Comparison of ordinary and Bayesian kriging techniques in depicting rainfall variability in arid and semi-arid regions of north-west India

Rainfall, a vital component of hydrological cycle, plays a key role in appropriate planning and sustainable management of water resources in water-short regions. However, its uneven distribution over the space and time necessitates getting knowledge about its variability. This study aimed at evaluating efficacy of ordinary and Bayesian kriging techniques in depicting spatial and temporal variability of annual rainfall in arid and semi-arid regions of north-west India. Using 35-year gridded rainfall datasets of India Meteorological Department (IMD at 0.5° resolution, 1971–2005) and Climate Forecast System Reanalysis (CFSR at 0.3° resolution, 1979–2013), geostatistical modelling was performed by employing five kriging techniques with novelty of exploring potential of empirical Bayesian kriging (EBK) technique, for the first time, for rainfall datasets in this study. Performance of the kriging techniques was evaluated by cross-validation based on five criteria. Both exponential ordinary kriging (EOK) and EBK revealed better performance over other kriging techniques. Thus, EOK and EBK were further examined by adopting goodness-of-fit criteria of correlation coefficient ( r ) and root-mean-square error (RMSE). Very high r values indicated equally excellent performance of both the techniques; however, less RMSE values in case of EBK suggested it as the best-fit technique, which was then used for developing spatially distributed rainfall raster. Spatially distributed 35-year mean annual rainfall (MAR) and coefficient of variation (CV) highlighted very high temporal variability of the annual rainfall (CV > 120%) in the western arid lands of the country with