基于MODIS NDVI数据的辽宁省春玉米物候期提取研究

摘要

遥感提取玉米物候信息可以提高田间精细化管理水平,对监测玉米季节性变化、确保粮食安全等具有重要意义.以辽宁省为研究区,基于2007,2008,2009年的MODIS植被指数(NDVI)数据,利用非对称高斯(AG)、Savitzky-Golay滤波(S-G)和双逻辑斯蒂克调合函数(DL)三种拟合方法重构春玉米NDVI时序曲线,并使用4次拟合平滑方法处理重构后的时序曲线.在此基础上,提出以动态振幅阈值法取代动态阈值法,并采用动态振幅阈值法、拐点法和最大值法提取春玉米关键物候期(出苗期、拔节期、抽雄期、成熟期),最后使用辽宁省12个农业气象站的地面物候观测数据对提取结果加以检验.结果表明:基于AG模型的重构NDVI时序曲线保真能力最强,其重构曲线与原曲线各年平均的均方根误差(RMSE)分别为0.035,0.050和0.041;基于S-G滤波方法的重构NDVI时序曲线平滑能力最强,其各年平均的拟合值(R2)分别为0.99,0.993和0.993;就春玉米拔节期、抽雄期、成熟期的提取而言,基于AG时序曲线的提取结果优于其他2种方法的提取结果,其3年提取结果与实际观测值的平均绝对误差(MAE)均小于5d,均方根误差(RMSE)均小于4d;而对春玉米出苗期而言,基于S-G时序曲线的提取结果明显优于其他两种方法的提取结果,其各年出苗期RMSE分别为4.1d、4.0d和2.9d;动态振幅阈值法优于前人所用的动态阈值法,能够有效地纠正异常低值点,使提取结果更接近真实日期.研究结果可为利用遥感资料提取春玉米物候期提供参考.%Having maize phenophases by remote sensing can improve its fine management level in field, which is significant for monitoring seasonal changes of maize and ensuring food security. In this paper,Liaoning province was taken as the studying area, based on the MODIS NDVI data of 2007, 2008 and 2009, and the fitting methods of Asymmetric Gaussians functions, Savitsky-Golay filter and Double Logistic functions are used to reconstruct the NDVI time series. The dynamic amplitude threshold method, inflection point method and maximum value method were used to extract the key phenophases (seedling, jointing, tasseling and maturity stage) of spring maize in Liaoning province, and the phenophases data observed at 12 agrometeorological stations in Liaoning were used to verify the extraction result. The results showed that the reconstruction based on the AG model was better compared with other two methods. The averages root mean square error (RMSE) of 2007, 2008 and 2009 were 0.035, 0.050 and 0.041, respectively. The smoothing effect based on the S-G method was better than those of other two methods. The averages R2 were 0.99, 0.993 and 0.993, respectively. The extraction result based on the AG model was better than those of other two methods on jointing, tasseling and maturity stages. The mean absolute error (MAE) of each year''s extraction result and the actual observation value were less than 5d, and the RMSE was less than 4d. On seedling stage, the extraction result based on S-G model was obviously better than those of other two methods, and the RMSEs on seedling stage of 2007, 2008 and 2009 were 4.1d, 4.0d and 2.9d, respectively. The dynamic amplitude threshold method was better than dynamic threshold method. It could effectively correct abnormally low values and make the extraction results better. This research could provide an effective technological approach to extract the spring maize phenophases of large area.