多层土壤观测数据同化的森林碳、水通量模拟

为充分考虑森林生态系统土壤水分的垂直运动及改善碳、水通量的模拟精度,利用Biome-BGC MuSo模型模拟了长白山森林通量站点的碳、水通量,该模型包含了多层土壤模块、物候模块以及管理模块;其次,利用集合卡尔曼滤波算法将站点观测的多层土壤参数同化到Biome-BGC MuSo模型中,并用站点涡动通量数据进行了验证。结果表明：与Biome-BGC模型模拟结果相比,Biome-BGC MuSo改善了站点净生态系统交换量（Net ecosystem exchange, NEE）、生态系统呼吸量（Ecosystem respiration, ER）和蒸散发（Evapotranspiration, ET）模拟精度,站点观测的时序土壤温度和水分数据同化到Biome-BGC MuSo后,碳、水通量模拟结果有了进一步的提升（NEE: R² = 0.70, RMSE = 1.16 gC·m^–2·d^–1; ER: R² = 0.85, RMSE = 1.97 gC·m^–2·d^–1 ; ET: R² = 0.81, RMSE = 0.70 mm·d^–1）。数据-模型同化策略为森林生态系统碳、水同量的模拟提供了科学的方法。     

多层土壤观测数据同化的森林碳、水通量模拟

为充分考虑森林生态系统土壤水分的垂直运动及改善碳、水通量的模拟精度,利用BiomeBGC MuSo模型模拟了长白山森林通量站点的碳、水通量,该模型包含了多层土壤模块、物候模块以及管理模块;其次,利用集合卡尔曼滤波算法将站点观测的多层土壤参数同化到Biome-BGC MuSo模型中,并用站点涡动通量数据进行了验证。结果表明:与Biome-BGC模型模拟结果相比,Biome-BGC MuSo改善了站点净生态系统交换量(Net ecosystem exchange,NEE)、生态系统呼吸量(Ecosystem respiration,ER)和蒸散发(Evapotranspiration,ET)模拟精度,站点观测的时序土壤温度和水分数据同化到Biome-BGC MuSo后,碳、水通量模拟结果有了进一步的提升(NEE:R2=0.70,RMSE=1.16 gC·m~(–2)·d~(–1);ER:R2=0.85,RMSE=1.97 gC·m~(–2)·d~(–1);ET:R2=0.81,RMSE=0.70 mm·d~(–1))。数据-模型同化策略为森林生态系统碳、水同量的模拟提供了科学的方法。     

融合支持向量机算法的数据驱动型数据同化方法研究

数据驱动建模是从数据中探究状态变量的时空演化关系。数据驱动型数据同化方法是探索使用数据驱动模型替代传统（基于物理的）模型,实现优化融合观测信息与模型模拟的同化方法。研究将数据驱动的支持向量机回归预测模型应用于集合卡尔曼滤波过程中,使用模拟预测方法对动力学系统进行非参数采样得到系统轨迹的代表性样本集,从样本集中重构动力学系统。提出一种支持向量机回归机器学习模拟预测策略的数据驱动数据同化方法,并将其应用于经典模式驱动同化系统。采用Lorenz-63和Lorenz-96非线性模型进行数值实验。通过改变样本集大小、噪声方差和观测步长等敏感性参数比较数据同化性能。结果表明：对于较大的样本集,该组合方法优于一般的顺序数据同化方法,从而证明新方法的有效性。     

一种耦合遗传算法与粒子滤波的积雪数据同化方法

为有效解决粒子滤波中粒子退化和贫化问题,实验将遗传算法耦合到粒子滤波中用于对粒子进行重采样,发展了基于遗传粒子滤波的积雪数据同化方案。以Noah-MP模型默认组合方案为模型算子搭建了积雪数据同化系统,分别在真实和合成同化试验情景下比较了集合卡尔曼滤波和粒子滤波的同化性能、不同重采样方法对粒子滤波同化性能的影响,探讨了遗传粒子滤波作为积雪数据同化方法的可行性。在理想试验情景中,遗传粒子滤波的整体同化性能次于系统重采样粒子滤波,但明显好于多项式重采样粒子滤波,并且遗传粒子滤波能够在积雪融化较快的阶段,更有效地利用观测信息对模型进行校正,在积雪消融期表现出了较好的同化性能。在真实试验情景中,遗传粒子滤波的同化性能明显好于采用其他重采样算法粒子滤波的同化性能。此外,两种同化情景的试验结果均表明粒子滤波的同化性能明显好于集合卡尔曼滤波。将遗传粒子滤波作为积雪数据同化方案是可行、有效的。     

观测误差协方差估计下的集合鲁棒滤波数据同化方法

在数据同化方法中,观测误差协方差矩阵是相关的,且与时间和状态有一定的依赖性.针对这种相关特性,将鲁棒滤波方法与观测误差协方差估计方法相结合,得到随状态时间变化的观测误差协方差,提出一种带有观测误差估计的鲁棒数据同化新方法,更新观测误差协方差,改善估计效果.从分析误差协方差,转移矩阵特征值放大等角度优化同化方法.利用非线...     

