Estimation of soil salinity in rice fields via satellite imagery and machine learning

Undoubtedly, one the most essential parts in agriculture nowadays is the ability of the farmer to access information about their field. There are plenty of solutions that a farmer can use nowadays, including ground or air sensors connected to IoT devices to receive data to desktop, laptop or mobile phones. As a result, farmers are given the opportunity to make decisions about their next moves in order to prevent potential damage on their property caused by either extreme weather phenomena or conditions including humidity, extreme temperature, moisture, pests. One critical and costly factor when discussing the cultivation of hectares of land is the use of water, which is crucial during irrigation for two reasons. The first one is that excessive usage of water means increased cost that could be avoided otherwise; and second, precision regarding the time of irrigation, because excessively saline soil in Rice fields can have a devastating effect on yield. Although there are techniques where in-field sensors can estimate soil salinity, it might be too costly and time-consuming in the cases of cultivating hectares of land. The current approach scrutinizes the use of remote sensing and more specifically processing of satellite image where the user can input images and receive a soil estimation on the farm through an application, without installing in-field sensors. Machine learning algorithms were used to estimate soil salinity in unseen images (tested) and use of Multivariate Linear Regression using IoT sensor reading in order to estimate soil salinity. This application has been designed with the use of statistical models, including single regression and mutlivariate regression. In order to get better results, min-max normalization in the satellite images were used. Moreover, cloud filtering thresholds were realized in the satellite images in order to completely remove clouds from them, as trying to get better accuracy.