叶面积指数 LAI (Leaf Area Index) 是一项重要的生物物理参数,它表征了单位地表面积上叶片的总量。植被的生长变化对气候、水文、生态等都会产生影响,因此 LAI 是众多陆面过程模型的输入参数。提高叶面积指数的估算精度,具有重要意义。利用遥感观测数据估算 LAI 是目前获取 LAI 的主要手段。覆盖区域甚至全球的 LAI数据常以数据产品形式呈现,MODIS LAI 产品覆盖全球,自 2000 年开始持续发布,是其中的代表。遥感估算 LAI 的模型有很多种,表现为从遥感观测计算 LAI 所采用的不同的算法。现有算法有一个重要的缺陷,就是对于历史数据中隐含的信息利用得不够。本研究提出基于数据机理的时间序列叶面积指数估算方法。MODIS LAI 产品已经积累多年,本研究假设 LAI 与反射率的函数关系隐含在这些历史数据中。通过使用基于数据机理的 (Data Based Mechanistic) 建模方法,从历史数据中提取 LAI 与反射率的关系模型,以 LAI_DBM 表示。LAI_DBM 模型立足于 MODIS 像元尺度 (1km),模型结构和系数完全从历史数据中提取,因此从建模机理上说,适用于各种地表覆盖类型的像元,也包括混合像元。本研究用 6 种不同地表覆盖的站点数据,进行了 LAI_DBM 建模和估算实验,建模数据和估算数据分离。估算结果表明,LAI_DBM 模型具有较强的的估算能力,对不同类型的像元都能适用。LAI_DBM 模型的估算结果在时间连续性上明显优于 MODIS LAI数据。LAI_DBM 模型模型的优点是模型结构简单、基于历史数据、不依赖先验知识,缺点是对历史数据的数量和质量存在依赖。针对此问题,本研究提出了基于数据机理的同化方法。对于一种或多种地表类型,使用数十个像元的历史数据构建通用的 LAI_DBM 模型,以 LAI_UDBM 表示。LAI_UDBM 模型可以直接用到同类的其它像元,由此减少对历史数据的依赖。但 LAI_UDBM 模型的估算结果不是作为最终结果,而是作为状态变量导入 PROSAIL 模型。PROSAIL 模型的输出变量再与 MODIS 反射率进行同化,通过集合卡尔曼滤波 EnKF (Ensemble Kalman Filter) 方法更新状态变量,得到最终的 LAI 估算结果。LAI_UDBM 模型与 PROSAIL 模型的耦合 (以 LAI_EnKF 表示),减少了 LAI_DBM 模型对历史数据的依赖,同时增强了模型的适用性。为了进一步验证 LAI_DBM 和 LAI_EnKF 模型,本研究最后使用 Bigfoot 的地面 LAI观测数据,对 MODIS LAI(以 LAIM ODIS 表示)、LAI_DBM 模型的估算结果 (以 LAIDBM表示) 和 LAI_EnKF 模型的估算结果 (以 LAIEnKF 表示) 进行了比较验证。在时间连续性上,LAIDBM 和 LAIEnKF 都远优于 LAIM ODIS 。与实测值的比较,LAIDBM 和 LAIEnKF略优于 LAIM ODIS 。需要注意的是,LAIDBM 是使用了相应站点的历史数据建模估算的结果,而 LAIEnKF 使用的是 LAI_UDBM 模型,因此完全没有使用相应站点的历史数据。相比于 LAIDBM ,LAIEnKF 对生长季特别是生长高峰的 LAI 变化特征表现得更为明显。

Leaf area index (LAI) is an important biophysical parameter, it is defined as total green leaf area per unit ground surface area. The dynamic change of vegetation has impact on climate, hydrology and ecosystem, so LAI is used as input variable to many land surface process models. It has great significance to improve the accuracy of LAI data.Currently LAI is mainly retrieved from remote sensing observation. LAI data are often provided as data products, which have local or global coverage. MODIS LAI product is one of these products that have global coverage, it has been continually published since 2000. There are a number of models to retrieve LAI from remote sensing observation, due to the specific algorithms that are applied. Most algorithms suffer from the same problem, which is the information in historicaldata is not fully utilized.This study propose a data based mechanistic approach to time-series LAI estimation. MODIS LAI product have been accumulated for over ten years, this study assume the relationship betweenLAI and reflectance is implicitly contained in historical data. A data based mechanistic (DBM) modeling approach is applied to extract this relationship. The extracted relationship model is denoted as LAI_DBM. LAI_DBM is based on the MODIS pixel size (1km), the model structure andmodel parameters are completely extracted from histroical data. This modeling philosphy means LAI_DBM can be applied to pixels of different land cover types, including mixed pixels. Six pixels of different land cover types are chosen to implenment LAI_DBM modeling and estimation. Data used for modeling are seperated from data used for estimation. The results show that LAI_DBM model can interpret the relationship between LAI and reflectance faily well. LAI_DBM model hasstrong applicability. The results of LAI_DBM estimation are better than the MODIS LAI in terms of time continuity.LAI_DBM model has the advantage of simplicity, data based, no prior knowledged needed. However, LAI_DBM model relys heavily on both the quantity and the quality of historical data.To overcome this problem, a data based assimilation method is proposed. For one or many kinds of land cover types, historical data from tens of pixels are used to build a universal LAI_DBM model, denoted as LAI_UDBM. LAI_UDBM model can be applied to other pixels of the same land cover type, thus have no dependency on historical data. The estimation result of LAI_UDBM is not the final result, it is inputted into PROSAIL model. The output of PROSAIL model is thenassimilated with MODIS reflectance. Ensemble kalman filter is applied to update state variables, yielding the final LAI estimation results. The coupling of LAI_UDBM and PROSAIL is denoted as LAI_EnKF. LAI_EnKF model improves the applicability of LAI_DBM model while removing the dependency of historical data.To further validate LAI_DBM and LAI_EnKF, LAI maps generated by Bigfoot project from field measurements are used to make comparison between MODIS LAI (denoted as LAIM ODIS ),estimation results of LAI_DBM model (denoted as LAIDBM ) and estimation results of LAI_EnKF model (denoted as LAIEnKF ). In terms of time continuity, LAIDBM and LAIEnKF are much betterthan LAIM ODIS . In terms of closeness to LAI field map, LAIDBM and LAIEnKF are slightly better than LAIM ODIS . It should be noted that, LAIDBM are estimated by models extracted from historical data of corresponding pixels. While LAIEnKF utilize the universal LAI_DBM model. LAIEnKF interpret the LAI changes during growth season faily better than LAIDBM .