| Accurate and fast acquisition of forest cover change information is of great significance for study of forest ecological environment changes and management.For change detection,passive remote sensing images have rich spectral information and texture information,which can better retain more detailed information of features,but its images are greatly affected by weather and sun illumination,therefore,it is difficult to obtain accurate and timely forest cover informstion.Active remote sensing,such as Synthetic Aperture Radar(SAR),can achieve all-weather and all-weather observation information.However,due to its own imaging characteristics,SAR images are greatly affected by terrain and speckle noise,and the false alarm rate and missing alarming rate are high in change detection.Therefore,syntheis of active and passive remote sensing technologies to detect forest cover changes at a high spatio-temporal scale has become a new trend.In this study,the typical representative artificial fast-growing forest areas in both the south-Guangxi Gaofeng Forest Farm(Gaofeng study area)and the northern natural forest area-Inner Mongolia Genhe City(Genhe study area),were selected as the research areas.Based on the data of Sentinel-1,using a log ratio method,improved log ratio method and time-series analysis method have been carried out for change detection in the both study areas.Based on Gaofen-1 panchromatic and multispectral data(GF-1 PMS),Multivariate change detection(MAD)algorithm and Iteratively Re-Weighted Multivariate Alteration Detection(IR-MAD)algorithm were carried out in the Gaofeng study area.Based on Gaofen-1 wide field view data(GF-1 WFV),MAD algorithm and IR-MAD algorithm were carried out in the Genhe study area.Field surveys were applied to verify the applicability and reliability of the above methods.Then the IR-MAD algorithm was used to explore the potential of Gaofen-6 wide field view(GF-6 WFV)data this topic.Then the GF-1 PMS was used to explore the applicability and effectiveness of the IR-MAD algorithm and the post-classification comparison method,therandom forest(Image RF)classifier embeded in the En MAP-Box toolkit to detect forest cover change in southern China’s artificial forest farms with high management intensity and complex terrain.Finally,the coarse to fine(CTF)decision fusion was used to obtain a better change detection by synthesizing results from both the active and passive remote sensing data,and the normalized vegetation index(NDVI)was used to determine the direction of forest cover.The results obtained by this study were as follows:(1)For the active remote sensing,the results of change detection in the two study areas using Sentinel-1 data showed that the time-series analysis based on SAR images has the highest change detection accuracy,followed by the improved log ratio method change detection,and the log ratio method has the lowest detection accuracy.(2)For the passive remote sensing,based on the GF-1 PMS and GF-1 WFV data,the change detection results showed that the IR-MAD algorithm has a more complete outline in the extraction of the change area,with less "salt and pepper" and higher accuracy,as compared with MAD algorithm.(3)Based on IR-MAD algorithm,the results of forest cover change detection in the two study areas using GF-1 WFV and GF-6 WFV data showed that the accuracy from GF-6 WFV data is higher than that of GF-1 WFV data.It implied that GF-6 WFV data with red-and yellow-edge bands has some advantages in forest cover change detection,and can play an important role in forest resource monitoring application.(4)The change detection results based on GF-1 PMS data in the Gaofeng study area showed that the IR-MAD algorithm was superior to the Image RF post-classification comparison method in accuracy and extraction effect on the changed areas.It shows that this method can quickly and accurately grasp the changes of forest cover in the study area,and is more suitable for this study area.(5)The accuracy of the change detection results of the CTF decision-level fusion method is better than the change detection accuracy of a single data source and a single method,which effectively limits the false alarm rate and the missing alarm rate.The overall accuracy of the change detection results in Gaofeng research area is 90.65%,and the Kappa coefficient is 0.86;the overall accuracy of the change detection results in Genhe research area is 78.54%,and the Kappa coefficient is 0.72.It shows that the CTF fusion change detection method in the two research areas has achieved good change detection results.In this study a high-spatial-temporal change detection method by incorporation of active and passive remote sensing was proposed in the typical representative forest areas in the both north and south China.The proposed method can quickly and accurately grasp the changes of forest cover in the study area. |
