Statistical Models in Estimating Air Temperature in a Mountainous Region of Greece

Abstract

The current work focuses on the estimation of air temperature (T) conditions in two high altitude (alt) sites (1580 m), each one at different orientation (southeast and northwest) in the mountain (Mt) Aenos in the island of Cephalonia, Greece, by using two well-known statistical models, simple linear regression (SLR) and multi-layer perceptron ( MLP), one of the most commonly used artificial neural networks. More specifically, the estimation of mean, maximum and minimum T in high alt sites was based on the respective T data of two lower alt sites (1100 m), the first at southeast and the second at northwest orientations, and was carried out separately for each orientation. The performance of both SLR and MLP models was evaluated by the coefficient of determination (R2) and the Mean Absolute Error (MAE). Results showed that the examined models (SLR and MLP) provided very satisfactory results with regard to the estimation of mean, maximum and minimum T, regarding southeast orientation (R2 ranging from 0.96 to 0.98), with mean T estimation being relatively better, as confirmed by the lowest MAE (0.83). Regarding northwest orientation, T estimation was less accurate (lower R2 and higher MAE), compared to the respective estimation of southeast orientation, but, the results were considered adequate (R2 and MAE ranging from 0.88 to 0.92 and 1.00 to 1.40, respectively). In general, the estimations of the mean T were better than those of the extreme ones (minimum and maximum T). In addition, better results (higher R2 and lower, in general, MAE) were obtained when T estimations were based on T data derived from sites located at areas with similar surroundings, as in the case of dense and tall vegetation of the sites at southeast orientation, irrespective of applied method.

Features of the formation and monitoring of the microclimate in non-insulated barns: unresolved issues

“…The drawback of such studies is that these models are not validated under production conditions (Wisnieski et al, 2019b). Although the coefficient of determination is considered a good criterion for evaluating the effectiveness of statistical models (Maniatis et al, 2017;Müschner-Siemens et al, 2020), reports on the accuracy of such models in experimental conditions are difficult to find.…”

“…It is not necessary to limit oneself only by monitoring the microclimate, although there are still many «gaps» (Hempel et al, 2018;Wang et al, 2018c). It is important to predict it using the capabilities of mathematical modeling (Maniatis et al, 2017).…”