NextGen and the End of the Monolithic National Water Model

NOAA's National Water Model has produced operational continental-scale hydrologic forecasts since August 2016. The current version, v3.0, has been operational since summer 2023 per NWS Service Change Notice 23-76. It is built on the WRF-Hydro architecture, runs a single physics formulation across the contiguous United States, and uses one calibrated parameter set tuned for that single formulation.

That architecture is being retired.

NOAA's Office of Water Prediction has built a new modeling infrastructure called the Next Generation Water Resources Modeling Framework, or NextGen, to replace the monolithic kernel. The next major operational NWM release, v4, will be the first to run on it. Practitioners who use NWM forecasts as a project input, whether as forcing for a downstream hydraulic model, as a boundary condition, or as a reference dataset for ungauged-basin work, should expect the operational forecasts to change in character as v4 enters operations.

This post is the institutional read on what is changing, why it matters, and what to plan for.

NextGen is not itself a hydrologic model. It is a modeling infrastructure designed to run other modeling formulations in a coordinated, operational way.

Three architectural choices define it.

First, it is model-agnostic. NextGen does not commit to a particular set of physics equations. Different formulations, including conceptual rainfall-runoff models, process-based physics, and machine-learning models, can be plugged in to handle different processes or different regions.

Second, it is BMI-based. The Basic Model Interface (BMI v2.0) is a published standard, maintained by the Community Surface Dynamics Modeling System (CSDMS), that lets independently-developed models be coupled into a common framework without re-coding each one. In practical terms: if a model exposes a BMI interface, NextGen can call it.

Third, it is hydrofabric-aligned. NextGen runs on a curated computational mesh distinct from the medium-resolution NHDPlus that anchored most modeling work over the last decade. This is a more significant change than it sounds. Most legacy NWM-adjacent project work used NHDPlus geometry to align catchments, reaches, and gauge IDs. The NextGen hydrofabric uses different aggregation rules and different reach definitions.

The implication of this architecture is that the operational federal hydrologic model is no longer required to use the same formulation everywhere. The same NWM run can, in principle, use one snow module in the Sierras and a different one in the Northeast, one infiltration formulation in karst and another in glacial till, one routing approach on a small headwater and another on a regulated mainstem. The framework is designed to support per-region or per-process formulation choice rather than forcing a single set of parameters to work everywhere.

What regions will actually run what model in the first operational v4 build has not been publicly finalized as of this writing.

The motivation for NextGen is empirical, not architectural fashion.

Peer-reviewed evaluations of NWM v2 and v3 against continental gauge networks have shown that the model's performance is heterogeneous. Johnson et al. (2023), in a comprehensive evaluation of NWM streamflow, explicitly framed the result as a call for heterogeneous formulations and diagnostic model selection. The paper's central finding: any single set of physics, calibrated as well as you like, cannot serve every hydroclimatic regime in the CONUS equally well. Cohen et al. (2023) corroborated region-dependent skill using a multi-decade retrospective dataset, and Timilsina and Passalacqua (2025), comparing NWM v2.1 and v3.0 across 610 USGS gauges in Texas including Hurricane Harvey, TS Imelda, and Hurricane Hanna, documented both advances and persistent regional challenges.

The weakness is not a tuning failure. It is the consequence of asking one set of physics to represent every regime at once.

A single rainfall-runoff formulation cannot be simultaneously optimal for a flashy semi-arid watershed in the Southwest, a snow-dominated mountain headwater, a heavily-regulated Mississippi tributary, and an urbanized coastal watershed. The tradeoffs that make a model good for one make it worse for another. Tuning narrows the gap; structural change is what is needed beyond tuning.

NextGen is the structural change. The architectural commitment to per-region, per-process formulation choice is the field's response to a regional-skill problem documented for years.

The transition will not be a one-day cutover. The operational story is more gradual, and three classes of project use NWM forecasts as a direct input.

The first is real-time flood forecasting. Local and state emergency managers, water utilities, and operations forecasters use NWM-derived flow and stage forecasts to support short-horizon decisions. As NextGen lands, the character of those forecasts is likely to change. Lead times may shift. Skill in specific regimes, such as urban flash flooding, snowmelt onset, and regulated rivers, may improve in some regions and produce different bias signatures than the v3 product. The Johnson et al. (2023) regional-skill literature is the strongest defensible basis for expecting this kind of redistribution.

The second is water-supply lookahead. Reservoir operators, water utilities, and agricultural water managers use NWM weekly-to-seasonal flow forecasts as one input into operational planning. The skill profile of those forecasts will reset when the underlying formulations reset. Operators who have calibrated their internal decision rules to v3 forecasts will need to recalibrate against v4 outputs as those become available.

The third is ungauged-basin discharge estimation. Engineers and modelers use NWM as a reference dataset for basins where local gauging is sparse. As the underlying model changes, the reference dataset changes. Comparisons of project-scale models against the federal product become non-stationary across the transition.

In each case, the project-side response is the same: identify the dependency, version-pin the analysis to the NWM release used, and document the assumption that the underlying federal product was held constant across the project life.

For project work in 2026 and 2027, three implications follow.

1. Projects that depend on NWM as forcing or boundary should explicitly version-pin. Document which NWM release was used, including the retrieval date if real-time. This is the analog of pinning a software dependency. It makes the analysis reproducible across the transition.
2. Do not transfer NWM-derived calibration parameters from a 2025 analysis into a 2027 analysis without re-evaluating. The federal product on which those parameters were tuned has shifted. Specific skill-delta numbers (for example, the v2.1-to-v3.0 deltas Timilsina and Passalacqua (2025) report for Texas) are not transferable to v3-to-v4, and citing them for that purpose would mislead.
3. In proposals that promise federal-forecast-derived deliverables in the 2026–2028 window, name the transition explicitly. Scoping language that says "NWM forecasts will be incorporated as available" without naming the v3-to-NextGen transition creates a defensibility gap if the federal product changes in mid-engagement.

The retirement of the monolithic NWM is the largest structural change to federal operational hydrology since NWM v1 entered production a decade ago. It is also a quiet change. The transition will not produce a press release; it will produce a series of operational updates that, in aggregate, leave the federal hydrologic forecasting infrastructure looking and behaving differently than it does today. Projects that recognize the structural change for what it is, and plan around it, will produce more defensible work across the transition than projects that treat the federal product as a stable reference.