10 Parameter optimisation, confidence and sensitivity
Contents
1. headiagi head2aqi dif h2aqi2 head4aqi dif h4aq23 167487 81 168625 96 168625 96 168892 23 168892 23 496741 53 496649 65 496649 65 496406 83 496406 83 1 1 2 1 2 9082 1 1 1 0745 1 3292 mi 6 x Td 1 3626 46438 98 46 55 17 219 1 9082 1 1 m PRI 314 1 4 7061 65185 219 1 e 18 c test triwaco trec all cal3 QrchFactor ado QRI1 Qriifactor ci test triwaco trec all cal3 Qrilfactor ado f luxRiv 1 5 2661 1238 5537 40 1 e 18 QRI1 Qriifactor ci test triwaco trec all cal3 Qrilfactor ado f lux rch 1 5 27661 63946 805 314 1 e 18 QRCH QrchFactor c test triwaco trec all cal3 QrchFactor ado END pntSaqi 169666 07 497469 99 f luxTot 2 QRCH QrchFactor Code Adore label Adore file with factors in FLO for each flux item NE 5 i 10 Parameter optimisation confidence and sensitivity analysis 5 Royal Haskoning Triwaco User s Manual 10 2 3 Output data description Calib cho The Calibration output file Calib cho contains a list of observation well names followed by the aquifer number the calculated and observed head and the difference between observed and calculated head At the end of this listing some statistical information is added both for the total of all aquifers and for the aquifers separately The calibration output file Calib cho can be viewed selecting Calibration View Output from the Calibration pull down menu or as a background map in Triplot2. 4 MonteCarlo Calibration Sensitivity and Confinence Analysis A complete description for using MonteCarlo will become available soon 10 4 1 Methodology When a MonteCarlo calculation is started it will run the model simulations many times Each time with different parameter values The program gets its parametervalues from a parameterdistribution according to the Latin Hypercubemethod Every run is compared with a set of measurements From the results the optimum parametervalues are selected 10 4 2 Running MonteCarlo After creating the calibration data set one extra calibration set should be created e g MCCAL Move the parameters to calibrate to the Tab sheet Modified Parameters Create for every modified parameter one user defined parameter For example for TX1 create MCTX1 If necessary create a zonation for the user defined parameters and give them a default value This is the label of the MonteCaloparameter Allocate all the parameters GEBRUIK 1 Maak in Triwaco voor windows een model en run het model in een CALIBRATIE set bijv CAL 2 Maak in triwaco een nieuwe calibratieset bijv MCCAL en modify die parameters die gecalibreerd dienen te worden Maak voor elke modified parameter een userdefined MC parameter bijvoorbeeld voor TX1 de parameter MCTX1 Maak voor deze parameters al of niet een zonering aan en geef voor de defaultwaarde of de zone een nummer op Dit nummer is het label van de MonteCarlo parame
3. Aquifer n 1 Where 10 Parameter optimisation confidence and sensitivity analysis 3 Royal Haskoning Triwaco User s Manual Item Description Format Code well 1 Name of the observation well A10 X coordinate X coordinate of the observation well F10 Y coordinate Y coordinate of the observation well F10 Cluster Cluster number l5 Aquifer code Aquifer number or code for head differences l5 Measured value The value to be compared with simulation F10 result head or head difference Time Stress input time for transient simulation F10 leave open when not used Weight Weight of the measured value Is used for F10 the overall calibration results Aquifer n Used for head flux difference 15 Aquifer n 1 Used for head flux difference 15 With the cluster number one can assign certain wells to a specific group for statistical analyses For phreatic heads you fill in a O for the number of the aquifer Code 9002 Calibration using head differences Head difference is specified as Measured value Aquifer n and Aquifer n 1 have to specified as well Code 9001 Calibration using fluxes See below for an explanation The well data has to be entered in a fixed format where A10 F10 l5 Fluxes a text with a maximum of 10 characters incl Spaces a number with a decimal point 10 characters long an i
4. Calib chp A print file Calib chp with information concerning the comparison can be viewed selecting Calibration View Log from the menu 10 2 4 Visualisation of calibration results in TriPlot The differences between the calculated and observed heads may be displayed in a map selecting Calibration View Map from the pull down menu This starts the graphical presentation tool Triplot and displays the model area and the computed differences The calibration results may also be displayed selecting Checkpoint Output from Triplot s View pull down menu In that case the user can specify whether or not the observation wells will be labeled or not T Triplot C MYMODE 1 0MDEL MODEL3 FEGRID NEW5 _ TPP D Eile View Param Window Help maj E SWEA aa al eel Contour map PHI S in metis Witego70 CN 1 gt 41 4 WYv29084 a Checkpoint Aquifer 1 0 100 0 10 0 m 21650 454945 Aw22078 y VU 6494 v 200 205 i M2 0490 inzage 45 5 Mago 95 08 hap 7 ia ST 62 NM es iin ae WWWZTNONZI39RVZ 049 M29771 0 100 29 Ww nds 13 12 4 WW29066 ue UN Ec Y ad M win g091 133 Esp 10 2 j N ms 563 LI e 7h 4 b E 511334 8471 Time series graphs can be used to evaluate transient model simulations Time series graphs will display the calculated heads and observations in one graph How to do this is explained under How to in chapter 9 TriPlot An example is given below 10 P
