研究目的
To estimate forest aboveground biomass using a combination of Sentinel-1 and Sentinel-2 data with Random Forest Regression, exploring the advantages of multi-temporal data, examining the regression approach, and validating estimates at forest stand and compartment levels.
研究成果
The combination of Sentinel-1 and Sentinel-2 data with Random Forest Regression effectively estimates forest biomass, with a saturation effect at 200 t/ha and best performance for biomass ranges of 100-200 t/ha. The approach supports large-scale biomass assessment but requires further analysis to improve accuracy in low and high biomass forests.
研究不足
The model shows saturation effects around 200 t/ha, overestimates low biomass values, and underestimates high biomass values. Limitations include potential inaccuracies due to backscatter mechanisms, soil moisture, and forest understory structure, with RMSE of 60 t/ha and variability across biomass ranges.
1：Experimental Design and Method Selection:
The study uses Random Forest Regression for biomass estimation, leveraging multi-temporal Sentinel-1 and Sentinel-2 data. Pre-processing includes calibration, geocoding, radiometric normalization, and speckle reduction.
2：Sample Selection and Data Sources:
Reference data from the National Forest Condition Inventory (2015-2016) and State Forest Inventory database were used, with sampling plots selected based on criteria like distance from forest edge and slope. EO data from Sentinel-1 and Sentinel-2 were downloaded via ESA Copernicus Data Access Hub.
3：List of Experimental Equipment and Materials:
Sentinel-1 and Sentinel-2 satellites, ESA SNAP toolbox for pre-processing, Sen2Cor package for atmospheric correction, and software for data analysis.
4：Experimental Procedures and Operational Workflow:
Data pre-processing, calculation of multi-temporal sums and medians for polarizations, preparation of cloud-free mosaics, training and validation of the RF model with 70% training and 30% validation data, and independent validation at forest stand level.
5：Data Analysis Methods:
Random Forest Regression with 200 trees, RMSE calculation for error assessment, and validation against reference data.
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
