1. Product Description:
The 30-meter rice coverage product for the Asian monsoon region is generated using the geospatial foundation model jointly developed by NASA and IBM.
The overall validation accuracy is 84.14%, with variations across different climate zones. The Continental zone exhibits the highest accuracy at 90.06%, while the Arid zone tends to misclassify other crops as rice. In the Tropical zone, errors arise due to cloud interference. Detailed descriptions can be found in the associated article.
We recommend that users evaluate the product using independently collected validation points before proceeding with further applications.
In the product, pixel value 0 represents non-rice areas, while 1 indicates rice.


2. Code Instructions:
This work is built upon the Prithvi model jointly developed by NASA and IBM, relying on their open-source program.
They provide detailed instructions regarding the setup of the runtime environment and related configurations; please refer to their code repository for more information.
The code we provide is adapted from theirs to suit rice mapping. The specific steps are as follows:
(1) Configure the environment according to the NASA/IBM open-source program;
(2) Replace the corresponding files in the you_path/hls-foundation-os/geospatial_fm directory with the code we provide;
(3) Refer to the rice_config.py we provide, modify your configuration file based on your data and model, and then perform model fine-tuning.
Fine-tuning command example: mim train mmsegmentation you_path/config.py --launcher pytorch --gpus 1


3. Contact Information:
If you have any questions regarding the product or code, please contact the author via email (fanghusheng@whu.edu.cn). 
We appreciate your feedback to help us further improve the results. Thank you.