5. al to an observed head The well data has to be entered in a fixed format where A10 a text with a maximum of 10 characters incl Spaces F10 a number with a decimal point 10 characters long I5 an integer number without a point 5 characters long The evaluation of the summed if more than one items are specified is in m3 day The fluxes of rivers line elements is in m3 d the vertical fluxes in the FLO simulation results file however is in m d Therefor the factor file has to be created It contains nodeinfluece areas for every node and if necessary can be made negative or positive For QRCH and QRIn a negative value represents an upward flux out of model and for QKWn a downward flux is negative The input file can be created by using TriPlot Open de model grid in TriPlot Load grid teo as parameter and select node influence area as parameter Param Load If necessery use the expression option in TriPlot to create a positive or negative factor Now choose Param Save in TriPlot and save the file in the appropiate file name Example input file calib chi Name Cluster XY coordinates for head difference amp of items for flux VEDIT PLUS Aquifer or code Value Time VVeight Aquifer O NPAK for Leave out head 9001 for flux if steady 9002 for head diff Aquifer 1 2 for head dif pitt C tmp Calb chi Ex mple CHI with heaa fluxes ard head differ uices
6. arameter optimisation confidence and sensitivity analysis 6 Royal Haskoning Triwaco User s Manual 10 Parameter optimisation confidence and sensitivity analysis 7 Royal Haskoning Triwaco User s Manual 10 3 TrCalCon Calibration Sensitivity and Confinence Analysis TrCalCon is the calibration and confidence module of the groundwater simulation package The package allows for a number of tasks in the setup and use of a model e Parameter sensitivity and correlation analysis Automatic parameter optimization e Confidence estimation based on linear variance calculation e Confidence estimation based on Monte Carlo analysis The confidence estimation can not only be performed on any output of the groundwater flow model but also on the result of any post processing program The module consists of a user interface and a number of calculation programs The user interface can be accessed from the Tools pull down menu in the TriShell the included in the TriShell However the package can also be used independently of Triwaco The full manual for TrCalCon on the use input and output files is available in appendix F A short description of the use is given in this chapter 10 3 1 Methodology The principle of calibration and confidence analysis can be described briefly as determination of the best values and confidence of parameter items based on measurements calibration and subsequently deriving the confidence of the
7. bration View Output from the Calibration pull down menu or as a background map in Triplot Calibration Calibration View Map A print file Calib chp with information concerning the comparison can be viewed selecting Calibration View Log from the menu 10 2 2 Input data description calib chi The name of the Calibration file is fixed to Calib chi and this file should be located in the Calibration dataset s directory where the simulation is carried out The calibration input file Calib chi can be viewed or edited selecting Calibration View Create Input from the Calibration pull down menu The file contains the names coordinates and observed heads for observation wells in the model s area For each observation well the aquifer number and a cluster number are indicated as well The use of a cluster number allows the user to discriminate between various areas of interest It is also possible to use the calibration file to calculate waterbalance calculations of a selected area In the following paragraphs the input data is described An example is given at the end of this paragraph Measured heads or head differences The file with the measured heads should have the following lay out title Code well 1 X coordinate Y coordinate Cluster Aquifer code Measured value Time Weight Aquifer n Aquifer n 1 Code well 2 X coordinate Y coordinate Cluster Aquifer code Measured value Time Weight Aquifer n
8. er s Manual In the optimization the best values are calculated for the selected parameter items Both steps also provide indications for improvement of the model setup and fixing of input errors Quick calculations of the confidence of the FLAIRS output sets can be carried out with TriConf and better calculations can be done with TraConf The latter program also provides the option to use a postprocessing program with FLAIRS It has options for capture zone analysis that can be used with TRACE as postprocessing program If we have an existing model that we do not want to calibrate further we can use the TrCalCon module to determine confidences of the model results The steps are e Sensitivity analysis TRICALB sensitivity e Confidence analysis TRICONF or TRACONF In this case the sensitivity analysis is used only to determine the variances of the parameter items and a model check that serve as the basis for the model output confidence calculation with TriConf or TraConf t m Parameter optimal values Parameter correlation amp sensitivity k mm Confidence determination without further calibration 10 3 2 Running the TRCALCON Shell It is strongly recommended to run TRICALB in a Triwaco for Windows environment Within version 2 1 the module for calibration and confidence can be added to the toolbox In this toolbox can be referred to TRCCshll exe which starts the module For an overview of prog
