Floods are a significant threat to communities around the world and require substantial resources and infrastructure to predict. Limited local resources in developing nations make it difficult to build and maintain dense sensor networks like those present in the United States, creating a large disparity in flood prediction across borders. To address this disparity, I operated the Iowa Flood Center Top Layer model to predict floods in Puerto Rico without relying on in-situ data measurements. Instead, all model forcing was provided by satellite remote sensing datasets that offer near-global coverage.
I used three datasets gathered via satellite remote sensing to build and operate watershed streamflow models: elevation data obtained by the Space Shuttle Endeavour through the Shuttle Radar Topography Mission (SRTM), rainfall estimates gathered by a constellation of satellites through the Global Precipitation Measurement Mission (GPM), and evapotranspiration rate estimates collected by Moderate Resolution Imaging Spectroradiometer (MODIS) sensors aboard the Aqua and Terra satellites. While these satellite remote sensing datasets make observations of nearly the entire world, their spatiotemporal resolution is coarse compared to conventional on-the-ground measurements.
Hydrologic models were assembled for 75 basins upstream of streamflow gages monitored by the United States Geologic Survey (USGS). Model simulations were compared to real-time measurements at these gages. Continuous simulations spanning 58 months achieve poor Nash Sutcliffe Efficiency and Klinge Gupta Efficiency of -112.0 and -0.5, respectively. The sources of error that influence model performance were investigated, underlining some limitations of relying solely on satellite data for operational flood prediction efforts.