Prediction model of undisturbed ground temperature using artificial neural network (ANN) and multiple regressions approach

Abstract

The prediction of ground temperature is of utmost importance for evaluating temperature fluctuations in both the surface and subsurface of the ground and is essential for making informed decisions. The prediction of daily ground temperature can be used in a variety of engineering fields, such as in environmental and geothermal systems. The objective of this study is to develop ground-temperature prediction models at various depths, using Multiple Linear Regression (MLR), Nonlinear Regression (NLR), and Artificial Neural Network (ANN) methods. Seven predictor variables, including ambient temperature, daily precipitation, wind speed, relative humidity, local atmospheric pressure, sunshine duration, and solar radiation, provided by the Korean Meteorological Administration (KMA), were utilized. Moreover, a database consisting of 2091 (winter) and 1034 (summer) data for each of the predictor variables and ground temperature were established for the model to predict hourly ground temperature at depths of 0, 1, 3, 5, 10, 15, and 20 m. To evaluate the performance of the prediction models, statistical metrics, including the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and average absolute relative error (AARE), were used. The performance results show that the value of R2 ranges from 65 % to 99 % in the validation phase of each model, indicating that the prediction models accurately predict the ground temperature. Based on statistical methods, the developed ANN model displays an excellent fit with observed field values when compared to the MLR and NLR models.