Preparation of Soil Database for Swat
2018 International SWAT Conference Brussels, Belgium
Understanding climate model uncertainty in streamflow projection
Vinod Chilkoti, Tirupati Bolisetti, Ram Balachandar Department of Civil and Environmental Engineering University of Windsor, Windsor, Ontario, Canada
Sep 19, 2018 Introduction 2
• Changing climate poses a crucial threat to the seasonal distribution of water availability
• Hydrological models forced with climate model data to project the future streamflow
Sep 19, 2018 2018 SWAT Conference Climate Impact Assessment – Modeling Chain 3 Model inputs Climate model •Climate data projections Climate •Topography Model Forcing •Soil Bias •Landuse Corrections
Hydrological Calibration Validated Model and model Development Validation
Climate change Climate Hydrological Model Impacts change assessment impact
Sep 19, 2018 2018 SWAT Conference Challenges 4 Suit of uncertainties inherent in the modeling chain is a major cause of concern • Climate models . GCM (Graham et al. 2007 ) . RCM (Bosshard et al. 2013, Chen et al. 2011a) • DownscalingNo consensus method over (Chen the et al.cause(s) 2011b) of uncertainty • ImportantHydrologic toModel understand the sources of uncertainty . Input (Renard et al. 2011) . Model Structure (Ludwig et al. 2009, Poulin et al. 2011) . Model parameters (Wilby 2005, Bastola et al. 2011) . Observed (output) data (mostly considered sacred)
Sep 19, 2018 2018 SWAT Conference Objectives 5 Major objectives of this research are to investigate the
• Effects of climate model uncertainty on streamflow projection
• Role of climate model ensemble members in the projection uncertainty
Sep 19, 2018 2018 SWAT Conference Study Area: Magpie River Watershed 6
CANADA
• Catchment area 2039 km2 • Length of river – 190 km
Sep 19, 2018 2018 SWAT Conference SWAT Model Setup 7
Topography (DEM) Landuse Forest – 70% Range land – 18% Water – 11% Delineated subwatersheds Urban – 01%
Sep 19, 2018 2018 SWAT Conference SWAT Model - Input 8 • Climate Data . Long term data available only at one station (Wawa A) . Gridded climate data is used (Ref: Hutchinson et al., Hopkinson et al.,)
• Flow data at Wawa is used for calibration and validation
Magpie River
Sep 19, 2018 2018 SWAT Conference SWAT Model Calibration 9 • 13 model parameters are calibrated . 4- surface water parameters (CN2, CH_K2,SOL_AWC & ESCO) . 3-ground water parameters (RCHRG_DP, GW_REVAP, ALPHA_BF) . 6-snow parameters (SFTMP, SMTMP, SMFMX, SMFMN, TIMP & SNOCOVMX) • Model Calibration . Calibrated SWAT model using multi-objective optimization framework . Borg algorithm • Falls under class of evolutionary algorithms • relatively newer algorithm
Sep 19, 2018 2018 SWAT Conference SWAT Model Calibration 10
Objective Functions n (O − S )2 ∑ i i Borg Algorithm Minimize(1− NSE) = Min i=1 1. NSE n 2 Parameter ∑(Oi − O) generation i=1 2. RSRLow SWAT_Ed it m Parameter ∑ Oi − Si Minimize(FDC ) = Min i=1 updating in SWAT 3. FDCsignature sign m ∑Oi SWAT Model i=1
Model Run NSE : Nash Sutcliffe Efficiency Objective Function RSR : Ratio of root mean square error to Statistical objectives evaluation standard deviation of observed data Hydrological FDCsign : Flow duration curve bias Signature objectives
Sep 19, 2018 2018 SWAT Conference Results: Model Calibration and Validation 11 Calibration Validation
simulated observed flow
Daily simulation Daily simulation
Statistic Calibration Validation
NSE 0.72 0.81 pBIAS 6.7% 2.7% KGE 0.75 0.83 p-Factor 0.61 0.73
Sep 19, 2018 2018 SWAT Conference Climate Change Projections 12 • Regional Climate Model (RCM) data is used • Data is extracted from CORDEX (Coordinated Regional Downscaling Experiment) • CORDEX – North America (NAM) Grid
Source: http://www.cordex.org/
Sep 19, 2018 2018 SWAT Conference Climate Change Projections 13 • Climate projection for two scenario periods . Mid-century : 2041 - 2070 . End-century : 2071 - 2100 • Multi-model climate ensemble for rcp4.5 scenario used Regional Climate Model Driving General Circulation Model Model (RCM) (GCM) No Modeling Modeling RCM GCM Agency* Agency* M1 CanRCM4 CCCma CanESM2 CCCma M2 RCA4 SMHI CanESM2 CCCma M3 CRCM5 UQAM CanESM2 CCCma M4 RCA4 SMHI EC-EARTH ICHEC M5 HIRHAM5 DMI EC-EARTH ICHEC M6 CRCM5 UQAM MPI-ESM-LR MPI-M
* CCCma- Canadian Center for Climate Modeling and Analysis ICHEC – Irish Center for High End Computing SMHI – Swedish Meteorological and Hydrological Institute UQAM-Université du Québec à Montréal DMI – Danish Meteorological Institute MPI –Max Planck Institute of Meteorology
