An investigation of the relative contribution of Lake Victoria to regional model bias in the customization of WRF for Eastern Africa
In order to understand the physical processes which govern climate variability and change over this region, the WRF model is being adopted at the NCSU climate laboratory. In this study we focus on the first step of the process which is to customize the model for the Lake Victoria Basin region.
The different options, parameters and schemes available through the WRF model are considered in detail. Multiple runs were conducted using different combinations and options. The short rains (OND) of 2009 were modeled as this was a year that represented the climatology. These runs included a large domain at a resolution of 50km, which covered the majority of the continent and a nested domain covering the lake region with a resolution of 10km. The runs were analyzed with respect to the TRMM satellite rainfall data by root mean square errors (RMSE) and standard deviations (SD) between the observations and the model output. Different options and parameters resulted in substantial variation within the model output. The RMSE and SD varied by several hundred millimeters between the different runs and also between the different domains. The best schemes and options are discussed with respect to both the location and length of the run.
One deficiency that was consistent in all runs was the inability for the model to reproduce the rainfall over the lake. As the domains decrease and focus on the lake, the RMSE and the SD increase significantly, implying that the model is not reproducing the conditions over the lake correctly. The lake is inaccurately represented regardless of which options and parameters are used.
The problems encountered with the lake are a source of bias within the rest of the domain as the analysis is partially weighted by the lake. In order to reduce this impact it is necessary to separate the lake basin and the surrounding area and analyze them individually. The lake, the surrounding area and the total domain were all used for analysis. The results show that the RMSE and SD of the surrounding area without the lake were much lower, with a substantial reduction in many cases. It became apparent that the lake significantly biased the total domain.
The results show that after the removal of the lake from the calculations, the combinations that appeared to give the most accurate results previously are not necessarily the same. Consequently, this implies that it may be necessary to customize the region independently before including the lake.
The results of this work show not only the importance of accurate customization but also that regions not accurately represented need to be focused on, in order to ensure that they are not causing a bias in the entire domain. Our analysis shows that the lake is incorrectly represented as the model does not represent the temperatures and circulation within the lake accurately. This has highlighted the necessity of using a coupled model, a focus of future work.
The results of this study illustrate the urgent need for improving climate monitoring and modelling capabilities over Lake Victoria which the proposed ten-year Hydroclimate research project for Lake Victoria (HYVIC) GEWEX Regional Hydroclimate Project (RHP) will address. On the regional scale, HYVIC will contribute to the understanding of the hydroclimatic variability over Lake Victoria Basin and reduce uncertainty for applications in agriculture, energy, disaster risk reduction especially in marine transportation, water resources and other social-economic sectors.