Modelling deficit irrigation of wheat in Zimbabwe.

Abstract

Wheat is grown in Zimbabwe during the relatively dry, cool winter with irrigation. On most large-scale farms, land resources exceed irrigation water resources. Consequently, the efficient use of water is of prime concern. This has led to the development and adoption of deficit irrigation techniques, with the aim of maximizing net financial returns per unit of applied water rather than per unit land area. This often requires that less water be applied than that required for maximum yields, which implies that water deficits are allowed to develop in the crop. Although the basic principles of deficit irrigation are known, there exists no systematic procedure for advising farmers on whether or not to, or how to, employ such a management option in Zimbabwe. This research was therefore undertaken to develop an interactive computer programme that would assist farmers in determining optimum irrigation strategies for wheat. The CERES-Wheat version 2.10 crop simulation model (WHV21) was chosen as the basis for this programme. In order to validate and modify, where necessary, WHV21, a series of field experiments were conducted at a number of wide-ranging locations in Zimbabwe during the period 1986 to 1992. These included sowing date x cultivar, sowing date x seeding rate, growth analysis and irrigation experiments. In all, 122 data sets were collected, of which 47 were used for model validation and 75 used for calibration and modification of WHV21. The initial validation of WHV21 showed that the model gave biased and imprecise predictions of phenological development, particularly under deficit-irrigated conditions. The simulation of tillering was poor and the model tended to over-predict dry matter accumulation and under-predict leaf area indices. The yield component and grain yield predictions were also generally imprecise. On the other hand, for most data sets, the simulated soil water contents were similar to measured soil water contents. These inconsistencies prompted a revision of the phenological and growth subroutines of the model. In the phenological subroutine, new thermal time durations and base temperatures (Tb ) for all growth phases were determined from regressions of the rate of phasic development on mean air temperatures. For growth phases one, two and three, a Tb of 4°C was established, whereas for growth phases four and five, a Tb of 3°C was used. The revised model included the prediction of leaf emergence (as apposed to leaf appearance) and first node appearance (Zadoks growth stage 31). In order to hasten plant development under conditions of soil water deficit stress, daily thermal time was made to increase whenever the actual root water uptake declined below 1.5 times the potential plant evaporation. These changes improved the prediction of crop phasic development: for example, the Index of Agreement for the prediction of physiological maturity was improved from 0.643 with WHV21 to 0.909 with the revised version. Many changes were made to the growth subroutine, inter alia: 1. the extinction coefficient in the exponential photosynthetically active radiation (PAR) interception equation was reduced from 0.85 to 0.45; 2. an allowance was made for the interception of PAR by the wheat ears during growth phases four and five, with the statement IPAR=1-EXP(-O.004*TPSM), where IPAR is the proportion of PAR intercepted and TPSM is the number of tillers m-2; 3. the area to mass ratio of leaves was increased from 115 cm2 g -l to 125 cm2 g -l during growth phase two and this was allowed to decrease under conditions of water deficit stress; 4. tiller production during growth phase one was made a function of daily thermal time, total daily solar radiant density and plant density, moderated by high air temperatures and a new soil water deficit factor that takes the dryness of the surface soil layer into account; 5. a cold temperature routine was added to reduce kernel numbers whenever the exposed minimum air temperature decreased below 0°C during the period ear emergence to the start of the linear kernel growth phase (cold temperatures during anthesis occasionally cause reductions to kernel numbers in Zimbabwe); and 6. the kernel growth rate was gradually increased during growth phase four, and the rate of kernel growth was increased under conditions of water deficit stress during growth phase five. The modifications made to the growth subroutine of WHV21 improved predictions of tillering, ear density, yield components and yield on the independent validation data set. The modified model (WHVZIM22) was used to evaluate wheat sowing date and irrigation strategies on ten-year sets of weather data from representative locations in Zimbabwe. The results indicated that the highest yields were obtained with sowings during the latter half of April and the early part of May at all tested locations. Yields were greater for each sowing date and irrigation regime at the high altitude (1480 m) location than at warmer, lower altitude locations. The response of wheat yield to irrigation application was typically curvilinear, particularly on the soil with a high water holding capacity. Maximum yields were attained with the application of 400 to 500 mm (net) water. Soils with low water holding capacities produced lower mean yields than soils with a high water holding capacity. Maximum financial returns tended to occur with the application of less water than that required for maximum yields, particularly on the soil with a high water holding capacity. However, the variance of financial returns increased with reductions in the amount of water applied. These simulation results corroborated field observations and, taken together with the improved predictive ability of WHVZIM22 over WHV21, provided sufficient justification to use the revised model as a basis for the development of a pre-season irrigation optimization computer programme. This programme seeks the intraseasonal irrigation regime that maximizes the total gross margin for a particular soil, cultural and weather scenario, within the constraints of land and water availability. The programme is written in Microsoft QuickBASIC 4.00 and can generate an optimized irrigation regime within 4 to 5 minutes when executed on an IBM AT-compatible 80486 computer running at 25 MHz. It is envisaged that the programme would be used as a pre-season management tool, but the literal application of the results in the field is not recommended in view of the fact that the WHVZIM22 model has a number of inherent limitations and is therefore not a perfect predictor of crop growth and yield. The optimum irrigation solution generated by the programme simply provides a basis from which a farmer can plan irrigation management strategies. The actual intraseasonal irrigation schedule would necessarily depend on the real-time crop, soil and weather conditions.