9. f the programscreen the user can indicate whether a sensitivity analysis should be carried out or a optimization 10 Parameter optimisation confidence and sensitivity analysis 11 Royal Haskoning Triwaco User s Manual Gm oy x File TriCalb ton TraCont Options Help HHT F3 Eil Available Parameters Parameters selected n Seb NAHE C Optimization Measurement Add Delete By using the pulldown menus the necessary inputfiles for TRICALB TRICONF and TRACONF can be generated Modules can be runned and results can be viewed TRICALB has to be run before either TRICONF of TRACONF can be used With the Options initial settings of TRCALCON can be changed The initial settings are stored in the file TRCALCON ini An example of this file is given in the appendix 10 3 3 Limitations of TrCalCon The following limitations are important to note e not more than one linked point can be used per river when a river parameter is selected this is due to the fact that the program uses IDs to refer to parameter items and the only ID in a linked point map is the river ID e the TraConf input has to be edited manually when using a post processing program e input for measurements other than heads fluxes interface elevations has to be edited manually e the contents of possible parameter weight files are not checked 10 Parameter optimisation confidence and sensitivity analysis 12 Royal Haskoning Triwaco User s Manual 10
10. model output for which the model has been built confidence analysis The calibration involves varying parameter values and distributions as well as improving of the model concept and fixing of input errors Only a part of this can be automated most remains manual hydrogeologic work Automatic parameter optimization gives the best values based on the given model concept and input plus measurements Moreover statistics are outputted about the sensitivity of the parameter items the correlation and the variances In this context a parameter item is a number of which the optimum value is determined during the automatic parameter optimization A parameter item can be anything that changes a value of a parameter e g e asingle value in a parameter map e a factor that is applied to all node values of the parameter e avalue in a map that is allocated and multiplied with an other parameter to get the required model parameter The last option allows for a multiplication factor that is not constant for all nodes or to optimize one contributing factor e g optimize the permeability and keep the thickness constant in stead of optimizing the resistance of an aquitard The sensitivity of the parameter items is an indication of the amount of information contained in the measurements for the determination of each parameter item A parameter item can be determined accurately if it is very sensitive And the calibration with measurements can no
11. nteger number without a point 5 characters long Heading not in use Items Cluster Code Measured value Time Weight Setname in FLO Setname factor Filename factor Heading not in use Items Cluster Code Measured value Time Weight Setname in FLO Setname factor Filename factor Setname in FLO Setname factor Filename factor Where Item Description Format Heading Name of the flux evaluation retrurns in A10 calib cho not in use Currently not in use Items Number of items to sum MO Cluster Cluster number 15 Code Code forfluxes should be 9001 15 Measured value The value to be compared with summed F10 items m3 day Time Stress input time for transient simulation F10 leave open when not used Weight Weight of the measured value Is used for F10 the overall calibration results Setname Setname in FLO file QRCH QKWn QRIn A10 Setname factor Setname of the factor file A10 Filename factor Filename of the factor file FREE With the cluster number one can assign certain wells to a specific group for statistical analyses choose differently from head clusters 10 Parameter optimisation confidence and sensitivity analysis 4 Royal Haskoning Triwaco User s Manual Code 9001 Calibration using fluxes Weight Use a weight which is small otherwise the weight is equ
