Browsing by Subject "Seasonal forecasts"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Downscaling of ECMWF SEAS5 seasonal forecasts over the Horn of Africa using the WRF model
    (2023) Mori, Paolo; Wulfmeyer, Volker
    Several studies have shown the potential for downscaling seasonal forecasts on a convection-permitting (CP) scale using limited-area models (LAMs). In most cases, such experiments initial and boundary conditions are derived from atmospheric and surface analyses, which use measurements to constrain the model evolution. For operational use, the boundary conditions are derived from global seasonal forecasts, which only evolve depending on numerical models. This difference will affect the downscaling process and potentially the results’ skill. In this work, the SEAS5 seasonal forecasts are downscaled to address this gap in our understanding. Specifically, the research questions are: What advantages of a CP simulation are present when dynamically downscaling ensemble seasonal forecasts with a LAM? How do boundary conditions and physics parametrization perturbations affect a LAM ensemble in terms of spread and reliability? What perturbations produce more ensemble spread for temperature and precipitation? The study area chosen is the Horn of Africa. The effects of climate change have become much more apparent in East Africa in the last decade: the rainy season has repeatedly failed, which has led to extreme droughts. Therefore, any improvements in this regions seasonal forecasts can help to develop adaptation strategies further. In addition, areas with complex topography benefit the most from increased spatial resolution, and the global models skill is higher in the tropics and subtropics than in middle latitudes. Thus, it is likely that downscaling can extract helpful information in this region. Four global ECWMF SEAS5 ensemble members were dynamically downscaled for summer 2018 over the Horn of Africa using the Weather Research and Forecasting (WRF) model to investigate the potential of a seasonal forecast on convection-permitting resolution (3 km). A total of 16 WRF ensemble members with varied initial and boundary conditions and different physical schemes were used to evaluate the impact of the downscaling. The analysis assessed the effects of perturbations on surface temperature and rainfall in terms of bias, spatial distribution, probability of extreme events, rain belt movements, and ensemble spread. The main findings of this work are the following: the WRF simulations reproduced the spatial distribution of the 2m temperature and precipitation patterns. The bias present in SEAS5 was transferred to the limited-area model, and the signal is even intensified in some areas. For example, while the four SEAS5 members deviated only by +0.2°C on average compared to the ECMWF analyses, the WRF ensemble bias was +1.1°C. The WRF ensemble simulated an average of 264 mm of rain, compared to 248 mm for SEAS5 and 236 mm for the GPM-IMERG satellite product. The convection-permitting resolution reproduced the precipitation probability density function slightly better than the global model and simulated extreme precipitation events missing in SEAS5. However, it overestimated their frequency compared to observations. In addition, WRF can reproduce the daily precipitation cycle well: the peak times coincide with measurements, showing an accurate representation of convection initiation in the area and the potential of dynamical downscaling at convection-permitting resolution. The boundary conditions limited the movement of the rain belt associated with the inter-tropical convergence zone in the downscaling. For example, the north extension of the tropical rain belt decreased in both models by 2 degrees of latitude compared to GPM-IMERG, and the global model timing strongly influenced the movements of the rain belt in WRF. The SEAS5 has shown moderate skill in precipitation forecasts in Ethiopia. Still, a better understanding of the yearly variability of the rain belt position is necessary, as it is a crucial factor in high-resolution downscaling in the region. The downscaling increased the ensemble spread for precipitation by an average of 60%, partially correcting the SEAS5 under-dispersion. In the Ethiopian highlands, perturbed boundary conditions are primarily responsible for the WRF ensemble spread. Their effect is often 50% greater than the variability resulting from the various physics parameterizations. The results show that boundary-conditions perturbations are necessary to generate adequate ensemble dispersion in a limited-area model with complex topography. The analysis partially confirmed the potential to improve seasonal forecasts through downscaling, especially concerning convective precipitation timing and heavy rainfall events. Some advantages of downscaling atmospheric analysis are lost due to the inaccuracies in the forcing derived from SEAS5 and model bias. It also highlights the necessity of further research on physics schemes or combinations suitable to convection-permitting resolutions.