26 posts tagged with "National Water Model"

On July 4, 2025, a flash flood on the Guadalupe River caused 119 fatalities in Kerr County, Texas. It was the deadliest flash flood in the United States since 1976, and a reminder that flash floods remain among the most dangerous weather hazards in the country. Forecasting them means working at the speed and scale of the storm: small basins, minutes to hours, and rainfall estimates that can make or break the prediction.

That challenge set the agenda for our theme at the 2026 Water Prediction Innovators Summer Institute: Advancing the Use of Quantitative Precipitation Inputs into the NextGen Framework. Ten graduate students in three teams (Don't Runoff!, FlashCast, and The Post-Processors) had seven weeks to bring sub-hourly radar rainfall, precipitation uncertainty, and ensemble forecasting methods into the NextGen Water Resources Modeling Framework. Seven weeks is not much, and the teams could not spend half of it compiling a modeling framework or downloading terabytes of radar data to laptops. Fortunately, they did not have to. The CIROH cyberinfrastructure team carried that load.

The framework fits in the Box​

NextGen is powerful but challenging to run. NextGen In A Box (NGIAB), CIROH's containerized distribution with more than 41,000 Docker pulls, turns that setup problem into a single pull command. This year's Summer Institute members were able to hit the ground running after just a single one-hour NGIAB workshop, demonstrating its sheer accessibility.

However, the teams did more than just run it. NGIAB is also an open box, and they took full advantage by extending it:

• FlashCast built a modified NGIAB Docker image that runs CFE and T-Route on a 15 minute timestep and forced it with 15 minute MRMS radar rainfall, expanding the framework's standard hourly configuration to the sub-hourly scales flash floods demand.
• The Post-Processors modified the NGIAB data preprocess library so NextGen could ingest MRMS radar estimates and HRRR forecast rainfall in place of the default AORC forcing. They then simulated 8,474 catchments around Ellicott City, Maryland.
• Don't Runoff! built a volume-preserving workflow that maps 2 minute, 1 km MRMS rainfall onto NextGen Hydrofabric catchments and used it to drive dHBV2-Flash, a new sub-hourly differentiable model designed for the framework.

Every modification lives in a container or a public repository. Any researcher can pull the same image, rerun the same experiment, and build on it. That is the point of the box: NGIAB enables maximal portability and reproducibility to maximize research's impact. For the complete set of shared resources, see each team's final report.

R2OHC Platform: NSF Jetstream2 and NGIAB did the heavy lifting​

None of this fits on a laptop. FlashCast alone ran 2,280 simulation scenarios per basin across three flash flood case studies (Asheville NC, Pittsburgh PA, and Kerrville TX) — 6,840 NextGen runs in total. The Post-Processors pushed 500 synthetic storms through NGIAB to train a neural network surrogate, on top of storm displacement experiments across thousands of catchments. Don't Runoff! processed radar rainfall for over 5,000 flash flood events across the country and trained a differentiable model on years of 15-minute data.

The computing resources came from CIROH cyberinfrastructure, which provided allocations on NSF Jetstream2, Indiana University's NSF-funded academic cloud, through CIROH's ACCESS allocation (EES240087). JetStream2 hosted virtual machines sized for each team's workload, from batch NGIAB runs to model training. The rainfall came from NOAA open data: 2-minute MRMS radar estimates, hourly AORC forcings, and HRRR forecasts, all pulled straight from public cloud archives. CUAHSI HydroShare closed the loop as the publication home for datasets and supplementary materials, and the NWC-CUAHSI-Summer-Institute GitHub organization hosts the code.

What the teams built​

Team Don't Runoff! delivered RUNOFF v1.0, a CONUS-scale catalog of over 5,000 flash flood events from the NOAA Storm Events Database, each mapped to hydrofabric catchments and paired with a USGS gauge and quality-controlled MRMS rainfall. RUNOFF is designed to conduct flash flood modeling experiments using high-quality precipitation estimates. They also delivered dHBV2-Flash, a 15-minute differentiable rainfall-runoff model that captured event peaks in the Upper Neuse basin with a median peak flow bias of 16% across 1,132 events. Supplementary materials are on HydroShare.

Team FlashCast addressed a question at the heart of flash flood prediction: how fine, and how accurate, does the precipitation input need to be? Their sensitivity pipeline converts thousands of NGIAB runs into critical uncertainty matrices, the precipitation uncertainty a forecast can tolerate before predictions stop being actionable. Early results from the three case studies point to a scale-dependent answer: for these events, resolution finer than about 7 km and 1 hour added little on its own, as long as storm volume was preserved. The pipeline is built to put that threshold to the test across many more basins and storms. Methods and figures are on HydroShare.

The Post-Processors built NCET, the Neighboring Catchment Ensemble Technique, which turns a single deterministic NextGen simulation into an ensemble streamflow forecast by borrowing hydrographs from neighboring catchments. Applied to the May 2018 Ellicott City flash flood with a real HRRR forecast, NCET recovered a flood the deterministic forecast missed, with the largest gains in probabilistic skill when the ensemble weights were aimed along the storm's displacement. That early result suggests storm location, more than catchment similarity, should steer the ensemble, and it points to a computationally inexpensive path toward flash flood forecasting within NextGen. The team also prototyped Replace and Route, a BMI component that lets NextGen borrow discharge from high-fidelity external models at hydrofabric nexus points. Evaluation materials are on HydroShare.

