Predicting readiness of arable land and scheduling of tillage operations using finite-horizon Markov decision process and Gaussian state space models

With the structural transformation of agriculture, it is getting increasingly important to consider the readiness of arable land when planning and scheduling machine operations. This is to ensure that the operations produce positive results to crops and soil, while not causing damage to soil. Taking the readiness of arable land parameters into account paves the ground for executing field operations at opportune times. Determining the readiness of arable land is widely affected by stochastic factors such as soil water content and meteorological information and some variables associated with trafficability, workability, and completion criteria along with soil characteristics. This paper presents the development of stochastic optimization and statistical learning models implemented as a decision model that assists the farmers in optimal timing of tillage operations on a day-to-day basis. The optimization model is a finite horizon Markov decision process modelling execution of tillage operations at opportune time based on information acquired from a mechanistic model simulation of agricultural fields and historical and forecast meteorological data. Moreover, a Gaussian state space model predicts the future soil water content of the field using a Bayesian updating approach, which is embedded into the optimization model. Numerical examples are given to show the functionality of the model. The prediction of optimal timing of two operations, tillage and seeding, has been well aligned with the practical experiences of the selected arable land.