Water & Soil
Reinforcement-Learned Irrigation Scheduling Under Constrained Water Budgets
K. Otieno, J. Lindqvist, F. Adeyemi · Farm Better Research Lab
August 2025 · 10 min read
62
Plots deployed
23%
Water saved
41%
Drainage loss cut
Abstract
A policy trained in a calibrated soil-water simulator schedules irrigation events for horticultural plots operating under weekly water quotas.
Method
The agent was trained in simulation and deployed on 62 plots with soil moisture probes at three depths. Control plots followed conventional fixed-interval schedules.
Findings
- Water use fell 23% with no statistically significant yield difference.
- Deep-drainage losses reduced by 41%, lowering nutrient leaching.
- The policy shifted most irrigation to pre-dawn windows without being instructed to.
- Benefits were smallest on sandy soils where buffering capacity is low.
Conclusion
Learned scheduling delivers real water savings, but soil type should gate deployment: heavy soils benefit most.
Note: this article is an illustrative mock-up created for the Farm Better site and is not a peer-reviewed publication.