Artificial intelligence to predict soil temperatures by development of novel model

Area of Science:

• Environmental Science
• Agricultural Science
• Data Science

Background:

• Soil temperature is crucial for understanding soil properties and achieving food sustainability.
• Developing regions face challenges in soil temperature data collection due to instrumentation issues and natural disasters.
• Uzbekistan's arid climate makes it particularly vulnerable to climate change impacts on agriculture.

Purpose of the Study:

• To develop an integrated model for predicting soil temperature at the surface and 10 cm depth in Nukus, Uzbekistan, using climatic factors.
• To identify the best-performing machine learning model for soil temperature prediction in arid environments.
• To enable soil temperature prediction without direct ground measurements for improved agricultural planning.

Main Methods:

• Utilized eight machine learning models, including Long Short-Term Memory (LSTM), to predict soil temperature.
• Trained models using climatic data and evaluated performance based on standard indicators.
• Focused on predicting soil temperature at the surface and 10 cm depth.

Main Results:

• The Long Short-Term Memory (LSTM) model demonstrated superior accuracy in predicting soil temperature at 10 cm depth.
• The developed models can predict 10 cm soil temperature using climatic data and predicted surface temperature.
• Soil temperature at 10 cm depth can be accurately predicted without requiring ground-based soil temperature measurements.

Conclusions:

• The developed machine learning models, particularly LSTM, offer a reliable method for predicting soil temperature in arid regions.
• This predictive capability can significantly aid in agricultural planning and enhance food production sustainability in data-scarce areas like Nukus.
• The models provide a valuable tool for environmental monitoring and climate change adaptation strategies in agriculture.