CN104656098B - A kind of method of remote sensing forest biomass inverting - Google Patents

Abstract

The invention discloses a method for remote sensing forest biomass inversion. On the basis of preprocessing the remote sensing data, the method respectively extracts the data from the LiDAR point cloud (including the three-dimensional spatial information of the canopy) and the multi-spectrum (including the canopy upper surface information). The characteristic variables of the vegetation canopy were extracted from the spectral information) data; the above-mentioned LiDAR point cloud and multi-spectral characteristic variables were screened through correlation analysis, and the aboveground and underground biomass were retrieved through a stepwise regression model combined with the ground-measured biomass information. The optimal inversion model of the northern subtropical forest biomass constructed by the present invention can improve the model "coefficient of determination" R by 3-24%; and can estimate the forest biomass with high precision, and reduce the "relative root mean square error" rRMSE by ² -10%. It can be applied in the fields of forestry survey, forest resources monitoring, forest carbon stock assessment and forest ecosystem research, and provide quantitative data support for sustainable forest management and comprehensive utilization of forest resources.

Description

A Method of Remote Sensing Forest Biomass Retrieval

technical field

The invention relates to the technical field of forest pest monitoring and pollution-free control, in particular to a remote sensing forest biomass inversion method.

Background technique

Accurate biomass estimation is of great significance for forest resource monitoring, forest carbon stock assessment and forest ecosystem research. At the same time, these information can also provide quantitative data support for sustainable forest management and comprehensive utilization of forest resources. The traditional biomass survey method is time-consuming and laborious, and can only obtain limited "point" information, which is difficult to be practically promoted in a large area; however, remote sensing technology can accurately and quickly obtain forest parameters at various scales, and has a very good practical value and application prospects.

The Landsat series of satellites can acquire forest multispectral (optical) information on medium and large scales. Its latest Landsat 8OLI sensor (the sensor is carried on the Landsat 8 satellite launched by NASA on February 11, 2013) is compared with previous sensors such as TM in terms of band setting and sensitivity to vegetation. There is a big improvement. However, optical remote sensing is still difficult to penetrate the forest canopy to obtain its vertical structure information, and it is easy to be saturated when obtaining forest biomass information in areas with high forest coverage (vegetative growth). LiDAR (LiDAR, Light Detection AndRanging) is an active remote sensing technology that has developed rapidly in recent years. The laser pulses it emits can penetrate the vegetation canopy to obtain its three-dimensional structure and energy information. Previous studies have shown that LiDAR can accurately estimate the biological characteristics of different forest types. It has great potential in terms of physical and structural properties.

In recent years, researches on inversion of forest biomass include: 1) Zheng et al published "Estimating aboveground biomass using Landsat7ETM+data across a managed landscape in northern Wisconsin, USA" in Volume 35 of "Remote Sensing of Environment" in 2004. Vegetation indices such as NDVI extracted from ETM+ (Landsat 7) images retrieved aboveground biomass information of coniferous forests in northern Wisconsin, USA. 2) Yang Cunjian et al. published "Correlation Analysis between Tropical Forest Vegetation Biomass of Different Age Groups and Remote Sensing Geoscience Data" in "Journal of Plant Ecology" Volume 28 in 2004. Using the original band method of TM (Landsat5) images to analyze the Estimation of vegetation biomass in Xishuangbanna tropical forest. 3) Ferster et al published "Aboveground Large Tree Mass Estimation in a Coastal Forest in British Columbia Using Plot-Level Metrics and IndividualTree Detection from LiDAR" in Volume 35 of "Canadian Journal of Remote Sensing" in 2009. : 0.1-2m) The point cloud height and canopy density information extracted from the LiDAR data were used to invert the aboveground biomass of temperate forests. 4) Saatchi et al. published "Benchmark Map of Forest Carbon Stocks in Tropical Regions across Three Continents" in the 12th issue of "Proceedings of the National Academy of Sciences of the United States of America" in 2011. This study was obtained from GLAS (Geoscience Laser Altimeter System ) large spot (diameter: 52-90m) LiDAR data to extract characteristic variables related to tree height information, and invert the biomass of tropical rainforest. However, the above methods only mine traditional "optical" and LiDAR data from a single angle, with low specificity, and the mining depth of feature variables is relatively shallow (that is, feature variables are not systematically extracted and screened from multiple angles), It is not yet possible to accurately invert the biomass.

