基于物理模型训练神经网络的作物叶面积指数遥感反演研究

叶面积指数(LAI)是估算作物生长的关键参数。基于物理模型的LAI反演,被认为是当前最为可靠的方法,但其反演复杂。本文提出了将物理模型和神经网络相结合,从地表反射率反演叶面积指数的算法,利用MOD IS地表反射率和4-scale模型反演作物LAI。(1)利用4-scale模型模拟不同LAI与地表反射率的关系,生成训练数据;(2)利用模型模拟的LAI训练神经网络;(3)以MOD IS地表反射率输入训练后的神经网络,反演LAI。估算的LAI与其他LAI产品进行了比较,结果表明,估算的作物LAI和MOD IS及CYCLOPES LAI产品空间和时间分布一致,均方根误差分别为0.4994和0.6558。以2004年衡水的作物LAI地面观测数据进行了直接验证,估算的LAI与研究区地表植被分布一致,但是,三种卫星LAI产品都小于地表测量,故需针对华北平原浓密作物设计模型参数化方案。     

同化叶面积指数和蒸散发双变量的冬小麦产量估测方法

同化遥感信息到作物生长过程模拟模型,是估测区域作物产量的重要方法之一。同化变量的选取对同化结果精度至关重要。本文在标定WOFOST作物模型参数的基础上,优化了WOFOST模型的默认灌溉参数。利用ET和LAI作为同化变量,分别构建了时间序列趋势信息的代价函数和四维变分代价函数;采用SCE-UA算法最小化代价函数, 重新初始化WOFOST模型初始参数——作物初始干物质重、作物35 ℃生命期和灌溉量。最后利用MODIS LAI产品（MCD15A3）、MODIS ET产品（MOD16A2）,同化到作物模型估测产量,并对比分析了水分胁迫模式下同化单变量（ET或LAI）和同化双变量（ET和LAI）的估产精度。结果表明：同化双变量ET和LAI的策略,优于同化单变量LAI或ET,双变量策略的冬小麦产量估测精度为R²=0.432,RMSE=721 kg/hm²;单独同化高精度LAI对提高估产精度具有重要作用,其冬小麦产量估测精度为R²=0.408,RMSE=925 kg/hm²;单独同化ET的趋势信息改善了WOFOST模型模拟水分平衡的参数,但是,产量估测精度（R²=0.013,RMSE=1134 kg/hm²）与模型模拟估测产量精度（R²=0.006,RMSE=1210 kg/hm²）相比改善效果有限。本研究为其他区域的遥感数据与作物模型的双变量数据同化的作物产量估测研究提供了参考价值。     

基于微波数据与光学数据集成的机器学习技术在作物产量估算中的应用

各类光学植被指数已成功地应用于各种植被监测与作物产量估算中,但这些指数易受大气状况的影响。由星载微波辐射计得到的植被光学厚度数据（VOD）与植被密度、含水量密切相关,数据可全天候获得,在农业遥感监测中呈现着巨大的潜力。作为来自不同传感器的遥感数据,微波遥感数据与光学遥感数据可以提供不同波长范围内的植被信息。为了更准确地进行作物产量估算,本研究提出将微波遥感数据与光学遥感数据共同应用于冬小麦单产估算中。研究选择L波段微波辐射计SMAP卫星的VOD数据与MODIS的标准归一化植被指数NDVI、增强型植被指数EVI、叶面积指数LAI、光合有效辐射分量FPAR数据作为研究变量,分别使用BP神经网络、GA-BP神经网络和PSO-BP神经网络建立冬小麦产量估算模型。结果表明： 3种神经网络回归模型的P值均小于0.001,通过了显著性检验。GA-BP神经网络回归模型的估算值与真实值在3种神经网络回归模型中表现了最高的相关性（R=0.755）与最低的均方根误差（RMSE=529.145 kg/hm²）,平均绝对误差(MAE=425.168 kg/hm²)和平均相对误差（MRE=6.530%）。为了分析多源遥感数据的结合在作物产量估算中的优势,研究同时构建了仅使用NDVI和LAI,使用NDVI、EVI、LAI、FPAR等光学数据进行冬小麦产量估算的3种GA-BP神经网络回归模型作为对比。结果表明,使用微波遥感数据与光学遥感数建立的GA-BP神经网络回归模型较上述3种作为对比的GA-BP神经网络回归模型的相关系数R值分别提高了0.163,0.229与0.056,均方根误差RMSE分别降低了122.334、158.462和46.923 kg/hm²,使用多源遥感数据的组合可以很好地提高作物产量估算的准确性。     

