In the heart of China’s Liaoning Province, a groundbreaking study is reshaping how we predict soil temperature, a critical factor in Earth science and agricultural production. The research, led by Zihan Yuan from the School of Physics and Electronic Technology at Liaoning Normal University, has been published in the IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. It leverages the power of long short-term memory (LSTM) models and remote sensing data to offer more accurate and stable soil temperature predictions at various depths.
Soil temperature is a key variable in numerous Earth science disciplines, influencing everything from plant growth to microbial activity. However, traditional soil temperature data from meteorological stations are often discrete and discontinuous, limiting their usefulness. To overcome this challenge, Yuan and his team turned to data from three Moderate Resolution Imaging Spectroradiometer (MODIS) products: normalized vegetation index, atmospheric precipitable water, and surface temperature. They combined this data with daily average soil temperature measurements at depths of 40, 100, and 200 cm from the ground surface, collected between 2017 and 2021.
The team established a soil temperature prediction model based on the LSTM model, a type of recurrent neural network particularly well-suited to time series prediction. To enhance the model’s accuracy and stability, they also developed an optimized LSTM model that considered the hysteresis factor of soil temperature relative to the surface temperature. “The hysteresis factor is crucial,” Yuan explains. “Soil temperature doesn’t change as quickly as surface temperature. Our optimized model accounts for this lag, leading to more accurate predictions.”
The results were impressive. The LSTM soil temperature prediction models achieved R² values of 0.86, 0.81, and 0.69 for depths of 40, 100, and 200 cm, respectively, with corresponding RMSE values of 0.30, 3.41, and 3.74 °C. The optimized LSTM models performed even better, with R² values of 0.90, 0.91, and 0.88, and the same RMSE values. Moreover, the standard deviation of the RMSE (SD_RMSE) decreased for the optimized model, indicating improved stability.
So, what does this mean for the agriculture sector? Accurate soil temperature predictions can inform planting and harvesting decisions, optimize irrigation schedules, and even help mitigate the impacts of climate change. As Yuan puts it, “Our model can achieve long-term daily temperature prediction of inter-annual soil temperature. This has significant implications for agricultural production, enabling farmers to make data-driven decisions that can enhance crop yields and sustainability.”
Looking ahead, this research could pave the way for more sophisticated models that incorporate additional factors, such as soil moisture and type. It also highlights the potential of machine learning techniques in Earth science, offering a powerful tool for understanding and predicting our planet’s complex systems. As the agriculture sector continues to grapple with the challenges posed by climate change, such innovations will be invaluable in ensuring food security and promoting sustainable practices.