Extraction of Spring Maize Planting Area by Combined Phenological Feature with Mixed Spectral Information
Obtaining planting area information of spring maize quickly and accurately is of great practical significance to national food security and the development of modern information agriculture. Remote sensing technology has unique advantages in crop planting area extraction, especially the method of combined crop phenology information with spectral data is one of the current study trends. Three counties in Liaoning province were selected as the study area, and MODIS-NDVI time series images from 2014 to 2015 were used as the main data source. Based on the Savitzky-Golay filtering method, MODIS-NDVI time series data were reconstructed by smoothing its noise, crop phenological features in growth season and NDVI curve change patterns of other major objects were extracted from the reconstructed data by TIMESAT, then NDWI and LSWI data during rice transplanting period and the near infrared reflectance data during soybean grain filling period in 2015 were jointly used to train classification threshold, construct the decision tree and extract the 2015 spring maize planting area. With consideration of the fragmentation of the plots and the diversity of land cover types, the spring maize endmember spectrum was extracted based on MOD09A1 reflectance image, and the mixed-pixel decomposition based on the linear spectral mixture model was used to obtain the spring maize abundance. Finally, the spring maize planting area was accurately extracted according to the classification result of the decision tree and the abundance ratio of spring maize. The OLI supervised classification image was used to evaluate the accuracy of the interpreted spring maize planting area, and the results showed that the precision of spring maize planting area was 81%. The statistical data was also used to evaluate the spring maize area accuracy respectively by the decision tree method and the decision tree combined mixed-pixel decomposition method, and the results showed that the precision was respectively 88.416% and 92.382%. The result could reflect the geographical distribution of spring maize, indicating that the reconstructed NDVI growth curve could accurately describe crop growth regularity. Using moderate-resolution time series images to extract crop acreage is feasible, and the mixed-pixel decomposition method could further improve the area extraction accuracy. The method of combined the mixed spectral information broke through the pixel limitation in traditional researches, enhanced the crop acreage extraction level to a sub-pixel scale, and was better than the traditional classification method based on pixel-scale MODIS time series information, which would effectively improve the crop acreage estimation accuracy, play an important role in accelerating digital agriculture process and enhancing the agricultural information level.
国家科技期刊平台
NSTL
万方数据
维普
