基于MODIS-EVI及物候差异免阈值提取黄淮海平原冬小麦面积

使用植被指数阈值法提取冬小麦种植面积时,通常需要根据区域间物候差异设置不同阈值。针对这一问题,该文以黄淮海平原为研究区,使用农业气象站生育期观测数据和气象再分析资料,利用逐步进入法模拟冬小麦播种期和成熟期,使用Savitzky-Golay(S-G)滤波重构的MODIS EVI数据逐像元计算播种期至成熟期EVI的峰值频数并结合光谱突变法构建了具有普适性的冬小麦种植面积提取模型。用统计数据验证提取结果表明:在市级尺度和县级尺度上R~2分别为0.91(RMSE 60.08×10~3 hm~2)和0.80(RMSE 8.97×10~3 hm~2)。该文改进的提取模型既考虑了区域间的物候差异,又避免了阈值设置问题,具有一定的普适性,能较好地在大范围内应用于冬小麦面积快速提取,可为大范围内冬小麦监测及估产提供参考。

Extracting winter wheat area in Huanghuaihai Plain using MODIS-EVI data and phenology difference avoiding threshold

Abstract: Huanghuaihai (HHH) Plain is the main cultivating area of winter wheat in China. Accurately detecting the winter wheat area in HHH Plain is of great importance and significance for grain yield estimation and national food security. Vegetation indices (VIs), such as normalized difference vegetation index (NDVI) or enhanced vegetation index (EVI), have been generally used to characterize the winter wheat cultivation area during some key growing stages. However, such method requires thresholds for VIs, which vary spatially due to the differences of local climates and phenology. In other words, if the study area covers wide range of climate gradients which cause significant phenology differences for winter wheat, one invariant threshold value is not suitable for the whole study area. Previous studies usually set different threshold values manually for different regions or provinces to solve this problem. Thus, there is no doubt that the usual solution increases the workload and introduces more uncertainties. For addressing the above issues, a new method was developed and tested in this study for extracting the winter wheat area over HHH Plain of China. The new method used a vegetation index decrease slope threshold to replace the vegetation index threshold at a specified growing stage, and added the peak number during winter wheat growing season as another decision condition. It avoided the issues of setting different threshold values and achieved good accuracy. First, the relationships between the sowing and maturity day of year (DOY) and climate factors were established by stepwise method. In this step, the ERA-Interim reanalysis data (air temperature, precipitation and solar radiation) and observed sowing and maturity date of winter wheat at 140 agro-meteorological sites were used. Among these sites, 70% of them (98 sites) randomly distributed were selected to build the relationships, and the other 30% (42 sites) were used to validate the established relationships. The ERA-Interim reanalysis data were used rather than the observed climate variables at meteorological sites due to the departure of the location of meteorological site from each other. The R2 values of simulated DOY versus observed DOY at agro-meteorological sites were 0.69 and 0.67 for winter wheat sowing and maturity stage, respectively, and the root mean square error (RMSE) values were 6.12 and 4.88 d, which indicated good reliability of the established relationships between sowing and maturity DOY of winter wheat and climate factors. Based on the established relationships, the gridded sowing and maturity DOY of wheat winter for the entire study area were calculated with ERA meteorological variables. Second, the MODIS EVI data (250 m) before and after maturity were used to calculate a decease slope for each pixel, and MODIS EVI data filtered by Savitzky-Golay (S-G) method were used to calculate the peak frequency of EVI curve between sowing and maturity stage of winter wheat for each pixel. Pixels with a decease slope less than ?0.02 and peak frequency equal to 2 were identified as winter wheat. Before these calculations, the sowing and maturity DOY were obtained by resampling to the same resolution with MODIS EVI data. It was unnecessary to set different thresholds for different provinces or regions. Finally, the winter wheat area was obtained after being masked by the 250 m resolution dry land extracted and resampled from land use and land cover data with 1 km resolution. The statistical data were collected from yearbooks and used to validate the statistical area at city and county levels, respectively. Validation results showed that the R2 (RMSE) values were 0.91 (60.08×103 hm2) and 0.80 (8.97×103 hm2) at city and county levels, respectively. The spatial distribution of winter wheat was also agreed well with the results of previous researchers. These demonstrated that the developed method produced a satisfactory accuracy and the result was reliable. The new approach considers the phenology difference and avoids the threshold set, and thus shows a good universal property to extract winter wheat area quickly over a large region.