Back to Library

Dealing with Climatic Risk in Agricultural Research: A Case Study Modelling Maize in Semi-arid Kenya

Published by:
Publication date
Number of Pages
Type of Publication:
Focus Region:
Sub-Saharan Africa
Focus Topic:
Climate / Weather / Environment
Type of Risk:
Weather & Climate related
Type of Risk Managment Option:
Risk assessment
B.A. Keating, B.M. Wafulat, J.M. Watikit, D.R. Karanja
CSIRO Division of Tropical Crops and Pastures, Kenya Agricultural Research Institute

Rainfall variability is a dominant feature of crop production in semi-arid regions. Soil fertility is also a major constraint, and much of the research effort has been directed at agronomic or genetic factors that impact on either or both the supply and demand for water or nitrogen. This paper reports on the application of models to research aimed at improving maize productivity under the highly erratic rainfall regimes of semi-arid eastern Kenya. Steps undertaken to test and adapt the CERES-Maize model are described, and a revised version called CM-KEN is shown to provide a realistic description of the major issues of concern in maize production in the region, Le. responses to plant population, planting time, location, nitrogen and water supply and the interactions between these factors.

The additional insight such a modelling approach provided in terms of the prospects for improving maize productivity in the region is examined. Current gennplasm is shown to be well adapted to the limiting rainfall regimes of the region. The major gains in productivity are likely to come from improved management of soil fertility and soil surface management. Indications are that nitrogen fertilisers should have a place in more productive systems in the region.

Insights pertaining 10 the conduct of agronomic research in regions of high climatic risk are also examined. Between 10 and 20 seasons of fertiliser rate trials were shown to be necessary to identify an optimum N fertilisation rate with any degree of confidence (i.e. to reduce coefficients of variation of the optimum rate to 25 and 15% respectively). In contrast, application of a validated model to the historical weather data enabled 63 seasons to be ‘sampled’ and coefficients of variation of optimum N rate to be reduced to approximately 1 %.