Seasonally Separated Logistic Models to Assess the Impact of Climate Variables on Occurrence of Rainfall over the Bagmati River Basin of Nepal

Aims: This paper aims to develop prediction models for forecasting rainfall occurrence over the Bagmati river basin of Nepal based upon climate related predictor variables.

Study Design: Time series design with statistical downscaling of large scale daily climate data and observed rainfall data.

Place and Duration of Study: Study was conducted at Central Department of Statistics, Tribhuvan University, Kirtipur, Nepal, between 2013 and 2015.

Methodology: A day is considered as a wet day if area weighted daily rainfall (AWDR) is more than 1 mm. Extreme rainfall is determined by the 98thpercentile of AWDR. Binary logistic regression models are built with 13 possible principal components (PCs) of 7 climate related predictor variables using daily data for 1981-2000 period. Thereafter, built models are validated for 2001-2008 period.

Results: Nine separate seasonal logistic models are fitted with Hosmer-Lemeshow tests having at least 0.207 p-values. The first PC of Air surface temperature has the greatest influence with odds ratio (OR) of 4.757 in predicting a wet day during post-monsoon across four models. It is followed by the first PC of Relative humidity with OR (4.112) in winter, first PC of Relative humidity with OR (3.443) in pre-monsoon and second PC of Relative humidity with OR (3.601) in monsoon. Similarly, second PC of Relative humidity has the highest contribution with OR (7.395) in predicting extreme rainfall in post-monsoon across all five models. It is followed by the first PC of Air surface temperature with OR (7.194) in monsoon, first PC of Relative humidity in winter with OR (6.820) and pre-monsoon with OR (5.076), and second PC of Relative humidity with OR (3.186) for the non-seasonal model.

Conclusion: The developed logistic regression models are applicable in forecasting rainfall occurrence seasonally in the Bagmati river basin of Nepal.