Telling one crop from another by satellite is the first step in mapping acreage, yield and risk. We tested three leading time-series models on grapes and strawberries in California, using Sentinel-2 images taken every few days across the season.
Research question
Grapes and strawberries are high-value crops, but most satellite crop models are built and tested on wheat, maize and rice. It was not clear which modern model works best for specialty crops, or whether one model is enough.
Methods
Three stages: label, cut, compare.
Results
AR-STAR was best on grapes, with SITS-BERT++ close behind.
Share of test samples classified correctly.
Data: Tables II and III of the paper.
Results
PSE-LTAE was best on strawberries but weakest on grapes.
Balance of precision and recall. Higher is better.
Data: Tables II and III of the paper.
Implications
The right model depends on the crop, so a mapping system should be able to switch models per crop. Satellite readings of each field also feed the environment inputs Jeeva uses to predict crop performance.
See JeevaThe paper
The full text, with methods and results, is on the publisher’s site.
Shravan, G. S., Raj S, A., Kulvanth S, S., Vidyasager, G. R., Srisailam, S., Pankajakshan, P., Rajamani, K. T., & Bhaskaran, V. (2025). Benchmarking SITS-BERT++, PSE-LTAE, and AR-STAR models for crop classification: A case study on grapes and strawberries in California. IEEE India Geoscience and Remote Sensing Symposium (InGARSS). https://doi.org/10.1109/InGARSS67683.2025.11583883
Related research
A spectral–spatial classifier trained on 1 to 10% of labels, with at most two tuned settings, that transfers across scenes without retraining.
Sensors (MDPI) · with IIT MadrasRead the deep dive
Two cultivars under four red–blue LED regimes at 16 and 18 h: crop duration fell by 17 days and flowering came up to 27 days earlier.
J. Advances in Biology & Biotechnology · with Malla Reddy UniversityRead the deep dive
Field screening of 158 accessions over two sowings identified resistant wild derivatives and the weather that drives disease.
J. Horticultural Sciences · with ICAR-NBPGRRead the deep dive
Tell us your crop and what you want the model to answer. We will reply with the model card and next steps.
