The snowpack is now being asked to keep its own water ledger
the snowpack is now being asked to keep its own water ledger: a physics guided lstm trained on globsnow v3.0 estimates daily snow water equivalent across the whole northern hemisphere at half a degree, a physically driven lstm learns continental u.s. swe from weather alone, and an lstm data assimilation matches an ensemble kalman filter on seven sites.
the snow now files its own water budget.
Context
An AGU paper builds a hybrid physics-guided model that augments Community Land Model 4.0 snow water equivalent (SWE) simulations with an LSTM, trained on GlobSnow v3.0 and forced with meteorological data, to estimate daily SWE at 0.5 degrees over the Northern Hemisphere. It says the hybrid outperforms both the physical model and a pure ML model on magnitude, timing and anomalies.
A NASA-affiliated arXiv paper builds an LSTM that estimates SWE over the continental United States from time series of physical and meteorological inputs plus static spatial factors, split into a snow presence classifier and a regression for SWE. It is trained on SNOTEL snow pillow measurements in the western United States and reports classification accuracy of at least 93%.
The Cryosphere paper uses LSTM networks for snow data assimilation across 7 study sites. It finds performance comparable to an Ensemble Kalman Filter, with a small RMSE drop of about 6 mm for SWE and 6 cm for depth, and a 70% reduction in computational time compared with a parallelized EnKF.
The first study matches 'physics guided lstm', 'globsnow v3.0', 'daily', 'northern hemisphere' and 'half a degree'. It is a hybrid that augments a land model, so 'physics guided' is accurate.
The note says a physically driven LSTM learns continental U.S. SWE 'from weather alone'. The paper's inputs include meteorological factors and static spatial and morphological factors, and it is trained on western US SNOTEL sites, so 'from weather alone' is not what the abstract says. That part is loose.
The note says the LSTM data assimilation 'matches an ensemble kalman filter on seven sites'. The paper says 'comparable' with a small performance drop and gains in speed, not equal performance. It also reports that all but 2 of 14 site-specific configurations improved on the open loop run.
Three separate studies on different data and scales. 'The snow now files its own water budget' is the author's framing.
Related work
- Observation-Constrained Physical Snow Water Equivalent Simulations Using a Physics-Guided Machine Learning Model (Water Resources Research, AGU) ↗Source for the hybrid Northern Hemisphere SWE model and its training data.
- A Physically Driven Long Short Term Memory Model for Estimating Snow Water Equivalent over the Continental United States (arXiv 2504.20129) ↗Source for the continental US LSTM, its inputs and the SNOTEL training data.
- Learning to filter: snow data assimilation using a Long Short-Term Memory network (The Cryosphere, 2025) ↗Source for the LSTM data assimilation results against an EnKF.
Watch next
- Check the inputs listed in the continental US LSTM paper against the 'weather alone' wording. Look for an evaluation of the hemisphere model outside its training period.
Sources
- Observation-Constrained Physical Snow Water Equivalent Simulations Using a Physics-Guided Machine Learning Model (Water Resources Research, AGU)agupubs.onlinelibrary.wiley.com
- A Physically Driven Long Short Term Memory Model for Estimating Snow Water Equivalent over the Continental United States (arXiv 2504.20129)arxiv.org
- Learning to filter: snow data assimilation using a Long Short-Term Memory network (The Cryosphere, 2025)tc.copernicus.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 21:48 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →