A method for estimating regional forest age
By constructing age-tree models related to tree species and slope direction, combining remote sensing technology and machine learning, the problem of insufficient accuracy in forest age estimation is solved, and higher spatial resolution and accuracy are achieved, suitable for complex terrain and multiple Forest age monitoring and estimation of species-tree areas.
Patent Information
- Application Number
- CN202410042055.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-11
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-01-11
AI Technical Summary
The existing technology has insufficient accuracy in forest age estimation, especially in natural forests with complex terrain and dense growth, and is difficult to observe in real time, which ignores the spatial heterogeneity and biomass allocation differences within the stands, which affects the Estimation accuracy of key functional parameters of ecosystems.
By constructing the age-tree model of the tree species and slope direction of the cell, the average age of trees of each cell is estimated, and combining remote sensing technology and machine learning models, the non-perturbation and perturbation forest areas are distinguished, and the spatial resolution of forest age distribution is improved. and estimation accuracy.
Achieve higher forest age estimation accuracy, is suitable for a variety of tree areas, meets the biomass and carbon storage estimation needs, can more accurately monitor and estimate the forest age distribution in the composite area of the mining area, and improves the accuracy of ecological environment quality assessment. .
Smart Images

Figure CN117975263B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of forestry, in particular to a method for estimating the age of regional forests. Background Art
[0002] Forest age is one of the important factors that determine the distribution of biomass and productivity of forest ecosystems, and is also an important reference indicator for forest management and administration. In terms of forest carbon sink estimation, there are significant differences in carbon density, carbon storage, and biomass of stands of different ages. The accuracy of forest age estimation is one of the key factors to improve the accuracy of forest carbon storage and carbon sink estimation.
[0003] Currently, forest inventory data are mostly generated through field surveys, such as growth cone sampling, afforestation records, etc., to conduct spatial statistics of a large number of forests. Field surveys require a lot of manpower and material resources, and are difficult to survey for natural forests with complex terrain and dense growth. In addition, forest inventories conducted every five years are not sufficient to provide real-time forest age observation data. The forest age information contained in forest inventory data is usually divided into age groups based on spatial distribution in small classes, ignoring the spatial heterogeneity of forest age within the forest stand and the differences in biomass distribution related to forest age, which to some extent affects the estimation accuracy of key functional parameters of forest ecosystems such as biomass, carbon storage, and productivity. Summary of the invention
[0004] In view of this, the main purpose of the present invention is to provide a method for estimating regional forest age, by determining the age-tree height model according to the tree species and slope aspect of the pixel to estimate the average age of trees in each pixel, thereby obtaining the forest age distribution of the area to be estimated, and improving the estimation accuracy of the regional forest age.
[0005] To achieve the above objectives, this application provides a regional forest age estimation method, including:
[0006] Obtain the image of the forest area to be estimated, and perform the following steps on each pixel of the image of the forest area to be estimated to obtain the forest age distribution of the area to be estimated:
[0007] From the age-tree height model set, find the age-tree height model of the tree species in the pixel at the slope of the pixel; substitute the tree height of the pixel into the found age-tree height model to obtain the average age of the trees in the pixel;
[0008] Among them, the construction of the age-tree height model set is: for any tree species, take the parameters of the trees growing on the shady slope, and construct the age-tree height model of the tree species on the shady slope; take the parameters of the trees growing on the sunny slope, and construct the age-tree height model of the tree species on the sunny slope.
[0009] In a possible implementation, the forest area image to be estimated is an image of a non-disturbed forest area, and the method further includes: estimating the forest age distribution of the disturbed forest area, which is:
[0010] Determine the pixels of disturbed forest area;
[0011] Execute for each disturbed forest area pixel separately:
[0012] Get the average initial age of trees in this pixel;
[0013] The average age of trees in a pixel is calculated by subtracting the recovery year of the disturbed pixel from the current year and adding the average initial age of trees in the pixel.
[0014] The average initial age of trees in the pixel refers to the average age of trees in the pixel in the recovery year of the disturbed pixel.
[0015] In another possible implementation, the disturbed forest area pixels are determined as follows:
[0016] The plant change tracking VCT method is used to track each pixel of the long-term multispectral satellite remote sensing image of the area to be estimated, and the comprehensive forest index IFZ time series of each pixel is obtained; if the IFZ value increases by more than the threshold compared with the previous year, the pixel corresponding to the IFZ time series is the disturbed forest area pixel; for the disturbed forest area pixel, the IFZ value decreases by more than the threshold compared with the previous year, and the corresponding year is the recovery year of the disturbed pixel; or,
[0017] The normalized burn rate index NBR of each pixel in the long-term multispectral satellite remote sensing image of the estimated area in each year was calculated respectively to obtain the NBR time series of each pixel. Among them, the pixel corresponding to the NBR time series with an inflection point of NBR decrease is the pixel in the disturbed forest area; for the pixel in the disturbed forest area, the year corresponding to the inflection point of NBR increase is the recovery year of the disturbed pixel.
[0018] In another possible implementation, the method further includes: spatially superimposing the forest age distribution of the undisturbed forest area and the forest age distribution of the disturbed forest area to obtain the forest age distribution of the area to be estimated.
[0019] In another possible implementation, the age-tree height model for any tree species is constructed as follows:
[0020] The average aboveground biomass, average tree height and average age of trees in various plots with the tree species as the dominant tree species on the shady slope are obtained respectively; the average aboveground biomass and average age are fitted to obtain the age-aboveground biomass function of the tree species on the shady slope; the average aboveground biomass and average tree height are input into the power function regression model to obtain the aboveground biomass-tree height model of the tree species on the shady slope; the aboveground biomass-tree height model of the tree species on the shady slope is substituted into its age-aboveground biomass function on the shady slope to obtain the age-tree height model of the tree species on the shady slope;
[0021] The average aboveground biomass, average tree height and average age of trees in various plots with the tree species as the dominant tree species on the sunny slope are obtained respectively; the obtained average aboveground biomass and average age are fitted to obtain the age-aboveground biomass function of the tree species on the sunny slope; the obtained average aboveground biomass and average tree height are input into the power function regression model to obtain the aboveground biomass-tree height model of the tree species on the sunny slope; the aboveground biomass-tree height model of the tree species on the sunny slope is substituted into its age-aboveground biomass function on the sunny slope to obtain the age-tree height model of the tree species on the sunny slope;
[0022] The sample plot is a forest land of a set size with only a single dominant tree species within its range and with only shady or sunny slopes.
[0023] In another possible implementation, the method of obtaining the average aboveground biomass, average tree height, and average age of trees in any plot is:
[0024] Use backpack / handheld laser radar to measure the diameter at breast height and height of each tree in the plot; use tree growth cones to sample the number of annual rings of each tree core in the plot as its age;
[0025] For any tree, substitute its DBH and tree height into the allometric growth equation corresponding to the dominant tree species in the plot to calculate the aboveground biomass of the tree;
[0026] The means of the aboveground biomass, tree height and age of the trees in the sample plot were calculated respectively to obtain the average aboveground biomass, average tree height and average age of the trees in the sample plot.
[0027] In another possible implementation, the tree height of each pixel of the forest area image to be estimated is determined as follows:
[0028] Construct the training set, recorded as:
[0029] D={(x i ,y i )}i=1,2,…,n
[0030] Wherein, D is a training set, and the independent variable sample data of D includes at least one of the multispectral image data, SAR image data and DEM data of the forest area; the dependent variable sample data is the RH95% / 98% band data of the satellite-borne lidar data of the forest area;
[0031] Use D to train a random forest regression model to obtain a tree height prediction model, or use D to train a gradient boosting regression tree model to obtain a tree height prediction model;
[0032] The pixel values of the image of the forest area to be estimated are respectively input into the tree height prediction model to obtain the tree height data of each pixel of the image of the forest area to be estimated.
[0033] In another possible implementation, the tree species of each pixel in the image of the forest area to be estimated is determined as follows:
[0034] Construct a sample set of the relationship between remote sensing index and tree species; use the constructed sample set to train a random forest model to obtain a tree species prediction model;
[0035] Extract the remote sensing index of each pixel of the image of the forest area to be estimated; input the remote sensing index of each pixel into the tree species prediction model to obtain the tree species of each pixel;
[0036] Among them, the independent variable data of any sample is at least one of the NDVI, EVI, spectral band information, and texture characteristics of the multispectral image data corresponding to a sample plot; the dependent variable data of the sample is: the dominant tree species of the sample plot.
[0037] In another possible implementation, the slope direction of each pixel of the image of the forest area to be estimated is also determined as follows:
[0038] Obtain the DEM data corresponding to the geographic coordinates of each pixel in the image of the forest area to be estimated, and determine the angle between the projection of the slope normal of each pixel on the horizontal plane and the clockwise direction of the true north direction through the obtained DEM data; determine the slope direction of the corresponding pixel based on the determined angles.
[0039] In another possible implementation, the step of obtaining the image of the forest area to be estimated is:
[0040] Using the forest distribution data in the land use data of the area to be estimated as a mask, masking at least one of the multispectral image, SAR image and DEM data of the area to be estimated to obtain an image of the forest area to be estimated; or,
[0041] Forest sample points are selected based on high-resolution images, and the random forest machine learning classification method is used to obtain images of the forest area to be estimated.
[0042] Based on the above, the regional forest age estimation method provided by the present invention also has the following advantages and characteristics:
[0043] 1. Using remote sensing image pixels as estimation units improves the spatial resolution of forest age distribution, making it suitable for estimating forest age in areas with multiple tree species, and meeting the data needs of more research, forest management, biomass and carbon stock estimation, and other work practices.
[0044] 2. Distinguishing between the estimation of forest age in undisturbed forest areas and disturbed forest areas can more accurately estimate the forest age distribution in mining-forest complex areas where the ecological environment quality is disturbed by mining, solving the current problem of the mixing of undisturbed forest areas and disturbed forest areas in mining areas, making forest age monitoring and estimation difficult.
[0045] 3. For the estimation of forest age in disturbed forest areas, the average initial age of trees is taken into account, which increases the accuracy of forest age estimation in disturbed forest areas. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 A schematic diagram of a flow chart of a method for estimating regional forest age according to an embodiment of the present invention;
[0047] Figure 2 This is a schematic diagram of the corresponding relationship between the angle and the slope direction in an embodiment of the present invention;
[0048] Figure 3 This is a schematic diagram of the NBR time series of a pixel in an embodiment of the present invention;
[0049] Figure 4 This is a schematic diagram of an image of a forest area to be estimated in a mining area in the middle reaches of the Yellow River according to an embodiment of the present invention;
[0050] Figure 5 This is a schematic diagram of RH95% band data of a satellite-borne laser radar in a mining area in the middle reaches of the Yellow River according to an embodiment of the present invention;
[0051] Figure 6 This is a schematic diagram of tree heights of each pixel in an image of a forest area in a mining area in the middle reaches of the Yellow River according to an embodiment of the present invention;
[0052] Figure 7 This is a schematic diagram of DEM data of various tree species in a mining area in the middle reaches of the Yellow River according to an embodiment of the present invention;
[0053] Figure 8 Based on Figure 7 Schematic diagram of the slope aspect of a forest area in a mining area in the middle reaches of the Yellow River using DEM data;
[0054] Fig. 9 This is a schematic diagram of forest age distribution in an undisturbed forest area in a mining area in the middle reaches of the Yellow River according to an embodiment of the present invention;
[0055] Fig.10 The annual average NBR index curve of the images of a mining area in the middle reaches of the Yellow River in the embodiment of the present invention;
[0056] Fig.11 This is a schematic diagram of forest age distribution in a disturbed area of a mining area in the middle reaches of the Yellow River in 2023 according to an embodiment of the present invention. DETAILED DESCRIPTION
[0057] Specifically, the process of a regional forest age estimation method according to an embodiment of the present invention is as follows: Figure 1 As shown, steps 101 to 104 are included.
[0058] Step 101: Acquire an image of the forest area to be estimated;
[0059] Step 102: Execute steps 103 and 104 for each pixel of the image of the forest area to be estimated, and obtain the forest age distribution of the area to be estimated.
[0060] Step 103: From the age-tree height model set, find the age-tree height model of the tree species in the pixel and the slope direction of the pixel.
[0061] Step 104: Substitute the tree height of the pixel into the found age-tree height model to obtain the average age of the trees in the pixel.
[0062] Among them, the construction of the age-tree height model set is: for any tree species, take the parameters of the trees growing on the shady slope, and construct the age-tree height model of the tree species on the shady slope; take the parameters of the trees growing on the sunny slope, and construct the age-tree height model of the tree species on the sunny slope.
[0063] Here, in step 101, the image of the forest area to be estimated is obtained as follows:
[0064] Using forest distribution data in the land use data of the area to be estimated as a mask, masking at least one of the multispectral image of the area to be estimated, obtaining synthetic aperture radar (SAR) image and obtaining digital elevation model (DEM, Digital Elevation Model) data, to obtain an image of the forest area to be estimated; or,
[0065] Forest sample points are selected based on high-resolution images, and the random forest machine learning classification method is used to obtain images of the forest area to be estimated.
[0066] Among them, multispectral image data include: long-term multispectral satellite remote sensing image data and / or drone aerial image data equipped with multispectral sensors.
[0067] To obtain the long-term multispectral satellite remote sensing images, you can choose NIR bands, SWIR1 bands, and SWIR2 bands of Landsat TM / ETM+ / OLI images, L1C-level multispectral data (MSI) of Sentinel-2, etc. Preprocessing the acquired images includes: time and cloud screening, radiation correction, geometric correction, band synthesis, cloud and shadow masks, and image clipping, etc., which can be achieved by selecting preprocessing platforms such as Google Earth Engine platform, remote sensing image processing (ENVI, The Environment for Visualizing Images) software, and Sen2Cor software.
