Advertisement

Downscaling temperature and precipitation: a comparison of regression-based methods and artificial neural networks

Authors

  • J.T. Schoof,

    Corresponding author
    1. Atmospheric Science Program, Department of Geography, Indiana University Student Building, 701 East Kirkwood Ave., Bloomington, IN 47405-7100, USA
    • Atmospheric Science Program, Department of Geography, Indiana University Student Building, 701 East Kirkwood Ave., Bloomington, IN 47405-7100, USA
    Search for more papers by this author
  • S.C. Pryor

    1. Atmospheric Science Program, Department of Geography, Indiana University Student Building, 701 East Kirkwood Ave., Bloomington, IN 47405-7100, USA
    Search for more papers by this author

Abstract

A comparison of two statistical downscaling methods for daily maximum and minimum surface air temperature, total daily precipitation and total monthly precipitation at Indianapolis, IN, USA, is presented. The analysis is conducted for two seasons, the growing season and the non-growing season, defined based on variability of surface air temperature. The predictors used in the downscaling are indices of the synoptic scale circulation derived from rotated principal components analysis (PCA) and cluster analysis of variables extracted from an 18-year record from seven rawinsonde stations in the Midwest region of the United States. PCA yielded seven significant components for the growing season and five significant components for the non-growing season. These PCs explained 86% and 83% of the original rawinsonde data for the growing and non-growing seasons, respectively. Cluster analysis of the PC scores using the average linkage method resulted in eight growing season synoptic types and twelve non-growing synoptic types. The downscaling of temperature and precipitation is conducted using PC scores and cluster frequencies in regression models and artificial neural networks (ANNs).

Regression models and ANNs yielded similar results, but the data for each regression model violated at least one of the assumptions of regression analysis. As expected, the accuracy of the downscaling models for temperature was superior to that for precipitation. The accuracy of all temperature models was improved by adding an autoregressive term, which also changed the relative importance of the dominant anomaly patterns as manifest in the PC scores. Application of the transfer functions to model daily maximum and minimum temperature data from an independent time series resulted in correlation coefficients of 0.34–0.89. In accord with previous studies, the precipitation models exhibited lesser predictive capabilities. The correlation coefficient for predicted versus observed daily precipitation totals was less than 0.5 for both seasons, while that for monthly total precipitation was below 0.65. The downscaling techniques are discussed in terms of model performance, comparison of techniques and possible model improvements. Copyright © 2001 Royal Meteorological Society

Ancillary