Uso de modelos preditivos para análise da presença de sódio no solo no município de Dom Pedrito-RS

Abstract

Agriculture is of utmost importance as it is an activity that affects the economy of the community as a whole. Due to this, various techniques have been developed over the years to observe and indicate potential problems in crops. The objective was to use remote sensing techniques combined with a predictive model based on Machine Learning to detect saline areas. Geology was a great ally in the analyses considering that these saline areas are sodium hotspots. A predictive model was implemented using a dataset extracted from Sentinel-2A satellite images with 8 bands. Each pixel was recorded containing the 8 layers of the MSI sensor, thus creating training data for two hypotheses where the model assigns "1" to saline areas and "0" to others. The results showed that 85% accuracy was able to identify some hotspots and may be useful in directing field visits to validate the information. More work needs to be done to verify if the traditional remote sensing method can be more effective in detecting salinized soils.