station or grid point can be provided which contains the regression parameters see Section 3 2 2 The archive files for testing tar gz contains not necessarily mean ingful example files which can be used for a test run under Windows for example WinZip is able to open this archive Unless otherwise stated files are expected to be located in the working directory i e the directory from which stars 2 4 exe is called and into which its results are written 3 1 Optional pre processing The program pre _ process exe can help identifying meaningful sets of rep resentative stations or grid points It is invoked as follows gt pre pro
19. prescribed exaggerated regression parameters check regression _parms cpp Check the regression parameters achieved so far and update the internally prescribed ones 4 6 output output cpp Various functions for output 19 4 7 Sim_charact Sim _charact hpp and Sim_charact cpp Initialise an object for the characterisation of the simulations pre_simulation cpp Preparatory functions and data post simulation cpp Synopsis of the individual simulation characteris tics and output 4 8 misc misc cpp Various arithmetic functions linear regression etc cluster as signment building of blocks and type conversions 4 9 pre_ processing pre_processing cpp Program which gives suggestions for meaningful choices for the set of representative stations 4 10 post processing post processing cpp All you need for the post processing 4 11 my SiMath Parts of the SiMath 1 0 library edited and corrected in parts by Boris Or lowsky Provides functionality for matrix and vector operations and cluster analysis 20 References J Lutz J Volkholz and Gerstengarbe F W Climate projections for south ern Africa using complementary methods Climate Change Strategies and Management 5 2 130 151 2013 B Orlowsky Setzkasten Vergangenheit ein kombinatorischer Ansatz f r regionale Klimasimulationen PhD thesis University of Hamburg Hamburg Germany 2007 http www sub uni hamburg de opus voll
20. r Unix and Linux environments which allows for an easy build is provided It has been tested in SuSE Linux and Debian Linux environments with the gcc 4 1 0 and 4 1 2 compiler Note that in this Makefile the boost library is supposed to be located in the directory where the program shall be compiled If the boost library is installed elsewhere adapt the boost related variables in the Makefile Once the boost library is installed please note that although boost is mainly a header only library the filesystem part needed for this application requires some compilation just place the archive stars 2 4 tar gz into a suitable directory unpack it by typing gt tar zxf stars 2 4 tar g2 and then after editing the Makefile compile it by typing gt make all This should give you the stars 2 4 exe the pre process exe and the post _process exe files Alternatively you can compile only one of them by typing gt make star or gt make post process or See http www silicos com simath html See http www boost org gt make pre_ process gt make clean removes the object files If the build is dynamically linking the boost library as it happens with the provided Makefile an gt export LD_LIBRARY_PATH lt path_to_stars2 4 bo0st boost_1_34_1 lib LD_LIBRARY_PATH or equivalent will be necessary 3 Usage As you can see from Section 1 in order to generate regional climate simula tions a small set
21. rief description This model generates simulated time series of climate data on a regional scale based on station or gridded data One of its most important features is that the simulated series are constrained only by the parameters of a linear regression line The annual means of the generated time series of a chosen climate variable have to match these parameters Inspired by Werner and Gerstengarbe 1997 the simulated series are assembled from segments of the observed series The approach generates a mapping from dates of a simulation period to dates of the observation period hence resampling the observation time series This resampling is constructed such that the corresponding series yield annual means of the chosen climate variable which feature the prescribed regression line Furthermore a set of heuristic rules makes sure that the resulting series exhibit realistic properties such as annual cycles or persistence See Figure 1 for an illustration of this resampling scheme for which temperature is chosen as the characteristic variable Here only a brief description of this construction is given For a detailed description the reader is referred to Orlowsky et al 2008 Observation period Simulation period Simulation period ARAMA ATE A 10 e 10 PAI A jt ill Hf 1980 1985 1990 1995 2000 1992 1994 1996 1998 2000 Figure 1 Assembling simulated series from segment
