Comprehensive Climate, Soil, Agronomic, and Management Dataset for Data-Driven Cotton Irrigation in NW China

This paper introduces a unique agricultural multimodal dataset that integrates climate and weather data, real-time soil sensor data (moisture, temperature, and fertility), crop management records, visible and multispectral images of cotton plant, spore and insect images, agronomic and yield data, specifically designed for cotton yield prediction and smart irrigation in the semi-arid climate of northwest China. To validate its utility, we employed multiple machine learning approaches (reinforcement learning, DNNs, XGBoost, MLP, and random forests) to model irrigation-yield dynamics and optimize irrigation strategies. Simulation experiments and field validation show that accurate quantitative irrigation can achieve maximum crop yield or maximum water productivity, underscoring the dataset's potential for data-driven agricultural research, smart decision-making, and sustainable agricultural practices.