基于时空数据融合的江汉平原水稻种植信息提取

PADDY RICE PLANTING INFORMATION EXTRACTION BASED ON SPATIAL AND TEMPORAL DATA FUSION APPROACH IN JIANGHAN PLAIN

及时、准确监测水稻种植面积,对区域粮食政策制定、粮食安全以及农业发展具有重要意义。然而我国南方地区水稻生长期内降水充沛的气候特点使得遥感影像"云污染"现象严重,为解决水稻种植信息遥感提取存在可用数据不足的问题,以江汉平原为例,利用时空数据融合模型(Spatial and Temporal Data Fusion Approach,STDFA)将Landsat 8 OLI与时序MODIS数据融合,重构出具有高时-空分辨率特征数据,然后采用面向对象的SVM分类方法对研究区内水稻种植信息进行提取,结果如下:融合后的红与近红外波段反射率与真实反射率的相关系数分别为0.84和0.81,研究区水稻提取精度为94.46%,Kappa系数为0.91。说明时空融合模型能够较好地重构出高时空分辨率数据,从而实现多云雨地区农作物种植信息遥感提取。

Paddy rice is an important crop in China. Extracting rice planting information timely and accurately is of great significance for food policy, food security and agricultural development. However, there are two difficulties in extracting rice planting information based on remote sensing in Southern China. One is the paddy rice growth period is accompanied by abundant precipitation. This makes the remote sensing imagery influenced by serious ＂cloud contamination＂. The other one is the cultivated land is not concentrated, making the crop classification result influenced by the phenomenon of ＂salt and pepper＂. To solve the problem of lacking available data in extracting paddy rice planting information based on remote sensing, we used Spatial and Temporal Data Fusion Approach （STDFA） to fuse the Landsat 8 and time-series MODIS images and gained the data which had the same temporal resolution with MODIS data and the same spatial resolution with Landsat 8 images. The correlation of reflectance about the red and near-infrared of fused data and true data is 0.84 and 0.81. To address the phenomenon of ＂salt and pepper＂, we used the object oriented image analysis method to derive the fusion result into several image objects and then classify and map the rice distribution in the study area. Using multi-temporal data has a higher accuracy relative to use single phase data in crop mapping, and that method has become an important way to crop classification based on remote sensing. Normalized difference vegetation index （NDVI） is widely used in vegetation classification. Based on the above two points, we used the time-series red and near-infrared data to calculate the time-series NDVI of each image object. Mostly, time-series NDVI data obtained by satellite included various noise components. To obtain change characteristics of NDVI before and after harvest of winter wheat, we used the HANTs filtering method for nosie reduction. Then we used the NDVI data to map the paddy rice fields though Support Vector Machine （SVM） method. We built the confusion matrix though the samples which came from field measurements and validated the extraction accuracy of rice. The Kappa index is 0.91 and the total classification accuracy is 93.16%. The result showed that：（1） Spatial and Temporal Data Fusion Approach has the ability to rebuild the time-series data which has had the high temporal resolution and the high spatial resolution in Southern China; （2） The object-orient method shows a high accuracy in mapping the paddy rice, suggesting that the object-orient classification method can also reduce the ＂salt and pepper＂ phenomenon in the clutter blocks.