Harnessing deep learning to forecast local microclimate using global climate data.

Area of Science:

• Environmental Science
• Agricultural Science
• Data Science

Background:

• Microclimate understanding is crucial for natural resource management and optimizing agriculture.
• Accurate microclimate monitoring relies on models integrating physical laws and data-driven algorithms.
• Local meteorological data is preferred, but global climate datasets are used in their absence.

Purpose of the Study:

• To introduce an innovative microclimate model combining physical laws and deep learning.
• To compare the accuracy of microclimate modeling using local meteorological data versus global climate data (ERA5).
• To assess the viability of global climate datasets for microclimate modeling when local data is scarce.

Main Methods:

• Developed a novel microclimate model integrating physical laws and deep learning techniques.
• Utilized meter-scale temperature and relative humidity data from the Lombardian foothills.
• Conducted a comparative analysis using local ARPA meteorological station data and the global ERA5 dataset as model inputs.

Main Results:

• Microclimate modeling achieved higher accuracy when using local meteorological data.
• The global climate dataset ERA5, while less accurate than local data, provided a viable alternative.
• The model successfully reproduced temperature and relative humidity variations at the meter-scale.

Conclusions:

• Local meteorological data yields superior accuracy for microclimate modeling.
• Global climate datasets like ERA5 are a reliable option for microclimate modeling in data-scarce regions.
• The developed deep learning model effectively captures microclimate dynamics.