Study On Estimating Methods Of Forest Canopy Closure Based On Multi-source Remote Sensing Data

Forest canopy closure is one of the essential factors in forest survey, and plays an important role in forest ecosystem management. Because of LiDAR has a high efficiency and high precision ability of obtaining forest vertical structure information, in recent years, it is widely used in studying on the remote sensing estimation of forest structure parameters(the above ground biomass, canopy closure, vegetation coverage, etc.). However, there is still not related research on remote sensing estimation of forest canopy closure in the domestic. Therefore, study on how to effectively combine the lidar data and other remote sensing data, and apply in remote sensing estimation of forest canopy closure is of great significance.In this study, we estimated canopy closure of temperate forest using LiDAR and multi-source remote sensing data(LANDSAT ETM+ image and ALOS PALSAR image). The canopy closure calculated from high density point cloud data of ALS(Airborne Laser Scanning) was used as model training data and validation data, the reflectance, vegetation indices and GLCM textural features calculated from LANDSAT ETM+ image and the backscattering coefficient, GLCM textural features calculated from ALOS PALSAR image were respectively used as independent variables, and Multi-variable stepwise regression(MSR), Random Forest(RF) and Cubist were applied to estimate canopy closure in Genhe forest of Daxinganling area, Inner Mongolia, China.Through compared with inversion results of three kinds of models based on one data source and predicted results of different data source based on one model, we got the following conclusions:(1) Through comprehensive Analysis of three models, Cubist model was optimal and the most stable, and its predictive ability was the strongest, and its inversion effect was the best: the modeling accuracy of Cubist model based on LANDSAT ETM+ image was, R2=0.956, RMSE=0.058, the inversion accuracy was, R2=0.977, RMSE=0.041, rRMSE=0.073, EA(%)=92.712; the modeling accuracy of Cubist model based on ALOS PALSAR image was, R2=0.846, RMSE=0.108, the inversion accuracy was, R2=0.960, RMSE=0.055, rRMSE=0.101, EA(%)=90.368. Random forest model existed the situation of high modeling accuracy and low inversion accuracy, its predictive ability is poorer, in addition, the situation of high canopy closure underestimated and low canopy closure overestimated is more serious. There were abnormal values in the prediction results of Multi-variable stepwise regression, the inversion accuracy was significantly higher than modeling accuracy, which showed the MSR model was not stable enough, and its generalization ability is not strong.(2) From the canopy closure inversion results, it can be showed that, the situation of high canopy closure underestimated and low canopy closure overestimated existed in the three kinds of models in different degrees, and also existed universally and inevitably in the forest parameters remote sensing inversion. The predicted results based on two kinds of different data sources(LANDSAT ETM+ image and ALOS PALSAR image) are in good consistency and of high accuracy, and the predicted values of canopy closure had a similar trend. The fitting accuracy of canopy closure estimation results from MSR, RF and Cubist model based on two kinds of data gradually improved, R2 was respectively equal to 0.9874, 0.9881 and 0.9883. But in the aspect of high value underestimated and low value overestimated, the inversion results based on ALOS PALSAR image was more obvious than based on the LANDSAT ETM+ image, the inversion accuracy based on the LANDSAT ETM+ image was higher, and its prediction ability was stronger. On the choice of remote sensing factors, GLCM texture feature selected in three models based on the LANDSAT ETM+ images and ALOS PALSAR image improved the modeling and inversion accuracy in a certain extent, can better predict the forest canopy closure, the effect of combining reflectance, vegetation index with GLCM texture feature and combining Backscattering coefficient with GLCM texture feature participated in modeling were the best.(3) LiDAR can effectively get the accurate data in a larger area, combined with LiDAR data and other remote sensing data can greatly improve the accuracy of forest parameters remote sensing estimation in regional scale. Therefore, from the domestic and foreign research progress and theoretical point of view, in this study, the canopy closure calculated from ALS-CHM was as the model of training data and validation data, and combining the LANDSAT ETM+ images and ALOS PALSAR image to estimate forest canopy closure in Gehe forest had the certain innovation and popularization. It was a potential remote sensing estimation method of canopy closure, and laid a foundation for the remote sensing estimation of canopy closure in larger regional and even national area in the future.