Combining hydrological modeling and remote sensing observations to enable data-driven decision making for Devils Lake flood mitigation in a changing climate

The water level of Devils Lake in North Dakota has been rising since 1993, reaching record highs in each of the past three years. Nearly $1 billion have already been spent in mitigating the flooding impacts. If the current wet cycle continues, Devils Lake, a terminal lake currently at 1452 ft, will likely overflow at 1458 ft and cause extensive downstream flooding, with devastating environmental and economic impacts at local, regional, and international levels. We have implemented a distributed rainfall-runoff model, HEC-HMS, to simulate the hydro-dynamics of the lake watershed, and used NASA's remote sensing data, including the TRMM Multi-Satellite Precipitation Analysis (TMPA) and AIRS surface air temperature, to drive the model. The entire watershed with an area of about 10,000 km2 was delineated into six sub-basins using 30 m DEM, with each sub-basin having several hundred thousand hydrological cells. We generated a fine-resolution weather data set, based on a combination of ground observations and remote sensing data, to drive the hydrological simulations. Compared with a very limited number of data series available from five meteorological stations located within the watershed (none belonging to the US Historical Climate Network), NASA data offer a uniform coverage and dense distribution. The satellite and ground observations of precipitation and temperature agreed well with each other. However, if only weather station data were used, the observed runoff was underestimated by at least 30%, regardless of the value of the snow melt-rate coefficient used. The inclusion of NASA data, on the other hand, greatly improved the accuracy of runoff estimates, to within 2% of observations. Better runoff estimates will enable better predictions of water levels. The watershed hydrological model is coupled with a reservoir model, HEC-ResSim. The calibration against the observed lake elevation and monthly evaporation estimates from 2001 to 2004 showed a lake seepage varying between 500 - 1300 cfs. The coupled models can reproduce water level of the lake at sub-feet accuracy, and will be driven by the downscaled CMIP-3 projections of future climate, to provide decision support for mitigation measures in response to the potential flooding.