植被覆盖度（Fractional Vegetation Cover，FVC）是刻画地表植被覆盖的重要参数，被广泛应用于生态环境监测、碳循环模拟和气候变化研究等方面。青藏高原是全球典型的生态脆弱区，准确估算青藏高原的FVC对于中国乃至亚洲的生态安全监测与评估具有重要意义。遥感由于其大范围的数据获取和连续观测能力，已成为区域尺度FVC估算的有效途径。目前，青藏高原的FVC遥感估算面临着两个主要问题：（1）多云天气条件导致遥感数据质量下降，从而直接影响FVC估算精度；（2）复杂地形和植被垂直地带性导致遥感像元较强的空间异质性，而现有的大多数FVC遥感产品空间分辨率较低，尤其未充分考虑地形的影响，难以准确、精细地反映青藏高原的植被覆盖状况。因此，有必要针对青藏高原气候、地形和植被等方面的特点，开展精细尺度的FVC估算方法研究，以提高FVC遥感估算结果的空间分辨率和准确性。

本研究旨在开展青藏高原FVC遥感精细估算方法研究，主要包括两方面的方法探究：（1）发展一种遥感植被指数（Vegetation Index，VI）时序数据重建方法，以降低多云天气条件对遥感数据质量的影响，为青藏高原FVC估算提供高质量的输入数据；（2）基于较高空间分辨率的Sentinel-2数据，发展一种适用于青藏高原的FVC遥感精细估算方法，以满足精细尺度和准确估算的需求。具体来说，首先基于MCD12Q1土地覆盖类型数据和MOD09GA V6地表反射率数据构建和评估本研究发展的VI时序数据重建方法，并利用该方法和现有的地表反射率时序数据重建方法分别对青藏高原2019 ~ 2021年的Sentinel-2 VI和各波段反射率时序数据进行重建。然后，利用随机森林回归模型从重建的Sentinel-2波段反射率和VI数据、基于数字高程模型数据获取的地形特征参数中，筛选出适用于青藏高原FVC估算的最佳特征组合，再结合随机森林回归模型发展一种适用于青藏高原的FVC遥感精细估算方法，并利用青藏高原3个典型测试区（祁连山、黄河源和横断山区）内基于无人机RGB影像获取的FVC数据和现有FVC遥感估算产品对该方法进行评估。本文主要研究内容与研究结论如下。

（1）发展了一种基于曲线特征加权的VI时序数据函数拟合重建方法（the reconstruction method based on Self-Weighting function fitting from Curve Features，SWCF）。该方法基于VI时序数据遵循植被生长和衰老的渐变模式以及多云天气条件通常会降低VI值的假设，根据VI时序数据的曲线特征为各数据点确定拟合权重，然后对其进行函数加权拟合重建。SWCF具有不依赖于辅助数据、重建质量高和自适应能力强等优势。在北半球中高纬度典型植被样点的VI重建结果对比中发现，与不加权函数拟合法以及Savitzky-Golay滤波法相比，基于SWCF的重建结果的均方根误差（Root Mean Square Error，RMSE）分别显著降低了26.82 ~ 52.47％（p < 0.05）和13.98 ~ 54.04% (p < 0.05)。在区域应用中，SWCF也展现出良好的适用性和鲁棒性。将SWCF应用于以VI时序数据上包络线为约束条件的Sentinel-2地表反射率时序数据重建方法中，克服了该方法不适用于不等时间间隔的地表反射率时序数据重建的局限性。基于新发展的SWCF方法，重建了青藏高原内64507个像元样点在2019 ~ 2021年每5天的Sentinel-2 VI和反射率时序数据，以用于青藏高原植被FVC遥感精细估算。

（2）发展了一种综合植被光谱特征和地形特征的青藏高原FVC遥感精细估算方法（Vegetation Spectral and Terrain Features-integrated Fractional Vegetation Cover estimation method，VSTF-FVC）。新方法的主要特色在于结合随机森林回归模型筛选出了适用于青藏高原FVC估算的最优光谱和地形特征组合，即：基于Sentinel-2红波段和近红外波段构建的归一化差异植被指数（NDVI）、利用Sentinel-2红波段分别与红边2、3和4波段构建的三种红边归一化差异植被指数（RENDVI）、高程和坡度。在青藏高原的3个测试区（祁连山地区、黄河源和横断山区）内，VSTF-FVC估算结果与基于无人机影像计算的FVC的决定系数R²为0.90，RMSE为0.11。相较于已有的MultiVI FVC数据集和GEO V3数据集，VSTF-FVC数据产品的RMSE分别降低了54.17%和16.67%。新发展的VSTF-FVC方法为青藏高原地区的植被FVC精细估算提供了方法支撑。

（3）生产了一套青藏高原典型区的高精度FVC产品。针对祁连山地区、黄河源地区和横断山区，采用本文新发展的VSTF-FVC方法生产了一期2020年8月中旬的FVC产品。相较于已有的数据产品（30米空间分辨率的MultiVI FVC数据集、300米空间分辨率的GEO V3数据集、500米空间分辨率的GLASS数据集），新FVC产品的空间分辨率提升至20米，估算精度的R²提升至0.90、RMSE降低至0.11。

Fractional Vegetation Cover (FVC) is an important parameter for characterizing the vegetation coverage, and is widely used in the monitoring of ecological environment, the modeling of carbon cycle and the research of climate change. The Qinghai-Tibet Plateau is a typical ecologically fragile area in the world, and the accurate estimation of FVC for the Qinghai-Tibet Plateau is vital for the ecological security monitoring and assessment in China and even Asia. Remote sensing has become an effective way to estimate FVC at regional scale due to its large range of data acquisition and continuous observation capability. At present, the remote sensing estimation of FVC in the Qinghai-Tibet Plateau faces two main challenges: (1) the degradation of remote sensing data quality resulted from the cloudy weather conditions, which directly affects the FVC estimation accuracy; (2) the strong spatial heterogeneity for the remote sensing pixel because of the complex topography and vegetation vertical zonality, while most existing FVC remote sensing-based products have low spatial resolution, and thus it is difficult for them to accurately and finely reflect the vegetation cover status of the Qinghai-Tibet Plateau. Therefore, it is necessary to carry out research on fine-scale FVC estimation methods to improve the spatial resolution and the accuracy of FVC estimation results, especially for the characteristics of the topography and vegetation in the Qinghai-Tibet Plateau.

