Update 2.2: Uncertainty in Projected Flow Simulations Created by: Scott Pokorny In Association with: University of Manitoba, Manitoba Hydro
Presentation Outline Review of project goals Hec-HMS model updates Dew point temperature estimation Forcing data standardization updates
Review of project goals To assess and quantify the uncertainty in projected flow simulations 3 models for the Lower Nelson River Basin (LNRB) were selected Hec-HMS, Watflood, and VIC Uncertainty is to be compared between models to assess how the input, model structure, parameter choice and output uncertainties relate Forcing data inputs are to be standardized to focus the uncertainty analysis on the models
Hec-HMS Updates Model updated from version 3.5 to version 4.2.1 for more evaporation method options Ungauged sub basins spatially close to each other have been grouped Total number of junctions and sub basins reduced
Hec-HMS Updates 4.2.1 simple vs 4.2.1 4.2.1 simple vs 3.5 4.2.1 vs 3.5 4.2.1 simple vs 4.2.1 4.2.1 simple vs 3.5 4.2.1 vs 3.5 Mean abs difference (m3/s) = 8.100 24.654 22.312 Max abs difference (m3/s) = 66.200 534.200 553.300 Correlation 1.000 0.998 The version 3.5 takes ~5 minutes to run The version 4.2.1 takes ~1 minute to run The version 4.2.1 simplified takes ~40 seconds to run Roughly 70 parameters are expected to be included in the GLUE runs after updates are finished
Dew Point Temperature Estimation Before the update, Hec-HMS handled evaporation with monthly average evaporation values The Priestly Taylor evaporation method will now be used but requires a dew point temperature time series as input This will need to be estimated for the climate projections from the available GCM data Hubbard, K. G., Mahmood, R., & Carlson, C. (2003). Estimating daily dew point temperature for the northern Great Plains using maximum and minimum temperature. Agronomy Journal, 95(2), 323-328.
Dew Point Temperature Estimation Coefficients have been estimated with regression at 13 stations nearby the LNRB This assumes that the regression constants will remain constant into the future To evaluate this assumption, coefficients will be regressed over a moving 10 year window where data is available There is less data available for regression when precipitation data is required Therefore, method 3 regression will be done twice, once with all available temperature data and once only using days where precipitation is also available
Dew Point Temperature Estimation
Dew Point Temperature Estimation Stn Name: FLIN_FLON Years 1994-2014 1994-2003 1995-2004 1996-2005 1997-2006 1998-2007 1999-2008 2000-2009 2001-2010 2002-2011 2003-2012 2004-2013 2005-2014 Temp only available data (%) 79.36 77.93 81.22 79.85 78.86 75.14 77.01 75.17 75.60 77.08 78.84 78.89 80.34 Precip required 44.60 57.04 58.58 54.75 54.57 50.55 48.48 41.94 35.68 34.06 34.16 33.34 29.98
Dew Point Temperature Estimation Method E MAE d-index r2 RMSE Lit Values 0.958217 2.212283 0.98943 2.860365 Lit Values (Pr clip) 0.951529 2.189827 0.987615 2.846792 Method 3 Values 0.967972 1.954186 0.991795 2.507426 Method 4 Values 0.96424 1.885736 0.990759 2.435094 Method 3 Values (Pr clip) 0.962505 1.93794 0.990298 2.497744
Forcing Data Standardization Updates
Forcing Data Standardization Updates Watch ERA Interim Watch ERA Precipitation Kendall Corr Spearman Corr Mean Yearly Sum Abs Diff max 0.45 0.53 97.87 mean 0.27 0.31 38.81 min 0.16 0.18 5.16 stdev 0.05 0.06 15.77
Forcing Data Standardization Updates North American Regional Reanalysis (NARR) NARR Precipitation Kendall Corr Spearman Corr Mean Yearly Sum Abs Diff max 0.49 0.62 245.78 mean 0.36 0.46 80.68 min 0.24 0.31 33.49 stdev 0.05 0.06 38.58
Forcing Data Standardization Updates Watflood Output Watflood Precipitation Kendall Corr Spearman Corr Mean Yearly Sum Abs Diff PBIAS (%) max 1.00 107.73 15.09 mean 0.67 0.77 29.95 -0.80 min 0.51 0.60 0.00 -34.90 stdev 0.08 0.07 16.20 6.81
Forcing Data Standardization Updates The most appealing data set is ANUSPLIN Already gridded at 10km resolution Gap filled data at station locations Watflood data requires a radius of influence and smoothing distance High performing values for these variables will be based station location and spacing Calibration would be needed to find the most acceptable values
Questions?