COMPARATIVE STATISTICAL VALIDATION OF 2 10-DAY WATER-USE MODELS AND OF 3 YIELD-REDUCTION HYPOTHESES FOR YIELD ASSESSMENT IN BOTSWANA

The 1978/1979-1985/1986 cereal yields of five regions in Botswana are statistically analysed as a function of two water-use models and three yield-reduction hypotheses. The first model does not include a root development model and assumes that all available soil moisture is readily available. The second model includes a simple root development function and assumes that only part of the maximum available soil moisture is readily available. Simultaneously, three hypotheses are tested regarding the effects of water stress on yield. The first hypothesis accepts that yield reduction is related to the cumulated deficit. The second hypothesis states that yield is affected differently by water stress, depending upon the development stage during which the stress occurs, and the final effect is additive. The third hypothesis is as the second, but assumes that the final effect is multiplicative. The results show that for the period under consideration: (1) the two water-use models equally account for yield variability in Botswana; (2) (sorghum + millet) yield variability is better accounted for when introducing different additive stage susceptibilities to water stress; (3) maize yield variability is significantly better accounted for when water deficit experienced during the reproductive stage is multiplied by the square of the deficit experienced during the vegetative stage; (4) subject to some restrictions, quantitative crop yield assessments are possible on a national scale; on a regional scale, qualitive crop state monitoring is indicated; (5) water stress during the 1978/1979-1985/1986 rainy seasons significantly depressed a positive time trend. © 1990.