From asking research questions to prototyping workflows, in seven weeks​

Each product is hydrofabric-native, open, and packaged into containers, BMI modules, and reproducible pipelines — the way that the Community NextGen ecosystem expects. This ease of access and reuse is research to operations in miniature, and it is what CIROH cyberinfrastructure is for. NGIAB collapsed setup time from weeks to minutes. NSF's Jetstream2 scaled experiments beyond what any student laptop could run. Open data archives fed the models, and HydroShare and GitHub made every result citable and reusable. The students arrived with research questions. They left behind open prototypes, datasets, and code for the community to build on.

The 2026 Water Prediction Innovators Summer Institute took place June 8 — July 22, 2026. 24 graduate student fellows from 21 unique U.S. institutions formed research teams under the guidance of seven Theme Leaders, spanning topics from precipitation forecasting to flood risk communication. Research outcomes are summarized in the 2026 WPI SI final report.

The teams thank the CIROH Cyberinfrastructure & DevOps team and Hydroinformatics team at the Alabama Water Institute (Arpita Patel, Josh Cunningham, Quinn Lee, Nia Minor, and colleagues) for cloud provisioning and NextGen support throughout the program, Dr. Fred Ogden for guidance on developments with the NextGen framework, and the Summer Institute coordinators, Megan Vardaman and Nana Oye Djan. This work used Jetstream2 through allocation EES240087 from the NSF ACCESS program. This research was supported by the Cooperative Institute for Research to Operations in Hydrology (CIROH) under award NA22NWS4320003 from the NOAA Cooperative Institute Program.

Flood Inundation Mapping (FIM) is the spatially explicit delineation of flood extent, turning hydrological forecasts and observations into actionable geospatial information. As the frequency and severity of flooding increase globally, accurate and scalable FIM has become essential to disaster preparedness, emergency response, infrastructure planning, and climate risk assessment. Operationally, flood maps are commonly generated from streamflow forecasts, such as those produced by NOAA's National Water Model in the United States. FIM, however, is only one part of the workflow. Before FIM can reliably inform decisions, it needs to be rigorously checked against high-quality observations of real flood events.

At the University of Alabama's Surface Dynamics Modeling Lab (SDML), with support from CIROH in partnership with NOAA, we have been building the CIROH FIM Community Ecosystem, an interconnected, open-source, Python-based suite of toolsets for large-scale flood inundation modeling, simulation, benchmarking, and evaluation. Below we describe what each component of the Ecosystem does, how they work together, and how to get started.

The problem: fragmented capabilities without a unified FIM framework​

This effort starts with a simple question: can flood inundation modeling, benchmarking, and evaluation be brought together into a seamless, unified, and, most importantly, scalable system that serves both operational forecasters and the research community? What exists today are small-scale studies and individual tools: models predict floods, methods that evaluate accuracy, observations that record real events, each built for a specific task and demanding its own painstaking setup. What does not exist is a unified framework spanning modeling to evaluation at scale, for one study area or a thousand, so that methodologies are consistent and ready to calibrate models and test hypotheses. For operational forecasters, the missing piece is scale: mapping a single basin is a tractable problem, but operational forecasting requires generating and trusting a modeling approach across numerous river reaches in real time.

Evaluation raises the bar further. A model's true potential and limitations emerge only when it is tested at scale across many events, diverse floodplains, topographic settings, and hydrologic regimes. Large-scale evaluation is what turns a case study into a defensible conclusion about where a model works, where it fails, and why. Yet such evaluation depends on an "answer key": high-quality records of flood inundation for real flooding events. Those benchmark data have historically been scarce, scattered, and inconsistent, and where benchmarks are unavailable or of poor quality, evaluations lose credibility, and even skillful models are met with mistrust. The result: research groups rebuild the same pipelines from scratch, evaluations cannot be compared across regions or scales, and promising models take longer to reach the people who need them.

Our effort: the CIROH FIM Ecosystem​

Our answer is an interconnected, interoperable system of open-source components designed to work together through simple programmatic interfaces. Most components are deployed as standalone Python packages: install only what you need, or chain them into a complete workflow.

(a) FIMserv: FIM generation tool using the NOAA OWP HAND-FIM operational framework​

The Office of Water Prediction's Height Above Nearest Drainage (OWP HAND) framework, the operational FIM system of the United States, is a deliberately simple terrain-based model run over Hydrologic Unit Code 8 (HUC8) watersheds (more about HUCs) using National Water Model streamflow predictions across 2.7 million river reaches nationwide. Configuring it, however, is not straightforward: setting up the full pipeline from watershed data to streamflow and reproducing it across computing systems have been persistent obstacles to applying, improving, or studying an otherwise scalable model. FIMserv, the modeling component of the CIROH FIM Ecosystem, removes that barrier. Given a watershed of interest and an event, it automatically configures the entire modeling setup, retrieving the OWP-precalculated watershed data from CIROH's AWS S3 cloud storage (which mirrors the NOAA OWP-precalculated HUC8 datasets for the HAND FIM run) and the NWM streamflow for the chosen event (a past event, also called retrospective data; a live National Water Model forecast; USGS gauge records; or global hydrological products such as GeoGLOWS). It then leverages the OWP inundation mapping framework to produce a flood extent map within minutes (Baruah et al., 2025). It covers NOAA's full forecast range, from hourly short-range runs to 30-day outlooks, and supports simulation of many watersheds and flood events. It can also evaluate the NWM streamflow against USGS gauge observations, providing a measure of confidence in a map's inputs. Users can also modify key configuration inputs, such as river slope, calibration coefficients, and river bathymetry, supporting experimentation well beyond the default modeling setup. It is wrapped as a Python package that runs just as well on a laptop as on cloud infrastructure, so what once took a long time now takes a single installation command.

