基于MODIS NDVI数据的辽宁省玉米种植面积提取研究

Research on Extraction of Spring Maize Planting Area in Liaoning Province Based on MODIS NDVI Data

基于2017年MODIS归一化植被指数(NDVI)数据,采用Savitzky-Golay(S-G)滤波方法和非对称高斯(AG)拟合方法重构辽宁省春玉米种植面积的NDVI时序曲线,使用4次拟合法平滑重构曲线,采用动态振幅阈值法、拐点法和最大值法提取了辽宁省春玉米的关键物候期(出苗期、拔节期、抽雄期),并联合决策树分类法估测了辽宁省春玉米的种植面积。结果表明:基于S-G重构法的MODIS NDVI时序曲线的综合性能更优,其重构曲线与原曲线间的平均均方根误差(RMSE)为0.196,平均平滑拟合度(R 2)为0.983;对春玉米出苗期和抽雄期的提取效果而言,基于S-G重构法的提取结果更优,平均误差分别为3.9和2.6 d;对春玉米拔节期而言,基于S-G重构法和AG重构法的提取效果相当,平均误差分别为2.9和2.8 d;基于S-G重构曲线提取的春玉米种植面积的精度为94.35%,优于基于AG重构曲线提取的精度89.29%。

The NDVI temporal curves of spring maize planting area in Liaoning Province were reconstructed by Savisky-Golay filtering method and Asymmetric Gaussians functions based on the MODIS NDVI data of 2017.The dynamic amplitude threshold method,inflection point method and maximum value method were used to extract the seedling,jointing and tasseling stage,and the spring maize planting areas in Liaoning Province were extracted by decision tree classification.The results showed that the comprehensive performance of the MODIS NDVI time-series curve based on the S-G method was better.The root-mean-square-error(RMSE)of reconstructed curve and origin curve was 0.196,and the average R 2 was 0.983.For seedling stage and tasseling stage,the results based on S-G method were better,with an average error of 3.9 d and 2.6 d.For jointing stage,the results based on S-G method and AG method were comparable,with an average error of 2.9 d and 2.8 d,respectively.The precision of spring maize planting area based on S-G method was 94.35%,which was better than that based on AG method(89.29%).