Artificial intelligence and deep learning models will be used to identify latent patterns in high-frequency sensor data. The sensor data will be aggregated in cells of a grid per timespan thus creating maps of sensor data. The first stage of the model uses convolution and recurrent layers to create dense representations of the input data, thus encoding both spatial and temporal patterns. The second stage of the model uses this representation to predict the salinization levels of the area in a grid of cells. This model can and will be used both for the long-term forecasting of salinization levels and for the short-term nowcasting of salinization in areas with sparse sensors. The framework that will be used is python/tensorflow/keras. Deep learning is very demanding in computations so new generation GPUs (like Nvidia TitanXP) will be used for faster processing of the data. Interaction with this model will be achieved by exposing the functionalities with a REST API.