NUMERICAL OPTIMIZATION OF NITROGEN APPLICATION TO RICE .1. DESCRIPTION OF MANAGE-N

A tool (MANAGE-N) to identify optimal nitrogen (N) management in irrig ated rice is presented. It combines a dynamic crop growth model (ORYZA -0) with a numerical optimization procedure and a user interface readi ng local weather and soil and crop characteristics as inputs to genera te site-tailored recommendations for N-fertilizer management. These ha ve the form of generalized logistic curves expressing the ideal tempor al pattern to apply any user-defined total N dose. ORYZA-0 simulates c rop N uptake, N allocation, growth and yield with only seven equations and a number of empirical constraint parameters. A single exponential relation describes crop growth rate as a function of daily incident g lobal radiation (R, MJ m(-2) d(-1)), bulk leaf nitrogen (N-L, g N per m(2) ground area) and a site calibration factor f(sv). Leaf area and l ight interception are not explicitly distinguished. Harvest index is c alculated from biomass at flowering and cumulative incident radiation after flowering. Values of model parameters are provided based on 16 d ata sets from China, India, The Philippines and Australia covering 94 nitrogen management treatments (doses, timing). The corresponding 94 g rowth curves are used to evaluate three aspects of the growth equation , using observed time series of R and N-L as boundary conditions: (i) goodness of fit of growth curves after calibration of f(sv); (ii) vari ation in f(sv) within and across locations; and (iii) errors in predic ted crop biomass at flowering and harvest, using cross-validation betw een treatments within each site. All curves were fitted with correlati on coefficients from 0.97 to 0.99. Parameter f(sv) varied more between sets (experiments) than within sets (p < 0.001), with preflowering va lues between 0.68 and 0.97 and postflowering values 0.38 to 0.96. Root mean squared errors of biomass prediction were 230 to 1600 kg/ha at f lowering and 280 to 3000 kg/ha at harvest, or 5 to 15% of observed val ues in most cases. Further tests of the model are presented in Part II (Ten Berge et al., Field Crops Research, this issue). (C) Elsevier Sc ience B.V.