Tropentag, September 10 - 12, 2025, Bonn "Reconciling land system changes with planetary health"

Estimating the optimum frequency of agricultural technology use using observational data and machine learning

Center for Development Research (ZEF), Univeristy of Bonn, Economcis and Technolgical Change, Germany

Abstract

Efficient use of agricultural technologies is critical for enhancing productivity and sustainability. This study investigates the optimal application frequency of a land preparation technology—laser Land Levelling (LLL) using observational data and machine learning.. LLL is a precision agriculture technology that enhances water use efficiency and crop productivity. Despite its proven benefits, the optimal frequency of LLL application remains unclear, with recommendations varying widely. This study addresses this knowledge gap by estimating the optimal frequency of LLL using observational data and a machine learning-based policy learning approach. We surveyed 1021 rice-farming households across four districts in Punjab, India—an area facing acute groundwater depletion. Plot-level and household-level variables were collected to analyse rice yield outcomes across LLL frequencies. Using the policytree algorithm from the generalised random forest framework, we assessed the average treatment effects of different LLL frequencies on rice yield. The findings suggest that the optimal LLL frequency is once every three years for the majority (71.8%) of the sample. While annual levelling slightly increased yield, it also incurred higher cumulative costs and time requirements, offering diminishing returns. The results indicate that overuse of LLL provides limited agronomic benefits while increasing financial and logistical burdens and potentially limiting access to shared equipment. This study demonstrates the utility of machine learning in technology adoption assessments where experimental data are unavailable. The findings provide actionable insights for farmers, extension agents, and policymakers aiming to balance productivity with resource efficiency. Validation through long-term agronomic trials is recommended to support the scalability of this approach.

Keywords: Frequency, machine learning, precision agriculture, technology adoption

Contact Address: Subash Surendran Padmaja, Center for Development Research (ZEF), Univeristy of Bonn, Economcis and Technolgical Change, Genscherallee 3, 53113 Bonn, Germany, e-mail: subashsparuni-bonn.de