Sep 19, 2018 2018 SWAT Conference Climate Change Projections 14 • Climate model data is forced into calibrated hydrological model • Large uncertainty is found in streamflow projection
Average Baseline Projected
Large uncertainty
Sep 19, 2018 2018 SWAT Conference Climate Change Projections 15 • Investigating the cause of streamflow uncertainty
Average Baseline Projected
• Models projecting higher value are always M1, M2 and M3 • Climate model ensemble is grouped into two, based on the driving GCM (boundary conditions)
Sep 19, 2018 2018 SWAT Conference Climate Model Grouping 16 • Multi-model climate ensemble for rcp4.5 scenario Regional Climate Driving General Circulation Model Model (RCM) Model (GCM) No Modeling Modeling RCM GCM Agency Agency M1 CanRCM4 CCCma CanESM2 CCCma Group-1 M2 RCA4 SMHI CanESM2 CCCma M3 CRCM5 UQAM CanESM2 CCCma M4 RCA4 SMHI EC-EARTH ICHEC Group-2 M5 HIRHAM5 DMI EC-EARTH ICHEC M6 CRCM5 UQAM MPI-ESM-LR MPI-M
Sep 19, 2018 2018 SWAT Conference Climate Model Grouping - Precipitation 17 • Precipitation and temperature data are key inputs for model simulation
Group-1 Models Group-2 Models Baseline
• Precipitation projections by the two model groups are not distinct
Sep 19, 2018 2018 SWAT Conference Climate Model Grouping - Temperatures 18 • Temperature projection by different model groups
Minimum Temperature Group-1 Models Maximum Temperature Group-2 Models Baseline Differences in the projections by the two model groups are identifiable
Sep 19, 2018 2018 SWAT Conference Projected Streamflow Comparison 19 • Group-1 model projects higher winter and spring temperature compared to Group-2 • This causes higher snow melt and occurring earlier
Group-1 Group-2
Comparison of projected streamflow by Group-1 models and Group-2 Models Sep 19, 2018 2018 SWAT Conference Projected Streamflow Comparison 20 • Mann-Whitney test on seasonal streamflow projection
p-value of Mann-Whitney test between projections by the two model groups
Winter Spring Summer Autumn Mid century 2.2 x 10-16 4.2 x 10-4 0.92 9.8 x 10-7 End century 2.2 x 10-16 2.2 x 10-16 0.34 2.4 x 10-3
• Results of the two groups are statistically similar only for summer
Sep 19, 2018 2018 SWAT Conference Projected Streamflow Comparison 21 • Change in streamflow w.r.t the baseline is thus variable for the two groups
Baseline Projected
Projection by Group-1 models Projection by Group-2 models
Sep 19, 2018 2018 SWAT Conference Conclusions 22 • Reasons for high uncertainty due to climate models has been investigated • Uncertainty is prevalent in the scenario streamflow projection • Uncertainty due to climate model ensemble has been highlighted • Driving GCM is the major cause of uncertainty • The presented idea needs to be affirmed using more number of climate models in other watersheds
Sep 19, 2018 2018 SWAT Conference Acknowledgments 23 Partial funding support by the following is gratefully acknowledged
• National Sciences and Engineering Research Council of Canada
• University of Windsor
• Ontario Graduate Scholarship
Sep 19, 2018 2018 SWAT Conference 24
Backup Slides
Sep 19, 2018 2018 SWAT Conference Results: Model Calibration 25 • Borg-SWAT optimization • Calibration period : 2003 to 2008 • Validation period : 2009 to 2012 • 22 optimal parameter sets are obtained • Parameters are equally likely simulator of the model
Pareto optimal front
Sep 19, 2018 2018 SWAT Conference Results: Model Calibration 26 • Flow Duration Curve (FDC)
simulated observed flow
Volumetric Efficiency Calibration Validation Flow Exceedance (2003-2008) (2009-2012) Segment (%) Monthly Daily Monthly Daily Peak 0 - 1 0.95 0.7 0.95 0.67 High 1 – 20 0.69 0.6 0.69 0.57 Mid 20 – 70 0.65 0.59 0.62 0.56 Low 70 - 100 0.5 0.3 0.51 0.41
Sep 19, 2018 2018 SWAT Conference Results: Model Uncertainty 27
Observed flow depth Simulated (Pareto optimal)
SurQ - Surface flow GwQ - Ground water flow ET - Evapotrasnpiration WY - Water yield
Sep 19, 2018 2018 SWAT Conference Climate Change Projection 28 • Precipitation
Average Baseline Projected End century scenario
Baseline : 1976 - 2005 Mid-century : 2041 - 2070 End-century : 2071 - 2100
Sep 19, 2018 2018 SWAT Conference Climate Change Projection 29 • Temperature Baseline : 1976 - 2005 Mid-century : 2041 - 2070 End-century : 2071 - 2100
Ensemble minimum temperature : End century Average change in minimum Temperature
Average seasonal change : Mid-century Average seasonal change : End-century Sep 19, 2018 2018 SWAT Conference