区域冬小麦生长模拟遥感数据同化的不确定性研究

发布时间：2018-03-02 02:34

  本文关键词： 作物生长模型 遥感数据同化 叶面积指数 不确定性 估产　出处：《中国农业科学院》2016年博士论文　论文类型：学位论文

【摘要】：及时、准确、大范围地进行区域作物生长监测和产量预测对于指导农业生产、保障粮食安全、促进农业可持续发展具有重要意义。作物生长模型遥感数据同化方法是解决区域作物生长模拟的有效途径。但作物生长模拟遥感数据同化模拟过程复杂,在实际应用中还存在很多不确定因素。深入研究和分析这些不确定性问题对作物模型区域化、提高区域同化模拟的精度具有重要理论研究价值和实际应用需求,有助于提高区域农情遥感监测的技术能力。本研究以作物模型遥感数据同化的不确定性为研究核心,从同化系统模型初始条件和模拟过程、同化算法、关键参数叶面积指数(LAI)遥感反演、气象驱动及遥感观测误差、时空尺度等方面出发,首先在作物模型本地化基础上,分别考虑模型模拟过程和初始条件扰动的不同模拟效果;然后,对比粒子滤波(PF)和本特征正交分解的四维变分(POD4DVAR)同化模拟结果,分析粒子/集合维数和扰动方差影响;接着,利用GF-1 WFV、HJ-1 CCD和Landsat-8 OLI影像数据,应用PROSAIL模型进行时序LAI遥感反演;最后进行区域同化模拟,并分析气象驱动、观测误差、时空尺度等方面的不确定性影响。主要研究结论如下:(1)通过对比初始条件优化、PF同化和初始条件扰动同步PF同化三种同化方案,得出第一种方案未考虑同化模拟过程不确定性,无法提高同化估产精度,其余方案估产精度较高,其中第三种同化方案最优,相对误差(RE)和均方根误差(RMSE)分别为6.00%和544kg/ha。研究发现观测误差的增加会使同化模拟精度降低;还发现同化时间点的选择对同化模拟的精度具有影响,在冬小麦孕穗期、抽穗期和拔节期同化观测数据可明显提高同化模拟精度。(2)PF和POD4DVAR两种同化方案均可以提高模拟精度。其中以POD4DVAR同化精度最高,RE为5.65%,RMSE为523kg/ha。研究发现粒子/集合维数从50增加到200时,同化模拟精度提高较小,但计算代价增加超过8倍;随着扰动方差降低,同化模拟精度增大。因此,在实际应用中选择合适同化算法、粒子/集合数和扰动方差,对于减少冬小麦生长同化模拟至关重要作用。(3)研究利用同步的GF-1 WFV、HJ-1 CCD和Landsat-8 OLI数据,在分析反射率和植被指数一致性的基础上,应用PROSAIL模型反演得到衡水冬小麦时序LAI结果,经人工和仪器测量两种实测LAI验证,总体RE分别为5.72%和9.44%,RMSE为0.26和0.39,表明时序LAI反演结果满足区域同化研究需求。(4)研究基于最优POD4DVAR同化方案进行区域同化估产,利用官方统计数据进行验证,估产精度较高,RE为8.32%,RMSE为452kg/ha。分析气象驱动不确定性影响,发现单气象站数据同化结果,无法反映各区域实际产量变化特性。研究还发现随着遥感LAI误差增大,估产精度降低,但减少趋势较小,同化在一定程度上消减了部分观测误差影响造成的同化模拟不确定性。时空尺度方面,同化观测频率增大,模拟精度提高;同化孕穗、抽穗和拔节三个生育期观测均可明显提高模拟精度;同化前、中期物候阶段观测也可显著提高估产精度。遥感观测同化空间分辨率降低,模拟精度降低,但计算效率提高。因此需综合考虑估产精度和计算代价,选择合理的时空尺度,以满足实际区域冬小麦生同化估产需求。
[Abstract]:Timely, accurate, large range of regional crop growth monitoring and yield prediction to guide agricultural production, food security, is of great significance to promote the sustainable development of agriculture. The crop growth model of remote sensing data assimilation method is an effective way to solve the regional crop growth simulation. But the crop growth simulation of remote sensing data assimilation process is complex, in the practical application there are still many uncertain factors. Research and analysis of these uncertain problems of regional crop model, has important theoretical value and practical application needs to improve the Regional Assimilation and the accuracy of the simulation, is helpful to improve the technical ability of regional agricultural remote sensing monitoring. This study focusing on crop model assimilation of remote sensing data uncertainty. From the initial conditions of model assimilation system and the simulation process, assimilation algorithm, key parameters of leaf area index (LAI) remote sensing inversion, The driving meteorological and remote sensing observation error, spatial scales and other aspects, first in the localization based on crop model, considering the model simulation of different simulation results of disturbance process and the initial conditions; then, compared with particle filter (PF) and the orthogonal decomposition feature of four-dimensional variational assimilation (POD4DVAR) simulation results, analysis of the particle / dimension set and then, the effect of perturbation variance; GF-1 WFV, HJ-1 CCD and Landsat-8 OLI image data, PROSAIL model was applied to temporal LAI remote sensing inversion; finally, Regional Assimilation and simulation, and analysis of meteorological observation error, drive, and other aspects of the impact of spatial and