This study aims to develop a fine-scale remote sensing estimation method for FVC of the Qinghai-Tibet Plateau, focusing on two aspects: (1) developing a reconstruction method for Vegetation Index (VI) time series to to alleviate the interference of the cloudy and rainy weather condition on the quality of remote sensing data and provide the high-quality input data for FVC estimation; (2) developing a fine-scale remote sensing estimation method for FVC based on Sentinel-2 data with high spatial resolution to meet the needs of fine-scale and accurate estimation in the Qinghai-Tibet Plateau. Specifically, the newly-developed reconstruction method for VI time series was firstly constructed and evaluated based on MCD12Q1 land cover type data and MOD09GA V6 daily surface reflectance time series. Then, the new method was employed to reconstruct the Sentinel-2 VI time series for the Qinghai-Tibet Plateau from 2019 to 2021. Additionally, it was combined with the existing reconstruction method for surface reflectance time series to reconstruct the time series for each reflectance band of Sentinel-2. Then, the random forest regression model was used to select the best combination of features applicable to FVC estimation on the Qinghai-Tibet Plateau from the reconstructed Sentinel-2 band reflectance and VI data, the topographic feature parameters obtained from the digital elevation model data. Subsequently, the fine-scale remote sensing estimation method was proposed by combining with the random forest model. The new method is evaluated by using the FVC data obtained from unmanned aerial vehicle RGB images and the existing FVC products in three typical test areas (Qilian Mountains, Yellow River Source and Hengduan Mountains) of the Qinghai-Tibet Plateau. The main research contents and findings of this paper are as follows.

(1) This paper proposed a reconstruction method for VI time series based on self-weighting function fitting from curve features (SWCF). SWCF is based on the assumptions that a yearly VI time series primarily indicates the seasonal dynamics of vegetation and that clouds and poor atmospheric conditions usually tend to depress VI values, causing sudden drops in VI time series. SWCF utilizes the curve features of VI time series to determine a fitting weight for each VI data point and then implements the weighted function fitting to reconstruct the VI time series. SWCF has the advantages of no dependence on ancillary data, high reconstruction quality, and high self-adaptive capability. Compared to the unweighted function fitting method and the Savitzky-Golay filtering, the Root Mean Square Error (RMSE) of the reconstruction results based on SWCF was significantly reduced by 26.82 - 52.47% (p < 0.05) and 13.98 - 54.04% (p < 0.05), respectively. In regional applications, SWCF also showed excellent applicability and robustness. By applying SWCF to the method incorporating upper envelopes of time series VIs as constraint conditions to Reconstruct time series of Sentinel-2 surface Reflectance, the limitation of inapplicability to surface reflectance time series with uneven time intervals has been overcome. Based on the newly developed SWCF method, the Sentinel-2 VI and reflectance time series of 64507 samples in three typical zones of the Tibetan Plateau every 5 days from 2019 ~ 2021 were reconstructed for the fine-scale estimation of FVC in the Qinghai-Tibet Plateau.

(2) This paper developed a Vegetation Spectral and Terrain Features-integrated Fractional Vegetation Cover estimation method (VSTF-FVC) for the Qinghai-Tibet Plateau. The main feature of the new method is to screen the optimal combination of features suitable for the FVC estimation in the Tibetan plateau using the random forest model. The optimal combination of features included the Normalized Difference Vegetation Index (NDVI) calculated from the Sentinel-2 red and near-infrared bands, three Red Edge Normalized Difference Vegetation Indices (RENDVI) constructed by the Sentinel-2 red band and red edge 2, 3, and 4 bands, respectively, elevation and slope. In the three test areas of the Qinghai-Tibet Plateau (e.g., the Qilian Mountains, the Yellow River source region and the Hengduan Mountains), compared to the FVC calculated from the unmanned aerial vehicle images, the coefficient of determination (R²) and RMSE of VSTF-FVC estimation results was 0.90 and 0.11, respectively. Compared to the existing MultiVI FVC dataset and GEO V3 dataset, the RMSE of the VSTF-FVC data was reduced by 54.17% and 16.67%, respectively. The newly developed VSTF-FVC method provides a methodological support for fine-scale estimation of FVC in the Qinghai-Tibet Plateau.

(3) This paper produced a high-precision FVC product for typical areas of the Qinghai-Tibet Plateau. The newly-developed VSTF-FVC method was used to produce a mid-August 2020 FVC product for the Qilian Mountains, the Yellow River Source and the Hengduan Mountains areas. Compared to the existing FVC products (e.g., 30m MultiVI FVC dataset, 300m GEO V3 dataset, and 500m GLASS dataset), the spatial resolution of the newly-produced FVC product was enhanced to 20m, and the R² and RMSE of the estimation accuracy was enhanced to 0.90 and 0.11, respectively.