Abstract

This research explores the efficacy of machine learning and deep learning models in predicting soil moisture, a critical factor in optimizing agricultural irrigation systems. Utilizing data from the Vantage Vue weather station and Watermark 200 SS soil moisture sensors, we conducted a comparative analysis of traditional models like RandomForest and MLP against advanced deep learning models, particularly LSTM and 1D Convolutional Neural Networks, enhanced with attention mechanisms. The study reveals that attention-augmented models, especially the CONV1D+Attention model achieving an R2 value of 0.51, excel in capturing the complex dynamics of soil moisture. These results underscore the potential of such models in handling complex timeseries data in contexts like soil moisture levels influenced by weather conditions, offering significant insights for improved water management and sustainable agricultural practices globally.