The field is now being asked to keep its own harvest forecast
the field is now being asked to keep its own harvest forecast: an attention model fuses years of satellite imagery with soil and weather to predict yield across space and time, an lstm reads sentinel ndvi and daily weather for a six week forecast within 6.3 percent error, and a usda pilot showed custom ai on fused satellite and drone data hitting 87.4 percent accuracy.
the farm now files its own yield estimate.
Context
The April 2026 Attn-CropNet preprint describes temporal attention over satellite, daily weather and soil inputs. Its methods text still contains placeholder study-region, climate and crop fields, plus an unspecified tile count. It is useful as a proposed architecture, but its reported performance cannot be treated here as a fully documented field result.
Codeworth’s HarvestIQ case-study page describes an LSTM using weekly Sentinel-2 vegetation indices, daily ERA5 weather, soil, agronomic inputs and prior yield. The vendor reports a six-week horizon and 6.3% mean absolute percentage error on a 2025 season holdout. These are vendor figures, not independently audited.
A separate September 27, 2024 Remote Sensing paper studies county-scale winter-wheat forecasting in China with Sentinel-2 and weather forecasts. It reports optimal performance 40 days before harvest using 25 days of future weather forecasts, with RMSE of 0.496 tonnes per hectare.
The 6.3% is mean absolute percentage error, not a guaranteed plus-or-minus error bound for every field. It is a vendor case study rather than the same experiment as the attention preprint or China wheat study. The latter uses an LSTM-GNN model and a different error metric.
The USDA 87.4% pilot claim was not established in the primary sources read and remains unsupported here, not refuted. A USDA ARS project page describes UAS-based biomass estimation research, while a satellite/UAS fusion paper with USDA coauthors studies cover-crop biomass and nitrogen. Neither establishes that specific pilot result or a harvest-yield classification accuracy.
Watch next
- A complete study protocol and data identifiers for Attn-CropNet. Independent HarvestIQ holdout results by field and crop. The primary source and exact metric behind the USDA 87.4% claim.
Sources
- Attention-based Multi-modal Deep Learning Model of Spatio-temporal Crop Yield Prediction with Satellite, Soil and Climate Data (preprint with method placeholders)arxiv.org
- Precision Agriculture with ML: Satellite NDVI, LSTM Models and Drone Analysis (Codeworth vendor case study)codeworth.uk
- A Deep-Learning Network for Wheat Yield Prediction Combining Weather Forecasts and Remote Sensing Datamdpi.com
- AI Innovations for UAS-Based Biomass Estimation and Regional Extrapolations in the Gulf Atlantic Coastal Plain (USDA ARS project)ars.usda.gov
- Learning to See More: UAS-Guided Super-Resolution of Satellite Imagery for Precision Agriculturearxiv.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 06:46 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 →