(b) FIMeval: FIM predictions evaluation framework​

A flood map is only as useful as the confidence placed in it, and that confidence comes from evaluation: comparing a model-predicted extent against a benchmark flood map from an actual or synthetic event. The traditional methods of evaluation, however, are manual, time-intensive, and error-prone; preparing maps in GIS platforms and/or doing it programmatically, and standardizing models and benchmarks (such as removing permanent water bodies and aligning projections) is so laborious that testing a model across many events becomes impractical. FIMeval automates, standardizes, and scales that step, evaluating any flood maps against benchmarks in a consistent, reproducible manner (Devi et al., 2026). It is deliberately model- and benchmark-agnostic: the map can come from any terrain-based, hydraulic, or data-driven model, and the benchmark from any trusted source. It computes several standard accuracy metrics, with built-in checks for how data imbalance between flooded and dry areas skews those scores. It then goes further, answering the impact-based question communities actually ask: which buildings does the model predict will be inundated, and was that prediction correct? It fits into any Python-based workflow: bring a flood map, whether produced by FIMserv or any other model, together with a benchmark you supply, and FIMeval scores the pair. What makes it especially powerful, though, is that users do not need to supply a benchmark at all: FIMeval connects seamlessly to the FIMbench database, so a single automated query pulls the right quality-controlled benchmark for your area and event and scores against it, provided the event date falls within our available benchmark coverage. FIMbench itself is our rigorously quality-controlled, multi-source benchmark database, described in detail in section (c) below. For FIMserv users, the evaluation can run embedded within FIMserv, generating and evaluating a map in a single pass. Demonstrations span major observed events and design floods alike across continental to global watersheds.

(c) FIMbench: an extensive FIM benchmarking database​

Trusting a model-predicted flood extent requires real-world evidence, precisely the benchmark data that has been hardest to obtain. FIMbench fills that gap with a rigorously quality-controlled library of multi-source, multi-sensor benchmark flood maps. Its current release includes 190+ maps (it will continuously keep growing with time) across the United States (Devi et al., in preparation), derived from aerial imagery acquired via NOAA's National Geodetic Survey (NGS), PlanetScope satellite imagery, Sentinel-1 radar, FEMA Base Level Engineering (BLE) 100- and 500-year return-period flow-based FIM products, and FIMs derived from USGS-surveyed high-water marks. Every map undergoes standardized processing and quality screening and is cataloged into quality- and product-based tiers; the guiding philosophy is "quality over quantity." FIMbench utilities then package each approved map into a database-compatible format and publish it to the CIROH AWS S3 database. Contributing to the database is currently limited to the SDML team, though we are exploring ways for the community to add data in the future; every published dataset, however, is openly available to everyone. Access follows two routes: the interactive FIMbench WebApp, developed with our partners at Aquaveo, supports visualizing, filtering, and downloading flood events; and the FIMbench Python API retrieves selected benchmarks programmatically from the cloud-hosted CIROH AWS S3 database. A cataloged, quality-assured, multi-sensor benchmark collection at this national scale has not, to our knowledge, been openly published before; paired with FIMeval, it makes testing a flood model against hundreds of real events practical rather than prohibitively laborious.

(d) FIMsens: flood extent and depth from satellite observations​

FIMsens, an upcoming Ecosystem component, adds an observation mode to the ecosystem: an end-to-end toolset that turns satellite imagery into error-corrected flood extent and depth products. Given a user-defined area of interest and event period, FIMsens discovers and previews suitable SAR or optical scenes, then processes them through linked modules for initial flood detection, reliability assessment, omission reconstruction, and depth estimation, beginning with Seeded Local Adaptive Thresholding (SLAT) (Wang et al., under review), which extracts the inundation directly observable in the imagery. The detections then pass to the Flood Object Reliability Filter (FOR-Filter) (Tian et al., in preparation), which uses image, geometric, terrain, and hydrologic evidence to remove or flag likely false positives before they propagate downstream. Finally, RS-FloodXDepth (Tian et al., 2026) combines retained flood evidence with surrounding topography to reconstruct hydrologically plausible inundation that would otherwise be missed due to vegetation, urban structures, clouds, or weak water signals, and to estimate depth from the inferred water surface and DEM elevation. A single pass delivers raw, filtered, and corrected extent layers together with a DEM-guided depth map.

