Empowering Precision Agriculture Practices through Machine Learning: Advanced Microclimate Forecasting

Código

I-EBHE0084

Autores

ULISES ADRIAN MARECOS VARGAS, Fi-Jonh Chang, Pu-Yun Kow, Wei Sun, Meng-Hsin Lee

Tema

WG 2.2: Participatory Water Systems

Resumo

Accurate and dependable microclimate forecasts are crucial for precision agriculture control systems, ensuring optimal conditions for crop growth. In precision agriculture, timely and precise microclimate information enhances decision-making, leading to improved crop yields and resource efficiency. Artificial Neural Networks (ANNs) are prominent in machine learning for environmental analysis due to their ability to model complex, non-linear relationships. Deep learning models, particularly, excel in identifying intricate patterns within spatial and temporal data. This study investigates and evaluates two advanced machine learning techniques: Long Short-Term Memory (LSTM) networks and Backpropagation Neural Networks (BPNN). LSTM networks, a type of recurrent neural network (RNN), are well-suited for time series forecasting due to their ability to capture long-term dependencies. BPNNs, known for their robustness in predictive tasks, also demonstrate significant potential. These models aim to deliver accurate and reliable hourly microclimate weather forecasts, including critical parameters such as internal temperature, relative humidity, and light levels. Accurate forecasting of these parameters is essential for managing greenhouse environments and open-field agricultural practices. Our data is sourced from a comprehensive gridded meteorological dataset collected from the Taiwan Agricultural Meteorological Observation Network Monitoring System between November 1, 2023, and January 10, 2024. This dataset provides a rich source of historical weather data essential for training our models. Additionally, we validate our models using real-time greenhouse internal IoT data from the Taiwan Agricultural Research Institute (TARI) in Central Taiwan. Integrating IoT data enhances the reliability and applicability of our forecasting models in real-world agricultural settings. Our study advances the understanding of machine learning techniques in microclimate weather forecasting. By leveraging advanced deep learning methodologies, we improve the predictive accuracy of microclimate forecasts, significantly benefiting agricultural practices. Moreover, these predictive models have the potential to play a crucial role in various climate-sensitive applications, including energy management, environmental monitoring, and urban planning. Enhancing microclimate forecasts contributes to more sustainable and efficient management of natural resources, ultimately supporting broader environmental and economic goals.