Establishing management zones for irrigation using soil properties and Remote Sensing

Soil texture influences many other soil attributes, including its physical, chemical, and biological characteristics. Soil texture dictates vital factors such as aeration, nutrient, water availability, and heat retention. These aspects collectively impact various aspects of plant life, encompassing growth, development, productivity, and quality. Agricultural soils are commonly classified into several categories based on their texture to facilitate effective agricultural practices like tillage, irrigation, fertilization, and pesticide applications. A growing call has recently been made for integrating machine learning (ML) techniques to enhance comprehension and insight into soil behaviour. However, it is essential to note that real-world datasets often exhibit inherent imbalances. In such cases, ML models tend to overemphasize the majority classes while simultaneously underestimating the minority ones. This study aimed to investigate the effects of imbalance in training data on the performance of a random forest model (RF). The original data used in this work was from La Chimenea farm station near Aranjuez (Madrid, Spain). The variables included were Electrical conductivity (EC), EC shape, EC depth, EC ratio, slope, curve, and NDVI derived from Sentinel-2. Clay and sand percentages were obtained with the exact spatial resolution, and the USDA classification was applied based on them. A descriptive statistics analysis was conducted to analyze the data. Then, Pearson's coefficient (r) of linear correlation was calculated to verify possible relations between the different variables. Then, a synthetic resampling approach using the Synthetic Minority Oversampling TEchnique (SMOTE) was employed to make a balanced dataset from the original data. The imbalance and balance data classification were compared to see SMOTE's benefits in better-classifying soil texture. Keywords: digital soil mapping; machine learning; soil texture; imbalance classification; data resampling    Acknowledgements This work has received support from projects PID2021-124041OB-C22 and PID2021-122711NB-C21, funded by the Ministerio de Ciencia e Innovación (Ministry of Science and Innovation).