基于相似性分析及线性光谱混合模型的双季稻面积估算

Estimation of double cropping rice planting area using similar index and linear spectral mixture model

为了解决大范围水稻种植信息提取时的混合像元问题,以便能够准确及时地获取水稻信息、指导水稻生产、保证粮食安全,该文提出了一种基于相似性分析和线性光谱混合模型复合的水稻提取业务化方法。以江西省为研究区,利用2010年4月15日至2010年10月31日的MODIS合成地表反射率数据(MODIS09A1),计算出时间序列MODIS-EVI指数,运用Savizky-Golay滤波方法对其进行平滑处理减少云等噪声的影响。根据双季稻的生长规律,结合野外调查和HJ-1ACCD2影像,确定双季稻样点,提取出标准双季稻EVI生长变化曲线,构建图像像元相似性指数,然后采用线性光谱混合像元分解模型对疑似双季稻像元进行混合像元分解,获得江西省双季稻种植面积信息的分布情况。结果显示,运用该方法提取的江西省双季稻种植分布情况与实际情况吻合,与江西省2010年统计年鉴中全省双季稻种植面积相比,提取精度为93%,精度较理想,与各地区统计面积相关性较好,R2=0.9659,可以为今后高精度水稻种植信息业务化的提取提供参考。

The planting of double cropping rice is a cropping system of planting and harvest twice a year in China. It is essential to get the planting area and spatial distribution of paddy rice at large scale for guiding rice production and regulating the regional balance of supply and demand. Currently, using remote sensing technology to monitor the rice planting in a wide range is becoming an increasingly important tool. However, the mixed pixel problem makes it imprecise to extract the double cropping rice paddy. In order to solve the mixed pixel problems on extracting planting area of paddy rice at large scale, a method was proposed based on similar index and linear spectral mixture model. The time-series of MODIS-EVI index was calculated by using multi-temporal MODIS09A1 data from 2010-4-15 to 2010-10-31 in Jiangxi province. The influence factors such as cloud were reduced by using Savizk-Golay filtering method. Combined with field works and HJ-1A CCD2 images, the rice field samples were identified according to the rice growth patterns. Then the standard double cropping rice EVI curve was extracted, and the similar index between each pixel’s EVI value of MODIS images, and standard double cropping rice EVI curve was calculated. To construct the similarity index map of rice, the suspected pixels in double cropping rice areas were extracted, and each mutually independent spectrum was got based on minimum noise fraction separating principal component and noise. The pixel points which pixel purity index larger than 3.0 were selected by calculating the image pixel purity index and extracting the high purity pixel, and the results of N-dimensional divergence were analyzed using N-dimensional visualization tools. In order to test the accuracy of the extracting method, the HJ-1A CCD2 datum was used to carry on the spatial contrast verification, and the result showed that it was substantially coincide in spatial distribution. Compared with the Statistical Yearbook of Jiangxi Province in 2010, the extraction accuracy was 93%, the correlation of the regional statistical was satisfied with R2=0.9659. The study can provide a reference for the extraction of high precision rice information in the future.