Location: Location not imported yet.
Title: Evaluation of reanalysis precipitation data and potential bias correction methods for use in data-scarce areasAuthor
![]() |
GARIBAY, VICTORIA - Purdue University |
![]() |
GITAU, MARGARET - Purdue University |
![]() |
KIGGUNDU, NICHOLAS - Makerere University |
![]() |
Moriasi, Daniel |
![]() |
MISHILI, FULGENCE - Sokoine University Of Agriculture |
Submitted to: Water Resources Management
Publication Type: Peer Reviewed Journal Publication Acceptance Date: 2/28/2021 Publication Date: 7/17/2021 Citation: Garibay, V.M., Gitau, M.W., Kiggundu, N., Moriasi, D.N., Mishili, F. 2021. Evaluation of reanalysis precipitation data and potential bias correction methods for use in data-scarce areas. Water Resources Management. 35(5):1587-1602. https://doi.org/10.1007/s11269-021-02804-8. DOI: https://doi.org/10.1007/s11269-021-02804-8 Interpretive Summary: Lack of reliable and accessible weather data often hinders efforts to develop sound scientific basis for water management policy decisions. Daily precipitation is a crucial input for water resources modelling and for storm-related planning. A common alternative to missing observed precipitation data is mathematically generated data. However, generated weather data are sometimes consistently overestimated or underestimated compared with measured data, which is often referred to as biased. This study was conducted to determine the suitability of generated precipitation data as a substitute for on-site measurements in data-scarce regions and to identify appropriate bias correction methods to improve the accuracy of generated data in three East African countries of Kenya, Tanzania and Uganda. Results showed notable differences between generated precipitation data and available observed data from weather stations in the three countries. A region-specific bias-correction method was determined and used to adjust generated precipitation data showing a marked improvement compared with observed data. For example, bias correction of generated total annual precipitation reduced under-prediction errors by 32% and over-prediction errors by 81%. The approaches used in this study are applicable to other global regions where precipitation data records are scarce or where complete and consistent data are not easily accessible. Technical Abstract: Data availability and accessibility often present challenges to resolving regional water management issues. One primary input essential to models and other tools used to inform policy decisions is daily precipitation. Since observed datasets are not always present or accessible, data from the Climate Forecast System Reanalysis (CFSR) have become a potential alternative. A comparison of CFSR precipitation data to available observed data from stations in the East African countries Kenya, Uganda, and Tanzania showed notable differences between the two datasets, particularly with respect to annual totals and number of days receiving rainfall. A sliding window bias correction approach evaluated using 3 methods with 8 different window length and timestep variations showed that empirical quantile mapping with a 30-day sliding window length and 1-day timestep achieved the best performance. A comparison of bias corrected CFSR precipitation data against observed data showed marked improvement in the similarity of the number of wet days and maximum daily rainfall between the two datasets. For annual totals, bias correction reduced under-prediction errors by 32% and over prediction errors by 81%. Results indicate that bias-corrected CFSR precipitation data provides an improved basis for water resources applications in the study region. Methodologies and approaches are extendable to other data-scarce regions and areas where complete and consistent data are not easily accessible. |