12. parameters have been allocated generate a flairs input file menu Calibration Generate Flairs input file and run flairs menu Calibration Run Flairs to check that all files are present and in working order Create a calibration file CHI as shown before in this chapter menu Calibration calibration view create input with the correct format including a weight in columns 61 70 use 1 0 unless you have reason to give a measurement more or less weight Run flairs again and check that all measurements are inside the model and have a valid time so that flairs produces a calculated value in the CHO file menu Calibration calibration view output or log The TRCALCON shell can be started now The following window will appear Gm Mul x File Options Help The programs and settings can be altered if necessary via the Options menu In order to start a model ini file has to be selected This is the model ini file of the calibation dataset e g CAL2 Use the Open option in the file menu or the Ctrl o shortcut to select and load the appropriate model ini file and the window changes In the available parameters screen the modified parameters will appear These parameters can be selected for a sensitivity analysis or optimization When a parameter is moved to the right the calibration mode ICPA can be changed by using the right mouse button In the measurements screen files like calib chi can be added At the lower end o
13. rams and files needed for running TRCALCON see the appendix In the TriShell project a first calibration dataset e g CAL 1 should be present This dataset contains a complete model that runs without errors This is the initial model with all apriori information For the application of TrCalCon a second calibration set should be added e g CAL2 The second calibration set e g CAL2 should be based on the first e g CAL1 The next step is to select the parameters for which items can be calibrated These should be transferred to Modified Parameters which can be done by means 10 Parameter optimisation confidence and sensitivity analysis 10 Royal Haskoning Triwaco User s Manual of the context menu under the secondary mouse button Switch to the modified tab and select for the parameters the appropriate allocator Parameter item to be optimized Allocator SERERE Constant factor _ Any only the allocated ADO file is used Value in parameter map Any external allocator En B Value in parameter map that is The external allocator AllocExp for the current parameter and any multiplied to get current parameter external allocator for the parameter with the map item to be optimized Allocate all parameters Now not only the ADO files are created but for external allocators also the CFG configuration files These files will be used by the TrCalCon programs to adapt the parameter items and allocate the parameters When all
14. rheads or waterbalance The model parameters are adjusted until the differences between calculated and observed heads are within acceptable limits This process is called calibration Triwaco provides several options and tools to calibrate a groundwater model Secondly two packages are available in the standard package for parameter optimisation TrcalCon and MonteCarlo 10 2 Calibration file 10 2 1 Introduction The calibration file is used to compare simulation results with observed heads fluxes and head differences In addition comparison of simulated waterbalances is now also possible To open view or edit the calibration file select Calibration Calibration View Create Input from the pull down menu If no calib chi is present the text editor creates an empty file The input file has a fixed format rameter Calibration Tools Window Help Options Generate Input Bun Simulation View Calibration View Create Input View Output View Log View Map 9 Comparison of observed and calculated heads will be performed automatically if a calibration file exists After selecting Run simulation from the Calibration pull down menu the groundwater simulation program Flairs is started and the calculation results are compared to the observed heads defined in the calibration file The calibration input and output files Calib chi and Calib cho can be viewed or edited selecting Calibration View Create Input or Cali
15. situation to translate this calculated confidence interval into a real confidence interval one has to assess how good the model is The variances and the correlations or covariances are used to determine the confidence of the model output The statistics are based on the assumption that the model describes the essential features This means that the remaining differences between measured and calculated values are so called white noise There may be no systematic differences no trends If the deviations from the measurements are not evenly distributed the model setup has to be improved This together with unreal optimized values are very clear messages for the need of adaptation that calibration without automatic parameter optimization does not give The use of the TrCalCon module can be illustrated with the following steps in a modeling study Setup of initial model Sensitivity analysis TRICALB sensitivity Parameter optimization TRICALB optimization Linear confidence analysis TRICONF Scenario calculations Non linear confidence analysis possibly with post processing TRACONF Use of TrCalCon module in modeling study The sensitivity analysis is the selection of the parameter items that will be optimized A large number of parameter items are considered and the sensitivities and correlations are checked in the TriCalb output 10 Parameter optimisation confidence and sensitivity analysis 9 Royal Haskoning Triwaco Us
