A Bayesian network approach for understanding the role of large-scale and local hydro-meteorological variables as drivers of basin-scale rainfall and streamflow

The present study examines the role of large-scale climate modes and local hydro-meteorological variables, at 1–6 months lead, as drivers of streamflow and rainfall at four river basins in peninsular India. Bayesian Network was used to develop the conditional independence structure between the target variables (streamflow and rainfall) and large-scale climate modes as well as local hydro-meteorological variables. From the conditional independence structure, variables possessing a ‘directed arc’ from the target variable were selected as the potential predictors for developing the prediction models. The most important potential predictors influencing streamflow were found to be rainfall, u-wind, and soil moisture, while those influencing rainfall were u-wind, air temperature, geo-potential height, precipitable water, vertical velocity, and relative humidity. The conditional independence structure reveals that the information provided by the large-scale climate modes is generally dispensable for all the basins given the local meteorological variables barring the Pacific Decadal Oscillation and El-Niño Southern Oscillation which exert substantial influence on the targets. The network structure further revealed that about 87–97% of the initial information are redundant. The prediction performance is better for rainfall (Refined index of agreement MD: MD GDV = 0.81, MD UMB = 0.80, MD UNB = 0.79 and MD CAV = 0.61) than for streamflow (MD GDV = 0.78, MD UMB = 0.75, MD UNB = 0.67 and MD CAV = 0.67) and comparable across all the basins. Further, dry, intermediate, and wet months are satisfactorily categorised in each basin using two drought indices—Standardised Drought Index for streamflow and Standardised Precipitation Anomaly Index for rainfall.