Preferences among Hydrologic Models for Studies involving Climate Change

Transcription

1 Preferences among Hydrologic Models for Studies involving Climate Change Levi Brekke, Reclamation Technical Service Center (Denver, CO) California Water and Environmental Modeling Forum, Annual Meeting Pacific Grove, CA, 23 February 2010

2 Questions Can we identify preferred models for simulating runoff in a climate that is different from the calibration climate? Can we attribute any preferences to structure aspects or application choices?

3 Acknowledgments Co-Investigators USGS (Lauren Hay, Greg McCabe) NWS CBRFC (Edward Clark) Reclamation (Nancy Parker, Tom Pruitt) Support Reclamation Research and Development Office Region Offices Mid-Pacific, Upper Colorado, Lower Colorado, Great Plains USGS National Research Program NWS Colorado Basin River Forecast Center

4 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

5 Models VIC: Variable Infiltration Capacity model Liang et al ( PRMS: Precipitation Runoff Modeling System Leavesley et al ( RDHM: NWS Hyd. Lab Research Distributed Hydrologic Model (gridded SacSMA/Snow17) Anderson et al. 1973, Burnash et al ( TMWB: Thornthwaite Monthly Water Balance McCabe and Markstrom 2007 ( *** Model structures vary, which leads to preferences among available basin-application datasets (weather, land cover, soils)

6 Model & Application Differences PRMS VIC RDHM TMWB Structural Differences Balances? water & energy water & energy water Water Vegetation? Up to 4 classes Up to 12 classes Up to 2 classes Implicit Soils? 3 stores per element 3 stores per element 6 stores per element 2 stores per element Sub- element portrayal? no Yes (elevation bands, veg fracs) no no Typical Resolutions? Weather? Vegetation? Typical Basin Application: Resolutions, Input Datasets daily, topo units <12km 2 custom stations distributed data NLCD 2001 single, prevalent class per cell (4 class options) daily, gridded (~50 to ~6km) Maurer2002, H&L2005, others Univ of MD 1km GLCD veg fracs per cell, 12 classes 6-hourly, gridded (~4km) NWS HRAP gridded (~4km) NLCD 2001 effective forest cover parameter Soils? STATSGO classes STATSGO classes STATSGO stores sizes and properties monthly, gridded (~12 to 4km) PRISM, or VIC input datasets No specification of vegetation controls on runoff STATSGO or Dunne & Wilmott soil moisture hold. cap.

7 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

8 Colorado River Basin Three Basins, varying characteristics East at Almont Animas at Durango Salt at Chrysotile

9 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

10 Resolutions and Weather Animas East Salt Spatial Resolution: HRAP grid (NWS CBRFC, ~4km) Time Step of Application: Daily: PRMS, VIC, RDHM Monthly: TMWB Gridded Weather Development: Daily Prcp, Tmin, Tmax from Station Data (T only, P only, T and P), two sources (SNOTEL = circle, COOP = triangle)

11 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

12 Calibration/Validation Design Classify input weather years (21 yrs, or WY ) by climate type Philosophy: target Cool/Wet to Warm/Dry transition; yearsets have >= 7 years Result: 3 equal size yearsets: Cool/Wet (C/W), Middle (M), Warm/Dry (W/D) Simulation Experiments: #1: Calibrate on C/W, Validate on W/D or {W/D & M} #2: Calibrate on C/W & M, Validate on W/D #3: Calibrate on W/D & M, Validate on C/W #4: Calibrate on W/D, Validate on C/W or {C/W & M}

13 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

14 Calibration Scheme Four Objectives (indicating Sim reproduction of Obs) Annual Flows, minimize RMSE Monthly Flows, minimize RMSE Period Monthly Mean Flows, minimize (1 correlation) Annual 2-day Maximum Flow, Minimize RMSE Priorities Weighted the objectives for ~equal emphasis Calibration Tool (parameter search) USGS implementation of UA-SCE (LUCA) Identify parameters to adjust, iterative simulation and parameter adjustment until stopping criteria are met number & types of estimated parameters varied by model

15 Approach 1. Consider four surface water hydrology models 2. Apply models to address questions a) choose three basins, range of climate and land characteristics b) apply models with consistent resolution and weather c) perform calibration-validation across contrasting climates d) calibrate models to reproduce several runoff aspects 3. Evaluate results for trends by basin type, runoff aspect Interpret results for model preferences

16 East Calibration Years (Observed = black, PRMS = dark blue, VIC = green, RDHM = red, TMWB = cyan)

17 Comparing Error Focused on Runoff Ideally would assess all water balance terms observations for other terms are limited Define Relative Error (RE) RE yearset, model, basin = 100 * RMSE / mean value Note: for Period-Monthly Mean (PMM), we calibrate to minimize (1-correlation of Sim & Obs); here we assess RE of RMSE between Sim & Obs

18 East

19 Comparing RE results (1) no clear trends, (2) TMWB did surprisingly well Experiment #1 Experiment #4

20 Uncertainties Weather: sparse station density, uncertainty? Vegetation and Soils: inconsistent source data and relation to classes and attributes connecting to model structure (e.g., ET, infiltration) how does this control results? Calibration: choice of objectives and weights? Model calibration arguably biased to reproduce wet conditions given objectives Maybe should have edited #3 to be minimize RMSE of Period Monthly Means rather than minimize (1-correlation), latter emphasizes phase matching but not amplitude elimination of 2dayMax objective? (equal playing field with TMWB) weights? Numerous others

21 Questions Revisited Can we identify preferred models for simulating runoff in a climate that is different from the calibration climate? Maybe but not based on results from this study. Perhaps an alternative experimental design could offer better insight Can we attribute any preferences to structure aspects or application choices? Hold that thought until first question is resolved.

22 Status Analyses is complete (i.e. we re broke) Preparing journal article for peer-review Sharing results and approach with others who may wish to assess other surface water hydrology models (e.g., WEAP, HEC-HMS)

23 Questions? Levi Brekke

24 Extras

25 Characteristics Overview

26 Example Calibration: PRMS, Experiment #1, Animas

27 Animas (Observed = black, PRMS = dark blue, VIC = green, RDHM = red, TMWB = cyan)

28 Salt (Observed = black, PRMS = dark blue, VIC = green, RDHM = red, TMWB = cyan)

29 Animas

30 Salt