16. t improve a value if the parameter item is very insensitive for the measurements It is important to keep in mind that the sensitivities are dependent on the type of measurements The addition of a few flux measurements may have much more impact than adding a lot of head measurements to a set of only head measurements The sensitivity is an important factor in the initial phase of the calibration in which the parameter items are selected that will be optimized automatically with the measurements The correlation between parameter items tells about the possibility to determine these parameter items together The standard example is the transmissivity As this is the determining factor for confined aquifers the permeability and saturated thickness are fully correlated and only one of these two can be determined in the calibration The other has to be inputted on the basis of so called a priori knowledge geology borings etc The variance of the parameter items is an indication of the accuracy with which the parameter item has been determined From the variance a confidence interval can be calculated One has to keep in mind that this confidence interval is based on the assumption that the model setup and all parameter items that are not 10 Parameter optimisation confidence and sensitivity analysis 8 Royal Haskoning Triwaco User s Manual optimized are known perfectly It requires good geohydrologie judgement and insight in the model and real
17. ter Alloceer de parameters 3 Voeg aan de dataset de calib chi toe Deze calibratie file heeft het volgende format Record 1 Title or heading of the calibration file Format A40 Record 2 to N Name X Y coordinate Cluster NR Aquifer NR Observation Time Weight Format A10 F10 2 F10 2 15 15 F10 2 F10 2 F10 2 with An alphanumeric string of n characters In integer value right side aligned of n digits Fn j real fixed number of n places with j decimals 4 Voeg in de dataset de parameters MCVIEWX toe met voor x de laagnummers van de modellagen Resultaten van een MonteCarloberekening kunnen worden ge xporteerd naar de de files MCVIEWX UNG MCVIEWx PAR en MCVIEWx MAP en zijn dan in triwaco te bekijken In MCCAL worden de berekeningen uitgevoerd De triwacoparameters bijv TX1 worden voor de verschillende berekeningen aangemaakt door voor elke berekening de parameters bijv TX1 uit de oorspronkelijk calibratieset CAL te vermenigvuldigen met een factor waarvan de zone opgegeven is in de parameter MCTX1 5 Start het programma MonteCarlo en kies optie new of open 6 Bij optie new dient het TRIWACOproject gezocht te worden prj file Vervolgens worden de dataset MCCAL gezocht en aangeklikt 10 Parameter optimisation confidence and sensitivity analysis 13 Royal Haskoning Triwaco User s Manual 7 Met de optie open wordt een bestaande MonteCarlofile mc ini gezocht 8 Vervolgens kunnen onder de menu op
18. tie MonteCarlo optimalisatie de genoemde stappen achtereenvolgens worden uitgevoerd Bij het genereren van de files wordt de flairsinvoerfile fli overgenomen uit de calibratiedirectory CAL De flairsinvoerfile wordt daarbij zodanig aangepast dat de verwijzingen naar de parameterfiles correct zijn Na wijzigen van de checkpnt invoerfile niet vergeten REFRESH WEEGFACTOREN uit te voeren Als eerste wordt een basisberekening uitgevoerd als onderdeel van de gevoeligheidsanalyse op basis van de gemiddelde waarden van de opgegeven verdelingen Vervolgens wordt een gevoeligheidsanalyse uitgevoerd waarbij steeds 1 parameter aangepast wordt gebruikt wordt de maximale waarde van de opgegeven verdeling Aan de hand van de resultaten van de gevoeligheidsanalyse kan de gebruiker die parameters selecteren waar het model het meest gevoelig voor is Parameters zijn onder de menuoptie INVOER MC PARAMETERS te deactiveren voor de optimalisatie Er wordt dan gerekend met factor 1 bij vermenigvuldigingsfactoren en factor O bij optelfactoren 10 Parameter optimisation confidence and sensitivity analysis 14
19. triwaco groundwater modelling software 10 Parameter optimisation confidence and sensitivity analysis ROYAL HASKONING Chapter 10 Parameter optimisation confidence and sensitivity analysis 10 1 Introduction m venten oet tn oet ot te a oreet diet Hn tiet at 10 3 10 2 Calibration fileira aaao even Seed eet fend Medved ete eee ed Ea ta veel Saeed eves 10 3 102 1 IntFOd CHON DE 10 3 10 2 2 Input data description calib chi a neeeneneeereeneneeeneeneneervennnnereeennneereennnneerennnnn 10 3 10 2 3 Output data description aiea aA aaan ee a aa aa aaa 10 6 10 2 4 Visualisation of calibration results in TriPlot ennn neenennervennnneervennen verven 10 6 10 3 TrCalCon Calibration Sensitivity and Confinence Analysis nnn eenneneereeneneerreenn 10 8 10 3 1 Methodology etterende net rd eto der e mt obe eie er 10 8 10 3 2 Running the TRCALCON Shell sess enne nnne nennen 10 10 10 3 3 lt Limitations Of TrGalG Oni ie ettet etie teo EP De len eue e benede le el ee 10 12 10 4 MonteCarlo Calibration Sensitivity and Confinence Analysis eseeeee 10 13 10 4 T LE oro Lo oo AE 10 13 10 4 2 Running Montecarlo neerstortte nnee den verdoen Pe ee en e Eee eenen deden 10 13 Royal Haskoning Triwaco User s Manual 10 1 Introduction After the first simulation run the first results of will be available In most cases model parameters need to be adjusted to fit the observed groundwate
Download Pdf Manuals
Related Search