CN114091613B - Forest biomass estimation method based on high-score joint networking data - Google Patents
Forest biomass estimation method based on high-score joint networking data Download PDFInfo
- Publication number
- CN114091613B CN114091613B CN202111419535.9A CN202111419535A CN114091613B CN 114091613 B CN114091613 B CN 114091613B CN 202111419535 A CN202111419535 A CN 202111419535A CN 114091613 B CN114091613 B CN 114091613B
- Authority
- CN
- China
- Prior art keywords
- forest
- data
- sample
- biomass
- satellite image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/80—Management or planning
- Y02P90/84—Greenhouse gas [GHG] management systems
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Physics (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a forest biomass estimation method based on high-resolution joint networking data, which belongs to the field of forest resource remote sensing monitoring and comprises the steps of firstly, acquiring high-resolution satellite image sample data and a digital elevation model; extracting terrain data from the digital elevation model; preprocessing high-resolution satellite image sample data; taking sample plot forest stand shifts as units, and extracting characteristic factors corresponding to the sample plot forest stand shifts; measuring and calculating the actually measured biomass of the sample plot forest stand shifts on the spot; carrying out correlation analysis and collinearity inspection on the actually measured biomass and the characteristic factors to obtain screened characteristic factors; establishing a random forest estimation model, and training the random forest estimation model by using the actually measured biomass and the screened characteristic factors; and acquiring high-resolution satellite image data of the area to be measured, inputting the corresponding screened characteristic factors into the trained random forest estimation model, and outputting a forest biomass estimation result, so that the estimation precision of the forest biomass can be improved.
Description
Technical Field
The invention relates to the field of forest resource remote sensing monitoring, in particular to a forest biomass estimation method based on high-resolution joint networking data.
Background
Forest biomass is an ecological term referring to the total amount of organic matter that lives in fact per unit area at a time in a forest. For the needs of economic utilization and scientific research, the aboveground biomass of forest trees and pasture is often investigated and counted, and accordingly, the proportion of biomass of various groups in a sample plot in the total biomass can be judged, and the forest resources can be better understood.
The forest ecosystem is used as the largest carbon sink system on the surface layer of the land, and plays an important role in balancing the ecological environment of the area and the global carbon cycle. The forest vegetation biomass accounts for 80% of the total amount of the land vegetation biomass, is an important index for accounting carbon sink and carbon cycle of a land ecosystem, and has important significance for analyzing the spatial distribution pattern and dynamic change of the carbon reserves of the forest vegetation. And, accurately estimating biomass, quantifying carbon sequestration and carbon flux in the area, clearing carbon inventory in the area, accounting for international carbon reserves, and mitigating rising CO 2 Pollution and the like are particularly important.
However, the existing forest biomass estimation method is poor in precision, and the obtained forest biomass estimation result is low in accuracy and poor in reliability. Therefore, how to improve the estimation accuracy of the forest biomass is an urgent problem to be solved in the current forest ecosystem development and environment sustainable development roads.
Disclosure of Invention
The invention aims to provide a forest biomass estimation method based on high-resolution combined networking data, which can improve the estimation precision of forest biomass and solve the problem of low estimation precision of the existing estimation method.
In order to achieve the purpose, the invention provides the following scheme:
on one hand, the invention provides a forest biomass estimation method based on high-score joint networking data, which comprises the following steps:
acquiring high-resolution satellite image sample data and a digital elevation model of an area to be detected; the high-resolution satellite image sample data comprises high-resolution first satellite image data and high-resolution sixth satellite image data;
extracting terrain data from the digital elevation model; the terrain data comprises DEM data, gradient data and slope data;
preprocessing the high-resolution satellite image sample data to obtain preprocessed high-resolution satellite image sample data;
extracting a wave band characteristic factor corresponding to each sample plot forest sub-class from the preprocessed high-resolution satellite image sample data by taking the sample plot forest sub-class as a unit; the characteristic factors comprise wave band characteristic factors and vegetation index characteristic factors, PCA principal component table transformation and texture characteristic extraction are carried out on the preprocessed images, and principal component analysis characteristic factors and texture characteristic factors are extracted;
sampling forest lands of the area to be measured, which belong to the same time as the high-resolution satellite image sample data, performing field measurement, and calculating the actually measured biomass of the forest stand shifts of the sample plot according to the field measurement data; carrying out correlation analysis and collinearity inspection on the actually measured biomass of the sample plot forest stand shifts and the characteristic factors to obtain the characteristic factors after the sample plot forest stand shifts are screened; establishing a random forest estimation model, and training the random forest estimation model by using the actually measured biomass of the sample plot forest sub-class and the screened characteristic factors to obtain a trained random forest estimation model;
acquiring high-resolution satellite image data of an area to be measured, preprocessing the high-resolution satellite image sample data according to the steps to obtain preprocessed high-resolution satellite image sample data, performing correlation analysis and collinearity inspection on actually measured biomass and characteristic factors of each sample plot forest stand sub-shift to obtain characteristic factors after screening of each sample plot forest stand sub-shift, processing the high-resolution satellite image data, inputting the screened characteristic factors corresponding to the high-resolution satellite image data into the trained random forest estimation model, and obtaining a forest biomass estimation result.
