3种NDVI时序数据重建方法对比研究

Comparative Study on Three NDVI Time Series Data Reconstruction Methods

受云、冰雪、气溶胶等因素的干扰,遥感时间序列数据集包含大量噪声。基于TIMESAT软件中非对称的高斯函数拟合法(AG)、双Logistic曲线拟合法(D-L)、Savitzky-Golay滤波(S-G)3种拟合方法重建了弗吉尼亚州韦兹县、陕西省神府地区、山东省兖州市不同类型的NDVI时序数据集,从空间格局和植被覆盖两个维度对比分析了3种方法的拟合效果。结果表明:①气候、海拔、地形和主要植被类型等因素将对数据重建结果产生影响,3种方法均有一定去噪能力,S-G方法的重建更关注“细节”,但保留了大量噪声,AG方法和D-L方法更加注重整体效果,重建后的时序曲线较平滑,符合植被生长变化的连续性规律,但会损失时序曲线的真实局部变化特征;②NDVI时序数据重建结果受植被覆盖类型影响较大,3种方法的耕地重建效果均较好且相差较小,AG方法和D-L方法对林地的重建质量高于S-G方法。

The remote sensing time series data set contains a lot of noise due to the disturbance of cloud,snow and ice,aerosol and other factors.In this paper,we used the TIMESAT software asymmetrical Gaussian function fitting method(AG),the double Logistic curve fitting method(D-L),the Savitzky-Golay filtering(S-G)three fitting methods to rebuild the different types of NDVI time series data sets inWeitz County,Virginia,Shenfu Region,Shaanxi Province,Yanzhou City,Shandong Province,and compared the fitting effects of the three methods from two dimensions of spatial pattern and vegetation coverage.The results show that①climate,altitude,topography and main vegetation types have influence on the results of data reconstruction.All the three methods have certain denoising ability.S-G method pays more attention to details,but it retains a lot of noise.AG and D-L methods pay more attention to the overall effect,and the time series curve after reconstruction is smooth,which is in line with the continuous law of vegetation growth and change,but it will lose the real local change characteristics of the time series curve.②The results of NDVI time series data reconstruction are greatly affected by the vegetation coverage type.The effects of the three methods are all good and the differences are small.The forest reconstruction quality by AG and D-L method are higher than that by S-G method.