Estimating monthly air temperature using remote sensing on a region with highly variable topography and scarce monitoring in the southern Ecuadorian Andes

Abstract

Monitoring of air temperature has implications in a wide range of environmental applications. Air temperature commonly measured with meteorological stations provides a high accuracy and temporal resolution for specific monitoring sites. However, in regions with highly variable topography and scare monitoring such as the case of the southern Ecuadorian Andes, these in situ data poorly describe the spatial variability of air temperature. Thus, remote sensing data has a great potential to estimate the spatial distribution of climatological variables due to the spatial continuity of the information. This research aims to estimate the spatial distribution of the monthly air temperature in the Paute river basin for the period 2014–2017, using statistical and geostatistical methods: linear regression (LR), random forest regression (RF), and regression kriging (RK), in addition to evaluate the use of altitude and other auxiliary variables (land surface temperature, latitude, and longitude). Cross-validation showed that RF performed better than LR as well as when using auxiliary variables compared to only the altitude (LR-altitude: RMSE=1.325 °C, P bias= −0.150%, r=0.775; LR-auxiliary variables: RMSE=1.265 °C, P bias=0.000% r=0.795; RF-altitude: RMSE=1.235 °C, P bias =0.200%, r=0.810; RF-auxiliary variables RMSE=1.205 °C, P bias =0.200%, r=0.820). The application of regression kriging was limited since less than 50% of the months had spatial autocorrelation in the regression model residuals. Nevertheless, in these months, regression kriging increased the estimation performance. The outcomes of this research work increase the understanding of the spatial distribution of monthly air temperature in the Paute river basin, which will improve hydrological modeling.