Optionally, the preprocessing the high-resolution satellite image sample data to obtain preprocessed high-resolution satellite image sample data specifically includes:
respectively carrying out radiometric calibration on the high-resolution first satellite image data and the high-resolution sixth satellite image data by using the radiometric calibration coefficient;
respectively carrying out atmospheric correction on the high-grade first satellite image data and the high-grade sixth satellite image data by adopting spectral response functions corresponding to the high-grade first satellite image data and the high-grade sixth satellite image data;
taking the DEM data as terrain correction data, selecting correction points based on Google non-offset images, and performing orthorectification on the top-scoring first satellite image data and the top-scoring sixth satellite image data respectively;
and respectively carrying out image fusion on the high-resolution first satellite image data and the high-resolution sixth satellite image data by adopting a SPEAR Pan Sharpening Pan Sharpening image fusion method.
Optionally, the extracting the feature factor corresponding to each sample plot forest stand shift from the preprocessed high-resolution satellite image sample data specifically includes:
extracting image band information of the preprocessed high-resolution satellite image sample data, and extracting band characteristic factors corresponding to sample plot forest stand shifts by applying Arcgis;
performing band operation on the image band information by adopting a band calculation method to obtain vegetation index characteristic factors corresponding to the small shifts of the forest stand of each sample plot;
adopting a PCA principal component analysis module in an ENVI software toolbox to carry out principal component analysis on the image waveband information to obtain principal component analysis characteristic factors corresponding to each sample plot forest stand shifts;
and extracting texture features in the image information by adopting a second-order gray level co-occurrence matrix module in an ENVI software toolbox to obtain texture feature factors corresponding to each sample plot forest stand shifts.
Optionally, the performing field measurement on the forest land of the sample plot to be measured, where the high-resolution satellite image sample data belongs to the same time, and calculating the actually measured biomass of each sample plot forest stand sub-shift according to the field measurement data and the topographic data specifically includes:
selecting sample points in a region to be measured to carry out actual measurement data acquisition, randomly and uniformly selecting four measurement points in a sample plot forest sub-class at the sample plot point to be measured by adopting an angle rule method, and respectively carrying out sample point distribution at the central position of the sample plot forest sub-class and the four measurement points;
carrying out on-site survey on the tree species at the sampling point by adopting a single-tree measurement mode, and recording tree species composition, dominant tree species, plant number, breast height, tree age, crown width, canopy closure degree, tree age, gradient, slope direction and elevation information of each sample plot in a forest stand class plot to obtain on-site measurement data;
checking each data in the field measurement data, and removing abnormal values in the field measurement data by applying a triple standard deviation principle to obtain the field measurement data after the abnormal values are removed;
and carrying out actual measurement biomass conversion on the field measurement data after the abnormal value is removed by utilizing a unitary volume table to obtain actual measurement biomass.
Optionally, the performing actually measured biomass conversion on the field measurement data from which the abnormal value is removed by using a single-volume table to obtain actually measured biomass specifically includes:
performing accumulation quantity inversion on each tree information in the field measurement data after the abnormal value is removed by adopting a monoblock volume table to obtain single tree accumulation quantity;
summing the single-tree accumulation amounts in all the forest stand shifts of the various fields to obtain the single-tree accumulation amount of the forest stand shifts of the various fields;
and calculating the actually measured biomass by adopting an accumulation-biomass conversion equation according to the small shift accumulation of the forest stand in the sample plot.
Optionally, the correlation analysis and collinearity test are performed on the actually measured biomass of each sample forest stand sub-group and the characteristic factor to obtain the screened characteristic factor, and the method specifically includes:
analyzing the correlation between the actually measured biomass of each sample forest stand sub-group and each characteristic factor by adopting SPSS analysis software to obtain a correlation coefficient of the correlation strength;
selecting characteristic factors which are obviously related at the 0.01 level; sorting the correlation coefficients corresponding to the characteristic factors, setting a correlation coefficient critical value, selecting the characteristic factors of which the correlation coefficients are greater than the correlation coefficient critical value in a sorting queue, and realizing the screening of the characteristic factors;
establishing a multivariate stepwise regression model according to the actually measured biomass and the screened characteristic factors to obtain a multivariate stepwise regression model;
and according to the VIF parameters corresponding to the characteristic factors of the multivariate stepwise regression model in the SPSS analysis software, realizing the co-linearity test of the characteristic factors and obtaining the characteristic factors after the sample plot forest stand shifts are screened.
Optionally, the method for estimating the random forest by using the actually measured biomass of the sample plot forest stand shifts and the screened feature factors includes:
sample division is carried out on the actually measured biomass of the sample plot forest stand shifts and the screened characteristic factors, 70% of the sample plot actually measured biomass and the characteristic factors are used as modeling training samples, and 30% of the sample plot actually measured biomass and the characteristic factors are used as verification samples;
and substituting the modeling training sample into a skleran module in Python for training, setting n _ estimators =475 and random _state =100 after carrying out parameter adjustment for multiple times, and obtaining a trained random forest estimation model.
And substituting the verification sample into the trained random forest estimation model to verify the precision of the trained random forest estimation model.
