Multivariate Statistical Approach and Correlation Analysis for Climate Variable Modelling in Regional Weather Prediction

Abstract

This study investigates the selection of predictor variables for temperature estimation using various statistical techniques across five districts in Bihar, namely Nalanda, Saran, Bhagalpur, Kaimur, and West Champaran. Bihar, located in the northern plains of India, has an agrarian economy that is highly sensitive to climatic and weather conditions. Understanding the variability and influence of different predictor variables is essential for effective management of agriculture, irrigation, hydrology, and sustainable development initiatives. The analysis is based on ten climatic predictor variables: specific humidity, maximum temperature, minimum temperature, mean sea level pressure, geopotential height, divergence, u-wind at 500 mBar, v-wind at 500 mBar, u-wind at 1000 mBar, and v-wind at 1000 mBar. These variables were sourced from the NCEP/NCAR Reanalysis dataset and subsequently converted from netCDF to tabular format using ArcMap (ArcGIS). Key statistical parameters such as mean, variance, and covariance matrices were computed, followed by the construction of correlation matrices for each district. Principal Component Analysis (PCA) was employed to examine the interrelationships among the variables and to identify dominant patterns influencing temperature variations. Visual tools including scatter plot matrices and histograms were used to further analyze variable interactions. The correlation monoplots helped determine the nature and strength of associations based on vector magnitudes and orientations. The outcomes of this study offer valuable insights for meteorologists and climate scientists in forecasting temperature and rainfall patterns. These findings can also support data-driven decision-making in climate-resilient agricultural planning and water resource management in the selected regions of Bihar.