temporal scales of uncertainty. The main conclusions are as follows: (1) by comparing the optimized initial conditions PF, assimilation and initial condition perturbations three assimilation scheme of synchronous PF assimilation, that the first scheme does not consider the assimilation process uncertainty, to improve the estimating accuracy of assimilation, The scheme of estimating precision is higher, of which third kinds of optimal assimilation scheme, the relative error (RE) and the root mean square error (RMSE) were 6% and 544kg/ha. study found increasing observation error will reduce the precision of the model is that assimilation; assimilation time point selection has an impact on the assimilation of simulation accuracy in winter wheat at booting stage observation, data assimilation heading stage and jointing stage can significantly improve the simulation accuracy of assimilation. (2) two PF and POD4DVAR assimilation scheme can improve the simulation precision. The POD4DVAR assimilation of the highest accuracy, RE 5.65%, RMSE 523kg/ ha. found that the particle / set dimension increased from 50 to 200, the smaller increase assimilation the accuracy, but the computational cost increases more than 8 times; with the perturbation variance reduction, assimilation simulation accuracy increases. Therefore, choosing appropriate Assimilation Algorithm in practical application, the particle / set number and perturbation variance, to reduce the winter Simulation of vital wheat growth. Assimilation (3) by GF-1 WFV HJ-1 and Landsat-8 CCD synchronization, OLI data, based on the analysis of reflectance and vegetation index consistency, PROSAIL model is applied to inversion of Winter Wheat in Hengshui time LAI results by artificial and instrument measuring two kinds of measured LAI verification, overall RE was 5.72% and 9.44%, RMSE was 0.26 and 0.39, indicating that the timing of LAI inversion results meet the requirements of Regional Assimilation research. (4) research on Regional Assimilation Scheme Based on the optimal estimation of POD4DVAR assimilation, verified by the official statistics, estimation accuracy, RE 8.32%, RMSE 452kg/ha. analysis of weather driven uncertainty, found data assimilation results the weather station, cannot reflect the actual change of output characteristics of each region. The study also found that with the increase of LAI remote sensing error, estimation accuracy decreases, but the decreasing trend is small, in a certain range of assimilation The degree by assimilation effects caused by partial observation uncertainty of error. Temporal and spatial scales, assimilation frequency increases, improve simulation accuracy; assimilation and jointing booting, heading three growth period of observation can significantly improve simulation accuracy; assimilation, mid stage phenological observations can significantly improve the estimation accuracy. The remote sensing spatial assimilation lower resolution, simulation accuracy is reduced, but the calculation efficiency is improved. Therefore the need to consider the estimation accuracy and computational cost, reasonable selection of temporal and spatial scales, to meet the actual regional winter wheat production yield assimilation needs.

【学位授予单位】：中国农业科学院
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S127;S512.11
，

本文编号：1554683