And extracting the screened characteristic factors of all sample plot forest stand shifts, and substituting the screened characteristic factor values into the trained random forest estimation model to obtain a biomass estimation result.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides a forest biomass estimation method based on high-resolution joint networking data, which is based on high-resolution joint networking data acquired by high-resolution first-order and high-resolution sixth-order satellites and a digital elevation model and provides a high-precision and multi-timeliness data base for forest biomass estimation. Firstly, acquiring high-resolution satellite image sample data of an area to be detected, and extracting topographic data from a digital elevation model; then, sampling data of the high-resolution satellite imagePreprocessing to obtain more accurate and clear high-resolution satellite image sample data; and then, taking the sample plot forest sub-class as a unit, acquiring the actually measured biomass of the sample plot forest sub-class and characteristic factors of all the sample plot forest sub-classes, and performing correlation analysis and co-linearity inspection on the actually measured biomass of the sample plot and the characteristic factors, so that the characteristic factors with weak correlation and strong co-linearity are removed, the preference of the characteristic factors is realized, the estimation precision of a random forest estimation model is improved, and the biomass estimation result is more accurate and reliable. And finally, the screened actually measured biomass and the characteristic factors are used for establishing a random forest estimation model, and the random forest estimation model is far superior to a linear regression model in the prior art in the aspects of estimation precision, generalization capability, overfitting resistance and the like, so that the estimation result R of the trained random forest estimation model is obtained 2 Can reach more than 0.9.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required in the embodiments will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts. The following drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.
Fig. 1 is a flowchart of a forest biomass estimation method based on high-score joint networking data according to embodiment 1 of the present invention;
fig. 2 is a schematic diagram of a forest biomass estimation method based on high-packet combined networking data according to embodiment 1 of the present invention;
fig. 3 is a structural diagram of input and output when the random forest estimation model provided in embodiment 1 of the present invention estimates forest biomass.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
As used in this disclosure and in the claims, the terms "a," "an," "the," and/or "the" are not intended to be inclusive in the singular, but rather are inclusive in the plural, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that steps and elements are included which are explicitly identified, that the steps and elements do not form an exclusive list, and that a method or apparatus may include other steps or elements.
The invention provides a forest biomass estimation method combining random forests and high-score combined networking data composed of a high-score first forest and a high-score sixth forest, which is more suitable for primary secondary forests with low and medium latitudes and higher canopy closure degrees, can improve the estimation precision of forest biomass, and solves the problems of low estimation precision, inaccurate estimation result and unreliability of the existing estimation method.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example 1
As shown in fig. 1 and fig. 2, the embodiment provides a forest biomass estimation method based on high-score joint networking data, which adopts high-score joint networking data based on high-score first-and high-score sixth-numbered satellites and estimates forest biomass in a region to be measured based on a random forest estimation model. The embodiment estimates the forest biomass of the watershed of the Yangtze river and the Huaihe river, and it is easy to understand that the region to be measured is not limited to the watershed region of the Yangtze river and the Huaihe river and other regions are also applicable, and accurate and reliable biomass estimation results can be obtained.
The method specifically comprises the following steps:
s1, acquiring high-resolution satellite image sample data and a digital elevation model of an area to be measured.
The high-resolution satellite image sample data comprises high-resolution first satellite image data and high-resolution sixth satellite image data. The domestic high-resolution one-number satellite image data and the high-resolution six-number satellite image data have the characteristics of high timeliness, high resolution and multispectral intervals, the domestic high-resolution one-number satellite image data and the domestic high-resolution six-number satellite image data are jointly used, the satellite image time resolution is guaranteed, the image spatial resolution reaches 2 meters, the red edge index of the high-resolution six-number satellite is introduced, the vegetation sensitivity is high, and the joint use of the high-resolution data provides a high-precision and multi-timeliness data base for forest biomass estimation. The digital elevation model is a solid ground model which realizes the digital simulation of the ground terrain through limited terrain elevation data, namely the digital expression of the terrain surface form and expresses the ground elevation in a group of ordered numerical value array forms.
In this embodiment, high-resolution satellite image sample data is used as a data basis for constructing a random forest estimation model, primary and secondary forests of the watershed of the river and the water of the Chinese yam are selected from the region to be detected, and high-resolution one-satellite image data and high-resolution six-satellite image data in the same time period are obtained, that is, forest biomass in the watershed region of the river and the Chinese yam is estimated by using two-spectrum satellite data in the same time period.
And S2, extracting terrain data from the digital elevation model. The terrain data includes DEM data (i.e., elevation data), grade data, and slope data, among others.
As shown in fig. 2, in the present embodiment, the terrain data is extracted from the digital elevation model, mainly extracting terrain factors including elevation, slope, gradient, and the like.
And S3, preprocessing the high-resolution satellite image sample data to obtain preprocessed high-resolution satellite image sample data.
In this embodiment, the high-resolution satellite image sample data after preprocessing is obtained by sequentially performing preprocessing such as radiometric calibration, atmospheric correction, orthometric correction, and high-resolution one-image fusion on the high-resolution satellite image sample data. The method specifically comprises the following steps:
and S3.1, respectively carrying out radiometric calibration on the first satellite image data and the sixth satellite image data by using the radiometric calibration coefficient.
And S3.2, respectively carrying out atmospheric correction on the high-resolution one-number satellite image data and the high-resolution six-number satellite image data by adopting spectral response functions corresponding to the high-resolution one-number satellite image data and the high-resolution six-number satellite image data.
And S3.3, taking the DEM data as terrain correction data, selecting correction points based on Google non-offset images, and performing orthorectification on the first satellite image data with high score and the sixth satellite image data with high score respectively to reduce geometric distortion caused by factors such as terrain.
