AI- and data-driven pre-crop values and crop rotation matrices

Abstract

Crop rotation planning, being an essential prerequisite for organic farming, involves determining the species and timing of crops on farmland to improve soil quality, crop yield, and resistance to pests and weeds. Pre-crop values and crop rotation matrices describe the effect of a crop on the next crop, mediated through the soil. The identification of these effects in traditional long-term field studies is resource intensive. Within this paper we present AI4CROPR, a method to identify pre-crop values and crop rotation matrices using Normalized Difference Vegetation Index (NDVI) data from remote sensing, clustering, and artificial intelligence. Our method uses 24.352 unique crop rotations prevailing on plots in Lower Austria from 2017 to 2021. We restricted the crop rotations to the 26 most used crop types, which represent about 95 % of the crops grown in the area. For each plot and year, we estimated yield potential using the Normalized Difference Vegetation Index (NDVI) from Sentinel-2 data. AI4CROPR enables the data-driven estimation of pre-crop values and creation of crop rotation matrices for entire regions based on their specific conditions and without the need to manually survey individual farms or plots. Validation has shown that results of the data- and AI-driven AI4CROPR method overlap to a great extent with recommendations from literature (28.20 % of the measured pre-crop values are identical to literature recommendations, 51.60 % deviate by one degree, and 19.67 % deviate by two degrees) and are suitable to extend the work to further regions and integrate them in crop rotation decision support systems.