Flood Forecasting Tools for Ungauged Streams in Alberta: Status and Lessons from the Flood of 2013 John Pomeroy, Xing Fang, Kevin Shook, Tom Brown Centre for Hydrology, University of Saskatchewan, Saskatoon www.usask.ca/hydrology
Why Forecast with Physically Based Models in Alberta? The Western Canadian environment has cold regions processes involving phase change and post-glacial topography that require rigorous application of physical principles for adequate simulation. Non-stationarity due to land use and climate changes adds uncertainty to the operation of empirical models. Extreme events such as droughts and floods require model forecasts for conditions outside of those from which they were derived. Physicallybased models can operate credibly and robustly in such extreme conditions
Cold Regions Hydrological Modelling Platform: CRHM Objected-oriented, modular and flexible platform for assembling hydrological models (Pomeroy et al., 2007, Hydrol. Process.) Modules from about 50 years of hydrology research at University of Saskatchewan and Environment Canada in prairie, mountain, boreal, arctic environments Purpose-built model by user from basin spatial configurations, spatial resolutions, and dominant hydrological processes in the basin. Hydrological Response Units (HRUs) based simulation Landscape units with characteristic hydrological processes Single parameter set Number of nature depending on variability of basin attributes and level of physical complexity chosen for model Sub-basins structure a series of representative basins with same physical process modules and HRUs but varying parameter values
CRHM Features Interpolation of weather data over a basin. Blowing snow redistribution and sublimation Forest canopy interception of snow and rain and subsequent sublimation or evaporation Radiation calculation to slopes and under forest canopies Energy balance snowmelt Infiltration to frozen and unfrozen soils Actual evapotranspiration coupled to soil moisture dynamics Depressional storage dynamics and routing Hydrological routing from basin characteristics Predict multiple endpoints: soil moisture, snowpack, streamflow, wetland storage, evaporative loss, etc Possible to obtain good results with no calibration
Marmot Creek Research Basin x x x x x x x
Marmot Creek Basin Hydrological Model
Hillslope Module
HRU Delineation and Model Structure
Forest Snow Dynamics Simulations Forest Clearing
North Face Alpine Snow Ridgetop Dynamics Simulations Upper South Face Lower South Face Snow redistribution from north face and ridgetop to south face and larch forest uncalibrated Larch Forest
Model Tests: Soil Moisture 2006-2011 Level Forest Site Uncalibrated
No observed discharge during 2013 due to damaged gauges during flood. NSE MB NRMSD Cabin 0.17-0.001 0.84 Middle 0.32-0.1 0.71 Twin 0.1-0.06 0.86 Marmot 0.47-0.01 0.7 Daily Discharge (m^3/s) 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 Uncalibrated Streamflow Test Daily Discharge (m^3/s) 0.45 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Marmot Creek Cabin Creek Simulation Observation Middle Creek Twin Creek
The Flood of June 2013
Marmot Creek Multi-elevation Precipitation Hay Meadow Upper Clearing Fisera Ridge
Hourly Rainfall Rate Marmot Creek 14 Rainfall Rate mm/hr 12 10 8 6 4 2 0 6/18/2013 6/19/2013 6/20/2013 6/21/2013 6/22/2013 6/23/2013 6/24/2013 6/25/2013 Hay Meadow (1436 m), Marmot Creek
Fisera Ridge Snowmelt
Clear-cut Soil Moisture Storage Upper Clearing, Marmot Creek
Flood Scenarios same time, different years
Flood Scenarios same time, different years
Scenario Flood Discharge over Last Eight Years Flood discharge: total discharge as result of the flood meteorology of 17-24 June 2013 put as scenario in seven earlier years. Antecedent precipitation: total precipitation from beginning of hydrological year to onset of flood, i.e. 1 October to 16 June Antecedent air temperature: average temperature in June to onset of flood, i.e. 1 June to 16 June
Application: Operational Forecasting of Ungauged Snowmelt Runoff Smoky River Basin is 46% ungauged Need to simulate spring streamflow from the ungauged basin area (23,769 km 2 ) in order to forecast Smoky River contribution to the Peace River Run model on a daily basis during flood forecast period update ungauged flows Use daily updates of meteorological model forecast data to run for the future Route ungauged with gauged flows for forecast
Smoky River Basin: 51,839 km2
Challenge: Reliable Meteorological Observations and Forecasts
Interpolate, Predict, Forecast GEM-WISKI-CRHM North America Ensemble Forecast System: 21 ensemble forecasts, 16 days in the future Options to adjust forecast and run scenarios M
Land Cover and Soil Parameters
Sub-basins for Modelling Modelled all ungauged and gauged basins without real time hydrometric stations Sub-basins grouped into types based on ecoregion Real time gauged basins are estimated from gauge measurements and routed outside of CRHM using SSARR
Module Structure within each HRU
HRU Classification of Smoky Basin HRU classification and interpretation of land cover, topography, drainage, soils to determine parameters was guided by sub-basin type which depended on ecoregion
Parameterisation Parameters measured where possible DEM, satellite-derived vegetation and soils maps Site visits for river roughness, interception Many parameters were brought in from research basin observations Hydraulic conductivity Soil depth Albedo, aerodynamic roughness One parameter calibrated from streamflow (sub-surface HRU travel time) for 3 HRUs
Routing between HRUs Routing sequence depends on sub-basin type (ecoregion)
Routing between Sub-basins Muskingum Routing used for river routing between sub-basins
Basin-scale Ungauged Prediction (a) (b) Great uncertainties in estimating local inflows for comparison to model results
(a) Basin-scale Prediction Evaluation (b) Predicted flows, Nash-Sutcliff Statistic: 0.41 (Little Smoky) and 0.87 (Smoky)
Predicted Spring Discharge Volume 15 March-31 May
Predicted Spring Peak Discharge
Conclusions Uncertainty in hydrological prediction is gradually being reduced after decades of research in Western Canada. Incorporating snow redistribution, snowmelt, infiltration to frozen soils and fill and spill runoff generation processes in models can provide the basis for predictability in the Prairie Provinces. Physically based predictive models can interface with advanced weather forecasting models and be incorporated in flood forecasting procedures.