In this embodiment, the terrain correction data is DEM data of 30m, and it is easily understood that 30m used in this embodiment is merely an example, and is not a fixed and unique value, and may be set by itself according to specific situations.
And S3.4, ensuring that the number of registration points is more than 350 while the registration points are uniformly distributed by adopting a SPEAR Pan Sharpening image fusion method, respectively carrying out image fusion on the high-resolution first-order satellite image data and the high-resolution sixth-order satellite image data, and improving the resolution of the fused image data to 2m.
S4, extracting characteristic factors corresponding to sample plot forest sub-shifts from the preprocessed high-resolution satellite image sample data by taking the sample plot forest sub-shifts as units; the characteristic factors comprise a wave band characteristic factor, a vegetation index characteristic factor, a principal component analysis characteristic factor and a texture characteristic factor. The method specifically comprises the following steps:
and S4.1, extracting image wave band information of the preprocessed high-resolution satellite image sample data to obtain wave band characteristic factors corresponding to the small shifts of the forest stand of each sample plot.
And S4.2, performing band operation on the image band information by adopting a band calculation method to obtain vegetation index characteristic factors corresponding to the forest stand shifts of each sample plot.
The calculation process of the band calculation method in the embodiment is as follows:
calculation formula of Ratio Vegetation Index RVI (Ratio Vegetation Index, RVI):
RVI＝ρ NIR /ρ RED
formula for calculating Normalized Vegetation Index NDVI (NDVI):
the Enhanced Vegetation Index (EVI) is calculated according to the following formula:
the calculating formula of the Difference Vegetation Index DVI (DVI) is as follows:
DVI＝ρ NIR -ρ RED
a calculation formula of Soil Adjusted Vegetation Index SAVI (SAVI):
wherein L represents a soil regulation index, and the value range is 0-1; when L is equal to 0, it indicates that vegetation coverage is zero; when L is equal to 1, it means that the vegetation coverage is high and the influence of the soil background is zero. In this example, L is 0.5 at the middle.
Formula for calculating Green Normalized Difference Vegetation Index GNDVI (GNDVI):
calculation formula of Infrared Percentage Vegetation Index IPVI (Infrared Percentage Vegetation Index, IPVI):
IPVI＝ρ NIR /(ρ NIR +ρ RED )
where ρ is BLUE Representing an image blueWave band, i.e. the first wave band of the image, p GREEN Representing the image green band, i.e. the image second band, p RED Representing the red band of the image, i.e. the third band of the image, p NIR The image near infrared band is represented as the fourth band of the image.
Aiming at a high-grade six-number satellite, the invention adds a red edge index, introduces two characteristic indexes of a ground chlorophyll index MTCI and a normalized difference red edge index NDRE1 which are obtained by calculating a red edge I wave band and a red edge II wave band in Sentinel-2 (Sentinel No. 2) data, and the formula is as follows:
MTCI＝(ρ 0.75 -ρ 0.71 )/(ρ 0.71 -ρ RED )
NDRE1＝(ρ 0.75 -ρ 0.71 )/(ρ 0.75 +ρ 0.71 )
where ρ is 0.71 Representing the red edge I band, i.e. the fifth band, rho, of the high-resolution six-number image 0.75 And the red edge II wave band of the high-resolution six-number image is represented, namely a sixth wave band.
The invention utilizes the characteristics of high aging, high resolution and multispectral of domestic high-resolution one-and high-resolution six-satellite data, and introduces the red edge index of the high-resolution six-satellite while ensuring the time resolution of satellite images by using the domestic satellite networking data of the high-resolution one-and high-resolution six-satellite, wherein the red edge index has high sensitivity to vegetation, and the red edge band is closely related to various physical and chemical parameters of the vegetation and is an important indication band for describing the plant pigment state and the health condition.
And S4.3, performing principal component analysis on the image waveband information by adopting an orthogonal principal component analysis module in the ENVI software toolbox to obtain principal component analysis characteristic factors corresponding to each sample plot forest stand shifts.
And S4.4, extracting the texture characteristics of the image by adopting a second-order gray level co-occurrence matrix module in an ENVI software toolbox to obtain texture characteristic factors corresponding to the small forest stand shifts of each sample plot.
And S5, performing field measurement on the forest land of the area to be measured, which belongs to the same time as the high-resolution satellite image sample data, and calculating the actually measured biomass of the forest stand shifts according to the field measurement data.
The process of calculating the measured biomass in this embodiment specifically includes the following steps:
s5.1, selecting actual measurement sample points in an area to be measured, randomly and uniformly selecting four measurement points in a sample plot sub-class of the actual measurement sample points by adopting an angle gauge method, respectively distributing sample plots at the central position of the sample plot sub-class and the four measurement points, and measuring sample points;
and S5.2, measuring longitude and latitude coordinate information of four boundary points of a sample plot by a handheld GPS instrument, surveying the tree species in the sample plot layout on the spot by adopting a single-tree measuring mode, measuring the tree height of each tree by a laser range finder, measuring the breast diameter of each tree by using a tape and a girth ruler, measuring the tree crown width of each tree by an angle gauge, measuring the data of the sample plot gradient, the slope direction and the like by using a compass, a gradient measuring instrument and the like, combining sample plot minor-stand vector data and topographic data in the existing forest land resource survey data, recording the information of the tree species composition, the dominant tree species, the tree height, the crown width, the canopy depth, the forest age, the gradient, the slope direction, the altitude and the like of the sample plot trees, and taking a picture and archiving the measured sample plot. Wherein, the breast height diameter of the tree species of the arbor in the sample plot is 5cm, meanwhile, the number of the tree plants in the sample plot within the small shifts is counted, and the measured data is summarized to obtain the field measured data.