基于Lorenz-96模型的顺序数据同化方法比较研究

顺序数据同化方法在数据同化系统中得到了广泛的应用,其性能各有优缺。选择3种典型的顺序数据同化算法,即集合Kalman滤波,集合转换Kalman滤波和确定性Kalman滤波,使用经典的Lorenz-96模型进行敏感性实验,研究不同的关键参数变化,如集合数目变化、观测数变化、误差放大因子变化和定位半径变化时对同化效果的影响。实验表明：集合数目和观测数目的多少直接影响3种方法的同化效果;协方差放大因子和定位半径的选择会提高同化精度。综合比较,确定性集合Kalman滤波算法是一种具有较强鲁棒性的滤波算法,能够在集合数较小的情况下达到较好的同化效果。     

基于多源遥感数据协同反演森林地表土壤水分研究

土壤水分在土壤监测中是一项重要的指标,对于农业生产、生态环境以及水资源管理有着重要的影响.随着遥感建模与反演理论的不断成熟,其逐渐成为分析土壤指标的重要技术与手段.因此,利用光学影像与雷达影像数据,以大兴安岭地区漠河市为研究区域,分别建立以Landsat 8为数据源的土壤水分反演模型和由Landsat 8影像数据与GF...     

面向大规模海洋数据同化算法的并行实现及优化

海洋数据同化是一种将海洋观测资料融合到海洋数值模式中的有效手段,经过同化的海洋数据更加接近海洋的真实情况,对人类理解和认识海洋具有重要意义。围绕海洋数据同化设计了一种基于区域分解的一般性并行实现方法。在此基础上,提出了一种基于IO代理的新并行算法。首先,IO代理进程负责数据的并行读取;接下来,IO代理进程对数据进行切块,然后将块数据发送给相应的计算进程;当计算进程完成局部数据同化后,IO代理进程负责收集计算进程的同化结果,并将其写入磁盘。该方法的主要优势在于：利用IO代理进程来负责IO,而不是像传统方法那样让所有进程都来参与IO（直接并行IO）,这样可以防止大量进程对磁盘的同时访问,有效避免进程排队所导致的等待。在天河二号集群上的测试结果表明,对于1度分辨率的数据同化,在核心数为425时,该并行实现的总运行时间为9.1 s,相对于传统串行程序的加速比接近38倍。此外,对于0.1度分辨率的数据同化,基于IO代理的并行同化算法在使用10 000核时依然具有较好的可扩展性,并且可将其IO时间最大限制在直接并行IO时间的1/9。     

多工况适应的车辆质心侧偏角观测算法研究

研究汽车高速运行稳定性优化控制问题,在车辆稳定性控制中,质心侧偏角是衡量稳定性的重要指标,观测对于稳定性控制非常重要。针对目前车载多传感器信息的观测条件,为解决质心侧偏角观测的准确性、快速性和多工况适应性问题,提出了一种融合卡尔曼滤波和信号积分的质心侧偏角观测算法。观测算法充分考虑了车辆动力学特性,采用车辆运行过程的多种工况进行了算法设计及切换。最后在Matlab/Simulink平台上搭建了质心侧偏角观测仿真实验平台,通过多工况下的仿真,对所提出的质心侧偏角观测算法进行了仿真验证,结果表明能快速准确地矫正质心侧偏角,使稳态误差减小。     

