Crop plot segmentation based on soil surface temperature dynamics

The world’s growing population demands more food with high nutritious content, even though the agricultural fields are already under too much stress. This paper introduces a method for segmenting crop plots meaningful in terms of moisture availability for plant growth. In this approach, we characterize the dynamics of soil surface temperature of a fallow crop plot during the sunlight hours using UAS and reference values from i-buttons. Afterward, based on its characterization, we conduct an unsupervised clustering method to segment the plot area. Next, we employ machine learning methods to infer the soil classes at the early stages of crop growth at high resolution from UAS and lower resolution from multispectral satellite images. The neural network classifiers using data obtained with UAS have a ROC AUC above 0.67. The regressors using data from satellite imagery predict the percentage of pixels corresponding to soil classes with an RMSE below 0.17. The results show that the proposed methodology is sound and allows for surface clustering based on the dynamics of soil surface temperature. [Display omitted]