Within the broader ecosystem, FIMsens complements rather than replaces the curated FIMbench reference databases. The same algorithm underpins FIMbench's Tier 2 and Tier 3 benchmarks, where its output is conservatively confirmed through qualitative assessment. Where the rigorously quality-controlled FIMbench database has a match, that curated map stays the trusted reference. Where no suitable benchmark exists, FIMsens can be used to generate event-specific, observation-derived flood extent and depth products from available satellite imagery, providing additional spatial and temporal coverage for FIMeval and related analyses. Because these event-specific products depend on the quality of the selected observations and processing conditions, they should still be reviewed through the toolset's QA outputs and, where appropriate, qualitative inspection before being used as reference information.

(e) FIMsim: cloud-based multi-model flood simulation​

FIMsim expands the ecosystem's flood-simulation engine, which models flooding using either terrain-based or hydraulic approaches: it runs the terrain-based NOAA OWP HAND-FIM through FIMserv, the ARC-Curve2Flood model, and extends the workflow to 2D hydraulic simulation using models with varying levels of physical complexity. Because no single flood model performs best everywhere, comparing alternatives has meant rebuilding a pipeline for each one; FIMsim removes that burden. In practice, the workflow is simple: the user defines a study area, and FIMsim acquires the inputs, builds the model configuration, and runs the hydrodynamic models of their choice, such as LISFLOOD-FP and TRITON. These hydrodynamic models provide GPU-enabled implementations that can substantially reduce execution time for computationally demanding simulations. Through the planned web application, users will be able to run the models remotely in a GPU-enabled Linux environment and download the results without installing the software locally. Different models can be run across catchments, or several models over a single catchment for agreement and ensemble analysis; integrating FIMeval and FIMbench provides evaluation for every output. Since FIMsim can be set up and run for every reach, a hypothesis can be tested across models rather than trusted to just one model. Simulation is not its only use. FIMsim can also serve purely as a preprocessor for hydraulic modeling, generating each input required for a 2D model separately. It can produce a mosaicked, clipped, reprojected, and resampled DEM at different spatial resolutions; generate Manning's n roughness grids from land-cover datasets of different sources, years, and resolutions; provide flowline data; and obtain inflow hydrographs from NWM, a USGS gauge, or an uploaded CSV file. The processed inputs can be exported in several commonly used formats, allowing modelers to use the ready-to-run files directly in their preferred hydraulic modeling engine. Today, FIMsim is available as a Python framework and a desktop application for macOS and Windows; a pip-installable Python library and a web application are also under development.

(f) FIMbox: a sandbox for large-scale, multi-model FIM development and analysis​

FIMbox is an open-source Python sandbox that treats flood inundation mapping as a controlled experiment: models, inputs, and parameters become variables that a researcher deliberately sets, holds fixed, or sweeps across in a single, reproducible run. A run begins with a boundary polygon for any area of interest rather than a precalculated HUC8, and a shared initialization step stages the common resources every model needs into one working directory: the DEM, NHD and NWM hydrography, FEMA NFHL flood hazard layers, NLD levees, OSM bridges and roads, and USGS gauge records. A configuration wrapper then exposes each subsequent stage, so a user-defined topography, hydrography, or roughness field is applied where it matters. The experiment setup determines which models to run, the terrain-based HAND workflow, a hydrodynamic model, or both. Each model triggers its own preprocessor, which builds only the datasets it needs and pulls inflows from the National Water Model or NextGen, or through the crosswalk module from global sources such as GEOGloWS.

Where FIMserv runs only the operational OWP HAND framework over precalculated HUC8 data, FIMbox opens the pipeline itself. Its HAND FIM modeling breaks the OWP HAND-FIM pre- and post-processing chain (from a DEM to a mapped extent) into small composable pieces: staging inputs, hydroconditioning the terrain, segmenting the stream network, generating HAND and its synthetic rating curves, calibrating them, generating the FIM, and evaluating it. Every stage exposes its parameters and writes intermediate products the user can inspect, replace, or feed back in, so a dataset or a coefficient can be substituted at one step and its consequences traced through to the mapped extent and depth. Because each dataset and parameter is a variable under the user's control, questions that are ordinarily laborious become routine: how much the resolution or the source of the river network, catchments, or DEM changes the predicted extent; how sensitive the result is to Manning's n roughness; and whether a better synthetic rating curve, a different stream segmentation, or an improved slope treatment sharpens extent and depth. Dask parallelization runs across the whole HAND pipeline, so the same workflow scales from a single basin to continental domains without rewriting a line of code, and the package runs the same way on a local workstation or on cloud computing resources.

What sets FIMbox apart is that a single experiment carries seamlessly from modeling through evaluation to benchmarking. Within one run it can drive a terrain-based HAND solution and, through the integrated FIMsim framework, a hydrodynamic model such as TRITON or LISFLOOD-FP, without altering their native configurations, then score each predicted extent with FIMeval against the best-matching FIMbench benchmark, selected by event date, geographic domain, spatial overlap, quality tier, or any combination of these; where no benchmark exists, FIMsens can attempt to derive one from satellite imagery. Because the whole chain, from fetching flows to scoring the result, is scriptable from a notebook or a scheduled job, FIMbox makes FIM research genuinely iterative: a hypothesis can be posed, tested, and scored at any scale, one basin or a thousand, with no manual handoffs between tools. That same design makes multi-model analysis natural, and work is now underway to extend the sandbox beyond a single model toward ensemble, multi-model FIM extents, repurposing results from a range of hydrologic models through the same workflow. Looking ahead, integrating NOAA's Next Generation Water Resources Modeling Framework (NextGen) will close the loop from meteorological forcing through hydrologic and inundation modeling to flood-extent assessment and evaluation, an end-to-end ecosystem simulation.

When should you use the FIM ecosystem?​

Each component can be used on its own; you do not need to adopt the entire ecosystem to benefit from a single component. However, integrating the components opens up research questions that no single component can answer alone. A few common starting points:

• "I need flood maps for a forecast or a past event or a synthetic scenario anywhere in the US; however, I need them fast and running on any local or cloud system." FIMserv can do that.
• "I built a flood inundation model, hydraulic, data-driven, or otherwise, and I have my own benchmark; I want to know how accurate the model is." FIMeval can do that.
• "I built a flood inundation model intended for regional, continental, or global application, and I want to investigate its potential and limitations across different hydroclimates and topographic settings." FIMeval, leveraging the FIMbench database, can do that.
• "I just need reliable benchmark flood maps for my study area." FIMbench, via the GUI or the Python API, can provide that.
• "I want to prepare the inputs for a 2D hydrodynamic model such as LISFLOOD-FP or TRITON, or run the model itself, wherever I need and without the usual setup hassle." FIMsim can do that, in a desktop app or Python framework, either handing you ready-to-run inputs and configuration to run on your own or running the simulation for you.
• "I want to run several flood models over the same watershed for ensemble and agreement analysis or assign a different model to each catchment according to what performs best there." FIMbox can do that, orchestrating the models through FIMsim and passing every result on to FIMeval and FIMbench for evaluation.
• "I want to experiment with the HAND workflow itself, swapping terrain, roughness, calibration, or model engines, feeding it custom or repurposed streamflow, and scaling from a single basin to continental domains." FIMbox can do that, breaking the workflow into flexible, Dask-parallel pieces that run straight through to evaluation.
• "I want to test a hypothesis on the terrain-based NOAA OWP HAND model, evaluating it across dozens or hundreds of events to judge whether the approach can be implemented across the US." The integrated ecosystem can enable that: FIMserv handles the modeling, and FIMeval, drawing benchmarks from FIMbench, handles the evaluation; the entire process is automated, programmatically and seamlessly.
• "I have no benchmark for my event and need to derive a flood extent and depth directly from satellite imagery." FIMsens can do that (once integrated), turning SAR or optical scenes into error-corrected extent and depth products.

Under the hood: technical dissemination​

Technically, the ecosystem is a modular, cloud-native suite of open-source Python packages and frameworks. What holds it together is a shared CIROH cloud backbone: every component reads and writes the same cloud-hosted data, so one tool's output is directly consumable by the next. FIMserv streams its inputs on demand: precalculated HAND hydrofabrics from the CIROH AWS S3 buckets, retrospective National Water Model (NWM) streamflow from NOAA's public S3 archive, and forecast NetCDF files from Google Cloud Storage, with the TEEHR framework's PySpark engine handling massive streamflow extractions. FIMeval builds on the standard geospatial Python stack (Rasterio, PyProj, GeoPandas, Shapely) for raster reprojection, map algebra, and dynamic bounding geometries based on different evaluation methods. The heavy layers it needs on the fly, such as permanent water bodies and Microsoft building footprints for impact-based assessment, are hosted on ArcGIS Online and other cloud services and fetched through REST endpoints for the evaluation domain only, so nothing oversized ever lands on the user's machine. FIMbench's dissemination rests on a decoupled web architecture: a Tethys Platform backend acts as a tile proxy, streaming compressed vector tiles of benchmark FIM extent straight from the CIROH AWS S3 storage to a React front end that renders interactive maps with MapLibre GL. The same S3 store serves the stateless FIMbench Python API, so the database remains a single, consistent source of truth, regardless of how it is accessed. FIMbox adds a parallel execution layer: it decomposes the HAND workflow into independent per-branch computations (hydroconditioning, branching, relative elevation, and synthetic rating curves) and uses Dask to distribute those branches across local cores or cloud workers. FIMsim moves computation to the cloud, provisioning the hydrodynamic engines on managed resources and assembling their inputs from public sources: USGS 3DEP terrain, land-cover-derived roughness, and NWM, USGS, or user-supplied hydrographs. Hardware is the one practical dividing line across the ecosystem: FIMbox's terrain-based HAND workflows run on ordinary local systems without a GPU, while the full FIMbox and FIMsim capability, terrain-based modeling coupled with 2D hydrodynamic simulation, is currently GPU-backed for accelerating the computation and Linux-based. FIMsens will add the observation path, query SAR and optical image catalogs, align downloaded scenes to a common DEM, and then hand them to the correction routine described above.

