Rice cultivation in Sri Lanka mapped with new optical remote sensing breakthrough

Researchers from the University of Wisconsin–Madison, MIT, and the World Bank have developed a pioneering remote sensing method to map rice fields and estimate yields in Sri Lanka with over 90% accuracy. The study reveals two decades of cultivation trends, including sharp yield declines during the 2021 fertilizer ban, offering policymakers powerful tools for food security planning.

Rice, the lifeblood of Sri Lanka’s food security and cultural identity, has become the focus of a groundbreaking study by researchers from the University of Wisconsin–Madison, the Massachusetts Institute of Technology (MIT), and the World Bank. Their collaborative work, published as a World Bank Policy Research Working Paper, introduces a novel method for mapping rice cultivation and estimating yields at the field scale using optical remote sensing and limited training data. In a country where rice consumption is nearly twice the global average and agriculture remains central to economic and political life, this innovation promises to transform how policymakers and farmers understand crop performance. The study covers a remarkable two decades, from 2000 to 2022, providing insights into how weather, management, irrigation, and even controversial policy shifts have shaped Sri Lanka’s rice sector.

From Surveys to Satellites: Overcoming Monitoring Gaps

While Sri Lanka’s government conducts annual paddy surveys that provide solid statistics at national and district levels, these surveys cannot capture the finer realities of smallholder systems that dominate the island’s landscape. Questions such as how rainfall variation, irrigation schedules, pests, or fertilizer availability affect yields at the village or field scale have long gone unanswered. These gaps limit understanding of agricultural water use, the impacts of climate change, and methane emissions from flooded fields. Remote sensing has been viewed as a solution, but Sri Lanka’s persistent monsoon cloud cover and its patchwork of tiny fields have made mapping difficult. The new study addresses these challenges by tailoring optical satellite data from Landsat and Sentinel-2 to the phenology of rice, isolating the signals of flooding during transplanting and greening during growth.

A Phenology-Driven System with Machine Learning Power

At the heart of the research lies a two-step methodology. First, rice mapping was achieved through a rule-based classification system that combined vegetation and water indices, then applied automated thresholds to separate rice fields from other land uses. This expert-driven approach allowed accurate identification even in fragmented landscapes. Second, yield estimation was performed using a random forest regression model trained with government crop statistics and a satellite-derived chlorophyll index, supplemented by climate variables such as rainfall, temperature, and vapor pressure deficit. The combination allowed the researchers to generate yield maps at 30-meter resolution nationwide for over twenty years. Accuracy reached more than 90 percent for area mapping, while yield predictions achieved error margins below 20 percent, a remarkable feat in such a complex environment.

Clear Patterns and Policy Lessons

The results highlight both expected and surprising patterns. National rice area estimates derived from satellites closely tracked official survey data, particularly during the Yala season when skies are clearer. District-level comparisons also showed strong consistency, although smaller districts with fragmented fields displayed larger divergences. Yield maps revealed average production ranging between 1,700 and 7,300 kilograms per hectare, reflecting geography and climate: higher outputs in irrigated lowlands and weaker yields in elevated interior regions. The analysis also provided rare insight into Sri Lanka’s 2021 ban on synthetic fertilizer imports. While cultivated area remained stable, since farmers had already committed to planting, yields collapsed by as much as 40 percent in some districts during the 2021–2022 Maha season. Detailed maps of Hambantota, for example, showed stable yields in 2021 but steep declines the following year, starkly illustrating the policy’s immediate impact.

Toward a New Era of Agricultural Monitoring

Validation confirmed the system’s reliability. Stratified samples checked against high-resolution imagery showed accuracies of 91 percent in the 2020–2021 Maha season and 93 percent in the 2021 Yala season. Errors were evenly distributed, preventing systemic bias. Comparisons with crop-cutting experiments across eight districts revealed strong overlaps between observed and modeled yield distributions, adding credibility. Beyond the numbers, the study demonstrates that even in cloud-heavy tropical regions, frequent optical observations combined with phenology-based algorithms can overcome traditional barriers. While deep learning methods have gained popularity in crop mapping, they demand vast training datasets rarely available in developing economies. By contrast, this approach delivers scalability, consistency, and historical depth with minimal ground data.

The implications extend far beyond Sri Lanka. Other nations in South and Southeast Asia face similar obstacles in monitoring rice, a crop vital to over half the global population. The ability to generate long-term, high-resolution, and spatially explicit data offers transformative possibilities for irrigation planning, fertilizer management, climate adaptation, and even carbon accounting linked to methane emissions. For policymakers, the availability of such granular insights means interventions can be better targeted, and the impacts of climate shocks or policy experiments more precisely assessed.

The study represents a rare achievement in agricultural science: the creation of a simple, cost-effective, and long-term system that not only reproduces official statistics but adds detail at the field scale. As Sri Lanka grapples with climate change and evolving agricultural demands, the marriage of optical satellite technology, expert-driven classification, and machine learning offers a new frontier in food security planning. It gives policymakers the power to see beyond national averages into the realities of individual farmers’ fields, helping to build resilience in one of the world’s most important staple crops.