And S5.3, checking each piece of data in the solid measured data, eliminating abnormal values in the solid measured data, and obtaining the solid measured data after the abnormal values are eliminated. And after the on-site measurement data is obtained, the abnormal value in the on-site measurement data is removed by applying a triple standard deviation principle, so that the accuracy of the modeling data is improved, and the accuracy and the reliability of the estimation result are further improved.
And S5.4, performing actual measurement biomass conversion on the field measurement data after the abnormal value is removed by utilizing a unitary volume table to obtain actual measurement biomass. The method specifically comprises the following steps:
and S5.4.1, performing accumulation quantity inversion on each tree information in the field measurement data after the abnormal value is removed by adopting a monoblock volume table to obtain the single-tree accumulation quantity. The unary Volume Table (One-Way Volume Table) is a Table for compiling and determining the Volume of the standing timber according to the breast diameter of the standing timber.
And S5.4.2, summing the single-tree accumulation amounts in all sample plot forest stand shifts to obtain the sample plot forest stand shift accumulation amount.
And S5.4.3, calculating the actually measured biomass by adopting an accumulation-biomass conversion equation according to the small shift accumulation of the forest stand of each sample plot.
Wherein, the accumulation-biomass conversion equation is as follows:
Bio＝aV+b
wherein Bio represents biomass per unit area in t/hm 2 (ii) a V represents the amount of accumulated area per unit area and has a unit of m 3 /hm 2 (ii) a and a and b are conversion parameters. Selecting an estimation model according to different forest stand types, wherein the corresponding a and b parameter values are shown in table 1, and the table 1 lists several common forest stand types:
TABLE 1 transition parameters for different forest stand types
And S6, carrying out correlation analysis and collinearity inspection on the actually measured biomass of the forest stand shifts of each sample plot and the characteristic factors to obtain the characteristic factors after screening of the forest stand shifts of each sample plot.
In brief, the step S6 is a process of screening the feature factors, a multivariate stepwise regression model is established by analyzing the correlation between each feature factor and the actually measured biomass value, a co-linearity test is performed, the feature factors with strong correlation are retained, the feature factors with weak correlation and possibly causing deviation to model estimation are removed, that is, the feature factors with poor influence on the biomass estimation precision are removed, and the abnormal value of the actually measured biomass in the modeling data is removed in the step S5.3, so that the random forest estimation model established and trained according to the actually measured biomass and the feature factors is more accurate and reliable, and the true level of the forest biomass in the region to be measured can be reflected more.
The step S6 comprises the following specific steps:
and S6.1, analyzing the correlation between the actually measured biomass of each forest stand sub-group and each characteristic factor by adopting SPSS analysis software to obtain a plurality of correlation coefficients representing the correlation strength.
And S6.2, selecting the characteristic factors which are obviously related on the 0.01 level (on both sides) according to the correlation coefficients corresponding to the characteristic factors. And sorting the correlation coefficients corresponding to the characteristic factors, setting a correlation coefficient critical value, and selecting the characteristic factors of which the correlation coefficients are larger than the correlation coefficient critical value in a sorting queue to realize the screening of the correlation of the characteristic factors. In this embodiment, the critical value of the correlation coefficient is between 0 and 1, and may be set to other values, and may be set according to the actual situation.
And S6.3, establishing a multivariate stepwise regression model according to the actually measured biomass and the screened characteristic factors to obtain the multivariate stepwise regression model with the optimal fitting result.
And S6.4, according to the VIF parameters corresponding to the characteristic factors in the multivariate stepwise regression model in the SPSS analysis software, the collinearity test of the characteristic factors is realized, wherein the VIF parameters refer to variance expansion coefficients, and the collinearity condition of each characteristic factor in the multivariate stepwise regression model can be obtained according to the VIF parameters of the multivariate stepwise regression model.
According to the method, the characteristic factors with strong correlation and relatively weak collinearity are screened out for modeling and model training through correlation analysis and establishment of multivariate stepwise regression, so that the accuracy of the random forest estimation model is guaranteed, and the accuracy of biomass estimation by applying the random forest model is improved.
And S7, establishing a random forest estimation model, and training the random forest estimation model by using the actually measured biomass of the sample plot forest stand shifts and the screened characteristic factors to obtain the trained random forest estimation model. The method specifically comprises the following steps:
and S7.1, carrying out sample division on the actually measured biomass and the screened characteristic factors of the forest stand shifts of each sample plot, wherein 70% of the actually measured biomass and the characteristic factors are used as modeling training samples, and 30% of the actually measured biomass and the characteristic factors are used as verification samples.
In the embodiment, 70% of samples are selected from the screened characteristic factors and the data of the actual measurement biomass without abnormal values as modeling training samples for establishing and training a random forest estimation model; selecting 30% of samples as verification samples for verifying the estimation accuracy of the trained random forest estimation model, and checking whether the random forest estimation model is overfitting, namely the number ratio of the modeling training samples to the verification samples is 7:3, it is easy to understand that this sample ratio is only a preferred value, and can be set according to the actual situation.
