High seasonality and interannual climate patterns drive the western U.S.’s water supply and demand variability. While the mean and variance of supply and demand drivers are changing with climate and urbanization, the metrics of reliability, resilience, and vulnerability (RRV) that guide urban water systems (UWS) seasonal management and operations tend to be built on assumptions of stationarity. In this research, we use documented performance of a real-world UWS as a testbed to investigate how RRV metrics – and therefore UWS planning and operations guidance – change in response to demands modeled with and without assumptions of stationarity during dry, average, and wet hydroclimate conditions. The results indicate an assumption of stationary demands leads to large differences between simulated and observed RRV metrics for all supply scenarios, especially in supply-limiting conditions when the peak severity is 129% from the observed. The management implications of relying on stationary demands are severe: if seasonal operational decisions were made on these model results, managers might over-estimate seasonal out-of-district water requests by 50%. In contrast, when using non-stationary demands, one can expect system performance error reduction between 30% to 60% for average and dry climate conditions, respectively, and accurate RRV metrics. Our results further indicate that this UWS is more sensitive to percent changes in per-capita demand relative to percent changes in supply, but because the supply variability is so much greater (158% vs. demand range of 28%), we suggest further work to examine the combined (and coupled) influence of both factors in overall system performance.

Machine-learning algorithms have shown promise for streamflow forecasts, reservoir operations, and scheduling, but have exhibited lower accuracy in predicting extended time horizons of peak storage volume (PSV). Deep learning algorithms exhibited improved inflow forecasting accuracy, but existing research has been mostly limited to real-time operation and short-term planning. We evaluate a new approach based on a hybrid ResCNN-LSTM Encoder-Decoder algorithm, enabling long-term multi-step reservoir forecasts. The proposed approach provides a three-month, weekly averaged prediction of reservoir storage volume (RSV) during the runoff season based on historical snow water equivalent (SWE). The optimal architecture and hyper-parameters for the model are configured through five-fold cross validation resulting in a twelve-layered residual convolutional neural network (ResCNN) as the encoder and a four-layered long short-term memory (LSTM) neural network as the decoder. We evaluate the algorithm using 30 years of RSV and SWE data at the Upper Stillwater Reservoir located in Utah. The most accurate long-term predictions occurred during periods of large runoff (in excess of 28,000 ac-ft). The periods where the model performed the worst were during small runoff and late-season SWE accumulation. We find that the ResCNN-LSTM consistently outperforms three widely used statistical models, with an average PSV absolute percent error of 2.66% for the proposed algorithm compared to SARIMA (14.22%), TBATS (13.82%), and VAR (18.14%).

Sparse observational data in developing regions leads to uncertainty about how hydro-climatic factors influence crop phases and productivity, knowledge of which is essential to mitigating food security threats induced by climate change. In this study, NASA Tropical Rainfall Measuring Mission (TRMM), Global Precipitation Measurement (GPM), and Global Land Data Assimilation System (GLDAS) data products bypass spatiotemporal limitations and drive machine learning algorithms developed to characterize hydro-climate-productivity interactions. Extensive feature engineering processes these products into nearly 4000 metrics, designed to decompose crop growing season hydro-climate conditions. Dimensionality reduction with bidirectional step-wise regression, Multi-Adaptive-Regression-Splines (MARS), and Random Forest algorithms are explored to determine key temporal hydro-climate drivers to agricultural productivity, with each method recognizing unique linear and non-linear predictors. Finally, multi-variate regression, MARS, and Random Forest models are trained on the drivers to predict seasonal crop yield. We apply this hydro-climate-productivity framework to investigate rabi wheat productivity on Pakistan’s Potohar Plateau. Here, we identify six of wheat’s ten phenological phases that display strong hydro-climate responses, with the shooting phase exhibiting sensitivity to precipitation intensity, minimum soil moisture, and sub-zero temperatures. In addition, the plateau’s heterogeneous climate-productivity connections are captured well by the calibrated models, strengthening their application for studying broader climate change impacts. The integration of remote sensing products and machine learning offers an effective framework to bypass in-situ data limitations and decompose climate-crop productivity relationships, thus improving drought onset recognition and food security forecasting.