Research On Identification Of Southern Forest Vegetation Types Based An GF-1 WFV And MODIS Spatiotemporal Fusion

It is difficult to obtain long time series of high spatial resolution remote sensing image in southern China because of the complex terrain and frequent cloudy and rainy weather.The spatio-temporal fusion can sychonorously obtain remote sensing data with high spatial-temporal resolution.Combining the fusion time series images and the phenology(PH)characteristics information,it is of great significance to explore the recognition effect of forest vegetation types with different resolution fusion time series images for effective supervision of forest resources.we took Xingguo County of Jiangxi Province as the study area,fused the Landsat8 OLI and GF-1 WFV images with high spatial resolution with high temporal resolution of MODIS09 A1 image respectively on the basis of enhanced spatial and temporal adaptive reflectance fusion model(ESTARFM),reconstructed the time series data of ESTARFM_Landsat8 EVI and ESTARFM_GF-1 EVI with 8d step of enhanced vegetation index(EVI),obtained the phenology characteristics and identified the forest vegetation types by using random forest classification(RFC)model.The research results can provide reference for the application of domestic high-resolution data in forest vegetation type identification.The main research results and conclusions are as follows:(1)Based on the ESTARFM model,it is found that the mean difference between the image predicted by IB fusion method and the real image are smaller than those of BI fusion,and the correlation is higher,which can better represent the spatial information of ground objects and the fusion effect is better.The correlation coefficients of ESTARFM_Landsat8 EVI and ESTARFM_GF-1 EVI generated by different spatial resolution fusion with real images are greater than 0.7,and they have good consistency in spatial distribution.ESTARFM fusion algorithm is also suitable for GF-1 and MODIS data,which can solve the problem of insufficient long-term sequence high spatial resolution images to some extent.(2)S-G filtering method has better reconstruction effect on forest vegetation,and the trend of A-G fitting and D-L fitting curve is similar.The fidelity of three reconstruction algorithms to original time series image is S-G>A-G>D-L.Overall,the correlation between fused Landsat8 image and original image based on S-G filtering reconstruction is higher than that of fused GF-1 image.The temporal reconstruction curves of different ground object types of fused images have different trends,and the distinction is obvious.The filtering curves of different forest vegetation types are mainly’几’ type,and the EVI value of the fusion Landsat8 filtering curve is slightly higher than that of the fusion GF-1 filtering curve.(3)Accuracy verification of phenological extraction from temporal images by standard deviation and root mean square error,and the results show that the phenological extraction results have high reliability.In general,the standard deviation of phenological extraction results of fused GF-1 temporal image is less than that of fused Landsat8 temporal image,and the deviation of data is smaller.The standard deviation of the fusion time series images and MODIS time series images on multiple phenological indicators are close,and the standard deviation is small and the dispersion is small.The root mean square error of phenological extraction results of fused GF-1 temporal image is smaller than that of fused Lansat8 temporal image,which is closer to the ’true value’ and more reliable.The phenological information of main forest vegetationtypes are different,which can reflect different forest types through phenology.(4)The classification accuracy of temporal images is fusion GF-1 temporal image>fusion Landsat8 temporal image>MODIS temporal image,and the classification accuracy of different combinations is EVI+PH>EVI>PH.The overall accuracy and Kappa coefficient of fusion timing images can reach 86.17%and 83.77%,respectively.43 variables were selected as the optimal features to participate in the classification.The classification accuracy of support vector machine(SVM)classification method is slightly lower than that of RFC method,and the model running time is much higher than that of RFC method.The overall accuracy and Kappa coefficient of the optimized feature image random forest classification results were 95.63%and 94.92%,respectively,including 37 temporal EVI values and 6 phenological feature information.The temporal EVI data contributed more to the identification of forest vegetation types,and phenological feature information was conducive to the improvement of classification accuracy.The fused GF-1 temporal image has higher accuracy than the fused Landsat8 temporal image in the identification of forest vegetation types in the southern region with cloudy and rainy terrain conditions.