Newly available satellite data from NASA’s Surface Water and Ocean Topography (SWOT) mission reveal significant flaws in many global river models, particularly for dammed rivers, arid- region rivers, and Arctic rivers.

A study led by hydrologist Colin Gleason of the University of Massachusetts Amherst, published in Geophysical Research Letters (title: “SWOT, Empiricism, and River Modeling”), compared state- of- the- art machine- learning river models against SWOT observations. After quality controls, the team analyzed 68,347 river reaches representing about 38% of global discharge.

Key findings include:

• Models struggle most with dammed rivers, rivers in arid climates (especially populated ones such as parts of Australia, Central Asia, the southwestern U.S., and Mexico), multi- channel rivers, and many Arctic rivers. Parts of Siberia and China are particularly difficult.⁠
• Less than 10% of reaches show “serious error”, but these are often among the most important for water resources, hydropower, irrigation planning, and climate- change projections.⁠
• Models perform better on wider rivers and on “normal” single- thread rivers without dams, glacial influence, or estuaries. However, only about 11% of Earth’s rivers fit that idealized description, “a weird river is the norm” .

Gleason notes that errors in these models directly undermine climate predictions and irrigation forecasts. Dammed rivers pose special challenges because models must account for human operations (e.g., pumped- storage hydropower that can change depth by more than a meter per day, as on the Connecticut River). Arid- region difficulties stem largely from delayed, non- obvious groundwater effects. Arctic modeling suffers from sparse training data for machine- learning systems.

The study frames SWOT as an “early microscope” for rivers: it encourages relying more on direct satellite measurements rather than models alone, while also identifying where models need structural improvements to incorporate realistic hydraulics and human influences.

The press release (dated 6 October 2026, by Julia Westbrook of UMass Amherst) and related coverage in outlets such as Science highlight how SWOT’s simultaneous measurements of water level and width expose the true complexity of rivers shaped by dams, braiding, freezing, and human water use.

SWOT, Empiricism, and River Modeling

Answers to questions about freshwater sustainability are answered by modeling rivers, as it is impossible to measure the millions of rivers on the planet in the field.

We use a recently launched satellite (SWOT) designed to measure rivers in a new way to assess an ensemble of global river models. We use SWOT to see which rivers are modeled in an opposite way to what the satellite observes: for example, when SWOT sees the river get deeper than it observed it last week, but the models say the river flow decreased from last week (or vice versa).

We see that models struggle in arid and populated regions, and do relatively well in areas with less human development. However, the less densely populated Arctic is also a challenge. This research reveals the basic spatial pattern of global river modeling errors for the first time, setting the stage for a new, improved generation of models.

The study uses an inductive approach grounded in SWOT’s simultaneous measurements of river water-surface height and extent (via radar interferometry) to evaluate global river models.

These models include traditional physics- based hydrologic and land- surface models coupled with routing schemes, machine -learning (e.g., LSTM) approaches that estimate discharge, and hybrids. Models were assessed only if they produce daily discharge (ideally also depth, width, and velocity) mapped to vector river channels in the SWORD database.

Main methodological approach

• High- quality SWOT height observations are compared with modeled discharge on the same day via the SWOT- ensemble- Spearman (SES) correlation (Spearman rank correlation, chosen because the height– discharge relationship is strong, monotonic, and nonlinear).
• An ensemble of machine- learning models is preferred; traditional physics- based ensembles fill gaps where ML output is unavailable.
• Models were run in a largely “naturalized” mode (human influences appear only indirectly via training gauges) so that satellite data could independently reveal human impacts rather than assuming them a priori.
• SWOT height skill is strong (68th- percentile error of ~16 cm in detecting height changes in the validation set).

Broader implications

Current global river models are limited by simplified channel geometry, incomplete representation of dams and water withdrawals, uncertain snowpack timing, and sparse data on human operations.

SWOT provides “detail at scale”, direct observations of real river complexity (dam operations, multi-channel partitioning, abrupt stage changes, infrastructure constraints, etc.) that most models cannot currently reproduce.

The authors argue for a revived empiricism: use SWOT measurements first and trust the observations, while also adapting model structures so they can assimilate realistic hydraulic states rather than relying on oversimplified assumptions.

The paper positions SWOT as a tool that both diagnoses existing model weaknesses on a reach- by- reach basis and supplies the primary data needed to build the next generation of more realistic global river models.

(The full article is open access on the Wiley Online Library page you linked. Supporting Information is also available there.)

Journal information: Geophysical Research Letters Volume 53, Issue 16, e2026GL124323

DOI: 10.1029/2026gl124323

Provided: University of Massachusetts Amherst

First published: 24 August 2026

Authors: Colin J. Gleason, Paul D. Bates, Michael T. Durand, Dongmei Feng, Augusto Getirana, Peirong Lin, Hind Oubanas, Tamlin M. Pavelsky, Chaopeng Shen, Laurence C. Smith, Jida Wang, Dai Yamazaki, Yuan Yang, Heejin An, Fiona B. Bennitt, Stephen J. Chuter, Chenqi Fang, Jonathan A. Flores, Elisa Friedmann, Sreelekha Jarugula, Haoyu Ji, Theodore Langhorst, Di Long, Dung T. Vu, Ziyun Yin, George H. Allen, Konstantinos M. Andreadis, Sylvain Biancamaria, Casey M. Brown, Cédric H. David, Ming Pan, Jeffrey C. Neal, Stephen P. Coss, Merritt E. Harlan, Peyman Saemian, Travis T. Simmons, Nikki Tebaldi, M. J. Tourian

Abstract

Given the insurmountable challenge of measuring all rivers in situ, global river models serve as the foundation of freshwater knowledge past, present, and future. We adopt an inductive empirical framework based on Surface Water and Ocean Topography (SWOT) satellite measurements to assess these models. After controlling for SWOT data quality, we examine 68,347 individual river reaches representing ∼38% of global discharge. We find river models currently struggle in areas of heavy economic development, multi-channel rivers, arid areas, and many Arctic rivers. After controlling for these expected errors, we find better skill as rivers get wider and that parts of Siberia and China are particularly difficult to model. We also find large variability and spatial heterogeneity to model performance, resisting oversimplification. Our results suggest that leveraging SWOT observations within river models will improve them, but river models must adapt their structure to represent realistic hydraulics to do so.