不同灌水水平下CROPGRO棉花模型敏感性和不确定性分析

Sensitivity and uncertainty analysis for CROPGRO- cotton model at different irrigation levels

基于过程的作物模型使用大量的品种和土壤参数来模拟作物生长和土壤水分变化。对于新的作物品种或新的环境,这些参数往往需要重新率定,然而许多参数难以通过实测获得。敏感性分析（sensitivity analysis,SA）可以量化模型输入参数对模型输出的影响,通过筛选出敏感性较大的参数进行率定,而把敏感性较小的参数设为固定值,可以极大简化参数率定过程,提高工作效率和模型模拟精度。为了给DSSAT-CROPGRO-Cotton模型应用于新疆地区进行棉花灌溉制度优化提供本地化的模型参数,对该模型进行了敏感性分析和不确定性分析。该文依据新疆石河子的棉花大田试验资料,应用Morris法和扩展傅里叶幅度敏感性检验（extend Fourier amplitude sensitivity test,EFAST）法对DSSAT-CROPGRO-Cotton模型3个灌水处理（60%ETC、80%ETC和100%ETC,ETC为作物蒸发蒸腾量crop evapotranspiration）下6个输出结果（初花天数、成熟天数、籽棉产量、地上干物质量、最大叶面积指数和蒸发蒸腾量）对于品种和土壤参数进行敏感性分析,并比较了2种方法的相关关系,最后对EFAST法的输出结果进行不确定性分析。相关分析结果显示,对于地上干物质量和最大叶面积指数,Morris法和EFAST法相关性介于0.87∽0.93,对于模型结果成熟天数、籽棉产量和蒸发蒸腾量,2种方法相关性介于0.66∽0.81。敏感性分析和不确定性分析结果显示,模型模拟灌水处理对初花天数无明显差异,且模拟初花天数和最大叶面积指数存在参数敏感性过于单一现象。模型参数敏感性随土层而不同：对于成熟天数,〉40∽80 cm土壤参数的敏感性更强;对于地上干物质量和蒸发蒸腾量,〉80∽120 cm土壤参数的敏感性更强,这可能是由于该地区气候干旱,下层土壤水分充足程度直接影响作物受到水分胁迫的程度,进而影响作物生长发育和蒸发蒸腾量。模型输出结果最大叶面积指数和蒸发蒸腾量存在一定程度的高估。该研究可提高CROPGRO-Cotton模型在新疆地区的模拟效率和模拟精度。

Process-based crop models use a large number of variety and soil parameters to simulate dynamic changes of crop growth and soil moisture. Many of the parameters are difficult to measure directly for different crop varieties or environments, recalibrations are often needed. Determining the importance of specific parameters to the model outputs is helpful to simplify the crop model calibrations. Sensitivity analysis（SA） can quantify the impact of input parameters on the model outputs and is helpful for model parameterizations. This study aimed to obtain model parameters of DSSAT-CROPGRO-Cotton model for irrigation schedule optimization of cotton in Xinjiang, China through sensitivity and uncertainty analyses. Based on the field cotton experiments in Shihezi Region of Xinjiang Uygur autonomous region, the Morris method and extended Fourier amplitude sensitivity test（EFAST） method were applied to analyze the sensitivity of six outputs of the CROPGRO-Cotton model to the variety and soil parameters at three irrigation levels. The model outputs included days of initial flowering and maturing, seed cotton yield, aboveground dry biomass, maximum leaf area index and evapotranspiration. In addition, the correlation between the two methods was analyzed and the uncertainty analysis was conducted for the model outputs from the EFAST method. Results showed that EFAST method was better than Morris method in sensitivity test. The Spearman rank correlation analysis showed that the correlation coefficient was between 0.87 and 0.93 for the aboveground dry biomass and maximum leaf area index, and between 0.66 and 0.81 for the days of maturing, seed cotton yield and evapotranspiration. The numbers of sensitive parameters was smaller from Morris method than EFAST method, indicating that Morris method may oversimplify sensitivity problem. Sensitivity and uncertainty analyses indicated that irrigation levels had no significant effects on the days of initial flowering and a simplistic parameter sensitivity issue existed for simulation of the days of initial flowering and maximum leaf area index. Soil parameters in different soil layers had different effects on the model outputs. The days of maturing were more sensitive to the soil parameters in soil layer of 40-80 cm, but the aboveground dry biomass and evapotranspiration were more sensitive to the soil parameters in soil layer of 80-120 cm. The maximum leaf area index and evapotranspiration were both overestimated to a certain extent, it was necessary to make an improvement so as to enhance the simulation accuracy before this model could be applied in Xinjiang. The most sensitive parameters for cotton mature days simulation was time between plant emergence and flower appearance（EMFL）, time between the first flower and first seed（FLSD） and time between the first seed and physiological maturity（SDPM）. The most sensitive parameters for seed cotton yield simulation was maximum fraction of daily growth that was partitioned to seed and shell（XFRT） and the most sensitive parameters for aboveground dry mass are drainage rate（SLDR） and field capacity in soil layer from 80 to 120 cm（SDUL3）. The most sensitive parameters for evapotranspiration simulation was SDUL3. The parameter of seed filling duration for pod cohort at standard growth conditions（SFDUR） was not sensitive for all outputs and thus could be set as constant values. Surface albedo（SALB）, runoff curve number（SLRO） and saturated hydraulic conductivity（SSKS1） were only sensitive to mature days. The results above would help to improve simulation efficiency and precision of CROPGRO-Cotton model in Xinjiang region.