Forest Above Ground Biomass（ AGB） estimation is important for ecosystem monitoring and carbon cycling studies.Accurately estimating regional and global AGB can reduce the uncertainty of carbon budgets.Over the last six years

regional and global forest AGB have been derived from various remote sensing data

including spaceborne LiD AR data（ height and vertical structure parameters）

optical multispectral data（ Vegetation Index（ VI）

Leaf Area Index（ LAI）

Absorbed Photosynthetic Active Radiation（ APAR）

image texture

Digital Surface Model（ DSM） and optical point cloud）

and microwave data（ backscattering coefficient

coherence

scattering phase center height

and DEM）. In this study

we reviewed the advantages and limitations of three kinds of inversion methods

i. e.

parametric method based on single sensor data

non-parametric method based on multi-sensor data

and a method based on physical mechanism models.First

parametric method mainly obtains multiple regression equations by analyzing the statistical relationship between AGB and various remote sensing variables. The method is simple but strongly dependent on site and time. Second

non-parametric methods were used to solve nonlinear and high-dimensional problems

including decision trees

k-nearest neighbors

artificial neural network

and support vector machine method. Such method is widely used in global and regional AGB estimation

but it lacks a physical mechanism and its accuracy depends on the number of training data sets. Third

the method based on mechanism models includes direct inversion using semi-empirical models and a look-up table method based on forest forward simulation model. Method usage is limited because of the contradiction between the accuracy and complexity of the model.As for remote sensing data used in AGB estimation

the spectral variables extracted from optical data have been widely applied. Radar is unaffected by weather conditions and it is capable of obtaining signal from branches

trunks

and even understories. Backscattering coefficient with SAR image

interferometric coherence with InS AR

vertical structure with Pol-InS AR

and backscattering contribution ratio of ground and vegetation with PCT technology are all closely related to AGB. Advances in LiD AR technology have demonstrated a capability to obtain the height and three-dimensional structure of forests

but its limitations include canopy species recognition and lack of spaceborne data.AGB estimation by combining multi-source remote sensing data has become a development trend because the data obtained from different portions of the electromagnetic spectrum and different observation configurations provide comprehensive information on forests. However

the retrieval accuracy did not meet the demands of ecosystem monitoring and carbon cycling study thus far. The uncertainties were attributed to the complexity of forest structures

mixed pixels and scale effect

as well as errors in allometric equations. The four potential aspects of biomass inversion studies to improve accuracy are presented: forest physical mechanism model study

multi-sensor synergy method

biomass seasonal and time variation study

and future data sources support.