Contents of the invention

Purpose of the invention: Aiming at the deficiencies in the prior art, the present invention proposes a remote sensing forest biomass inversion method, which has the characteristics of strong specificity, low cost, and easy popularization and application.

Technical solution: In order to realize the above-mentioned purpose of the invention, the technical solution adopted in the present invention is:

A remote sensing forest biomass inversion method: on the basis of preprocessing the remote sensing data, extracting The characteristic variables of the vegetation canopy; the above-mentioned LiDAR point cloud and multispectral characteristic variables were screened through correlation analysis, and combined with the measured biomass information on the ground, the aboveground and underground biomass were inverted through a stepwise regression model.

The method for inversion of forest biomass by remote sensing comprises the following steps:

1) OLI image preprocessing: First, radiometrically calibrate the original image with the radiometric calibration parameters of the OLI sensor; convert the original DN value into a pixel radiance value; Then the image is geometrically corrected, and the point of the same name is selected, and the quadratic polynomial is used for correction. The correction error is controlled within 0.1 pixels, and the nearest neighbor pixel method is used for resampling.

2) LiDAR data preprocessing:

a) Noise level estimation and data smoothing: first convert the original data to the frequency domain, and then use the low-value part with higher frequency as the judgment standard of the noise level; then use the Gaussian filter for smoothing, because the Gaussian filter is effective in While smoothing the data, it can also maintain the trend of the original curve to the greatest extent;

b) Gaussian fitting (decomposition) and point cloudization of waveform data: Based on the assumption that the echo data is the accumulation of multiple Gaussian functions, the waveform data is fitted using the nonlinear least squares method; and then filtered through local maximum peak detection The algorithm extracts discrete point clouds from the processed waveform data, and the energy and amplitude information of the returned signal is recorded in each discrete point;

c) Generating digital terrain: The purpose of height normalization of LiDAR data is to obtain the "true" vegetation height without the influence of terrain, which is usually obtained by subtracting terrain height from the original LiDAR data height information; therefore, the accurate generation of digital terrain model (DTM ) is an important prerequisite for calculating the normalized vegetation height; firstly, classify the discrete point cloud extracted from the waveform data, then perform Kraus filtering on the last echo to remove non-ground points, and finally use the filtered last echo Data and generate a digital terrain model by means of natural proximity method interpolation;

d) Use DEM to normalize the elevation of the vegetation echo points, even if the height value of the vegetation point is converted into a height value relative to the ground, and cut each plot by the coordinates of the lower left corner and upper right corner ;Finally, extract the normalized point cloud data corresponding to the coordinate positions of 55 sample plots through GIS analysis tools;

3) OLI feature variable extraction: By performing band combination, tasseled cap transformation, texture information extraction, principal component analysis, minimum noise separation transformation and various vegetation index transformations on OLI images, five groups of characteristic variables were extracted (see Table 1 for details), That is, the original single-band variables, band combination variables, information-enhanced group variables, vegetation index variables, and texture information variables; where texture analysis is performed on the first principal component of principal component analysis;

4) Extracted LiDAR point cloud feature variables: LiDAR point cloud feature variables are based on three-dimensional normalized LiDAR point cloud values to calculate four groups of feature variables (see Table 2 for details), namely height variable, height percentile variable, crown Layer density variable, canopy coverage variable;

5) Feature variable screening: Perform Pearson's correlation analysis on the extracted LiDAR feature variables and OLI feature variables and the parameters to be predicted, and select the feature variables whose absolute value of Pearson's correlation coefficient is higher than 0.2 as modeling candidate variables; Pearson's correlation coefficient The calculation method is:

In the formula, x _i is the measured characteristics of a forest stand on the ground, y _i is a LiDAR feature variable, is the average value of _xi , is the average value of y _i ;