Assimilating ASAR data for estimating soil moisture profile using an ensemble Kalman filter

Active microwave remote sensing data were used to calculate the near-surface soil moisture in the vegetated areas. In this study, Advanced Synthetic Aperture Radar （ASAR） observations of surface soil moisture content were used in a data assimilation framework to improve the estimation of the soil moisture profile at the middle reaches of the Heihe River Basin, Northwest China. A one-dimensional soil moisture assimilation system based on the ensemble Kalman filter （EnKF）, the forward radiative transfer model, crop model, and the Distributed Hydrology-Soil-Vegetation Model （DHSVM） was developed. The crop model, as a semi-empirical model, was used to estimate the surface backscattering of vegetated areas. The DHSVM is a distributed hydrology-vegetation model that explicitly represents the effects of topography and vegetation on water fluxes through the landscape. Numerical experiments were con- ducted to assimilate the ASAR data into the DHSVM and in situ soil moisture at the middle reaches of the Heihe River Basin from June 20 to July 15, 2008. The results indicated that EnKF is effective for assimilating ASAR observations into the hydrological model. Compared with the simulation and in situ observations, the assimilated results were significantly improved in the surface layer and root layer, and the soil moisture varied slightly in the deep layer. Additionally, EnKF is an efficient approach to handle the strongly nonlinear problem which is practical and effective for soil moisture estimation by assimilation of remote sensing data. Moreover, to improve the assimilation results, further studies on obtaining more reliable forcing data and model parameters and increasing the efficiency and accuracy of the remote sensing observations are needed, also improving estimation accuracy of model operator is important.     

LAI scale effect research based on compact airborne spectrographic imager data in the Heihe Oasis

As one of the key parameters for characterizing crop canopy structure, Leaf Area Index(LAI) has great significance in monitoring the crop growth and estimating the yield. However, due to the nonlinearity and spatial heterogeneity of LAI inversion model, there exists scale error in LAI inversion result, which limits the application of LAI product from different remote sensing data. Therefore, it is necessary to conduct studies on scale effect. This study was based on the Heihe Oasis, Zhangye city, Gansu province, China and the following works were carried out: Airborne hyperspectral CASI(Compact Airborne Spectrographic Imager) image and LAI statistic models were adopted in muti-scale LAI inversion. The overall difference of muti-scale LAI inversion was analyzed in an all-round way. This was based on two aspects, "first inversion and then integration" and "first integration and then inversion", and on scale difference characteristics of three scale transformation methods. The generation mechanism of scale effect was refined, and the optimal LAI inversion model was expanded by Taylor expansion. By doing so, it quantitatively analyzed the contribution of various inversion processes to scale effect. It was found that the cubic polynomial regression model based on NDVI(940.7 nm, 712 nm) was the optimal model, where its coefficient of determination R2 and the correlation coefficient of test samples R reached 0.72 and 0.936, respectively. Combined with Taylor expansion, it analyzed the scale error generated by LAI inversion model. After the scale effect correction of one-dimensional and twodimensional variables, the correlation coefficient of CCD-LAI(China Environment Satellite HJ/CCD images) and CASI-LAI products(Compact Airborne Spectro graphic Imager products) increased from 0.793 to 0.875 and 0.901, respectively. The mean value, standard deviation, and relative true value of the two went consistent. Compared with onedimensional variable correction method, the twodimensional method had a better correction result. This research used the effective information in hyperspectral data as sub-pixels and adopted Taylor expansion to correct the scale error in large-scale and low-resolution LAI product, achieving large-scale and high-precision LAI monitoring.     

