站点CERES-Rice模型区域应用效果和误差来源

作物区域模拟是利用有限的空间数据,最大限度地反映出生育期、产量等作物性状的时空变化规律.由于目前的作物模型大多是田间尺度的站点模型,把它运用到区域水平的效果如何研究甚少.文章利用CERES-Rice模型,对作物模型在我国的区域应用效果进行了分析.首先利用田间观测数据在各实验点上对模型进行了详细的站点校准,以验证模型在我国的模拟能力;然后以我国水稻生态区(精确到亚区)为单位,运用均方根差(RMSE)法进行了区域校准和验证;最后利用区域校准后的CERES-Rice模型,模拟了1980～2000年的网格(50km×50km)水稻产量,并与同期农调队调查产量进行统计比较,以验证区域应用的效果,为区域模拟的推广和应用提供参考.结果表明:经过空间校准后的CERES-Rice模型,在水稻的主产区1～4区(占种植面积的95%)模拟的平均产量与调查产量相对均方根差在22%以内,两者的符合度也较好,个别区域(5、6) RMSE%在24%～30%之间;1980～2000年水稻各产区模拟的平均产量与调查产量随时间变化趋势也具有一定的一致性;全国1896个网格中,大部分网格(71.01%)模拟的21年水稻年产量与调查产量的RMSE%在30%之内,且大部分分布在水稻主产区,考虑到水稻种植面积的权重后,认为利用区域校准和验证后的CERES-Rice模型进行水稻区域模拟,可以反映出产量的时空分布特征,能够为宏观决策提供相应的信息.但目前区域模拟中还存在着一定的误差,有待今后进一步研究.

The performance of regional simulation of CERES-Rice model and its uncertainties

One primary purpose of regional simulation is to predict the spatial yield variation and temporal yield fluctuation, with presently available geographical database. Most crop models are site-specific, and some researches have attempted to up-scale the crop models for regional simulation. However, few have emphasized on their performance and uncertainties. The performance of regional simulation in China was evaluated in this study, and CERES-Rice model was employed in the simulation. In order to assess the suitability of model application for environments in China, we first calibrated the crop model at plot scale over main rice areas. Second, we calibrated and validated the CERES-Rice model at regional scale using the statistic of RMSE, cultivar coefficients and management practices were aggregated into each sub Agro-Ecological Zone (AEZ). Last, for analyzing the performance of the regional simulation, we simulated the historical (1981-2000) rice yield at 50 km 50 km grid scale, and compared the simulated with census values. Results show: the pattern of yield variation captured by model in most of the rice areas, especially in the main rice planting regions, i.e. AEZ 1-4 (over 95% of the total rice cultivation area), with RMSE< 22%. Some regions showed bad results, i.e. AEZ 5 and 6 with RMSE% 24%-30%. Simulated year-to-year yields matched well with the census values. 71.01% of simulation grids showed a bias less than 30%, most of them concentrated in the main rice planting areas. Therefore, the regional simulation is able to produce a reasonable estimation in spatial yield variation and temporal yield fluctuation, it can be used as a tool to provide information for policy makers at macro scale. There were various sources of uncertainty in the study and were analyzed in the discussion section.