The Open Atmospheric Science Journal

2015, 9 : 9-22
Published online 2015 July 31. DOI: 10.2174/1874282301509010009
Publisher ID: TOASCJ-9-9

Logistic Model as a Statistical Downscaling Approach for Forecasting a Wet or Dry Day in the Bagmati River Basin

Rajendra Man Shrestha , Srijan Lal Shrestha and Azaya Bikram Sthapit
Central Department of Statistics, Tribhuvan University, Kirtipur, Nepal.

ABSTRACT

A binary logistic model is developed for probabilistic prediction of a wet or dry day based upon daily rainfall data from 1981 to 2008 taken from 25 stations of Bagmati River basin. The predictor variables included in the model are daily relative humidity, air surface temperature, sea level pressure, v-wind which are expressed as principal components of 9 grids of the National Centers for Environmental Protection (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis data with resolution of 2.50×2.50. Principal component analysis is used to reduce the dimension of the predictors in the presence of spatial correlations between grids and thus reduce their multicollinearity effect. The result depicts that the model has 86.4 percent predictive capability in the analysis period (1981-2000) and 86.1 in the validation period (2001-2008) along with support of receiver operating characteristic (ROC) analysis. The results demonstrate that the first two principal components of relative humidity are the key predictor variables with respective odds ratios (ORs) of 4.18 and 3.61, respectively. The other statistically significant predictors are the second principal component of v-wind with OR 1.43, the second and first principal components of air surface temperature with ORs 1.38 and 0.76, respectively and the first principal component of sea level pressure with OR 0.44. Goodness-of-fit test, ROC analysis and other main diagnostic tests showed that the fitted logistic model is characterized by good fits for analysis as well as validation period.

Keywords:

Binary logistic model, climate change, principal components, rainfall, statistical downscaling, weather variables.