基于多变量和数据同化算法的冬小麦单产估测

发布时间：2018-01-16 23:13

  本文关键词：基于多变量和数据同化算法的冬小麦单产估测　出处：《中国农业大学》2017年博士论文　论文类型：学位论文

  更多相关文章： 冬小麦 产量估测 CERES-Wheat模型 数据同化 条件植被温度指数

【摘要】：作物长势监测及产量的准确估测、预测是粮食安全的重要保障。遥感技术的发展为大面积、实时、动态的作物产量估测、预测提供了有效途径,其中,基于数据同化算法耦合作物生长模型和遥感数据,能充分考虑到作物生长的内在机理过程以及环境因素对作物生长发育的影响,同时能有效地解决作物生长模型区域参数获取困难的问题。以陕西省关中平原为研究区域,获取冬小麦生长季的Landsat遥感数据,利用Landsat数据计算归一化植被指数(NDVI)和反演条件植被温度指数(VTCI),分析NDVI和实测叶面积指数(LAI)、NDVI和实测地上生物量(β)以及VTCI和实测0～20 cm层土壤含水量(θ)间的相关性,从而构建回归模型以估测区域LAI、β和θ。通过田间实测和调查的LAI、β、θ、小麦单产和收获日期对CERES-Wheat模型的遗传参数进行标定,对标定的CERES-Wheat模型的模拟结果进行验证,结果表明,模拟β与实测β的平均相对误差(MRE)及模拟θ与实测θ的MRE均小于10%,模拟单产和实测单产的MRE小于15%,模拟小麦收获日期与实际日期的偏差小于4 d,说明标定的CERES-Wheat模型的模拟精度较高。利用四维变分(4DVAR)、集合卡尔曼滤波(EnKF)和粒子滤波(PF)算法同化CERES-Wheat模型模拟的和Landsat数据反演的LAI、β和θ,以获取冬小麦主要生育时期的LAI、β和θ同化值,通过田间实测数据分别检验和对比3种算法同化LAI、β和θ的精度,结果表明,基于3种算法的同化变量均能有效地表达CERES-Wheat模型模拟LAI、β和θ在小麦不同生育时期的变化特征,同时在遥感反演变量的影响下,同化变量比模拟变量更接近实测值。PF算法同化LAI和β的精度高于EnKF和4DVAR算法同化LAI和β的精度,且基于PF算法能较好地表达模拟LAI和β的变化特征,因此,PF算法同化LAI和β的效果最优。EnKF算法在表达模拟LAI和β的变化特征方面,效果优于4DVAR算法的同化效果,然而,EnKF算法同化LAI和β的精度低于4DVAR算法同化LAI和β的精度。4DVAR算法在表达模拟LAI和β的变化特征方面效果较好,且同化LAI和β的精度也较高,然而,4DVAR算法由于需要设置同化时间窗口而缺少同化时间段后10天的同化值。将小麦各生育时期的LAI、β和θ同化值与实测单产进行线性回归分析,构建单个生育时期的估产模型,利用熵值的组合预测法计算单个生育时期估产模型的权系数,构建组合估产模型,通过实测单产检验和对比不同估产模型的精度,结果表明,在4DVAR、EnKF和PF三种算法中,基于PF算法的估产模型的精度大于EnKF和4DVAR算法的估产模型的精度。进一步对比和分析小麦不同生育时期的LAI、β和θ同化值与实测单产间的相关性,选取与实测单产相关性较大的变量作为最优同化变量,分别在小麦各生育时期同化双变量(LAI和β、或LAI和θ、或θ和β)、同化多变量(LAI、β和θ)和同化最优变量构建估产模型,结果表明,在返青期θ为最优同化变量,在乳熟期β为最优同化变量,在拔节期和抽穗-灌浆期LAI、β和θ为最优同化状态量。在小麦各生育时期同化最优变量的估产模型的精度(R2=0.81,RMSE=317.85 kg·ha-1)大于同化多变量的估产模型的精度(R2=0.76,RMSE=348.64 kg·ha-1),同时同化LAI、β和θ的估产模型的精度大于同化双变量的估产模型的精度,且同化双变量的估产模型的精度大于同化单个变量的估产模型的精度。因此,在作物不同生育时期同化和籽粒产量相关性较大的变量,能够有效地提高作物单产估测精度。关中平原冬小麦地分为灌溉地和旱地,区分灌溉地和旱地的状态变量(LAI、β和θ)与实测单产间的相关性大于不区分灌溉地和旱地的状态变量与实测单产间的相关性,因此,在灌溉地和旱地分别构建同化估产模型,其单产估测精度(R2=0.85,RMSE=287.55kg·ha-1)大于不区分灌溉地和旱地的估产模型精度。利用区分灌溉地和旱地的模型估测2007—2008年以及2013—2016年的关中平原冬小麦单产,分析小麦单产的区域分布特征,结果表明,关中平原中部和西部的小麦地分布密集,且小麦平均单产较高;关中平原北部和东部的小麦地分布较零散,且小麦平均单产低于中部和西部的小麦平均单产。
[Abstract]:The accurate estimation of crop growth monitoring and yield prediction, is an important guarantee of food safety. The development of remote sensing technology for large area, real-time, dynamic estimation of crop yield, which provides an effective way of prediction, data assimilation algorithm coupled crop growth model and remote sensing data based on, can fully take into account the intrinsic mechanism and the process of crop growth environmental factors influence the growth and development of crops, and can effectively solve the regional crop growth model parameters acquisition problem. The Guanzhong Plain in Shaanxi Province as the study area, Landsat remote sensing data acquisition of winter wheat growth season, the calculation of normalized difference vegetation index (NDVI) by using Landsat data and inversion of vegetation temperature condition index (VTCI) analysis. NDVI and the measured leaf area index (LAI), NDVI and measured aboveground biomass (beta) and the VTCI and measured 0 ~ 20 cm soil moisture (0) and the relationship between. Build a regression model to estimate the regional LAI, beta and theta. Through field measurement and investigation of LAI, beta theta, genetic parameters of the CERES-Wheat model of wheat yield and harvest date of calibration, the simulation results of CERES-Wheat model calibration is verified, the results show that the average relative error of simulated and measured beta beta (MRE) and the simulated and measured theta theta MRE were less than 10%, the yield and yield of the simulated measured MRE less than 15%. The deviation of the simulated wheat harvest date and the actual date of less than 4 D, indicating a higher simulation accuracy of CERES-Wheat model calibration. Using the four-dimensional variational (4DVAR), a collection of Calman filter (EnKF) and particle filter (PF CERES-Wheat) algorithm assimilation model simulation and inversion of Landsat data LAI, beta and theta, to obtain the main growth stages of winter wheat LAI, and beta theta value by field measured data assimilation, respectively test and comparison of 3 algorithms of LAI assimilation, and beta 0 precision, results show that the 3 algorithms can effectively assimilate the variable expression of CERES-Wheat model based on LAI, beta and theta changes in the