6) Statistical analysis: The biomass information collected from ground measurements was used as the dependent variable, and the feature variables extracted by remote sensing methods were used as independent variables to establish a multiple regression model. Use the stepwise method (stepwise) and test the change of the coefficient of determination (R ² ) to select the appropriate variables to enter the model; if there are independent variables that make the statistic F value too small and the T test cannot reach the significant level (P value>0.1 ), then it will be eliminated; if the F value is large and the T test reaches a significant level (P value <0.05), then it can be entered; the coefficient of determination (R ² ), root mean square error (RMSE) and relative root mean square error (rRMSE) are used Evaluate the accuracy of the regression model;

In the formula, x _i is the characteristics of a forest stand measured on the ground, is the average value of _xi , is the characteristic of a forest stand estimated by the model, and n is the number of sample plots;

It can be seen from formula (3) that rRMSE is RMSE (root mean square error) and (measured value mean) percentage.

The method for remote sensing forest biomass inversion of the present invention, on the basis of preprocessing LiDAR and OLI data, respectively from the point cloud (comprising three-dimensional space information of the canopy) and multispectral (comprising the spectral information of the upper surface of the canopy) Extract the characteristic variables of the vegetation canopy from the data, and then combine the measured biomass information on the ground to invert the aboveground and underground biomass through the stepwise regression model on the basis of screening the characteristic variables. information and the spectral information on the upper surface to fully exploit the vegetation biophysical characteristics information contained in the two dimensions; 2) use the correlation analysis to screen the above extracted characteristic variables and use them in the final inversion model, This is conducive to mechanism interpretation and method transplantation; 3) Since OLI data is free data, this method also greatly saves the cost of improving the accuracy of biomass inversion on multiple scales.

Beneficial effects: Compared with the prior art, the present invention fully excavates the vegetation biophysical characteristic information contained in the three-dimensional space information of the canopy and the spectral information dimension of the upper surface by integrating LiDAR point cloud and OLI multi-spectral characteristic variables , and with the help of correlation analysis to screen the above extracted feature variables, and finally determine the inversion model. This method is conducive to mechanism interpretation and method transplantation; and because OLI data is free data, this method also greatly saves the cost of improving the accuracy of biomass inversion on multiple scales. Preliminary experimental verification results show that the optimized inversion model of the northern subtropical forest biomass constructed by the present invention (compared with the method based on a single multispectral data source) can improve the model "coefficient of determination" R ² by 3-24%; And it can estimate the forest biomass with high precision (reduce the "relative root mean square error" rRMSE by 2-10%). It can be applied in the fields of forestry survey, forest resources monitoring, forest carbon stock assessment and forest ecosystem research, and provide quantitative data support for sustainable forest management and comprehensive utilization of forest resources.

Description of drawings

Figure 1 is the true-color aerial photo of the study area and the distribution map of 55 plots;

Figure 2 is the technical roadmap of remote sensing forest biomass retrieval methods;

Figure 3 is the Pearson's correlation coefficient diagram between each characteristic variable and aboveground biomass;

Figure 4 is a graph of Pearson's correlation coefficient between each characteristic variable and underground biomass;

Figure 5 is a scatter diagram of the OLI model-comparison of the measured values of aboveground biomass plots and the predicted values of the model;

Figure 6 is a scatter diagram of the OLI model - the comparison between the measured value of the underground biomass sample plot and the predicted value of the model;

Figure 7 is a scatter diagram of the LiDAR model-comparison of the measured values of aboveground biomass plots and the predicted values of the model;

Figure 8 is a scatter diagram of the LiDAR model-comparison of the measured value of the underground biomass sample plot and the predicted value of the model;

Figure 9 is a scatter diagram of the comprehensive model-comparison of the measured values of the aboveground biomass sample plots and the predicted values of the model;

Figure 10 is a scatter diagram of the comprehensive model-comparison of the measured values of underground biomass plots and the predicted values of the model.

Detailed ways

The present invention will be further described below in conjunction with specific examples.

Example 1

A method for remote sensing forest biomass inversion. The technical route is shown in Figure 1. This method is a method for remote sensing forest biomass inversion that integrates LiDAR point cloud and OLI multispectral feature variables. In the preprocessing of remote sensing data On this basis, the characteristic variables of the vegetation canopy were extracted from the LiDAR point cloud (including the three-dimensional spatial information of the canopy) and the multi-spectral (including the spectral information of the upper surface of the canopy) data respectively; Spectral characteristic variables, combined with ground-measured biomass information, were used to invert aboveground and belowground biomass through a stepwise regression model. details as follows:

The research area is located in the state-owned Yushan Forest Farm in Changshu City, Jiangsu Province (120°42′9.4″E, 31°40′4.1″N), which belongs to the subtropical monsoon climate, with a mild climate and an average annual precipitation of 1054mm. The area is about 1103hm ² The height is 20-261m. Yushan Forest Farm belongs to the northern subtropical secondary mixed forest. The main forest types are coniferous forest, broad-leaved forest and mixed forest. The main coniferous tree species are Pinus massoniana (Pinus massoniana), Chinese fir (Cunninghamia lanceolata), Slash pine (Pinuselliottii) and Black pine (Pinus thunbergii), etc.; the main broad-leaved tree species are Quercus acutissima (Quercus acutissima), sweet maple (Liquidambar formosan) and some evergreen broad-leaved tree species, such as Fagaceae, Lauraceae and camellia Plants of the family Theaceae.

55 square plots (size: 30×30m, setting time: July-August 2012 and August 2013). According to the proportion of tree species, the plots were divided into three types: coniferous forest (n=13), broad-leaved forest (n=16) and mixed forest (n=26). During the plot survey, for trees with a diameter at breast height greater than 5 cm, the tree species, diameter at breast height (measured with a girth), tree height and height under branches (measured with a Vertex IV laser altimeter) and crown width (that is, two The projection distance in the main direction, measured with a tape measure), counts the dead trees with a DBH less than 5cm, but does not participate in the calculation of biomass. The coordinates of the southwest corner of the sample plot were determined using differential GPS (Trimble GeoXH6000 Handheld GPS units), and the positioning accuracy was better than 0.5 meters by receiving JSCORS wide-area differential signals (Figure 2).

According to the individual tree survey data, the relevant forest parameters at the plot scale were summarized, including aboveground and belowground biomass per unit area (t·hm ^-2 ) at the plot scale. Biomass information Calculate the biomass of a single tree through the allometric growth equation (following the principle of proximity), and summarize the aboveground biomass (W _A ) and belowground biomass (W _B ) per unit area of each plot.

OLI image preprocessing. The study used the 2-7 bands in the Landsat 8OLI image (strip number 119/38) on July 19, 2013. The spatial resolution of the image is 30m, the radiation resolution is 12bit, and the spectral range covers 11 bands. First, radiometric calibration is performed on the original image with the help of the radiometric calibration parameters of the OLI sensor. Convert the original DN value to the pixel radiance value. Then use the FLAASH model to perform atmospheric correction on the image (atmospheric model: Tropical; aerosol model: Urban; aerosol inversion method: 2-Band (K-T); initial visibility: 40; spectral response function: ldcm_oli.sli), so that The radiance value is converted to the actual reflectance of the ground surface. Then the image was geometrically corrected, and 40 points of the same name were selected, and the quadratic polynomial was used for correction. The correction error was controlled within 0.1 pixel, and the nearest neighbor pixel method was used for resampling.

