Comparison of Season-ahead Prediction Techniques on Regionalized Grid-level Precipitation: Application to Western Ethiopia

Abstract:

Season-ahead precipitation predictions offer utility in decision-making relative to water resource utilization and management, including agricultural planning and reservoir operation, particularly for regions with highly variable spatial-temporal precipitation patterns. Preprocessing precipitation by objective regionalization has the potential to improve prediction by defining appropriate scales of homogenous clusters. Statistical prediction techniques and downscaling approaches are evaluated over western Ethiopia, including principal component and hierarchical Bayesian approaches, at the cluster and grid scales. Predictors are drawn from large scale climate indices and variables and local drivers (e.g. soil moisture, elevation, spring rains, etc.). Preliminary results indicate substantial improvements in prediction skill when applying regionalization and, for locations/grids with more complex geographic characteristics, through the addition of local variables. Grid-scale screening of prediction techniques and suitable predictors is undertaken to identify optimal model combinations.