towards improved seasonal rainfall predictions over east africa
Margaret Kimani is a PhD student in the department of Water Resources. Her supervisor is prof.dr. Z. Su from the Faculty of Geo-information Science and Earth Observation.
Empirical seasonal rainfall predictions over East Africa can be improved by use of the most consistent and representative rainfall data that characterize its variability to determine atmospheric-sea interactions that govern the seasonal to inter-annual variabilities. Rainfall variability is externally induced by large-scale circulations and locally by the heterogeneity of the terrain like topography and inland water bodies that modify its processes. The General Circulation Models (GCM) used for seasonal predictions have challenges characterizing the sub-grid processes because of their coarse resolutions and models initialization challenges. The unpredictable variability has led to the loss of lives and destruction of properties over this region. Out of the two main rainfall season of March-May (MAM) and October-December(OND), MAM is more challenging because of its different mechanisms not well understood. It is the most dependent on for agricultural and water resource management because of its large intensity and longest rain days. This thesis is composed of six chapters and chapter which is introduction details the scientific and societal importance of the research. It empirically uses the satellite rainfall data to explore the processes leading to rainfall variability by identifying the atmospheric-ocean interactions to improve the predictions. This requires consistent and long climatological rainfall data.
Thus, the second chapter following thesis introduction deals with assessments of seven satellite-derived rainfall products (TARCAT, CHIRPS v2, TRMM-3B43, CMORPH v1, PERSIANN-CDR, CMAP and GPCP) using locally gridded rain gauge data over the region. This was aimed at identifying the most suitable product that can consistently characterize rainfall variability spatially and temporally. A grid-based statistical comparison between the two datasets was employed, based on pixel values located at the rainfall stations. Further, the impact of topography on the performance of the products was assessed in areas of the highest negative bias mainly observed in October-December (OND) rainfall months. All the products could replicate rainfall patterns but exhibited systematic errors, especially on high elevated areas which decreased with an increase in temporal resolution from a monthly to yearly scale. CMORPH, CHIRPS, and TRMM showed consistently high performance during both seasons, attributed to their ability to retrieve rainfall of different rainfall regimes but CHIRPS was chosen based on its long climatology.
In the third chapter, the Bayesian bias correction was applied to CHIRPS v 2 and the rain gauge data used as a reference. Assessments of the approach’s influence on the rainfall estimate spatially and temporally were explored. Significant performance of the approach was observed during years of low rainfall from shallow convections but decreased in areas of sparse rain gauge network that insufficiently represented rainfall variabilities. The locally bias corrected CHIRPS was then used to identify atmospheric processes linked with rainfall variability.
In the fourth chapter, moisture distribution influence on rainfall variability during different phases of IOD over the region was assessed. Since the region is in the equatorial tropics, convective variables were utilized to identify moisture convergence areas in relation to convective areas. Results showed advected moisture as the source of precipitable moisture and contribution from local evaporation is minimal mainly affecting the Lake Victoria region. Moisture is dominantly advected from the Indian Ocean through low-level convergence mainly during ITCZ overpass. During the southwest monsoon, mid-tropospheric moisture convergence is experienced from the west attributed to Congo Basin. The Kenya highlands act as moisture sinks separating the west and the eastern part of the region and the locally induced convection over Lake Victoria are suppressed by the fluxes. Empirical Orthogonal Functions (EOF 1,2) of moisture were utilized to determine the impact of the Dipole Mode Index (DMI) and Nino 3.4 to moisture distributions during different phases of IOD.DMI has the largest influence during ITCZ overpass explaining > 60% and less during southwest monsoon months. EOF2 contributes 31% influence in May attributed to DMI with little impact from Nino3.4. EOF1 is dominantly influenced by DMI but during early onset of southwest monsoon corresponding with El Niño year suppression of moisture influx from the Indian Ocean associated with Nino 3.4.
In the fifth chapter, Real-time Multivariate MJO (Madden–Julian oscillation), or RMM, amplitudes were utilized to link the atmospheric variability to Ocean forcing, related to rainfall variability (especially extremes). The analysis was based on ERA-Interim reanalysis data at different timescales. For the oceanic influence, the Sea Surface Temperature (SST) forcing on 2-m surface temperature related to MJO convections was used to test the predictability of the MAM rainfall season, covering a period of 33 years (1981–2013). The study observed that MAM MJO modulates low-level moisture influx from the Indian Ocean enhancing Convergence Zone that extends over to the western part of the region bordering the Congo Basin. The SST time series are derived from 3 areas, namely the Maritime Continent, southwestern and central Indian Ocean and utilized as predictors of moisture budget as proxy of rainfall. A Stepwise regression model shows the highest skill at lag 4, with a Brier skill score (BS) of 0.02 and an Anomalies of Correlation Coefficients (ACC) of 0.82. Whereas the influence of MJO on rainfall during OND (and JJA) rainy seasons is fairly minor, MJO has a major impact on the rainfall dynamics during the MAM rainy season. Consequently, the inclusion of the MAM MJO magnitude (notably for MJO 1 and MJO 8) can help improve seasonal prediction for the MAM rainy season in East Africa.
Finally, Chapter 6 gives the synthesis detailing conclusions and recommendation from the study.