Who is it for? Try it today​

The CIROH FIM Community Ecosystem is completely open source and already in active use across the community: as of mid-2026, PyPI records put FIMserv past 45,000 downloads and FIMeval past 20,000, with the FIMbench API and the FIMbox testbed more recently released. How you start depends on the component: four of them are installed as Python packages with a single command (Astral uv makes it faster still), FIMsim ships as a desktop application, and FIMsens will join once released.

uv/pip install fimserve   # flood map generation

uv/pip install fimeval    # flood map evaluation

uv/pip install fimbench   # benchmark database access

uv/pip install fimbox     # FIM testbed access

Each framework is provided with step-by-step documentation and example notebooks on its GitHub repository: FIMserv, FIMeval, FIMbench, FIMbox and so on, so any user can go from installation to a first flood map and its evaluation without leaving the documentation. Run them wherever you work, from a laptop to Google Colab or CIROH's 2i2c JupyterHub. Prefer a point-and-click route? The FIMbench web portal lets you browse the benchmark database, and FIMsim runs today as a desktop application.

Who benefits from the FIM ecosystem? Broadly, anyone who works with flood hazard modeling. From researchers testing new models to forecasters, planners, engineers, and students, each takes something different from it: a place to experiment, options they can defend in operations, and flood information that has been checked before it is trusted. Our goal is simple: anyone, whatever their background, can make a flood map, check it against reality, and trust what it tells them.

Get involved​

The FIM ecosystem is a living effort, and feedback will improve it. We will appreciate your feedback and suggestions about technical problems, proposed utilities, and additional components. We are committed to incorporating that input into each release, and collaboration is what we need most. The ecosystem is still very much in motion: not every framework yet has its own graphical interface, and some are not yet packaged for PyPI, but many avenues and dimensions are advancing in parallel. Keep an eye out for new releases, and email us about collaborations, ideas, or anything you would like to build with us. All the FIM components code lives in the SDML GitHub repository, and each repository has its own contribution guidelines: open an issue with a new idea or a bug, submit a contribution, or simply reach out to the SDML team. We would genuinely love to hear from you; it is how this ecosystem will keep evolving. And this is only a start: the CIROH FIM Community Ecosystem is not limited to the components described here. We want it to grow beyond SDML, with colleagues across CIROH institutions and the broader community contributing new components, datasets, and ideas that strengthen it as a shared, community-owned capability.

Acknowledgments​

This work is supported by the National Oceanic and Atmospheric Administration (NOAA) through the Cooperative Institute for Research to Operations in Hydrology (CIROH) under the NOAA Cooperative Agreement with The University of Alabama (NA22NWS4320003). We thank the Aquaveo team for collaborating on the development of the Graphical User Interfaces for these tools.

CIROH cyberinfrastructure used: CIROH AWS S3 cloud storage hosts the OWP-precalculated HUC8 HAND hydrofabric datasets streamed on demand by FIMserv, and hosts the FIMbench benchmark database; the CIROH Tethys Platform / CIROH Portal hosts the FIMbench WebApp, developed with our partners at Aquaveo (we are also thinking of hosting a curated virtual environment/image on 2i2c services soon). This cyberinfrastructure is the backbone that makes the ecosystem scalable and interoperable: because every component reads from and writes to the same CIROH cloud-hosted data, one tool's output is directly consumable by the next, and users never need to download oversized national datasets. CIROH's Tethys hosting and 2i2c JupyterHub lower the barrier further, providing point-and-click and zero-install routes into the tools, while CIROH compute resources make evaluation across hundreds of events and continental-scale runs practical rather than prohibitive.

Additional Links​

• Understand the CIROH FIM ecosystem visually: https://sdml.ua.edu/fim-ecosystem/
• Explore the CIROH FIM ecosystem more interactively with our NotebookLM Mind Map

Publications​

1. Baruah, A., Dhital, S., Cohen, S., Tran, T.N.D., Elhaddad, H., Watts, C.L., Devi, D., Chen, Y., & Pruitt, C. (2025). FIMserv v.1.0: A tool for streamlining Flood Inundation Mapping (FIM) using the United States operational hydrological forecasting framework. Environmental Modelling & Software, 192, 106581. https://doi.org/10.1016/j.envsoft.2025.106581
2. Devi, D., Dhital, S., Munasinghe, D., Cohen, S., Baruah, A., Chen, Y., Tian, D., & Pruitt, C. (2026). A framework for the evaluation of flood inundation predictions over extensive benchmark databases. Environmental Modelling & Software, 196, 106786. https://doi.org/10.1016/j.envsoft.2025.106786
3. Devi, D., Munasinghe, D., Tian, D., Dhital, S., Cohen, S., Raghavan, R., Swain, N., Nikrou, P., Baruah, A., & Chen, Y. (in preparation). FIMbench: A benchmark database of multi-tier flood inundation maps for large-scale evaluation.
4. Tian, D., Liu, H., Wang, L., Cohen, S., & Mandal, T. (2026). RS-FloodXDepth: Enhancing remote sensing-derived flood extent and estimating flood depth using a hydrologically guided region-growing method and high-resolution DEMs. Water Resources Research, 62, e2025WR042384. https://doi.org/10.1029/2025WR042384
5. Tian, D., Liu, H., Cohen, S., & Wang, L. (in preparation). Commission error analysis and correction in remote sensing flood maps using an object-oriented, hydrologically informed approach.
6. Wang, L., Liu, H., Tian, D., & Cohen, S. (2025, under review). Rapid automated flood mapping using SAR imagery and locally adaptive thresholds derived from multivariate gamma probability distributions. International Journal of Applied Earth Observation and Geoinformation.