SMAP L2多尺度土壤水分数据与ISMN实测数据的差异特征研究

SMAP卫星的二级（L2）土壤水分数据是直接反演结果,能够从模型、算法、参数等多方面体现其对土壤水分反演的综合能力。在这一级别下,SMAP设计了包括L2＿SM＿P(36km)、L2＿SM＿P＿E(9 km)和L2＿SM＿SP(3 km和1 km)在内的多种尺度的土壤水分数据,能满足不同的实验和应用需求。以ISMN地面实测土壤水分数据作为对比参照,以偏差（Bias）、均方根误差（RMSE）、无偏均方根误差（ubRMSE）和相关系数（R）作为分析指标,分析了SMAP L2土壤水分数据和ISMN实测数据间的差异表现。结果显示：在不同静态条件下（气候类型、土壤性质和植被类型）,植被对差异的影响最大,土壤性质的影响最小;在不同动态条件下（土壤水分、植被光学厚度和地表温度）,植被光学厚度和土壤水分对差异影响较大,地表温度的影响较小;在4种SMAP L2土壤水分数据中,9 km数据与ISMN实测数据的差异最小,其次是36、3、1 km尺度的数据;结合静态条件和动态条件来看,36 km和9 km尺度的数据与ISMN实测数据的差异情况类似,3 km和1 km数据差异情况类似。     

Experiments of one-dimensional soil moisture assimilation system based on ensemble Kalman filter

Ensemble Kalman filter is a new sequential data assimilation algorithm which was originally developed for atmospheric and oceanographic data assimilation. It can be applied to calculate error covariance matrix through Monte-Carlo simulation. This approach is able to resolve the nonlinearity and discontinuity existed within model operator and observation operator. When observation data are assimilated at each time step, error covariances are estimated from the phase-space distribution of an ensemble of model states. The error statistics is then used to calculate Kalman gain matrix and analysis increments. In this study, we develop a one-dimensional soil moisture data assimilation system based on ensemble Kalman filter, the Simple Biosphere Model (SiB2) and microwave radiation transfer model (AIEM, advanced integration equation model). We conduct numerical experiments to assimilate in situ soil surface moisture measurements and low-frequency passive microwave remote sensing data into a land surface model, respectively. The results indicate that data assimilation can significantly improve the soil surface moisture estimation. The improvement in root zone is related to the model bias errors at surface layer and root zone. The soil moisture does not vary significantly in deep layer. Additionally, the ensemble Kalman filter is predominant in dealing with the nonlinearity of model operator and observation operator. It is practical and effective for assimilating observations in situ and remotely sensed data into land surface models.     

Improving Soil Moisture Estimation by Joint Assimilationof MODIS Land Surface Temperature and Airborne L band Microwave Brightness Temperature

In this work,a novel soil moisture data assimilation scheme was developed,which was based land surface model (CoLM,Common Land Model),microwave radioactive transfer model (L MEB,L band Microwave Emission of the Biosphere),and data assimilation algorithm (EnKS,Ensemble Kalman Smoother).This scheme is used to improve the estimation of soil moisture profile by jointly assimilatingMODIS land surface temperature and airborne L band passive microwave brightness temperature.The ground based data observed at DAMAN superstation,which is located at Yingke oasis desert area in the middle stream of the Heihe River Basin,are used to conduct this experiment and validate assimilation results.Three LAI products are used to analyze the influence of LAI on soil temperature.Three assimilation experiments are also designed in this work,including assimilation of MODIS LST,assimilation of microwave brightness temperature,and assimilation of MODIS LST and microwave brightness temperature.The results show that the uncertainties in LAI influence significantly soil temperature simulations in different soil layers.MODIS LAI product is seriously underestimated in this study area,which results soil temperature overestimation about 4~6 K.Three assimilation schemes can improve soil moisture estimations to different extend.Joint assimilation of MODIS LST and microwave brightness temperature achieved the best performance,which can reduce the RMSE of soil moisture to 31%~53%.     

Retrieving soil temperature profile by assimilating MODIS LST products with ensemble Kalman filter

Proper estimation of initial state variables and model parameters are vital importance for determining the accuracy of numerical model prediction. In this work, we develop a one-dimensional land data assimilation scheme based on ensemble Kalman filter and Common Land Model version 3.0 (CoLM). This scheme is used to improve the estimation of soil temperature profile. The leaf area index (LAI) is also updated dynamically by MODIS LAI production and the MODIS land surface temperature (LST) products are assimilated into CoLM. The scheme was tested and validated by observations from four automatic weather stations (BTS, DRS, MGS, and DGS) in Mongolian Reference Site of CEOP during the period of October 1, 2002 to September 30, 2003. Results indicate that data assimilation improves the estimation of soil temperature profile about 1 K. In comparison with simulation, the assimilation results of soil heat fluxes also have much improvement about 13 W m^− 2 at BTS and DGS and 2 W m^− 2 at DRS and MGS, respectively. In addition, assimilation of MODIS land products into land surface model is a practical and effective way to improve the estimation of land surface variables and fluxes.     

