Predicting Soil Temperature: A Machine Learning Battle

Soil temperature (ST) plays a crucial role in understanding environmental conditions and supporting crop growth. In a recent study, scientists used four machine learning methods to estimate daily ST at different depths (5 cm, 15 cm, and 30 cm) in Bathinda, India, between 2016 and 2019. The methods included random forest (RF), radial basis neural network (RBNN), multi-layer perceptron neural network (MLPNN), and co-active neuro-fuzzy inference system (CANFIS).

Four meteorological factors were considered: mean air temperature (Tmean), relative humidity (RH), wind speed (WS), and bright sunshine hours (SSH). Researchers combined these factors in different ways and chose the best one using the gamma test (GT). The machine learning models were then evaluated using performance metrics like mean absolute error (MAE), root mean square error (RMSE), and others. Interestingly, the CANFIS model performed the best at all depths. It had the lowest errors and highest efficiency scores. This shows that CANFIS, with inputs from Tmean, RH, WS, and SSH, is very effective in estimating daily soil temperature at different depths.

Actions