Regional Streamflow Drought Forecasting In The Colorado River Basin Using Deep Neural Network Models

Streamflow drought is a prominent concern for water resource managers throughout the United States and development of models capable of prediction and early warning of streamflow drought is a high priority for society. Process-based, large-scale (e.g., Conterminous United States [CONUS]) hydrologic models have struggled to achieve reliable streamflow performance in arid regions and for low-flow periods. Deep learning has recently seen broad adoption in streamflow prediction and forecasting applications throughout the world with performance equaling or exceeding that of process-based models. Deep learning models offer an approach to increase the accuracy of streamflow drought predictions and to expand the spatial coverage of river locations with available streamflow drought forecasts.

In this work, we present preliminary results of a deep learning model for the Colorado River Basin (CRB) capable of predicting streamflow drought occurrence at ungaged locations. A long short-term memory (LSTM) neural network model was trained on 40 years (1980-2020) of streamflow data from 425 stream gages within and surrounding the CRB using both static watershed attributes as well as climate and remotely sensed dynamic forcing inputs. Model tests were performed to evaluate model accuracy for now-casting streamflow drought conditions at ungaged locations and for forecasting drought conditions at lead times ranging from 3 days to 30 days. Comparisons of LSTM model performance for predicting drought using static drought thresholds (calculated over all days and years) and variable drought thresholds (unique threshold calculated for each day of the year) identify notable differences in model skill between locations with implications for model design.