基于LTDR AVHRR和MODIS观测的全球长时间序列叶面积指数遥感反演

叶面积指数是描述土壤-植被-大气之间物质和能量交换的关键参数,获取大区域长时间序列叶面积指数有助于研究气候变化条件下植被的响应及反馈。本文利用MODIS观测和经过重新处理的地表长时间数据集（Land Long Term Data Record）LTDR AVHRR数据,生成了全球1981-2012年叶面积指数数据。算法通过建立二者之间像元级关系,利用高质量MODIS观测约束历史AVHRR数据的反演,这有助于减小2种存在显著差别传感器反演结果的不一致性,也有助于提高AVHRR反演质量。首先算法利用高质量MODIS地表反射率反演2000-2012年叶面积指数,然后利用多年每8 d的LTDR AVHRR地表反射率数据计算简单比植被指数（Simple Ratio,SR）,利用SR平均值和MODIS LAI平均值建立像元级AVHRR SR-MODIS LAI关系。在此基础上,实现1981-1999年AVHRR LAI反演,最终得到全球1981-2012年叶面积指数数据。本算法反演的AVHRR和MODIS LAI与全球植被的空间分布吻合,能表征主要生物群系类型的季节变化特征,2个数据集一致性较好,并且与NASA MODIS LAI标准产品（MOD15A2）的空间分布和季节变化曲线吻合较好。     

Multiple cropping intensity in China derived from agro-meteorological observations and MODIS data

Double-and triple-cropping in a year have played a very important role in meeting the rising need for food in China.However,the intensified agricultural practices have significantly altered biogeochemical cycles and soil quality.Understanding and mapping cropping intensity in China′s agricultural systems are therefore necessary to better estimate carbon,nitrogen and water fluxes within agro-ecosystems on the national scale.In this study,we investigated the spatial pattern of crop calendar and multiple cropping rotations in China using phenological records from 394 agro-meteorological stations(AMSs)across China.The results from the analysis of in situ field observations were used to develop a new algorithm that identifies the spatial distribution of multiple cropping in China from moderate resolution imaging spectroradiometer(MODIS)time series data with a 500 m spatial resolution and an 8-day temporal resolution.According to the MODIS-derived multiple cropping distribution in 2002,the proportion of cropland cultivated with multiple crops reached 34%in China.Double-cropping accounted for approximately 94.6%and triple-cropping for 5.4%.The results demonstrat that MODIS EVI(Enhanced Vegetation Index)time series data have the capability and potential to delineate the dynamics of double-and triple-cropping practices.The resultant multiple cropping map could be used to evaluate the impacts of agricultural intensification on biogeochemical cycles.     

A MODIS time series data based algorithm for mapping forest fire burned area

Burned area mapping is an essential step in the forest fire research to investigate the relationship between forest fire and climate change and the effect of forest fire on carbon budgets. This study proposed an algorithm to map forest fire burned area using the Moderate-Resolution Imaging Spectroradiameter (MODIS) time series data in Heilongjiang Province, China. The algorithm is divided into two steps: Firstly, the ‘core’ pixels were extracted to represent the most possible burned pixels based on the comparison of the temporal change of Global Environmental Monitoring Index (GEMI), Burned Area Index (BAI) and MODIS active fire products between pre- and post-fires. Secondly, a 15-km distance was set to extract the entire burned areas near the ‘core’ pixels as more relaxed conditions were used to identify the fire pixels for reducing the omission error as much as possible. The algorithm comprehensively considered the thermal characteristics and the spectral change between pre- and post-fires, which are represented by the MODIS fire products and the spectral index, respectively. Tahe, Mohe and Huma counties of Heilongjiang Province, China were chosen as the study area for burned area mapping and a time series of burned maps were produced from 2000 to 2011. The results show that the algorithm can extract burned areas more accurately with the highest accuracy of 96.61%.     

面向GF-1 WFV数据和MODIS数据的时空融合算法对比分析

对城市热岛效应、植物覆盖指数、叶面积指数等地表参数的高频次高精度反演,能更好地实现基于遥感手段的地表特征动态监测。然而,目前单一数据源的遥感影像还很难实现高时空分辨率数据的同步获取,时空融合技术是解决这个时空分辨率矛盾的有效方法。根据原理不同,时空融合算法可以分为基于线性模型的融合算法、基于光谱解混的融合算法等。高分卫星产品是近几年中国高分辨率对地观测系统重大专项天基系统中的首发星,对于该类数据的时空融合研究仍然较少。因此,本文拟采用4种常见的时空融合算法（STARFM、FSDAF、STDFA、Fit_FC）实现GF-1 WFV数据与MODIS数据的时空融合,分析这几种方法对GF-1 WFV数据时空融合的有效性和精度,从而为后续的研究提供一定依据。     

Practical Model of Sea Ice Thickness of Bohai Sea Based on MODIS Data

Sea ice thickness is one of the most important input parameters in the studies on sea ice disaster prevention and mitigation. It is also the most important content in remote sensing monitoring of sea ice. In this study, a practical model of sea ice thickness (PMSIT) was proposed based on the Moderate Resolution Imaging Spectroradiometer (MODIS) data. In the proposed model, the MODIS data of the first band were used to estimate sea ice thickness and the difference between the second-band reflectance and the fifth-band reflectance in the MODIS data was calculated to obtain the difference attenuation index (DAI) of each pixel. The obtained DAI was used to estimate the integrated attenuation coefficient of the first band of the MODIS at the pixel level. Then the model was used to estimate sea ice thickness in the Bohai Sea with the MODIS data and then validated with the actual sea ice survey data. The validation results showed that the proposed model and corresponding parameterization scheme could largely avoid the estimation error of sea ice thickness caused by the spatial and temporal heterogeneity of sea ice extinction and allowed the error of 18.7% compared with the measured sea ice thickness.     

玉米叶面积指数的CHRIS/PROBA数据反演分析

叶面积指数(LAI)是衡量植被生态状况和估算作物产量的一个重要指标。LAI的反演是定量遥感研究的重要内容。传统的经验统计反演方法基于单一观测角度的遥感数据进行,忽略了地物反射率的方向性。若在反演中加入多观测角度的信息,则有可能提升LAI反演的精度。以2008年甘肃省张掖市玉米实验区为研究区,利用欧空局的CHRIS/PROBA多角度高光谱数据对比分析了传统植被指数NDVI、RVI、EVI的变化规律及其反演玉米叶面积指数LAI的精度,并根据NDVI随观测角度的变化规律,构造出新型多角度归一化植被指数MNDVI,分别对实测叶面积指数进行线性回归并利用实测数据对估算LAI进行精度验证,结果表明：新型MNDVI指数相比于传统NDVI、RVI、EVI对LAI的反演精度有了显著提升,估算模型决定系数R²达到0.716,精度验证均方根误差为0.127,平均减小了33.3%。     

基于DCT-PLS算法的MODIS LST缺值填补方法研究

受云层、传感器误差等因素影响,中分辨率成像光谱仪（MODIS）获取的地表温度产品（LST）在时间和空间上存在大量缺失数据,严重影响了基于时间序列数据的分析与应用。本文引进一种基于离散余弦变换与惩罚最小二乘的多维数据快速平滑方法（DCT-PLS）,利用数据集自身的时间和空间信息填补缺值。本文在粤港澳大湾区开展实证研究,将DCT-PLS算法用于填补该地区2001年1月—2017年12月的月值MODIS LST数据缺值,并引入人工模拟缺值对算法进行误差分析与精度验证。算法误差分析结果表明,填补误差主要来源于三维算法对数据集中有偏LST时间信息的使用,并因此产生显著高估或低估的填补结果。基于此,本文提出了利用MODIS LST数据集自身时间序列信息自动计算获取有效辅助LST信息的优化策略,从而实现填补算法计算效率和精度的提升：平均计算时间从12.0 s提高至1.7 s,平均R从0.94提高至0.97,平均RMSE从1.94 K提高至0.74 K（相较于三维算法）。在大湾区的填补结果表明（日间结果：R=0.98、RMSE=0.79 K;夜间结果：R=0.99、RMSE=0.56 K）,优化后的DCT-PLS算法可以快速鲁棒地填补MODIS LST月值数据产品中的缺值,并且具有稳定性强、不依赖外部数据集的计算特性,能够适应长时间序列MODIS LST缺值填补。     

全球36 km格网土壤水分逐日估算

土壤水分是陆面生态系统和能量循环的核心变量之一,利用微波遥感技术获得的土壤水分产品的时间分辨率一般是2-3 d,因此精确地获得具有较高时间分辨率的土壤水分成了人们关注的焦点。本文尝试将SMAP (the Soil Moisture Passive and Active）土壤水分和MODIS光学数据相结合,利用广义回归神经网络进行全球36 km土壤水分的估算,提升SMAP土壤水分的时间分辨率。结果显示,广义回归神经网络估算土壤水分与SMAP保持了高相关性(r = 0.7528),但其却保留了较高的误差 (rmse = 0.0914 m³/m³)。尽管如此,估算的土壤水分能够很好地保持SMAP土壤水分的整体空间变化,并且提升了土壤水分的时间分辨率（1 d）。此处,本文研究了SMAP土壤水分与MODIS光学数据之间的关系,这对今后利用机器学习进行SMAP土壤水分降尺度研究提供了重要的参考价值。     

Kriging插值和序贯高斯条件模拟算法的对比分析

本文对Kriging插值与序贯高斯条件模拟值的算法联系进行了推导,并将两种计算结果和原始数据的统计参数作了对比。结果表明,Monto-carlo方法求得的序贯高斯条件模拟值经数学变换后等同于已知数据和此前模拟数据共同参与的Kriging插值结果与一个随机偏差的和,该随机偏差的均值为0,方差为Kriging误差方差。最优性条件导致Kriging插值结果的方差较原始数据降低了1个Kriging误差方差,造成Kriging平滑效应,其空间变异函数值降低,但自协方差函数值不变。序贯高斯条件模拟避免了平滑效应,其方差、变异函数和自协方差函数均不变,而其模拟值的误差方差较Kriging误差方差增加了1倍,表明1次随机模拟值的误差比Kriging插值大。然而,多次随机模拟值的平均值与Kriging插值的地理制图效果近似,可以弥补局部估值误差大的不足。因此,在应用中,Kriging插值是提供局部最优估计的方法,但却低估了全局的空间变异。而序贯高斯条件模拟的优点,在于提供若干等可能概率的模拟结果以进行估值的不确定性评价,并再现全局的空间可变性。     

2001—2017年中国不同耕作区重建MODIS LAI时空动态

叶面积指数Leaf Area Index （LAI）作为植被生物量指标之一,耕作区LAI不仅能反映作物的长势动态,且与农业生态、作物产量密切相关。本文通过对2001—2017年中国农田区域的MODIS-LAI长时序数据进行重建,利用Mann-Kendall检验、变异系数、重心迁移模型等方法分析了中国耕作区LAI的时空变化特征。结果表明：① 中国耕作区LAI在2001—2017年显波动式上升,且与农作物单产相关系数高达0.91;② 不同耕作区季节差异显著,夏季>秋季>春季>冬季,夏季平均为1.54,生长季平均为1.13,秋季平均为0.78,春季平均为0.63,冬季平均为0.31;③ 2001—2012年二熟、三熟区LAI变化平缓,2012年后有上升趋势但未发生明显突变;一熟区2006年之前处于平稳上升状态,2006年之后发生突变上升趋势显著;④ 研究时段内我国长江以北的耕作区LAI变异程度较为突出,最高达4.12; 农田面积重心经历了先向西南迁移,后再向西北迁移过程,农田生长季LAI重心相对于农田面积重心变幅较大,经历了南北波动式向西部迁移过程,迁移距离分别为82.78 km、90.53 km。     

Crop classification using MODIS NDVI data denoised by wavelet: A case study in Hebei Plain, China

Time-series Moderate Resolution Imaging Spectroradiometer (MODIS) Normalized Difference Vegetation Index (NDVI) data have been widely used for large area crop mapping.However,the temporal crop signatures generated from these data were always accompanied by noise.In this study,a denoising method combined with Time series Inverse Distance Weighted (T-IDW) interpolating and Discrete Wavelet Transform (DWT) was presented.The detail crop planting patterns in Hebei Plain,China were classified using denoised time-...     