And S7.2, substituting the modeling training sample into a sklern module in Python to generate the random forest estimation model, training the random forest estimation model, and continuously adjusting model parameters during training. Finally, determining the number of decision trees in the sklern module as n _ estimators =475 and random _state =100, wherein the precision of the random forest estimation model is the highest, and the random forest estimation model at the moment is used as a trained random forest estimation model; estimation result R of trained random forest estimation model 2 A rmse value of 10.521 was reached at 0.904.
S7.3, substituting the verification sample into the trained random forest estimation model, verifying the precision of the trained random forest estimation model, and verifying a result R 2 And the RMSE is 9.163, the verification result shows that the random forest estimation model has no phenomenon of overfitting, if the random forest estimation model has no phenomenon of overfitting, the step S8 is executed, otherwise, the model training is continued, and the verification is paid attention to avoid the phenomenon of overfitting of the random forest estimation model for biomass estimation.
S8, acquiring high-resolution satellite image data of an area to be detected, preprocessing the high-resolution satellite image sample data according to the step to obtain preprocessed high-resolution satellite image sample data, performing correlation analysis and collinearity inspection on the actually measured biomass and the characteristic factors of the sample plot forest class to obtain characteristic factors after the sample plot forest class is screened, processing the high-resolution satellite image data, namely, repeating the steps S1 to S6 by taking the high-resolution satellite image data at the moment as a data base to obtain screened characteristic factors corresponding to the high-resolution satellite image data. And finally inputting the screened characteristic factors corresponding to the high-resolution satellite image data into the trained random forest estimation model to obtain a forest biomass estimation result.
As shown in fig. 3, since the correlation and the collinearity between the measured biomass and the feature factor are analyzed before, and the establishment, the training and the verification of the random forest estimation model also use a data set based on the measured biomass and the feature factor, the input of the trained random forest estimation model is the screened feature factor corresponding to the high-score satellite image data, and the output is the estimated forest biomass corresponding to the feature factor, so as to obtain an accurate and reliable forest biomass estimation result.
The invention provides a forest biomass estimation method based on high-resolution combined networking data, which provides a high-precision and multi-timeliness data base for forest biomass estimation based on high-resolution one-number satellite image data, high-resolution six-number satellite image data and a digital elevation model. Firstly, acquiring high-resolution satellite image sample data of an area to be detected, and extracting topographic data from a digital elevation model; then, preprocessing the high-resolution satellite image sample data to obtain more accurate and clear high-resolution satellite image sample data; and then, by taking the sample plot forest stand shifts as units, acquiring the actually measured biomass and the characteristic factors of the sample plot forest stand shifts, and performing correlation analysis and co-linearity test on the actually measured biomass and the characteristic factors, so that the characteristic factors with weak correlation and strong co-linearity are removed, the characteristic factors are screened, the estimation precision of the established random forest estimation model is improved, and the biomass estimation result is more accurate and reliable. And finally, the screened actually-measured biomass and the characteristic factors are used for establishing a random forest estimation model, and the characteristic that the random forest estimation model is far superior to a linear regression model in the aspects of estimation precision, generalization capability, overfitting resistance and the like in the prior art is utilized, so that an accurate and reliable forest biomass estimation result can be output by utilizing the trained random forest estimation model, and the problem of low estimation precision of the conventional estimation method can be solved.
Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The foregoing is illustrative of the present invention and is not to be construed as limiting thereof. Although a few exemplary embodiments of this invention have been described, those skilled in the art will readily appreciate that many modifications are possible in the exemplary embodiments without materially departing from the novel teachings and advantages of this invention. Accordingly, all such modifications are intended to be included within the scope of this invention as defined in the claims. It is to be understood that the foregoing is illustrative of the present invention and is not to be construed as limited to the specific embodiments disclosed, and that modifications to the disclosed embodiments, as well as other embodiments, are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.
