Chapter 18 - Extremely randomized trees versus random forest, group method of data handling, and artificial neural network

Abstract

Understanding the relationship between soil temperature (Ts) and air temperature (Ta) is become of great importance, and a great deal of research is undertaken to demonstrate the strong correlation between the two variables. In a major part of the studies conducted previously, the Ts was linked to the Ta via a large amount of variables in the presence of other ones, that is, relative humidity, solar radiation, wind speed, and their accuracy were enhanced. However, modeling Ts using only the Ta is rarely reported in the literature. This chapter describes new and original approach for estimating soil temperature using only air temperature. The proposed models require two variables: soil temperature measured at the same time, and the periodicity calculated as the year, month, day, and hour of day. Four machines learning were proposed and compared: (i) extremely randomized trees (ERT), (ii) random forest regression, (iii) group method of data handling, and (iv) the standalone artificial neural network. The models were developed using data collected at the USGS website, and measured at 15-min time interval. The accuracies of the models was evaluated using several error criteria namely, mean absolute error, root mean square error, Nash-Sutcliffe efficiency (NSE), and correlation coefficient (R). The results demonstrated that the soil temperature values calculated using the machines learning models were highly correlated with the measured in situ soil temperature, and the ERT model exhibited high accuracy, slightly better than the other models, with R and NSE nearly equal to 0.999 and 0.999, during the validation phase. However, even though the Ts calculated using the linear regression model (MLR) were not highly correlated with the in situ Ts , the accuracy of the MLR model was still acceptable (R = 0.891, NSE = 0.793). Furthermore, in overall, the ERT model was the most accurate approach.