干旱区草原地上植被生物量估算——以乌图美仁大草原芦苇植被为例

草原是干旱区生态系统中重要的可再生资源。本文基于草本植被的结构特征,利用ASAR和TM数据,结合MIMICS模型,提出了一种估算干旱区草原地上植被生物量的方法。该方法将光学遥感数据容易反演的叶面积指数（LAI）作为反演生物量模型的参数之一,并利用LAI成功估算了单位面积内的草本植被密度。将地上生物量作为输入变量代入改进的MIMICS模型,利用查找表方法,计算出地上植被生物量。然后,将该方法应用于乌图美仁草原的地上植被生物量的反演。结果表明,该方法能够成功地反演干旱区草原草本植被地上生物量,精度达到R²=0.8562,RMSD=0.6263。最后,分析了该方法估算植被生物量的误差来源。     

Assimilation of high frequency radar data into a shelf sea circulation model

High Frequency (HF) radar current data is assimilated into a shelf sea circulation model based on optimal interpolation (OI) method. The purpose of this work is to develop a real-time computationally highly efficient assimilation method to improve the forecast of shelf current. Since the true state of the ocean is not known, the specification of background error covariance is arduous. Usually, it is assumed or calculated from an ensemble of model states and is kept in constant. In our method, the spatial covariances of model forecast errors are derived from differences between the adjacent model forecast fields, which serve as the forecast tendencies. The assumption behind this is that forecast errors can resemble forecast tendencies, since variances are large when fields change quickly and small when fields change slowly. The implementation of HF radar data assimilation is found to yield good information for analyses. After assimilation, the root-mean-square error of model decreases significantly. Besides, three assimilation runs with variational observation density are implemented. The comparison of them indicates that the pattern described by observations is much more important than the amount of observations. It is more useful to expand the scope of observations than to increase the spatial interval. From our tests, the spatial interval of observation can be 5 times bigger than that of model grid.     

福建省森林生态系统NPP的遥感模拟与分析

利用MODIS遥感影像,结合气象资料等数据,采用BEPS过程模型对2004年福建省的森林生态系统植被净初级生产力(NPP)进行了模拟验证。研究结果表明,2004年福建省森林生态系统NPP平均值为578.97gC/m²·a,NPP总量累计达到46.18×106tC;不同林地NPP全年平均值大小依次为:竹林≈阔叶林>杉木>马尾松,其值分别为:788.6gC/m²·a,780.0gC/m²·a,519.8gC/m²·a,437.3gC/m²·a;时空分析结果表明,2004年6-8月NPP形成较为明显的"坑"形分布形态,主要的原因之一很可能是有效降水量偏少;在空间分布上,福建省森林生态系统NPP与海拔高程显著相关,体现了该地区森林生态系统NPP空间分布的地域特征,这在一定程度上表明随着海拔上升,山高坡陡,人类对森林生态系统的干扰活动减少,有助于森林生态系统生产力的提高和维持。最后,分析了应用BEPS过程模型模拟福建省森林生态系统净初级生产力的不确定性问题。     

交通路况基态修正模型及其效率分析

 交通路况在时间上和空间上具有连续变化的特征,在时空维度上对交通路况进行高分辨率采样得到的数据,对研究交通路况的时空动态十分有利。但长时间大范围的高分辨率交通路况信息数据量巨大,给数据的组织和管理带来了困难。目前,尚没有一种成熟的时空数据模型对高时空分辨率交通路况数据进行高效(顾及数据存储与访问效率)的组织管理。本文提出一种基于线性参照系统的交通路况基态修正模型。此模型应用基态修正模型的基本思想,在时间维度上对交通路况数据进行无损压缩,又引入动态分段技术和线性参照系统,以路划作为交通路况载体,在空间维度上对交通路况数据进行压缩存储。利用成都市区真实交通路况数据,本文验证了此模型的有效性,比较了6种不同参数下交通路况基态修正模型的存储和访问效率,给出了最佳模型建议。