LiDAR data preprocessing. 1) Noise level estimation and data smoothing: first convert the original data to the frequency domain, and then use the low-value part with higher frequency as the judgment standard of the noise level. Then the Gaussian filter is selected for smoothing, because the Gaussian filter can maintain the trend of the original curve to the greatest extent while effectively smoothing the data. 2) Gaussian fitting (decomposition) and point cloudization of waveform data: Based on the assumption that the echo data is the accumulation of multiple Gaussian functions, the nonlinear least square method is used to fit the waveform data. Then the discrete point cloud is extracted from the processed waveform data through the local maximum peak detection filtering algorithm, and the energy and amplitude information of the returned signal is recorded in each discrete point. 3) Generating digital terrain: The purpose of normalizing the height of LiDAR data is to obtain the "real" vegetation height without the influence of terrain, which is usually obtained by subtracting the terrain height from the height information of the original LiDAR data. Therefore, accurate generation of digital terrain model (DTM) is an important prerequisite for calculating normalized vegetation height. Firstly, the discrete point cloud extracted from the waveform data is classified, and then Kraus filtering is performed on the last echo to remove non-ground points. Finally, the digital terrain model is generated by interpolating the filtered last echo data with the help of natural proximity method. Then use the DEM to normalize the elevation of the vegetation echo points, that is, the height value of the vegetation point is converted into a height value relative to the ground, and cut each plot by the coordinates of the lower left corner and upper right corner. Finally, the normalized point cloud data of the corresponding coordinate positions of 55 sample plots were extracted by GIS analysis tools.

By performing band combination, tasseled cap transformation, texture information extraction, principal component analysis, minimum noise separation transformation and multiple vegetation index transformations on OLI images, 5 groups (53 in total) of feature variables were extracted, 6 original single-band variables, 10 band combination variables, 10 information enhancement group variables, 18 vegetation index variables, and 9 texture information variables. Texture analysis is performed on the first principal component of principal component analysis, with a window size of 3×3 and a lag distance of 1 image Yuan. The meaning and calculation formula of 53 variables (see Table 1). LiDAR feature variables are based on 3D normalized LiDAR point cloud values to calculate 4 groups (34) of feature variables: 10 height variables, 13 height percentile variables, 10 canopy density variables, and 1 canopy coverage degree variable. The meaning and calculation formula of 34 variables (see Table 2).

Table 1 OLI characteristic variables and description

A total of 87 extracted variables (34 LiDAR feature variables and 53 OLI feature variables) were analyzed by Pearson correlation with the parameters to be predicted, and the feature variables whose absolute value of Pearson's correlation coefficient was higher than 0.2 were selected as modeling candidate variables. The calculation method of Pearson's correlation coefficient is:

In the formula, x _i is the measured characteristics of a forest stand on the ground, y _i is a LiDAR feature variable, is the average value of _xi , is the average value of y _i ;

The study selected 34 candidate variables, and the absolute values of these candidate variables and their Pearson's correlation coefficients with the parameters to be predicted are shown in Figure 3 and Figure 4. The Pearson's correlation coefficients between these 34 variables and biomass are high and significant, indicating that there is a good linear relationship between them. Therefore, these 34 characteristic variables are regarded as the final modeling candidate variables.

Table 2 LiDAR characteristic variables and description

A multiple regression model was established by using the biomass information collected from ground measurements as dependent variables and the feature variables extracted by remote sensing methods as independent variables. Appropriate variables to enter the model were selected by using the stepwise method (stepwise) and examining the variation of the coefficient of determination (R ² ). If there is an independent variable that makes the F value of the statistic too small and the T test cannot reach the significant level (P value>0.1), it will be eliminated; if the F value is large and the T test reaches the significant level (P value<0.05), it will be included. The accuracy of the regression model was evaluated by coefficient of determination (R ² ), root mean square error (RMSE) and relative root mean square error (rRMSE);

In the formula, x _i is the characteristics of a forest stand measured on the ground, is the average value of _xi , is the characteristic of a forest stand estimated by the model, and n is the number of sample plots;

It can be seen from formula (3) that rRMSE is RMSE (root mean square error) and (measured value mean) percentage.

Firstly, the OLI biomass estimation model (OLI model for short) and the LiDAR biomass estimation model (LiDAR model for short) were constructed using OLI characteristic variables and LiDAR characteristic variables respectively, and then a comprehensive biomass estimation model (comprehensive model for short) was constructed based on these two types of characteristic variables. ). When constructing these three models, the analysis was divided into two situations. Firstly, the statistical analysis of all sample plots was performed without distinction, and then the sample plots were divided into coniferous forest, broad-leaved forest and mixed forest according to the composition of tree species for analysis. The accuracy evaluation of each model for different forest types is shown in Table 3. See Figure 5-10 for the comparison scatter points between the measured values and the predicted values of the models (OLI model, LiDAR model and comprehensive model) based on aboveground and underground biomass of different forest types.

Table 3 Model accuracy evaluation table for different forest types

Among them, R ² is the coefficient of determination of the model; RMSE is the root mean square error; rRMSE is the relative root mean square error.

The above results show that the inversion result of the integrated model is better than that of the single characteristic variable model (ie, OLI model and LiDAR model), and the model fitting effect is improved: R ² increased by 3-24%; the estimation accuracy is improved: rRMSE decreased by 2 -10%.

Claims (4)

1. a method for remote sensing forest biomass inversion, it is characterized in that, comprises the following steps: 1) OLI image preprocessing: Firstly, radiometric calibration is performed on the original image with the radiometric calibration parameters of the OLI sensor, and the original DN value is converted into a pixel radiance value; then, atmospheric correction is performed on the radiometrically calibrated image using the FLAASH model , so as to convert the radiance value into the actual reflectance of the surface; then carry out geometric fine correction on the image after atmospheric correction, select the point of the same name, and use the quadratic polynomial for correction, the correction error is controlled within 0.1 pixel, and adopt The nearest neighbor cell method is used for resampling; 2) LiDAR data preprocessing: first perform noise level estimation and data smoothing, then Gaussian fitting and point cloudization of waveform data to generate digital terrain, and then use DEM to normalize the elevation of vegetation echo points; 3) OLI feature variable extraction: By performing band combination, tasseled cap transformation, texture information extraction, principal component analysis, minimum noise separation transformation and multiple vegetation index transformation on OLI images, five groups of characteristic variables are extracted, which are the original single-band variables , band combination variables, information enhancement group variables, vegetation index variables, and texture information variables; wherein texture analysis is performed on the first principal component of principal component analysis; 4) Extracted LiDAR point cloud feature variables: LiDAR point cloud feature variables are four groups of feature variables calculated based on three-dimensional normalized LiDAR point cloud values, which are height variable, height percentile variable, canopy density variable, canopy density variable, and canopy density variable. layer coverage variable; 5) Feature variable screening: Perform Pearson's correlation analysis on the extracted LiDAR feature variables and OLI feature variables and the parameters to be predicted, and select the feature variables whose absolute value of Pearson's correlation coefficient is higher than 0.2 as modeling candidate variables; Pearson's correlation coefficient The calculation method is: <mrow><mi>r</mi><mo>=</mo><mfrac><mrow><munderover><mi>&amp;Sigma;</mi><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><mrow><mo>(</mo><mrow><msub><mi>x</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>x</mi><mo>&amp;OverBar;</msub>mo></mover><mi>i</mi></msub></mrow><mo>)</mo></mrow><mrow><mo>(</mo><mrow><msub><mi>y</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>y</mi><mo>&amp;OverBar;</mo></mover><mi>i</mi></msub></mrow><mo>)</mo></mrow></mrow><msqrt><mrow><munderover><mi>&amp;Sigma;</mi><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><msup><mrow><mo>(</mo><mrow><msub><mi>x</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mi>i</mi></msub></mrow><mo>)</mo></mrow><mn>2</mn></msup><mo>&amp;CenterDot;</mo><munderover><mi>&amp;Sigma;</mi><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><msup><mrow><mo>(</mo><mrow><msub><mi>y</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>y</mi><mo>&amp;OverBar;</mo></mover><mi>i</mi></msub></mrow><mo>)</mo></mrow><mn>2</mn></msup></mrow></msqrt></mfrac><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>1</mn><mo>)</mo></mrow></mrow> In the formula, x _i is the measured characteristics of a forest stand on the ground, y _i is a LiDAR feature variable, is the average value of _xi , is the average value of y _i ; 6) Statistical analysis: use the biomass information collected by the ground