Towards seasonal forecasting of maize yield in eastern Africa: skill in the forecast model chain as a basis for agricultural climate services

Climate variability is an important driver for regionally anomalous production levels of especially rainfed crops, with implication for food security of subsistence farmers and economic performance for market oriented agriculturalists. In large parts of the tropics, modern seasonal ensemble forecast systems have useful levels of skill, that open up the possibilty to develop climate services that assist agriculturalist and others in the food chain (farm suppliers, commodity traders, aid organisations) to anticipate on expected anomalous conditions. In this thesis we explore the forecast skill at various steps in the modelling chain for seasonal maize yield anomalies in East Africa. First, we analyse the skill of ECMWF System-4 (S4) climate forecasts for primary meteorological variables against gridded observations and find both potential and real skill for rainfall and temperature in typical cropping seasons in eastern Africa. However, forecast skill is a function of geographical region, season, climate variable (i.e. higher skill in temperature, rainfall, downwelling shortwave radiation in that order) and forecast lead-time, as such skill assessment should not be generalized over a large geographical area. Next we analyse correlations between reported production and anomalous weather conditions, using a range of climate indicators relevant for arable farming, such as growing and killing degree days, and rainfall amount, evenness, random independent events (unevenness), and timing during consequent maize growth phases in two case study regions. In this case significant levels of correlation and skill are revealed  that open up the potential for statistical forecasting by use of climate forecast derived variables. Sensitivity of yields to climate  indicators depend on geographical  location, for example, higher sensitivity to rainfall is found in northern Ethiopia while in a location in equatorial-western Kenya, there is higher sensitivity to temperature indicators. At the next level of complexity we explore the use of full process based crop models forced by seasonal climate forecasts to forecast anomalous water-limited maize yield in the region, and find again potentially useful levels of skill with at least two months lead before planting, in most agricultural regions. But this again depends on regions, for example yield forecasts in Tanzania, in the season starting October did not, have  skill. Finally, we try to attribute skill levels to physiographic characteristics (soils, maize cultivars, geographical region etc.) and address some issues of scale of aggregation for two case study regions in Kenya and Ethiopia. The results showed that skill assessment at national boundaries and high resolution crop simulation units may inform both maize production related policy decisions at regional or national levels, and also support maize  production decisions at specific cropping locations such as farm management decisions made by farmers, We conclude with a synthesis discussing further on found skill levels in relation to potential climate services aimed at the agricultural sector in East Africa and beyond