Estimation of air temperature and reference evapotranspiration in the Minas Gerais state by different methods

Abstract

Consistent weather data is obtained by weather stations. These data are important for different fields of science such as climatology, irrigation, and hydrology. The meteorological element and the agrometeorological parameter air temperature and evapotranspiration (ET), respectively, are fundamental for studies in these fields. The temperature indicates the amount of energy available in the water-soil-atmosphere system. This energy can influence various processes on the Earth's surface, among them the growth and development of plants. ET is the process of water transportation from a vegetated surface to the atmosphere including the evaporation and transpiration process. These meteorological variables and agrometeorological parameter can be monitored daily in weather stations, however, in the Minas Gerais State, the coverage of the weather stations network is limited. Besides, interruptions and errors in the database are quite common. In this sense, this research aimed to develop models that can reliably estimate air temperature and evapotranspiration through easily obtained input data such as geographic coordinates. As described in paper 1, the aim was to develop models of multiple linear regression (MLR), artificial neural network (ANN), and random forest (RF) to estimate the mean (Tmean), maximum (Tmax), and minimum (Tmin) monthly air temperatures as a function of geographic coordinates, altitude, and month for different localities in the Minas Gerais State, Brazil, with Köppen’s climatic classification Cwa or Cwb. The Tmax, Tmean and Tmin data were extracted from national network of climatological stations (INMET). MLR was implemented using the data analysis tool in Microsoft Excel®. ANN and RF were implemented using the WEKA. The results showed that the algorithms RF and ANN were used to estimate Tmean, Tmax, and Tmin with high accuracy. The best results were obtained using the RF model. The MLR did not present a good accuracy. In paper 2, the aim was to evaluates the performance of ANN, RF, Support Vector Machine (SVM) and MLR to estimate the monthly mean reference evapotranspiration (ET0) with four different input data combinations (I8, I6, I3 and I2) and in three scenarios: (SI) at the state level, where all climatological stations were used; and at regional level (SII and SIII), where the Minas Gerais state was divided into two areas according to the climatic classification of each climatological stations. The climatic classifications proposed by Thornthwaite (SII) and by Köppen (SIII) were used. All models were implemented by WEKA. The results showed that ANN and RF performed better in SI, II, III with the I8 (latitude, longitude, altitude, month, Tmean, Tmax, Tmin, and relative humidity) or I6 (latitude, longitude, altitude, month, Tmean, and relative humidity) input data. The SVM and MLR performed better in all scenarios when only two input variables were used (I2 - mean temperature and relative humidity). Although dividing into scenarios results in less input data for models training, SII and SIII showed a slightly better result in the southern areas of the Minas Gerais state.