Can the ecosystem mimic hypotheses be applied to farms in African savannahs? M. VAN NOORDWIJK 1, * and C. K. ONG 2 1 International Centre for Research in Agroforestry (ICRAF), P.O. Box 161, Bogor 16001, Indonesia; 2 ICRAF, United Nations Avenue, Gigiri, Nairobi, P.O. Box 30677, Kenya (*Author for correspondence: m.van-noordwijk@cgiar.org) Key words: competition, fertility, microclimate, risk, stability, water Abstract. The first ecosystem mimic hypothesis suggests clear advantages if man-made land use systems do not deviate greatly in their resource use patterns from natural ecosystems typical of a given climatic zone. The second hypothesis claims that additional advantages will accrue if agroecosystems also maintain a substantial part of the diversity of natural systems. We test these hypotheses for the savannah zone of sub-Saharan Africa, with its low soil fertility and variable rainfall. Where annual food crops replace the natural grass understorey of savannah systems, water use will decrease and stream and groundwater flow change, unless tree density increases relative to the natural situation. Increasing tree density, however, will decrease crop yields, unless the trees meet specific criteria. Food crop production in the parkland systems may benefit from lower temperatures under tree canopies, but water use by trees providing this shade will prevent crops from benefiting. In old parkland trees that farmers have traditionally retained when opening fields for crops, water use per unit shade is less than in most fast growing trees introduced for agroforestry trials. Strong competition between plants adapted to years with different rainfall patterns may stabilise total system productivity – but this will be appre- ciated by a farmer only if the components are of comparable value. The best precondition for farmers to maintain diversity in their agroecosystem hinges on the availability of a broad basket of choices, without clear winners or ‘best bets’. 1. Introduction 1.1. Two mimicry hypotheses Only 8% of Africa’s land area comprises fertile soils and the remaining 92% can be defined as marginal lands, which are either too dry, too wet or too saline (Yaker, 1993). Inherent soil fertility is low as most of the continent is geologically old and highly eroded and recent volcanic activity strictly localised. Large areas of apparently ‘unsuitable’ marginal lands are cleared each year for agricultural production, leading to further land degradation. Yet recent studies of some of these natural ecosystems, especially tropical grass- lands, have shown that they can be highly productive (Solbrig et al., 1996). If current agricultural interventions are indeed inappropriate, would it then not be better to mimic savannah ecosystems, seen today in the East African Serengeti, where large mammals harvest and convert the primary production of the vegetation whether grass or trees. Tropical savannahs have long Agroforestry Systems 45: 131–158, 1999.  1999 Kluwer Academic Publishers. Printed in the Netherlands.