Advanced Remote Sensing for Climate-Smart Rice Practices: Mapping DSR and TPR in India

• Published in: 2024 IEEE India Geoscience and Remote Sensing Symposium (InGARSS) pp. 1 - 4
• Main Authors: Konkathi, Preethi, Torbick, Nathan, Kumar, Vivek, Huang, Xiaodong, Mishra, Suyash, Pandey, Shivang, Ajmera, Ishan, Adhvaryu, Chinmay, Munoz-Basta, Edgar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 02.12.2024
• Subjects: Accuracy; Adaptation models; Data models; DSR; Farming; Measurement; Normalized difference vegetation index; Predictive models; Random forests; Remote sensing; rice cultivation; seeding practices; Sentinel-1; sustainability; TPR
• DOI: 10.1109/InGARSS61818.2024.10984185

India is known for its vast cultivation of rice crops, especially during the Kharif season. The type of seeding-Direct Seeded Rice (DSR) or Transplanted Rice (TPR) typically influences both crop yield and environmental sustainability. This research presents the characterization of seeding practices adopted in Punjab using a combination of remote sensing data from various sources and advanced machine learning technologies. Based on the field data gathered from Mitti Labs projects sites, comprehensive analysis is offered on various seeding techniques and their associated agricultural practices observed across diverse paddy field environments. The remote sensing datasets used consist of Sentinel-1 C band Synthetic Aperture Radar polarimetric information and Sentinel-2 optical indicators like Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVl), Land Surface Water Index (LSWI) and SoilAdjusted Total Vegetation Index (STAVl). These indices offer information on soil moisture levels and the health of and metrics related to crop growth. We included factors such as soil characteristics from SoilGrids V2 and information from the Digital Elevation Model (DEM) to gain a deeper insight into the factors that impact decisions, on seeding practices.We trained and tested a Random Forest model to differentiate between two types of practices - DSR and TPR using data collected during the Kharif season starting from May 20, 2024, to August 20, 2024. The model showed results with an Out of Bag (OOB) accuracy score of 0.709, an overall accuracy of 0.709, and an F1 score of 0.708, indicating that the developed model could effectively classify seeding practices based on the collected data. This study highlights the accuracy of predictive models by using multi-sourced remote sensing with machine-learning technologies for sustainable and climate-smart rice farming in the Punjab region. Further developments are in the pipeline to improve the model by the end of the season to adapt to changing crop growth patterns over time while incorporating advanced deep-learning methods for different parts of India and South Asia.