Claims (5)
1. A forest biomass estimation method based on high-score combined networking data is characterized by comprising the following steps:
acquiring high-resolution satellite image sample data and a digital elevation model of an area to be detected; the high-resolution satellite image sample data comprises high-resolution first satellite image data and high-resolution sixth satellite image data;
extracting terrain data from the digital elevation model; the terrain data comprises DEM data, gradient data and slope data;
preprocessing the high-resolution satellite image sample data to obtain preprocessed high-resolution satellite image sample data;
extracting characteristic factors corresponding to each sample plot forest sub-class from the preprocessed high-resolution satellite image sample data by taking the sample plot forest sub-class as a unit; the characteristic factors comprise a wave band characteristic factor, a vegetation index characteristic factor, a principal component analysis characteristic factor and a texture characteristic factor;
the method comprises the following steps of performing field measurement on a forest land of an area to be measured, which belongs to the same time as the high-resolution satellite image sample data, and calculating the actual measurement biomass of a forest stand shift of a sample plot according to the field measurement data and the topographic data, and specifically comprises the following steps:
selecting actual measurement sample points in an area to be measured, randomly and uniformly selecting four measurement points in each sample plot forest sub-class at the actual measurement sample points by adopting an angle rule method, and respectively distributing sample points at the central position of the sample plot forest sub-class and the four measurement points;
carrying out field survey on sample plots with different forest ages and different tree species at the distributed sample points by adopting a single-tree measurement mode, and recording tree species composition, dominant tree species, tree height, tree trunk number, crown width, canopy density, forest age, gradient, slope direction and elevation information of each tree by combining the topographic data to obtain field measurement data;
checking each piece of data in the field measurement data, and eliminating abnormal values in the field measurement data to obtain field measurement data with the abnormal values eliminated;
performing accumulation-biomass conversion on the field measurement data after the abnormal value is removed by using a unitary volume table to obtain actually measured biomass;
carrying out correlation analysis and collinearity test on the actually measured biomass of the sample plot forest stand shifts and the characteristic factors to obtain the characteristic factors after the sample plot forest stand shifts are screened, and specifically comprising the following steps:
analyzing the correlation between the actually measured biomass of the forest stand shifts and each characteristic factor by adopting SPSS analysis software to obtain a plurality of correlation coefficients representing the correlation strength;
selecting feature factors which are obviously related on the level of 0.01, sorting the correlation coefficients corresponding to the feature factors, setting a correlation coefficient critical value, selecting the feature factors of which the correlation coefficients are greater than the correlation coefficient critical value in a sorting queue, and realizing the screening of the feature factors;
establishing a multivariate stepwise regression model according to the actually measured biomass and the screened characteristic factors;
according to the VIF parameters corresponding to the multiple stepwise regression model in the SPSS analysis software, carrying out co-linearity inspection on each characteristic factor in the multiple stepwise regression model, realizing co-linearity screening of the characteristic factors, and obtaining the characteristic factors after small screening of forest stands of each sample plot;
establishing a random forest estimation model, and training the random forest estimation model by using the actually measured biomass of the sample plot forest stand shifts and the screened characteristic factors to obtain a trained random forest estimation model;
acquiring high-resolution satellite image data of a region to be measured, preprocessing the high-resolution satellite image sample data according to the steps to obtain preprocessed high-resolution satellite image sample data, performing correlation analysis and collinearity inspection on actually measured biomass and characteristic factors of sample plot forest stand shifts to obtain characteristic factors after sample plot forest stand shifts are screened, processing the high-resolution satellite image data, inputting the screened characteristic factors corresponding to the high-resolution satellite image data into the trained random forest estimation model, and obtaining a forest biomass estimation result.
2. The forest biomass estimation method based on high-score joint networking data according to claim 1, wherein the preprocessing is performed on the high-score satellite image sample data to obtain preprocessed high-score satellite image sample data, and specifically comprises:
respectively carrying out radiometric calibration on the first satellite image data and the sixth satellite image data by using radiometric calibration coefficients;
respectively carrying out atmospheric correction on the high-grade first satellite image data and the high-grade sixth satellite image data by adopting spectral response functions corresponding to the high-grade first satellite image data and the high-grade sixth satellite image data;
taking the DEM data as terrain correction data, selecting correction points based on Google non-offset images, and performing orthorectification on the top-scoring first satellite image data and the top-scoring sixth satellite image data respectively;
and respectively carrying out image fusion on the high-resolution first satellite image data and the high-resolution sixth satellite image data by adopting a SPEAR Pan Sharpening Pan Sharpening image fusion method.
3. The forest biomass estimation method based on the high-resolution joint networking data according to claim 1, wherein the extracting of the feature factors corresponding to each sample plot forest stand shift from the preprocessed high-resolution satellite image sample data specifically comprises:
extracting image band information of the preprocessed high-resolution satellite image sample data to obtain band characteristic factors corresponding to the small forest stand shifts of each sample plot;
performing band operation on the image band information by adopting a band calculation method to obtain vegetation index characteristic factors corresponding to the small shifts of the forest stand of each sample plot;
performing principal component analysis on the image waveband information by adopting an orthogonal principal component analysis module in an ENVI software toolbox to obtain principal component analysis characteristic factors corresponding to each sample plot forest stand shifts;
and extracting texture features in the image band information by adopting a second-order gray level co-occurrence matrix module in an ENVI software toolbox to obtain texture feature factors corresponding to each sample plot forest stand shifts.
4. The forest biomass estimation method based on the high-score joint networking data as claimed in claim 1, wherein the accumulating-biomass conversion is performed on the field measurement data after the abnormal values are removed by using a single volume table to obtain measured biomass, and specifically comprises:
performing accumulation quantity inversion on each tree information in the field measurement data after the abnormal value is removed by adopting a monoblock volume table to obtain single tree accumulation quantity;
summing all single-tree accumulation amounts in each sample plot forest stand shift to obtain sample plot forest stand shift accumulation amounts;
and calculating the actually measured biomass by adopting an accumulation-biomass conversion equation according to the small shift accumulation of the forest stand of each sample plot.
