Inter-annually varying vegetation improves seasonal forecasts of near-surface temperature in East Africa

  Interactions between vegetation and the overlying atmosphere, mediated by changes in surface moisture availability and energy flux partitioning, exert significant influence on near-surface temperature and other atmospheric variables. At seasonal timescales, these vegetation-atmosphere interactions have the potential to enhance forecast predictability. However, current operational seasonal forecast systems, such as ECMWF’s SEAS5, prescribe vegetation to a fixed climatological state. This does not fully capture vegetation-atmosphere interactions and thus the potential predictability is not fully exploited. In this presentation, we investigate the atmospheric response to prescribing inter-annually varying Leaf Area Index (LAI) in seasonal hindcasts and assess its effect on seasonal forecast skill.  This work focuses on Africa, where seasonal forecasts are crucial for agricultural planning and extreme weather preparedness.

  A series of seasonal hindcasts run for the period 1993-2019 using ECMWF’s coupled Integrated Forecasting System are used. We compare two experiments – a control experiment, which uses climatological LAI and a non-varying land cover map, and an experiment which implements a dataset of inter-annually varying LAI and land cover maps produced by merging multiple satellite products. In general, prescribing inter-annually varying LAI increases African near-surface air temperatures by up to 0.2K compared to a fixed climatological LAI across Africa. To evaluate temperature changes associated with LAI variations, we perform a Seasonal-reliant Empirical Orthogonal Function analysis (see Wang and An, 2005) on the driving LAI dataset. We find a mode of variation that is correlated with the Indian Ocean Dipole (IOD) index for the September-November-December season (correlation coefficient of ~0.75); thus, we view this mode of LAI variation as the vegetation response to increased East African rainfall during active IOD events. Results show a consistent near-surface temperature response across East Africa when inter-annually varying LAI is prescribed. The temperature response is shown to be consistent with simulated changes in the surface energy balance. Forecast skill of temperature, measured as bias compared to ERA5 values, is shown to be improved when vegetation varies inter-annually. Improvements in bias are largest following extreme IOD events and for areas where the control hindcasts’ bias is largest, with a maximum in temperature bias reduction of 0.6K and an average bias reduction of 0.2K. Thus, we find that increased complexity in vegetation representation in seasonal forecasts leads to improvements in forecasted temperature through better representation of land-atmosphere interactions influenced by the IOD.