Influence of meteorological variables on reference Evapotranspiration in the State of Alagoas, Brazil, based on multivariate analysis

Reference evapotranspiration (ETo) is an important agro-meteorological parameter used in several meteorological, climatological and hydrological studies. This study aimed to diagnose the effect of the meteorological variables and large-scale variability modes in ETo, based on logistic regression models. Monthly data from five conventional meteorological stations located on the State of Alagoas, Brazil, were used: agua Branca, Maceio, Pao de Acucar, Palmeira dos indios and Porto de Pedras, from 1960 to 2016 period. The variables used were as follows: temperatures (air-Tair, maximum-Tmax and minimum-Tmin), relative humidity (RH), sea level pressure, sunshine, wind speed (WS), rainfall, evaporation and also two climatic proxies: Oceanic Nino Index (ONI) and Tropical South Atlantic (TSA). ETo was estimated using the Penman-Monteith FAO method. The binary logistic regression procedure evaluated ETo values below the percentile 75 (ETo75 characterized as 1). Regarding the variables used the percentile 75 (except Tair and rainfall that used the percentile 30). The temperatures were the main variable for ETo75 events, except in agua Branca. The Tmax presented the highest odds ratio (OR) in Palmeira dos indios with value of 21.10 times on occurrence of ETo75. The wind is the most important variable in Pao de Acucar, due the heat capacity of the Sao Francisco River, with OR value of 3.72 times. The ONI did not integrate the models. Based on AIC and ROC curve values, Palmeira dos indios presented the best model with a value of 195.95 and 0.979, respectively.