Applying REC analysis to ensembles of particle filters

Particle filters (PF) are sequential Monte Carlo methods based in the representation of probability densities with mass points. Although currently most researches involving time series forecasting use the traditional methods, particle filters can be applied to any state-space model and generalize the traditional Kalman filter methods, providing better results. Furthermore, it is well-known that for classification and regression tasks ensembles achieve better performances than the algorithms that compose them. Therefore, it is expected that ensembles of time series predictors can provide even better results than particle filters. The regression error characteristic (REC) analysis is a powerful technique for visualization and comparison of regression models. The objective of this work is to advocate the use of REC curves in order to compare traditional Kalman filter methods with particle filters and analyze their use in ensembles, which can achieve a better performance. This work is an extended version of the paper presented at the 2007 International Joint Conference on Neural Networks (IJCNN) [1].     

Assimilation of remotely sensed data for improved latent and sensible heat flux prediction: A comparative synthetic study

Predicted latent and sensible heat fluxes from Land Surface Models (LSMs) are important lower boundary conditions for numerical weather prediction. While assimilation of remotely sensed surface soil moisture is a proven approach for improving root zone soil moisture, and presumably latent (LE) and sensible (H) heat flux predictions from LSMs, limitations in model physics and over-parameterisation mean that physically realistic soil moisture in LSMs will not necessarily achieve optimal heat flux predictions. Moreover, the potential for improved LE and H predictions from the assimilation of LE and H observations has received little attention by the scientific community, and is tested here with synthetic twin experiments. A one-dimensional single column LSM was used in 3-month long experiments, with observations of LE, H, surface soil moisture and skin temperature (from which LE and H are typically derived) sampled from truth model run outputs generated with realistic data inputs. Typical measurement errors were prescribed and observation data sets separately assimilated into a degraded model run using an Ensemble Kalman Filter (EnKF) algorithm, over temporal scales representative of available remotely sensed data. Root Mean Squared Error (RMSE) between assimilation and truth model outputs across the experiment period were examined to evaluate LE, H, and root zone soil moisture and temperature retrieval. Compared to surface soil moisture assimilation as will be available from SMOS (every 3 days), assimilation of LE and/or H using a best case MODIS scenario (twice daily) achieved overall better predictions for LE and comparable H predictions, while achieving poorer soil moisture predictions. Twice daily skin temperature assimilation achieved comparable heat flux predictions to LE and/or H assimilation. Fortnightly (Landsat) assimilations of LE, H and skin temperature performed worse than 3-day moisture assimilation. While the different spatial resolutions of these remote sensing data have been ignored, the potential for LE and H assimilation to improve model predicted LE and H is clearly demonstrated.     

Computing point estimates from a non-Gaussian posterior distribution using a probabilistic k-means clustering approach

The Kalman filter algorithm gives an analytical expression for the point estimates of the state estimates, which is the mean of their posterior distribution. Conventional Bayesian state estimators have been developed under the assumption that the mean of the posterior of the states is the ‘best estimate’. While this may hold true in cases where the posterior can be adequately approximated as a Gaussian distribution, in general it may not hold true when the posterior is non-Gaussian. The posterior distribution, however, contains far more information about the states, regardless of its Gaussian or non-Gaussian nature. In this study, the information contained in the posterior distribution is explored and extracted to come up with meaningful estimates of the states. The need for combining Bayesian state estimation with extracting information from the distribution is demonstrated in this work.     

增强粒子滤波船舶组合导航算法

针对粒子滤波作为非线性/非高斯估计方法存在的粒子退化和贫化的问题,提出了一种基于集合卡尔曼滤波(Ensemble Kalman filter,EnKF)和马尔可夫蒙特卡罗(Markov Chain Monte Carlo,MCMC)的增强粒子滤波算法。首先,使用EnKF分析代替先验密度对PF的建议密度进行定义,从而降低粒子退化的风险;其次,当发生粒子退化时,通过MCMC方法进行重采样,以增加粒子的多样性,从而降低了粒子贫化的可能性,提高滤波器的精度;最后,将提出的方法应用到GPS PPP/INS组合导航系统中,实验结果均表明,增强粒子滤波算法能提高估计精度,其性能优于标准粒子滤波。     

基于卡尔曼滤波器的数据融合方法在高温炉检测系统中的应用

将数据融合方法引入高温炉窑温度检测系统 ,充分利用现有检测系统的能力 ,在不增加任何设备的情况下提高检测精度 ,并且证明该方法具有理论简单、易于实现、精度高 ,适用于各种高温炉窑温度检测系统。