22. s are to be calcu lated from the general trend in delta temp If not they are read sentative station or grid point tolerances A list of floating point numbers giving the tolerated de viations from the prescribed regression line One tolerance per station separated by blanks Units are the same as the unit of the chosen characteristic variable We recommend values of 10 20 tol jd integer number The allowed difference of the Julian day when replacing blocks see Section 1 1 Defaults to 20 no conca integer number Number of rearrangements of entire calendar years the best of which is used as the first approximation A 11 reasonable number is 50000 However for tests one can use a lesser number for example 5000 to save computation time simparm dat could look like basz_path data elbe_ data char _var 5 blocklength 12 normalise n output block n internal _cluster n no_clusters 50 no_sims 3 stats 17007 21005 delta temp 1 2 calce trends y tolerances 0 2 0 19 tol _jd 20 3 2 4 The calendar for the simulation period The program also looks for a text file containing the calendar dates of the sim ulation period in the working directory It has to be named datum_ sim dat and contains the dates in the format day month year e g 1976 1976 1976 1976 1976 1976 1976 O Ot Ha wd HA ee T The simulation period must not consist of more years than the observation period 3 3 Run
23. s of observed series corre sponding to a prescribed regression line Left observation series and prescribed regression line for temperature Right simulated series with annual means dots featuring the prescribed regression line 1 1 Outline for a single location station or grid point If only a series for a specific location meteorological station or grid point is needed no spatial dependencies have to be taken into account Figure 2 summarises the generation of the date to date mapping for this case It comprises two steps operating at different time scales The first step operates on the time scale of years parts of the flow chart concerned with this time scale are highlighted in grey It generates a first approximation to the mapping which consists of a simple rearrangement of entire calendar years from the observation period This rearrangement is chosen out of a large random sample of shuffled calendar years such that its corresponding temperature series is as close as possible to the prescribed regression line and to the observed inter annual variance This series is guar anteed to exhibit realistic annual cycles and weather sequences within the single years as they are simply copied from the observed series Choosing the rearrangement which is the closest to both the prescribed temperature regime and the inter annual variance from the observation period helps to pre vent a loss of long term variability to which resampling sch
24. texte 2007 3316 B Orlowsky and K Fraedrich Upscaling European surface temperatures to North Atlantic circulation pattern statistics International Journal of Climatology 29 6 839 849 2009 B Orlowsky F W Gerstengarbe and P C Werner A resampling scheme for regional climate simulations and its performance compared to a dynamical RCM Theoretical and Applied Climatology 92 209 223 2008 B Orlowsky O Bothe K Fraedrich F W Gerstengarbe and X Zhu Fu ture climates from bias bootstrapped weather analogs An application Journal of Climate 23 13 3509 3524 2010 P C Werner and F W Gerstengarbe Proposal for the development of climate scenarios Climate Research 8 171 182 1997 21
25. the corresponding file would look like 8 7 9 9 Note that the second number has to be followed by a blank line 3 2 3 Simulation parameter files The program will look for a file called simparm dat in the working directory In this a number of options for the simulations have to be set always in the form OPTION VALUE The options are e basz_ path string Contains the absolute path of the observations basis scenario files Defaults to the current directory e char var integer number The column number in which the char acteristic variable is stored Must range between 4 and the number of e blocklength integer number The length of the blocks in days For Central Europe 12 days has proven to be a reasonable length which is also the default e normalise y n Whether or not the used data should be normalised This is important if the variability varies much across the studied re gion If normalise y the data is normalised to have a mean of 0 and an unit standard deviation e output block y n As you can see from Section 1 1 and Figure 2 the input data is organised in blocks of blocklength days sliding blocks These blocks are classified using a cluster analysis If you want to use an own clustering algorithm set output block y The program then only builds the sliding blocks outputs them to block dat which can be used for the external cluster analysis and exits Otherwise the blocks are bui
Download Pdf Manuals
Related Search