研究目的
To develop a new approach to map time-series sub-pixel forest cover with Landsat and LiDAR data, integrating these data for continuous monitoring, comparing statistical modeling algorithms, and providing evidence for forest cover change in the Three North Shelter Forest Program area.
研究成果
The integration of LiDAR and Landsat data with temporal trajectory fitting significantly improves forest cover estimation. Random forest with temporal fitting performs best (R2=0.82, RMSE=5.19%). Forest cover in Youyu county increased from 15.5% in 1987 to 37.8% in 2012, demonstrating the success of the Three North Shelter Forest Program. Future research should focus on better sigmoid curve fitting and climate response studies.
研究不足
Lack of Landsat imagery in the 1990s due to cloud cover and SLC failure, challenges in automatic parameter selection for logistic curve fitting, and potential issues with sampling strategy independence and size affecting regression results.
1：Experimental Design and Method Selection:
The study integrates airborne LiDAR and Landsat time-series imagery to derive forest cover change. LiDAR data are used to extract sub-pixel forest cover for a reference year (2009), and the Landtrendr algorithm is applied to Landsat spectral data for temporal noise reduction and trajectory fitting. Four modeling algorithms (stepwise linear regression, quantile regression neural network, support vector machine, and random forest) are compared for forest cover estimation.
2：Sample Selection and Data Sources:
The study area is Youyu county in Shanxi province, China. Airborne LiDAR data were acquired in September 2009 with a Leica ALS60 system. Landsat TM and ETM+ imagery from 1986 to 2013 were downloaded from USGS. Field data from 78 plots (30m x 30m) collected in 2003-2004 are used for validation.
3：List of Experimental Equipment and Materials:
Airborne LiDAR system (Leica ALS60), Landsat satellites (TM and ETM+ sensors), software including RiSCAN PRO for LiDAR processing, and R for statistical modeling.
4：Experimental Procedures and Operational Workflow:
Steps include extracting forest cover from LiDAR using a Beer's Law modified model, preprocessing Landsat imagery (radiometric calibration, atmospheric correction, cloud masking), applying Landtrendr for temporal fitting, building statistical models between forest cover and spectral indices, and validating with cross-validation and field data.
5：Data Analysis Methods:
Performance of models is evaluated using R2, RMSE, and mean error. Ten-fold cross-validation is used for uncertainty analysis.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
