We are excited to share the publication of our latest paper “The Future of Vineyard Irrigation: AI-Driven Insights from IoT Data”, published in Sensors (MDPI). The study leverages Internet of Things (IoT) and artificial intelligence (AI), aiming to contribute to sustainable agricultural practices by enhancing irrigation management.
Authors: Simona Stojanova, Mojca Volk, Gregor Balkovec, Andrej Kos in Emilija Stojmenova Duh
Link to the paper: https://www.mdpi.com/1424-8220/25/12/3658#sec4-sensors-25-03658
Paper highlights
This research investigates the impact of integrating the IoT and AI-driven predictive models on improving irrigation management. It demonstrates how deep learning (DL) techniques can be effectively applied in agro-environmental contexts. Specifically, it presents a Long Short-Term Memory (LSTM)-based machine learning (ML) model, designed to forecast future irrigation needs by integrating diverse datasets, collected from a small-scale commercial estate vineyard in southwestern Idaho, the United States of America (USA). More precisely, the dataset consists from:
- Soil moisture
- Climatic variables: air temperature, relative humidity, solar radiation, wind speed and wind direction metrics.
- Historical irrigation records
Through this analysis, the research shows several contributions on this topic:
- DL techniques are better than traditional ML models in regard to processing time-series sequential data, hence showing better overall performance results;
- the performance of the LSTM model was affected by the number of inputs, i.e., the dataset, meaning that integrating a more diverse dataset can affect the performance of the model;
- unlike many studies, the focus is on the irrigation factor and irrigation prediction, instead of other influencing factors.
By developing, training, and testing an LSTM model, we achieve a balance between the simplicity and performance of the model, achieving high predictive accuracy. The reduced model complexity provides possibilities for easy implementation, making it suitable for real-world use. This work proves that effective irrigation forecasting can be achieved without the need for complex analyses, offering an efficient solution for farmers.
