Cartografía predictiva de variables climáticas: comparación de distintos modelos de interpolación de la temperatura en España peninsular

Predictive cartography of climatic variables: Comparing interpolation models of temperature in peninsular Spain Models of three families of interpolation techniques: tred surfaces, multiple linear regresion with predictors derived from a digital elevation model (DEM) and kriging, were tested to interpolate temperatures in peninsular Spain. The models were validated with an independent random sample of meteorological stations. Results indicate that it is possible to estimate temperatures in points where there are no recording stations with an error arround 9%. Linear or polinomial trend surfaces, and kriging (always correcting temperature for altitude) performed significantly poorer as interpolators than smoothing spline surfaces, local regression surfaces, or multiple regression models with predictors derived from a DEM. The better models did not differ significantly among themselves. Models that perform better interpolating 30 years mean temperatures are also the best for a 1 year mean. Also, the same subset of topographic predictores derived from the DEM tend to be the better estimators of mean temperature independently of the month or the duration of the period interpolated. One of the best multiple regression models with predictors from the DEM was used to produce maps of actual monthly temperatures for the period 1965-1995 in peninsular Spain.