NUMERICAL OPTIMIZATION OF NITROGEN APPLICATION TO RICE .2. FIELD EVALUATIONS

The MANAGE-N model (Part I; Ten Berge et al., this issue) was tested b y comparing predicted and measured final crop biomass production for 4 8 rice cultivars under application of 0, 30-40, 60-80 and 90 to 120 kg urea-N per ha at Cuttack, India, during seven consecutive wet seasons . The overall coefficient of correlation between predicted and measure d values was 0.93 (p < 0.01). The root mean squared prediction error w as 1110 kg biomass per ha in long, 769 kg/ha in medium and 1210 kg/ha in short duration cultivars. Data sets from three other experiments at Nanchang (Jiangxi, China), Jinhua (Zhejiang, China) and Cuttack were used to (a) evaluate model-recommended N management (dose D-M, timing T-M) against local state-recommended management (dose D-S, timing T-S) ; and (b) determine the yield prediction accuracy of the model. Under dose D-S, timing T-M increased grain yields significantly (p < 0.05) b y 5 to 10% at all sites, relative to T-S. Application of dose D-M resu lted in further yield increases at Jinhua and Cuttack. Negative effect s of high N dose on yield were observed but not predicted at Nanchang and Jinhua. Agronomic efficiency (mean yield increment per kg N applie d) was at all sites higher in DSTM and DMTM than in DSTS. Predicted gr ain yields differed from observed values by less than the least signif icant difference between field treatments (LSD, p = 0.05) at Nanchang and Jinhua. Differences were larger than LSD at Cuttack. Simulations f urther indicate that the time window for N application narrows as N in put decreases. Simulated effects of crop duration on yield, N response and optimum N dose are discussed. (C) Elsevier Science B.V.