Our Team​

Supath Dhital

Geography Researcher II

The University of Alabama

Dipsikha Devi

Postdoctoral Researcher

The University of Alabama

Anupal Baruah

Research Scientist

The University of Alabama

Dinuke Munasinghe

Research Scientist

The University of Alabama

Parvaneh Nikrou

PhD Candidate

The University of Alabama

Dan Tian

Postdoctoral Researcher

The University of Alabama

Yixian Chen

Research Scientist

The University of Alabama

Sagy Cohen (PI)

Professor

The University of Alabama

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.

Opening New Doors for Research with the NextGen Framework​

The NextGen framework holds great potential for hydrologic modeling, but is often inaccessible due to its strenuous setup and requirements. As such, embedding it within a cloud-based framework offers a natural solution to this problem by removing some of the administrative and technical requirements for compute resource setup and computational library configuration, thus opening the door for a wider audience to tke advantage of the strengths of the framework.

With the CIROH Community NextGen Hub (CCNH), we’ve created a cloud-based environment that addresses exactly those setup challenges, so users can focus on science instead of software.

A Preconfigured, Ready-to-Use Cloud Environment​

CCNH is a containerized, cloud-based modeling environment hosted on the CIROH-2i2c JupyterHub. It packages everything a researcher needs to run end-to-end NextGen workflows — from input preprocessing through model execution, calibration, evaluation, and output visualization — into a single, ready-to-use JupyterHub image. Built on the same containerization patterns as NGIAB, CCNH leverages a Pangeo base image and includes:

• Pre-compiled NextGen framework binaries from NGIAB based docker image
• NGIAB data preprocessing tools for automated retrieval and subsetting of hydrofabric and meteorological forcing datasets
• T-Route routing components for streamflow simulation
• SPOTPY(Statistical Parameter Optimization Tool for Python) for model calibration
• TEEHR(Tools for Exploratory Evaluation in Hydrologic Research) for performance evaluation
• PyNGIAB, a Python wrapper that lets you run NextGen simulations directly from Jupyter notebooks
• HydroShare integration tools (nbfetch, hs_files-jupyter, hsclient) for seamless data exchange to save results in HydroShare for collaboration, reproducibility and publishing
• JupyterLab with distributed computing capabilities for interactive, scalable workflows

The result: researchers can go from zero to running a calibrated NextGen simulation in a fraction of the time previously required.

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.

A screenshot of the HydroShare resource page for Jupyter Notebooks for the Retrieval of AORC Data for Hydrologic Analysis.

The Analysis of Record for Calibration (AORC) dataset is recognized as a high-value resource for the CUAHSI and CIROH community. This dataset is hosted by NOAA via Amazon Web Services (AWS) and is available in two primary formats: a latitude-longitude gridded dataset and the National Water Model (NWM) projected dataset, part of the NWM Retrospective archive. To enhance accessibility and illustrate analysis capabilities, we developed four user-friendly Jupyter Notebooks that enable data retrieval for both specific points of interest and spatial domains defined by shapefiles:

A poster presented by the I-SMART team at the CIROH Developers Conference, held at the University of Vermont in Burlington from May 28 to 30, 2025.

The densely populated Hackensack River watershed lies within the New York City Metropolitan Area, which spans northern New Jersey and southern New York. Accurate streamflow forecasting within this region is therefore essential to enable effective water resource management, flood prediction, and disaster preparedness.

Precipitation data is critical for effective hydrological modeling, making the identification of reliable data sources a key priority. This is why the Integrated Spatial Modeling and Remote Sensing Technologies Laboratory (I-SMART), an interdisciplinary research unit within the Davidson Laboratory at Stevens Institute of Technology in Hoboken, New Jersey, uses the latest developments in both atmospheric and hydrological modeling to address flood risks in the Hackensack Watershed with solutions that could be expanded to the entire New York City Metropolitan Area.

Last week, I had the incredible opportunity to co-present a keynote at the CIROH Developers Conference (DevCon 2025), which attracted over 200 attendees. This presentation, which I presented alongside Dan Ames, focused on "CIROH HydroInformatics and Research Cyberinfrastructure." It was a fantastic experience to share insights into the powerful tools and technologies that CIROH engineers, students, researchers have been developing to advance hydrological research and operations.

CIROH-AWI Science and Technology Team.
Left to right: Sagy Cohen, Steven Burian, Manjila Singh, Saide Zand, Savalan N. Neisary, Arpita Patel, Nia Minor, Trupesh Patel, Sifan A. Koriche, Jonathan Frame, Reza S. Alipour, Hari T. Jajula, Chad Perry; Josh Cunningham.

May was a pivotal month for representing the Cooperative Institute for Research to Operations in Hydrology (CIROH) and our collective work in advancing water science. As one of CIROH's Ambassadors, I had the privilege of connecting with the broader scientific community at two key events: the Environmental and Water Resources Institute (EWRI) Congress in Anchorage, Alaska, and the 2025 CIROH Developers Conference in Burlington, Vermont.

Predicting water flow with precision across the vast U.S. landscape is a complex challenge. That's why Song et al. 2024 developed δHBV2.0, a cutting-edge hydrologic model. It’s built with high-resolution modeling of physics to deliver seamless, highly accurate streamflow simulations, even down to individual sub-basins. It's already proven to be a major improvement, performing better than older tools at about 4,000 measurement sites. We also provide a comprehensive 40-year water dataset for ~180,000 river reaches to support this.

