Analysis of Error in a Model Predictive Irrigation Controller

Abstract

Significant portions of the world’s agricultural land are vulnerable to desertification, leading to water shortages and changing climate conditions. Smart irrigation controllers could be part of the solution by helping farmers save water and adapt to changing climate without sacrificing yield. This thesis presents an analysis of sensitivity to crop model parameters in the MIT GEAR Lab’s new POWEIr irrigation controller with the goal of making it cheaper and easier to deploy and therefore more accessible. The analysis shows that, of the four crop parameters, the controller is most sensitive to the crop coefficient (K subscript c), moderately sensitive to the maximum rooting depth (Zᵣ), less sensitive to depletion fraction (f subscript d), and almost completely independent of the the yield response factor (K subscript y). This result is potentially useful for designing calibration procedures for the deployment of the POWEIr Controller, especially where there may be limited ability to calibrate the controller.