Machine Learning and ocean variables for improved predictions of water availability in the US

Salient Predictions Inc. proposes to develop new technology for accurate subseasonal to seasonal (S2S) water availability forecasts so that communities may manage water resources in preparation for hazardous events like floods and droughts. Current hydrological models face challenges related to data inputs, as climate change has been presenting weather patterns atypical from historical data. Without quality data inputs, the accuracy of the output from hydrological models suffers. To overcome these challenges, Salient Predictions proposes to use improved S2S forecasts as the weather input to Variable Infiltration Capacity (VIC) hydrological forecasts, thus providing improved water availability forecasts spanning from 2- to 52-weeks into the future. Salient’s base S2S forecast technology uses machine learning and various oceanic, atmospheric, and land-based variables to make improved predictions of weather up to a year in advance. This core technology evolved from decades of research at Woods Hole Oceanographic Institution and the Massachusetts Institute of Technology. Phase I will address challenges related to 1) the spatio-temporal gap between S2S forecast variables and those required to capture more heterogeneous hydrological processes, and 2) the need to include climate change projections into S2S forecasting, as many recent hydrological events are being at least partially attributed to climate change.