Machine Learning for Sustainable Rice Production: Region-Scale Monitoring of Water-Saving Practices in Punjab, India

TL;DR

The study tackles the challenge of monitoring water-saving rice practices at regional scales by introducing a dimensional classification framework that separately models sowing (DSR vs PTR) and irrigation (AWD vs CF) using Sentinel-1 SAR time series. Ground-truth data from 1,400 fields in Punjab enables robust evaluation, with results showing the dimensional approach achieving higher $F1$ scores ($F1_{sowing}=0.800$, $F1_{irrigation}=0.740$) than a combined 3-class model, and foundation-model embeddings (Presto, AlphaEarth) further boosting performance. Scaling to ~3 million fields demonstrates spatial heterogeneity in adoption and strong district-level agreement with government records ($\rho=0.69$, $RBO=0.77$), providing a practical tool for policy targeting and climate-action planning. The work highlights the potential of SAR-based, data-efficient monitoring with low ground-truth requirements to inform water conservation, while acknowledging temporal-resolution and boundary-detection limitations and outlining avenues for future SAR missions and regional generalization.

Abstract

Rice cultivation supplies half the world's population with staple food, while also being a major driver of freshwater depletion--consuming roughly a quarter of global freshwater--and accounting for approx. 48% of greenhouse gas emissions from croplands. In regions like Punjab, India, where groundwater levels are plummeting at 41.6 cm/year, adopting water-saving rice farming practices is critical. Direct-Seeded Rice (DSR) and Alternate Wetting and Drying (AWD) can cut irrigation water use by 20-40% without hurting yields, yet lack of spatial data on adoption impedes effective adaptation policy and climate action. We present a machine learning framework to bridge this data gap by monitoring sustainable rice farming at scale. In collaboration with agronomy experts and a large-scale farmer training program, we obtained ground-truth data from 1,400 fields across Punjab. Leveraging this partnership, we developed a novel dimensional classification approach that decouples sowing and irrigation practices, achieving F1 scores of 0.8 and 0.74 respectively, solely employing Sentinel-1 satellite imagery. Explainability analysis reveals that DSR classification is robust while AWD classification depends primarily on planting schedule differences, as Sentinel-1's 12-day revisit frequency cannot capture the higher frequency irrigation cycles characteristic of AWD practices. Applying this model across 3 million fields reveals spatial heterogeneity in adoption at the state level, highlighting gaps and opportunities for policy targeting. Our district-level adoption rates correlate well with government estimates (Spearman's $ρ$=0.69 and Rank Biased Overlap=0.77). This study provides policymakers and sustainability programs a powerful tool to track practice adoption, inform targeted interventions, and drive data-driven policies for water conservation and climate mitigation at regional scale.

Figures (16)

• Figure 1: DSR predictions for Punjab in 2024. A) Map of Punjab with predicted DSR adoption distributions, shown with district boundaries; darker shades denote higher DSR plot-density. B) Comparison between model predictions and Punjab government estimates of DSR adoption by district, ordered by acres of DSR activity. C) District-level adoption compared to government records; evaluated using Spearman’s $\rho$=0.69, R$^2$=0.46, Rank Biased Overlap=0.77 and MAE$\approx12$ thousand acres. Key districts are labeled.
• Figure 2: Map of Punjab showing the study area, Sentinel-1 satellite coverage (ascending and descending orbits), and locations of surveyed plots. Insets show sample plots at different zoom levels, with the innermost inset displaying delineated fields from three classes: Control (plots that use puddled transplanting and continuous flooding), alternate wetting and drying (AWD), and direct seeded rice (DSR). Background satellite imagery from Bing Maps.
• Figure 3: Water management practices across classes, created in collaboration with rice agronomy experts. The $y$-axis represents water level in a rice paddy, and $x$-axis shows growth stages. The growing season naturally separates into two phases: initial (left of dotted line) and main (right), which have distinct irrigation needs. Blue arrows indicate irrigation events, while yellow arrows denote sowing. The different practices create distinct temporal signatures, with the vertical dotted line marking the boundary between sowing and irrigation dimensions.
• Figure 4: Dimensional divisions of dataset by practice type. A) Plots divided by sowing and irrigation dimensions, with number of plots in parentheses. B) Classification of irrigation practices (grouping Control and DSR as CF). C) Classification of sowing practices (grouping Control and AWD as PTR). Dotted red lines in B) and C) show classes are separated in a task-dependent manner.
• Figure A1: Planting date distributions across water management practices of interest.