measurement as the dependent variable, and the characteristic variables extracted by the remote sensing method as the independent variable to establish a multiple regression model; use the stepwise method and check the change of the coefficient of determination R ² to select the entry model If there is an independent variable that makes the F value of the statistic too small and the T test does not reach a significant level, it will be eliminated; if the F value is large and the T test reaches a significant level, it will be entered; the coefficient of determination R ² , the mean square The root error RMSE and the relative root mean square error rRMSE evaluate the accuracy of the regression model; <mrow><msup><mi>R</mi><mn>2</mn></msup><mo>=</mo><mn>1</mn><mo>-</mo><mfrac><mrow><munderover><mo>&amp;Sigma;</mo><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><msup><mrow><mo>(</mo><msub><mi>x</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>x</mi><mo>^</mo></mover><mi>i</mi></msub><mo>)</mo></mrow><mn>2</mn></msup></mrow><mrow><munderover><mo>&amp;Sigma;</mo><mrow><mi>i</mo>mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><msup><mrow><mo>(</mo><msub><mi>x</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mi>i</mi></msub><mo>)</mo></mrow><mn>2</mn></msup></mrow></mfrac><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>2</mn><mo>)</mo></mrow></mrow> In the formula, x _i is the characteristics of a forest stand measured on the ground, is the average value of _xi , is the characteristic of a forest stand estimated by the model, and n is the number of sample plots; <mrow><mi>R</mi><mi>M</mi><mi>S</mi><mi>E</mi><mo>=</mo><msqrt><mrow><mfrac><mn>1</mn><mi>n</mi></mfrac><munderover><mi>&amp;Sigma;</mi><mrow><mi>i</mi><mo>=</mo><mn>1</mn></mrow><mi>n</mi></munderover><msup><mrow><mo>(</mo><mrow><msub><mi>x</mi><mi>i</mi></msub><mo>-</mo><msub><mover><mi>x</mi><mo>^</mo></mover><mi>i</mi></msub></mrow><mo>)</mo></mrow><mn>2</mn></msup></mrow></msqrt><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>3</mn><mo>)</mo>mo></mrow></mrow> <mrow><mi>r</mi><mi>R</mi><mi>M</mi><mi>S</mi><mi>E</mi><mo>=</mo><mfrac><mrow><mi>R</mi><mi>M</mi><mi>S</mi><mi>E</mi></mrow><msub><mover><mi>x</mi><mo>&amp;OverBar;</mo></mover><mi>i</mi></msub></mfrac><mo>&amp;times;</mo><mn>100</mn><mi>%</mi><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>4</mn><mo>)</mo></mrow><mo>;</mo></mrow> In step 2), Gaussian fitting and waveform data point cloudization are as follows: nonlinear least squares method is used to fit the waveform data; then the discrete point cloud is extracted from the processed waveform data through the local maximum peak detection filtering algorithm, and each The energy and amplitude information of the returned signal is recorded in discrete points; In step 2), noise level estimation and data smoothing are as follows: first convert the original data to the frequency domain, and then use the low-value part with higher frequency as the judgment standard of noise level; then use Gaussian filter for smoothing. 2. the method for remote sensing forest biomass inversion according to claim 1, is characterized in that: step 2) in, generating digital topography is: at first extracting discrete point cloud from waveform data is classified, then last return Kraus filtering is performed on the waves to remove non-surface points, and finally the digital terrain model is generated by interpolating the filtered last echo data with the help of natural proximity method. 3. the method for remote sensing forest biomass inversion according to claim 1, is characterized in that: in step 2), utilize DEM to carry out normalization process to the height of vegetation echo point: make the height value conversion of vegetation point is the height value relative to the ground, and cut each sample plot through the coordinates of the lower left corner and upper right corner; finally, extract the normalized point cloud data of the corresponding coordinate positions of 55 sample plots through GIS analysis tools. 4. the method for remote sensing forest biomass inversion according to claim 1, is characterized in that: in step 6), T inspection does not reach significant level and is P value > 0.1, and T inspection reaches significant level and is P value < 0.05 .