高时空分辨率的植被指数VI（Vegetation Index）数据是农业和生态研究的重要基础数据集，目前常用的VI数据的时空分辨率存在不可调和矛盾。考虑VI时序变化对数据融合的影响，提出一种新的VI数据时空融合模型VISTFM（Vegetation Index Spatial and Temporal Fusion Model），VISTFM采用模糊C聚类算法，对存量时序VI数据按土地利用类型划分为若干子类，从高低分辨率影像中随土地覆被类的变化规律提取子类，结合低分辨率影像提取的土地覆被类变化规律融合生成高时空分辨率的VI数据。用常用的Landsat和MODIS数据验证该算法，测试表明，VISTFM能够较好的捕获VI的中间变化过程，与常用的基于线性混合模型的模型和时空自适应反射率融合模型及其改进模型相比，利用VISTFM获得的植被指数数据集具有更高的时空分辨率。

Vegetation Index (VI) data with high spatial and temporal resolution are highly important in the use of remote sensing technology to observe the earth. Owing to technological limitations, obtaining VI data that exhibits both high spatial and temporal resolution is impossible. The arrival of multi-source remote sensing data spatial and temporal fusion model enables the retrieval of data with high spatial and temporal resolution. The commonly used model does not have the ability to capture the intermediate change process of pixel value, and a certain regularity occurs in the change of the VI of the landmark. This study proposes a new multi-source data fusion model (FCMVISTFM) based on Fuzzy C-Mean algorithm (FCM).
Making pixels that group together have similar VI values and law of VI changes throughout the period. FCMVISTFM uses FCM to divide land-cover types into certain categories based on multi-phase VI data, which are defined as subclasses of each land-cover class. Each land-cover class average value is calculated by using the linear unmixed model, and subclass average value is calculated by the law between land-cover class and subclass. The VI data are fused by Landsat8 OLI and MODIS data based on the assumption that the average VI value of subclass S is the same as the VI value of pixels that belong to subclass S.
Results show that FCMVISTFM can achieve relatively high accuracy. The average values of correlation coefficient (R), RMSE, ERGAS, and variance are 0.9057, 0.0674, 1.9795, and 0.0045, respectively. With this level of accuracy, VI data can be used for vegetation research and observations of the earth. Commonly used line unmixed models, spatial and temporal adaptive reflectance fusion model (STARFM), and its improved models have the problem of uncertain ability to capture the intermediate change process of VI. Thus, FCMVISTFM is more accurate compared with STDFA and ESTARFM.
FCMVISTFM is developed for obtaining high spatial and temporal resolution VI data, making it easier to capture the intermediate changes of VI, which can be applied where high spatial and temporal resolution VI data are needed. In this study, the accuracy of the multi-source data fusion model can be increased by the following aspects. (1) The models based on line unmixed model, regardless of pixel classes or subset S average value calculations, are based on the entire image. However, in the STARFM model and improved models based on STARFM, the data fusion based on high and low resolutions pixels in a certain window, cloud cover only affects the calculation of pixels near its coverage area. The acquisition of multiphase cloudless images is especially difficult when the study area is large. In this case, the STARFM model and improved models based on STARFM have more application advantages. (2) All of the multi-source remote sensing data fusion models are based on certain assumptions, though these assumptions are based on a certain theoretical and have certain rationality. Errors are mainly caused by assumptions. A complete model assumption is proposed as the main way to improve the accuracy of the multi-source remote sensing data fusion model. (3) The time sequence laws of the VI of various landmark are not disordered, but a certain regularity, such as the specific laws of the crop's VI, occurs. If the multi-source remote sensing data fusion model is established based on these laws, then it can also improve the accuracy of fusion results to some extent.