characteristics of wheat in different growth stages, while the influence of remote sensing variables, assimilation variables than analog variables is more close to the measured value of.PF algorithm has higher accuracy than LAI and beta assimilation EnKF and 4DVAR algorithm and the accuracy of the assimilation of LAI beta, and based on the characteristics of PF algorithm can better simulate the expression of LAI and beta, therefore, the PF algorithm and the effect of assimilation of LAI beta.EnKF optimal algorithm in the expression of LAI and beta simulation, but the effect is better than 4DVAR algorithm, EnKF algorithm, assimilation, assimilation LAI and beta is less precise than the accuracy of the.4DVAR algorithm 4DVAR algorithm LAI assimilation and beta good results in changes of expression of LAI and beta simulation, and the assimilation of LAI and beta, the precision is higher, however, because of the need to set the same 4DVAR algorithm The time window of assimilation and lack of assimilation time after 10 days. The value of wheat in different growth periods of LAI, beta and theta value assimilation by the linear regression analysis with the measured yield estimation model, construction of single growth period, yield coefficient of single period model using entropy combination forecasting method, constructs the combined estimation model through the measurement, analysis and comparison of different yield estimation model accuracy, results showed that in 4DVAR, EnKF and PF three algorithm, the estimation model estimation model PF algorithm accuracy is greater than EnKF and 4DVAR algorithm. Based on the accuracy of further comparison and analysis of different growth stages of wheat LAI, beta and theta value assimilation the correlation between yield and measured, and the measured yield significantly larger correlation variables as the optimal variables respectively in the assimilation and assimilation of wheat at different growth stages of two variables (LAI and beta, or LAI and theta, or theta and beta), and assimilation The amount of (LAI, beta and theta) and assimilation of optimal variables to construct yield estimation model, the results show that the optimal assimilation in the regreening period theta is variable in the milk stage beta is the optimal assimilation variable, in the jointing stage and heading filling LAI, beta and theta is the optimal state estimation model. The assimilation of wheat at different growth stages the precision of the optimal assimilation variables (R2=0.81, RMSE=317.85 kg HA-1) estimation model of multi variable precision than assimilation (R2=0.76, RMSE=348.64, kg, HA-1) and LAI assimilation, yield estimation model and the estimation model of beta theta assimilation accuracy is higher than the double variable precision estimation model and estimation model of a bivariate assimilation the accuracy is greater than the single variable assimilation accuracy. Therefore, the crop in different growth period of assimilation and grain yield correlated variables, can effectively improve the accuracy of crop yield estimation of Winter Wheat in Guanzhong Plain. Divided into irrigation and dry land irrigation, distinguish The state variable irrigation and dry land (LAI, beta and theta) is greater than the correlation between state variables, and the measured yield does not distinguish between irrigation and dry land between the measured and correlation between the yield estimation models were constructed, assimilation in irrigated and dry land, the yield estimation accuracy (R2=0.85, RMSE=287.55kg, HA-1) is greater than the yield estimation accuracy the model does not distinguish between irrigation and dry land. The distinction between irrigation and dryland estimation model of 2007 - 2008 and 2013 - 2016 winter wheat yield in Guanzhong Plain, the analysis of regional distribution, the results showed that the wheat yield in Guanzhong Plain, central and Western wheat distribution intensive, and wheat average yield of wheat in Guanzhong Plain in northern high; and the eastern part of the distribution is scattered, and the average yield of wheat is lower than the average yield of wheat in the central and western regions.

【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：S512.11
，

本文编号：1435240