光谱信息与作物生长模型数据同化中的时间尺度优化

Temporal scale optimization for assimilation of spectral information and crop growth model

光谱信息与作物生长模型同化的效率提升是同化方法区域应用研究的一个重要方面。该文通过设置不同步长的光谱观测值同化时相,开展针对光谱信息与作物生长模型WOFOST(world food studies)同化的时间尺度优化研究,以提高同化效率。基于长春地区水稻生长周期,该文设置了4个等距时间尺度(步长分别为5,10,20和30 d)和一个关键时相时间尺度(同化时相对应水稻生长关键时期),在不同时间尺度下利用光谱信息计算的修正叶绿素吸收比值指数MCARI1(modified chlorophyll absorption ratio index)同化WOFOST模型,通过比较不同时间尺度下的同化精度和效率,优化同化时间尺度。结果表明:随着同化时间尺度增大,同化效率逐渐提高,而同化精度逐渐降低。在平衡精度和效率的前提下,选择步长介于10~20 d的时间尺度或关键时相尺度作为光谱信息与作物生长模型的同化时间尺度是合理的。该文提出的优化同化时间尺度方法为提高光谱信息与作物生长模型同化的区域应用效果提供了参考。

The improvement in the efficiency (running time) of the assimilation of spectral information into the crop growth model is an important researching aspect of applying the assimilation method at the regional scale. In this study, for reducing the running time while maintaining the performance, the temporal scale optimization of the assimilation was carried out by setting the different step sizes of time phases at which remote sensing observed values were assimilated into the coupling model of crop growth model WOFOST (world food studies) and radiative transfer model PROSPECT+SAIL. Based on the growth cycle of rice in Changchun, Jilin Province, China, four equidistant temporal scales (the step sizes of them were 5 , 10, 20 and 30 days, respectively) and a crucial temporal scale (corresponding to the crucial growth period of rice) were set for assimilation. The time phases of crucial temporal scale were selected by taking the derivative of the time series curve of the leaf area index (LAI). The time phases correspond to the extreme points and inflection points of LAI or LAI growth rate curves were also selected, which were the crucial periods of the growth process or the demarcation points of different growth stages. Then the vegetation indices-modified chlorophyll absorption ratio index (MCARI1) were calculated from the spectral information on the corresponding time phases of each temporal scale, and then assimilated into the coupling model WOFOST+PROSPECT+SAIL to optimize the input parameters day of transplanting (IDTR) and temperature sum from sowing to transplanting (TSUMST) by using the assimilation algorithm - particle swarm optimization (PSO). Finally, the assimilation temporal scale was optimized by comparing the assimilation efficiency and the simulated accuracy of crop parameters, i.e., LAI, total above ground production (TAGP) and dry weight of storage organs (WSO), at the five different temporal scales. The results showed that the assimilation efficiency was raised and the accuracy gradually decreased as the step size of equidistant temporal scales increased. Although the number of the assimilation time phases at crucial temporal scale was greatly reduced comparing with that at equidistant temporal scale with the step size of 10 days, the assimilation efficiency at the crucial temporal scale was promoted and the assimilation accuracy at the crucial temporal scale was close to it. Compared with the equidistant temporal scale with the step size of 20 days, the increasing rate of the assimilation efficiency at the crucial temporal scale was faster than the decreasing rate of the assimilation accuracy at the crucial temporal scale. On the premise of balancing the assimilation efficiency and accuracy, the temporal scale with the step size ranging from 10 days to 20 days, and the crucial temporal scale were regarded as the reasonable temporal scales of assimilating spectral information into the crop growth model for the growth simulation of rice. In this study, we proposed a novel method of selecting the time phases for assimilation by setting the equidistant temporal scales or extracting the crucial periods of crop growth process. It provided a reference for improving the application effect of the assimilation method at regional scale by optimizing the assimilation temporal scale.