抗植被物候影响的中等分辨率密集时序遥感数据构建

Time series remote sensing data are important for forest cover change detection. However, regular time series remote sensing data are deficient in spatial resolution; remote sensing data with higher resolution cannot form regular time series, and change detection with these data is heavily affected by the vegetation phenology. To provide data for near realtime forest cover change detection at landscape level, we proposed a procedure for generating time series moderate resolution data suppressing vegetation effect, which involves four steps. First, time series filtering will be adapted to remove noise, and fill data gaps, thus the data series smoothness will be improved. Second, time series data will be decomposed into trend, seasonal, and remainder components, this will help to reduce the effect of vegetation phonology by removing the seasonal components. Third, automatic radiometric normalization for time series remote sensing imagery by iterative slow feature analysis will be adapted to reduce the radiometric variance irrelevant to forest cover change. Finally, moderate spatial resolution remote sensing data and low spatial resolution remote sensing data with will be fused by the sparse representation-based spatiotemporal reflectance Fusion Model(SPSTFM), and dense, regular, vegetation phenology effect removed time series moderate resolution remote sensing data will be reconstructed. This method will enhance temporal, spatial, and spectral consistency among time series remote sensing data, providing high quality input data for automatic forest cover change detection.

时序遥感数据对森林覆盖变化监测很重要。但是，规则时间序列的遥感数据空间分辨率不足，空间分辨率满足要求的遥感数据不能形成规则的时间序列，且植被物候对覆盖变化监测的精度影响显著。为提供景观尺度森林覆盖变化近实时监测所需数据，我们提出抗植被物候影响的中等分辨率密集时序遥感数据构建方法。该方法包括四个步骤：首先利用时序滤波排除噪声、填充数据，提高时序平滑性；第二是通过光谱分解，将时序分解为趋势、季节、残差等组分，通过季节组分的降低植被物候影响；第三利用迭代式慢特征分析方法，自动选择伪不变象元，实现光谱的自动归一化，减少与覆盖无关的光谱变化；最后采用基于稀疏表达的时空反射比融合模型对中等和低空间分辨率遥感数据进行融合，构建密集的、规则的、消除植被物候效应的中等分辨率时序遥感数据。该方法将增强时序遥感数据时间、空间、光谱一致性，为森林覆盖变化自动监测提供高质量的输入数据。

植被物候对覆盖变化监测的精度影响显著。为提供景观尺度森林覆盖变化近实时监测所需数据，我们提出抗植被物候影响的中等分辨率密集时序遥感数据构建方法。利用2013-2016年所有可用的Landsat8 OLI影像，并以Landsat7 ETM影像作为补充数据，基于时序地表反射率预测模型，开展空间分辨率较高的密集时序遥感数据构建研究。对不同地表反射率预测模型拟合中等分辨率影像的效果对比和精度评价。结果表明，只基于Landsat8时序数据，自适应模型与基本模型的拟合结果相似，只能拟合基本的季相变化；将Landsat7作为补充数据，并对Landsat7数据进行辐射归一化，可以提高拟合精度。采用自适应模型可以更好的拟合细节以及年内的多峰变化；植被区域拟合精度最高，大部分波段RMSE的值都处于0.01左右；但水体和建设用地，采用自适应模型误差增大。以16天为间隔构建了2013-2016年30米分辨率的密集时序遥感数据。从预测影像与真实影像对比、数据集时间变化合理性分析、典型地物的时空趋势、对变化监测的作用四个方面对构建结果进行评价。结果表明，拟合影像与真实影像相差较小，拟合精度较高；构建的数据集能反映福建省高植被覆盖区域的季节和年度变化特点；数据集夏秋冬三个季节的变化趋势与实测光谱基本一致；数据集能够实现快速变化监测，提高变化监测的频率，构建结果较好。