Comparación de técnicas de interpolación espacial de temperatura ambiente de los modelos atmosféricos CMIP6 y de las estaciones meteorológicas

Abstract

Interpolation is essential to estimate data in areas without information. This study addresses the scarcity of data in areas without meteorological stations in the province of Chimborazo, which makes it difficult to understand the behavior of the temperature variable. Therefore, the objective of this degree work was to compare spatial interpolation techniques of ambient temperature data from CMIP6 atmospheric models and weather stations. The methodology was quantitative, with a non-experimental cross-sectional design. Data were downloaded from the SAM (523 points) period 1998-2020, detecting and completing missing information with missForest in RStudio. On the other hand, temperature data from meteorological stations period 2014-2020 were provided by the Alternative Energies and Environment Group (GEAA). Convenience sampling was used, determining that 2015 shows higher reliability and accuracy due to fewer missing data. Statistics show an increase in the mean and median over time. Kolmogorov-Smirnov and Shapiro-Wilks normality tests, indicated that the MAR data do not follow a normal distribution unlike the station data. IDW, Spline and Kriging interpolation techniques were then implemented and evaluated using validation metrics. The best fitting method that presented lower errors was Kriging, for the SAM data the Universal Kriging with exponential semivariogram and constant trend presented an EM of -0.01, SEM of 1.64 and RSME of 1.10, while for weather station data the Ordinary Kriging method with Gaussian semivariogram and constant trend presented an EM of -0.16, SEM of 2.74 and RSME of 2.93, the study shows that the Kriging interpolation is the most effective, since it uses spatial correlation and trend analysis.