A machine learning and model-agnostic approach to identify the factors potentially explaining the use of agricultural land

Keywords: Modelling, Machine learning, Model agnostics, Agriculture
Abstract Type: Paper Abstract

Abstract

To effectively plan and manage the use of agricultural land, it is crucial to identify and evaluate the multiple human and environmental factors that influence it. In this study, we propose a model framework to identify the factors potentially explaining the use of agricultural land for wheat, maize, and olive grove plantations at the regional level. By developing a machine-learning model coupled with a model-agnostic approach, we provide global and local interpretations of the most influential factors. We collected nearly 140 variables related to biophysical, bioclimatic, and agricultural socioeconomic conditions. Overall, the results indicated that biophysical and bioclimatic conditions were more influential than socioeconomic conditions. At the global interpretation level, the proposed model identified a strong contribution of conditions related to drainage density, slope, and soil type. In contrast, the local interpretation level indicated that socioeconomic conditions such as the degree of mechanisation could be influential in specific parcels of wheat. As demonstrated, the proposed analytical approach has the potential to serve as a decision-making tool instrument to better plan and control the use of agricultural land.