4 posts tagged with "NRDS"

Technological advances are evolving water prediction capabilities at a ludicrous pace. From revolutionary machine learning algorithms to dramatic advances in computational hardware, the potential for making accurate hydrologic predictions has never been higher. To meet this new potential, the hydrologic community continuously generates models and approaches based on cutting edge research that could potentially benefit operational systems. However, many of these innovations lack a path to operational deployment.

The NextGen Research Datastream (NRDS) provides a mechanism by which these ideas can be refined and make their way into operations.

Developed by Lynker and the Alabama Water Institute (a Cooperative Institute for Research to Operations in Hydrology partnership), the NRDS facilitates the actualization a research idea from the community in a scalable and deployable numerical prediction system. To evaluate each of these modeling concepts, NRDS deploys prototype models to generate a continuous “datastream”. These outputs can then be evaluated and made more accurate. This cycle of streamlined deployment and iterative design lets these prototypes mature into a product that can be picked up by an operational forecasting team.

To enable this process to be done rapidly and smoothly, the entire system is designed with reproducibility and iterative improvement as core principles. The NRDS is an automated numerical prediction system generating regular stream flow forecasts that uses the NextGen Water Resources Modeling Framework (NextGen) as the core modeling engine and NextGen In A Box (NGIAB) as the simulation environment. This system generates forecasts across the contiguous United States (CONUS) on CIROH's operational cyberinfrastructure backbone: the research-to-operations (R2O) Hybrid Cloud (R2OHC) platform, with deployment on the AWS cloud. What makes the NRDS exciting is that the entire system is open-sourced, reproducible, publicly browsable, and potentially editable by anyone in the hydrologic community.

In October 2025, Penn State's Multi-scale Hydrology, Processes and Intelligence group (MHPI), led by Dr. Chaopeng Shen, and the Alabama Water Institute (AWI), led by Steve Burian and Arpita Patel, achieved a milestone R2O effort: the preliminary integration of δHBV 2.0 [4] -- a daily-scale, high-resolution, distributed differentiable model -- into a NextGen ecosystem. This resulted in the first adoption of a differentiable model into NextGen In A Box (NGIAB) [2] and provided an opportunity for CIROH researchers to fine-tune the δHBV 2.0 architecture for NextGen operation.

Having proven viability for daily timescale predictions on high-resolution river networks [4], MHPI researchers recently adapted δHBV 2.0 into a multi-timescale architecture designed to parameterize HBV and simulate streamflow at hourly intervals, at scale, across the NextGen HydroFabric. This new model, δHBV 2.0 MTS (Multi-TimeScale) [5], is a fusion of a daily and hourly δHBV 2.0 model designed to efficiently handle ML training with high geospatial and temporal complexity. (See MTS Architecture for more details about this construction.)

With δHBV 2.0 MTS maintaining similar forecasting skill compared to its daily-scale counterpart [5], Penn State and AWI were once again reunited in a joint effort to embed hourly scale differentiable modeling within AWI's operational ecosystem as a demonstration of model viability and to facilitate open access to its runtime.

Differentiable Models​

δHBV 2.0 and δHBV 2.0 MTS differentiable model constructions are briefly outlined here to contextualize the development efforts. For further detail, see each model's respective citation. At their core, differentiable models embed traditional process-based equations (here, the HBV rainfall-runoff model) inside a machine learning training loop. Because these models are designed to be differentiable (e.g., in PyTorch), gradients flow end-to-end from the loss function back through the physical equations and into the neural networks that supply their parameters. This lets the model learn optimal parameterizations directly from observed data while still obeying mass-balance and storage constraints encoded in HBV -- combining interpretability and physical consistency of process-based hydrology with the flexibility of deep learning.

Last week, at the invitation and expert coordination of James Halgren, teams from RTI International (Sam Lamont and Matt Denno) and the University of Calgary (Darri Eythorsson, Cyril Thebault, and Martyn Clark) met at AWI for an intensive working session focused on weaving recent CIROH research into AWI’s fork of the NOAA Office of Water Prediction (OWP) Next Generation Water Resources Modeling Framework (nicknamed “NextGen”). James took the lead in developing the agenda, lining up the right scientific and technical expertise and ensuring that the week targeted the most critical software integration challenges. Throughout the visit, the RTI and UCalgary teams collaborated closely with AWI software engineers Quinn Lee, Josh Cunningham, hydrologic scientist Sifan A. Koriche, and James himself. The days were filled with whiteboards, deep technical conversations, and strategic planning around the future of NextGen water prediction. This recap captures the key themes and the momentum that carried through the week.

Almost from the start, 2025 has been a banner year in hydrologic modeling, with advancements in capabilities on both sides of the aisle of CIROH's research-to-operations (R2O) pipeline.

• From the research skunkworks, Penn State's MHPI group, led by Dr. Chaopeng Shen introduced a new generation of distributed, differentiable hydrologic models spearheaded by δHBV 2.0. Capable of high-resolution, continental-scale streamflow forecasting across the CONUS Hydrofabric, δHBV 2.0 fuses process-based modeling and machine learning to enable efficient parameter calibration and interpretable predictions at scale -- with demonstrated viability as a National Water Model 3.0 successor.