国产高分一号数据估算草地植被覆盖度方法研究——以呼伦贝尔草原露天煤矿区为例

Estimation Method of Fractional Vegetation Coverage for Grassland Based on Chinese GF-1 Satellite Image:Taking Hulun Buir Prairie Open-Pit Coal Mine as an Example

为了探讨与分析国产高分一号(GF-1)数据在北方露天煤矿区草地植被覆盖度估测中的精度及适用性,该文基于GF-1与SPOT6多光谱影像数据,以多个植被指数为自变量,利用像元二分模型、偏最小二乘(PLS)回归、支持向量机(SVM)回归3种模型对区内植被覆盖度进行估算,结合野外同步实地植被样方数据,对比分析不同估算模型的精度及适宜性,并通过蒙特卡洛模拟多尺度交叉建模的误差传播,分析空间分辨率不同对植被覆盖度估测的精度影响。结果表明:GF-1数据基于增强型植被指数的SVM回归模型(R^2=0.8149,RPD=2.336,RMSE=8.694%)与SPOT6数据基于归一化植被指数的SVM回归模型(R^2=0.8755,RPD=2.870,RMSE=7.032%)估算效果较好。不同分辨率数据交叉传递过程中SVM回归模型的精度高于PLS回归模型。因此,基于GF-1数据构建的SVM回归模型可以高精度地估算区域草地植被覆盖度。

The aim of this study is to indicate the potential ability of Chinese GF-1 satellite imagery in grassland cover retrieval at the surface mine area of north prairie. The Chinese GF-1 and SPOT6 multi-spectral imageries were selected. The Dimidiate Pixel Model,Partial Least Squares Regression （PLSR） and Support Vector Machine Regression （SVMR） models based on the different vegetation indices were established to retrieve grassland cover. The field observation data from sampling plots were used to assess the accuracy of different models. Furthermore, Monte Carlos simulation was conducted to evaluate cross-scale er- ror propagation from SPOT6 to GF-1 data. The results showed that Enhanced Vegetation Index （EVI） from GF-1 can produce high accuracy （R^2 =0. 8149,RPD= 2. 336,RMSE= 8. 694%） based on SVM model, meanwhile, the Normalized Difference Vegetation Index （NDVI） from SPOT6 data can produce high accuracy（R^2 =0. 8755,RPD=2. 870,RMSE= 7. 032%）based on SVM model as well. In terms of the cross-scale error propagation from SPOT6 to GF-1 data,the SVM model outperformed PLS model in grassland cover retrieval. Therefore, the Chinese GF-1 data can provide grassland cover with high accuracy based on SVM model.