5. The forest biomass estimation method based on the high-score joint networking data as claimed in claim 1, wherein the random forest estimation model is established by using the actually measured biomass of each sample plot forest stand class and the characteristic factors after screening, and is trained to obtain the trained random forest estimation model, and the method specifically comprises the following steps:
carrying out sample division on the actually measured biomass and the screened characteristic factors of all sample plot forest stand shifts, taking 70% of the actually measured biomass and the characteristic factors as modeling training samples, and taking 30% of the actually measured biomass and the characteristic factors as verification samples;
substituting the modeling training sample into a sklern module in Python to generate the random forest estimation model, training the random forest estimation model, and continuously adjusting model parameters during training to obtain a trained random forest estimation model;
and substituting the verification sample into the trained random forest estimation model to verify the precision of the trained random forest estimation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111419535.9A CN114091613B (en) | 2021-11-26 | 2021-11-26 | Forest biomass estimation method based on high-score joint networking data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111419535.9A CN114091613B (en) | 2021-11-26 | 2021-11-26 | Forest biomass estimation method based on high-score joint networking data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114091613A CN114091613A (en) | 2022-02-25 |
| CN114091613B true CN114091613B (en) | 2023-03-24 |
Family
ID=80304898
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111419535.9A Active CN114091613B (en) | 2021-11-26 | 2021-11-26 | Forest biomass estimation method based on high-score joint networking data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114091613B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114819737B (en) * | 2022-05-26 | 2023-10-17 | 中交第二公路勘察设计研究院有限公司 | Method, system and storage medium for estimating carbon reserves of highway road vegetation |
| CN114937029B (en) * | 2022-06-21 | 2023-01-31 | 西南林业大学 | Forest carbon storage amount sampling estimation method, device, equipment and storage medium |
| CN115115948B (en) * | 2022-07-26 | 2024-03-29 | 云南大学 | Forest land information refined extraction method based on random forest and auxiliary factors |
| CN116433748B (en) * | 2023-06-14 | 2023-08-22 | 南开大学 | Forest land multisource data fusion forest carbon reserve determination method and system |
| CN116935238B (en) * | 2023-07-07 | 2024-02-23 | 滁州学院 | Forest disturbance monitoring method, system, equipment and medium based on deep learning |
| CN117423011A (en) * | 2023-11-09 | 2024-01-19 | 滁州学院 | Forest carbon reserve remote sensing estimation method, system, equipment and medium |
| CN117911896A (en) * | 2024-02-19 | 2024-04-19 | 西北师范大学 | Grassland carbon sink monitoring method based on satellite remote sensing |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105913017A (en) * | 2016-04-08 | 2016-08-31 | 南京林业大学 | Corresponding period double high resolution remote sensing image-based forest biomass estimation method |
| CN109459392B (en) * | 2018-11-06 | 2019-06-14 | 南京农业大学 | A kind of rice the upperground part biomass estimating and measuring method based on unmanned plane multispectral image |
| CN110162872B (en) * | 2019-05-17 | 2022-11-29 | 中国科学院城市环境研究所 | Biomass estimation method fusing sample plot data and forest resource clearing data |
| CN112434617B (en) * | 2020-11-26 | 2021-08-13 | 南京观微空间科技有限公司 | Forest biomass change monitoring method and system based on multi-source remote sensing data |
-
2021
- 2021-11-26 CN CN202111419535.9A patent/CN114091613B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN114091613A (en) | 2022-02-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114091613B (en) | Forest biomass estimation method based on high-score joint networking data | |
| Castillo et al. | Comparing the accuracy of several field methods for measuring gully erosion | |
| Sun et al. | Forest vertical structure from GLAS: An evaluation using LVIS and SRTM data | |
| Simard et al. | Mapping forest canopy height globally with spaceborne lidar | |
| Morsdorf et al. | Assessment of the influence of flying altitude and scan angle on biophysical vegetation products derived from airborne laser scanning | |
| Nelson et al. | Estimating Quebec provincial forest resources using ICESat/GLAS | |
| CN107389036A (en) | A kind of large spatial scale vegetation coverage computational methods of combination unmanned plane image | |
| CN105913017A (en) | Corresponding period double high resolution remote sensing image-based forest biomass estimation method | |
| CN115561181B (en) | Water quality inversion method based on unmanned aerial vehicle multispectral data | |
| CN114241331B (en) | Remote sensing modeling method for ground biomass of reed in wetland by taking UAV as ground and Septinel-2 medium | |
| CN113408111B (en) | Atmospheric precipitation inversion method and system, electronic equipment and storage medium | |
| Seier et al. | UAV and TLS for monitoring a creek in an alpine environment, Styria, Austria | |
| CN109946714A (en) | A kind of method for building up of the forest biomass model based on LiDAR and ALOS PALSAR multivariate data | |
| Koma et al. | Quantifying 3D vegetation structure in wetlands using differently measured airborne laser scanning data | |
| CN111144350B (en) | Remote sensing image positioning accuracy evaluation method based on reference base map | |
| CN113466143A (en) | Soil nutrient inversion method, device, equipment and medium | |
| Tassinari et al. | Wide-area spatial analysis: A first methodological contribution for the study of changes in the rural built environment | |
| Xi et al. | Quantifying understory vegetation density using multi-temporal Sentinel-2 and GEDI LiDAR data | |
| Drolon et al. | Monitoring of seasonal glacier mass balance over the European Alps using low-resolution optical satellite images | |
| Bohlin et al. | Deciduous forest mapping using change detection of multi-temporal canopy height models from aerial images acquired at leaf-on and leaf-off conditions | |
| CN113901348A (en) | Oncomelania snail distribution influence factor identification and prediction method based on mathematical model | |
| Maan et al. | Tree species biomass and carbon stock measurement using ground based-LiDAR | |
| Agarwal et al. | Development of machine learning based approach for computing optimal vegetation index with the use of sentinel-2 and drone data | |
| Weifeng et al. | Multi-source DEM accuracy evaluation based on ICESat-2 in Qinghai-Tibet Plateau, China | |
| Smith et al. | Forest canopy structural properties |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |