Applying Statistical and Artificial Intelligence Methods to Deliver New Meteorological Forecasts for the Next Generation of Ensemble Streamflow Prediction Capabilities

Objective:

In recent years, the NWS Office of Water Prediction (OWP) and River Forecast Centers (RFCs) have highlighted several impactful gaps in existing weather model inputs to operational streamflow forecast models. Specifically, the existing capabilities suffer from biases and difficulties in predicting extremes, and are relatively limited and inflexible with respect to the information they incorporate. Our work aims to address these gaps by developing and testing a flexible, probabilistic weather model processing package for ensemble streamflow forecasting that can leverage input from multiple sources and improve performance for extreme events.

Approach:

During the first year, we have gathered existing and available data and technical resources to help broaden but also streamline our work. This includes leveraging existing testbed resources (at CW3E, CSM, and OWP), and identifying and gathering historical NWP datasets and hindcasts produced through ongoing CIROH projects. We have also begun developing standardized evaluation protocols for probabilistic predictions of forcings for hydrologic models. These verification strategies will be aligned with the larger CIROH testbed program, and will be used to benchmark the performance of baseline predictions across testbed basins. During the second year, we will explore flexible strategies for training our methods with various weather model inputs, and will investigate mechanisms for bolstering the space-time and cross-variable cohesion of the processed outputs. During the second year, we will seek to identify combination(s) of method and predictors that outperform existing methods according to our standardized evaluation approach. This work will center around collaborations with our NWS partners, to ensure that our experiments encompass potential predictors with operational and programmatic forecasting value. Finally, we will produce gridded ensemble outputs across the available period(s), demonstrate the methods in near real-time, and share our software package and documentation with RFCs and OWP.

Impact:

Abstract:

The Center for Western Weather and Water Extremes (CW3E) and Colorado School of Mines (CSM) will expand upon existing collaborative efforts using state-of-the-science ensemble processing and deep learning methods to develop and deliver an enhanced meteorological ensemble forecast processing package that can drive more skillful probabilistic streamflow predictions in the National Weather Service (NWS) Hydrologic Ensemble Forecast Service (HEFS) and Next Generation (NextGen) forecasting platforms using computationally efficient and flexible algorithms. Our work is expected to (1) improve overall streamflow forecasting skill (notably for extreme events, using probabilistic information derived more directly from full ensembles); (2) explore and assess alternative training strategies that enable more flexible use of additional meteorological forecasting datasets; and (3) incorporate more model/method flexibility using skillful and efficient algorithms (which is well suited for changing climates and landscapes). Importantly, our work aims to introduce additional congruence between the operational probabilistic forecasting paradigms of the NWS weather and water enterprises via the inclusion of large, dynamic ensemble weather models in the forcing engine(s) of NWS streamflow ensemble forecast system(s).