Estimation Of Forest Above-ground Biomass And Net Primary Productivity Using Multi-source Remote Sensing Data

As the largest carbon pool in terrestrial ecosystems,forests have a powerful carbon sink function and play an irreplaceable role in the regional and global carbon cycle.Accurate estimation of high-resolution forest biomass and net primary productivity at regional and global scales is of great significance for the studies of global climate change and impact.For forest Above-Ground Biomass(AGB)estimates,the result of AGB estimation based on the traditional approach for forest resource survey can provide accurate and reliable results.However,the approach requires many field investigations,which is time consuming,laborious and destructive.Moreover,it is difficult to implement in areas with poor geographical location and poor natural conditions.In addition,due to the spatial heterogeneity of forest ecosystems,and differences in sampling methods,observation time and strategies selected during measurement,there is a great uncertainty when obtaining forest AGB at large regional and national scales using the traditional approach.For forest Net Primary Productivity(NPP)estimates,commonly used estimation models include statistical-based model and process-based model.The statistical-based model is simple and easy to obtain the parameters,but its ecological and physiological mechanism is not clear,and there is a big uncertainty in the estimation results.The process-based model has an unambiguous mechanism,but the model has some problems such as difficulty in obtaining parameters,high model complexity,and difficulty in performing regional spatio-temporal scale conversion.To obtain a high-resolution,large-scale and high-precision estimation result for forest ecological parameters,remote sensing(RS)technology is widely used to build the model which is simple,efficient and with unambiguous ecological and physiological mechanism.The rapid development of remote sensing technology and the emergence of high-resolution multi-source remote sensing data provide a strong technical and data support for the large-scale forest AGB and NPP estimates.Optical remote sensing imagery can provide horizontal structure information of forest canopy,which can be used to estimate ecological parameters such as leaf area index and canopy density.Synthetic Aperture Radar(SAR)data can penetrate the forest canopy and obtain the vertical structure information of the forest canopy,which can be used to estimate the tree height,volume and other vertical structure parameters.Light Detection and Ranging(Li DAR)can directly obtain high-precision three-dimensional structure information from trees,which can be used to estimate the forest stand height and biomass.The fusion of multi-source remote sensing data can take full use of their advantages,providing more comprehensive and abundant information for the estimation of Forest AGB.In addition,the forest ecological parameters obtained from remote sensing imagery can be used as the input of process-based model for NPP estimation,which can not only combine the physiological growth mechanism,but also make full use of the spatio-temporal scale of remote sensing data,thus providing a reliable technical means to study the spatio-temporal dynamics of forest NPP.Taking different forest types in Yichun City as research object,this paper has explored the methods for estimation of forest AGB and NPP,providing a reliable technical support for the regional climate change and forest resource management.The main researches are:(1)A method of estimating forest stand mean height based on empirical model is proposed,which combines Sentinel-1B radar data and Sentinel-2A multi-spectral image information to realize spatial continuous estimation of stand mean height.Two models are constructed using VH,VV polarization backscatter coefficient from Sentinel-1B and the forest coverage variable(FVC)from Sentinel-2A,respectively,and then the optimal estimation model is found through comparative analysis.The results showed that the estimation model constructed by the VH and FVC is more robust and reliable,and the estimation accuracy is higher.This method can achieve high resolution and spatially continuous estimation results of forest stand mean height in large area,providing a reliable data input for the AGB estimation.(2)A refined forest type extraction method based on object-oriented random forest algorithm is proposed,which combines Sentinel-2A,multi-temporal Landsat-8,Sentinel-1A and DEM remote sensing data and solves the problems of low precision,insufficient refinement and low resolution of forest type extraction.By comparing and analyzing the results of forest type identification,we found out the best features combination.The results showed that the approach of forest type identification based on Sentinel-2A,multi-temporal Landsat-8,VV polarization backscattering from Sentinel-1A,and DEM data could achieve the highest precision.The method can achieve large-scale and high spatial resolution refinement map of forest type,which provides a typical demonstration for research on the refinement of forest types in other areas,and basic data for forest carbon stocks and carbon sink estimates.(3)A method for estimating forest AGB based on machine learning algorithm is proposed,which combines multi-source remote sensing information and effectively solves the problems of scale conversion,signal saturation,and being greatly affected by terrain and wavelength of usual methods.In order to obtain a better model for forest AGB estimates,models were built according to two strategies:(a)To build models according to three different forest types(coniferous forest,broad-leaved forest and mixed conifer–broadleaf forest);(b)To construct model based on all forest types grouped together.By comparing and analyzing the AGB estimation results,we concluded that the AGB estimation model constructed by(a)is better,more stable and accurate.In addition,the AGB estimation models for coniferous forest and broad-leaved forest have similar performance,and are superior to the model of mixed forest types,which may be due to the more complicated forest community structure of mixed forest types.(4)A remote sensing-3PG coupling model is proposed for estimating forest NPP,which considers both forest physiological mechanism,ecological processes and spatio-temporal scalability of ecological parameters estimated based on remote sensing data.The method overcomes the problems of the high complexity of process-based model,numerous input parameters and difficulty in obtaining,and difficulty in expanding and transforming in spatio-temporal scale.