STARFM 算法生成湿地类型 TM 反射率数据的应用评价

摘要

Multispectral remote sensing data with both high temporal and high spatial resolutions can hardly be obtained directly under current data acquisition conditions.We proposed to us STARFM (spatial and temporal adaptive reflectance fusion model)to synthesise the high time series data with high spatial resolution.Nevertheless,the applicability of this model for the geographic areas in China and its prediction precision validation have not yet been substantially done.Accordingly,the Hulun Wetland Nature Reserve in Inner Mongolia is selected as the study area,and with the help of Landsat-5 TM data and high time series MODIS reflectance products,we used STARFM algorithm to generate the TMdata with high time series feature,and then made the comparative validation analysis on it with the actual TM reflectance data.Result showed that the STARFM algorithm has a higher prediction capability on the phase features of different terrain categories in wetland area varying along with time phases under the condition of preserving certain prediction accuracy in space,it is particularly suitable for the spatiotemporal fitting or data prediction research in those wetland areas where the reflection characteristics change little along with the time phases.%当前数据获取条件下，很难直接获得兼具高时间与高空间分辨率的多光谱遥感数据，提出利用 STARFM（Spatial and Temporal Adaptive Reflection Fusion Model）算法来合成高时间序列的高空间分辨率数据。该算法在我国地理区域的适用性与预测精度验证等工作尚未充分展开。为此，以内蒙古呼伦湖湿地自然保护区为研究样区，并借助于 Landsat-5 TM（Thematic Mapper）与高时序 MODIS 反射率产品，利用 STARFM算法生成具有高时序特征的 TM数据，进而将其与真实 TM数据进行对比验证分析。结果表明：STARFM算法能够在空间上保持一定预测精度的条件下，对湿地区域内不同地物类别随时相的变化特征具有较好的预测能力，尤其适用于对反射特征随时相变化较小的湿地区域进行时空拟合或数据预测研究。