Penn State research group pushed δHBV2.0 further, training it with even more detailed river data and integrating other trusted models, aiming to make it a key part of the NextGen national water modeling system (as a potential NWM3.0 successor). But here’s a common hurdle: making powerful scientific tools like this easy and reliable for everyone to use within a larger framework can be tough. Setup issues, runtime errors, and inconsistent results can frustrate users.

NGIAB stepped in to solve exactly this problem. Team has taken the complexity out of using the operations-ready models within NextGen by creating one unified, reliable package. Thanks to NGIAB, users don't have to worry about tricky setups or whether the model will run correctly. NGIAB ensures that our models are compatible everywhere and, most importantly, that they run exactly as designed, consistently and faithfully, every single time, no babysitting required. This means users get the full power of our advanced modeling, without the headaches.

🌍 AWI News​

The Alabama Water Institute (AWI) at the University of Alabama (UA) recently published an article highlighting how NextGen In A Box (NGIAB) could transform hydrological modeling. This article provides great insight into NGIAB's real-world impact:

• 🚀30-minute setup vs days/weeks of configuration
• 📖 Provo River Basin Case Study demonstrating rapid deployment

➡️ Read the full press release here!

Pennsylvania State University (PSU) researchers have been leveraging CIROH Cyberinfrastructure to tackle complex hydrological modeling challenges. This post highlights their innovative approach using the Wukong computing platform in conjunction with Amazon S3 bucket storage to efficiently process and analyze large-scale environmental datasets. 🚀

The Community NextGen framework has seen significant advancements in November 2024, with major updates across multiple components and exciting new resources for users. Let's dive into the key developments that are making hydrologic modeling more accessible and powerful than ever.

Image Source: https://github.com/BYU-Hydroinformatics/api-nwm-gcp

Several important historical and ongoing National Water Model (NWM) datasets are now available on Google Cloud BigQuery, which makes them queryable through SQL using Google Cloud console. Some of these data sets are also accessible through an API (e.g. using Python). These datasets and their current status are as follows:

This month, we are excited to showcase two case studies that utilized our cyberinfrastructure tools and services. These case studies demonstrate how CIROH's cyberinfrastructure is being utilized to support hydrological research and operational advancements.

1. ngen-datastream and NGIAB​

We're excited to share some recent developments and updates from CIROH's Research CyberInfrastructure team:

Cloud Infrastructure​

• CIROH's Google Cloud Account is now fully operational and managed by our team. You can find more information here.
• We're in the process of migrating our 2i2c JupyterHub to CIROH's Google Cloud account.
• We've successfully deployed the Google BigQuery API (developed by BYU and Google) for NWM data in our cloud. To access this API, please contact us at ciroh-it-support@ua.edu. Please refer to NWM BigQuery API to learn more.

CIROH Developers Conference 2024

The CIROH team recently participated in the 2nd Annual CIROH Developers Conference (DevCon24), held from May 29th to June 1st,2024. The conference brought together a diverse group of water professionals to exchange knowledge and explore cutting-edge research in the field of hydrological forecasting.

AWRA 2024 Spring Conference

The CIROH CyberInfrastructure team recently participated in the AWRA 2024 Spring Conference, co-hosted by the Alabama Water Institute at the University of Alabama.

Themed "Water Risk and Resilience: Research and Sustainable Solutions," the conference brought together a diverse group of water professionals to exchange knowledge and explore cutting-edge research in the field.

Google Cloud Next '24

Hello everyone, and thanks for stopping by!

I recently had the incredible opportunity to attend Google Cloud Next 2024 in person for the first time, and it was truly an amazing experience. From insightful keynote presentations and workshops to vibrant booths buzzing with connections, the event was a whirlwind of innovation and inspiration.

The CIROH team has been diligently accelerating research cyberinfrastructure capabilities this month. We're thrilled to share key milestones achieved in enhancing the Community NextGen project and our cloud/on-premises platforms.

Welcome to the February edition of the CIROH Hub blog, where we bring you the latest updates and news about the Community NextGen project and CIROH's Cloud and on-premise Infrastructure.

Our team has been hard at work enhancing CIROH's Infrastructure and Community NextGen tools. Here are some highlights from February 2024:

1. We successfully launched our new On-premises Infrastructure, which is now fully operational. You can find documentation for it here.

Welcome to the January edition of the CIROH Hub blog, where we share the latest updates and news about the Community NextGen project monthly. NextGen is a cutting-edge hydrologic modeling framework that aims to advance the science and practice of hydrology and water resources management. In this month's blog, we will highlight some of the recent achievements and developments of the Community NextGen team.

We've released NGIAB v1.1.0! This release fixes issues:

• #21
• #67
• #44

More info: https://github.com/CIROH-UA/NGIAB-CloudInfra/releases/tag/v1.1.0

Visit NGIAB News

NextGen Framework Forcings​

A new forcing processor tool has been made public. This tool converts any National Water Model based forcing files into ngen forcing files. This process can be an intensive operation in compute, memory, and IO, so this tool facilitates generating ngen input and ultimately makes running ngen more accessible.