Soil Moisture Prediction Using Machine Learning Techniques

In the Texas High Plains, water is mainly obtained through groundwater pumping from the Ogallala Aquifer. However, this source of water is being depleted faster than it can be replenished creating a groundwater shortage. The main draw from the Ogallala Aquifer in the Texas High Plains region is for irrigation. This research aims to begin creating a series of models with the end goal of producing a tool producers can use to predict when irrigation is necessary or not using an economic threshold for irrigation. By using a soil moisture predictive model in a tool producers can use to determine irrigation necessity, the goal is to allow producers to become more profitable while focusing on water use and allocate their resources more effectively. The soil moisture predictions will allow the final tool to determine whether irrigation is needed to help maintain the growth of the crop or if it is not timely and will cost more than the producer will profit from irrigation use. Initial outcome of the model created in this research would be to create a baseline for the irrigation threshold under West Texas environments, thus, allowing for irrigation use to be optimized on farms based on profitability. The first step in building this tool is to utilize machine learning techniques to develop models using data on weather, soil characteristics, and existing stocks of soil moisture that best predict future soil moisture. The value of additional data is analyzed to determine best predictions for use in optimization and management decisions. In the long term, the developed model will be further tested using different management systems on various fields to better deliver a robust and flexible model for the enhancement of regional water use efficiency.