[0068] Acquire the aerial images of the UAV equipped with the multispectral sensor; preprocessing includes: image registration, radiation correction, geometric correction, image stitching and cropping, etc.
[0069] To obtain the SAR image, Sentinel-1 dual-polarization SAR image can be selected, for example, VV-dual-polarization SAR image for vertical transmission and vertical reception, or VH-dual-polarization SAR image data for vertical transmission and horizontal reception; Google Earth Engine platform can be used to obtain and pre-process dual-polarization SAR image, or Generic Mapping Tools for Synthetic Aperture Radar (GMTSAR), Sentinel Application Platform (SNAP), etc. can be used to pre-process dual-polarization SAR image.
[0070] The relative height (RH) band data of the satellite-borne lidar data can be the ATL08 data of the ICESAT-2 (Ice, Cloud and Land Elevation Satellite-2) laser satellite, or the L2A data of the Global Ecosystem Dynamics Investigation (GEDI), including the spot scale canopy height and profile indicators.
[0071] After step 102, the tree species of each pixel in the forest area image to be estimated is determined as follows:
[0072] Construct a sample set of the relationship between remote sensing index and tree species; use the constructed sample set to train a random forest model to obtain a tree species prediction model;
[0073] Extract the remote sensing index of each pixel of the image of the forest area to be estimated; input the remote sensing index of each pixel into the tree species prediction model to obtain the tree species of each pixel;
[0074] The independent variable data of any sample is at least one of the normalized vegetation index (NDVI), enhanced vegetation index (EVI), spectral band information, and texture characteristics of the multispectral image data corresponding to a sample plot; the dependent variable data of the sample is the dominant tree species of the sample plot. 70% of the sample set is used as training data, and 30% is used as the basis for the accuracy evaluation of the training model.
[0075] Here, the remote sensing index needs to be normalized.
[0076] After step 102, the slope direction of each pixel of the forest area image to be estimated is determined as follows:
[0077] Obtain the DEM data corresponding to the geographic coordinates of each pixel in the image of the forest area to be estimated, and determine the angle between the projection of the slope normal of each pixel on the horizontal plane and the clockwise direction of the true north direction through the obtained DEM data; determine the slope direction of the corresponding pixel based on the determined angles.
[0078] like Figure 2 As shown, the slope direction of the corresponding pixel is determined based on any angle, specifically:
[0079] Taking the north direction as the starting point 0, 9 areas of 0-1-2-3-4-5-6-7-8-0 are divided along the north→east→south→west→north directions at intervals of 40°; if the angle is in the 0-1-2 or 7-8-0 area, the corresponding pixel is a shady slope; if the angle is in the 2-3, 6-7 or 3-4-5-6 area, the corresponding pixel is a sunny slope.
[0080] Here, determining the angle and judging the slope direction of the corresponding pixel can be achieved in ArcGIS.
[0081] After step 102, the tree height of each pixel of the forest area image to be estimated is determined as follows:
[0082] Construct the training set, recorded as:
[0083] D={(x i ,y i )}i=1,2,…,n
[0084] Wherein, D is a training set, and the independent variable sample data of D includes at least one of the multispectral image data, SAR image data and DEM data of the forest area; the dependent variable sample data is the relative height metrics at 95% / 98% (RH95% / 98%) band data of the satellite-borne laser radar data of the forest area;
[0085] Use D to train a random forest regression model to obtain a tree height prediction model, or use D to train a gradient boosting regression tree model to obtain a tree height prediction model;
[0086] The pixel values of the image of the forest area to be estimated are respectively input into the tree height prediction model to obtain the tree height data of each pixel of the image of the forest area to be estimated.
[0087] Here, the tree height prediction model obtained by using the D gradient boosting regression tree model is specifically:
[0088] The loss function is calculated using the negative binomial log-likelihood function, and the expression is as follows:
[0089] L(y,f(x))=log(1+exp(-2y·f(x)));
[0090] Among them, y is the sample data of the dependent variable, and f(x) represents the estimated value of tree height;
[0091] Initialize the weak learner and calculate the initial leaf node output parameters that minimize the loss function; the calculation method for initializing the weak learner is:
[0092]
[0093] N is the number of samples of input data; c is the output parameter of the initial leaf node; L(y i ,c) represents the loss function obtained by training with the i-th sample;
[0094] Based on the negative gradient of the loss function, the residual approximation of the next regression tree fitting uses the current model value to iteratively update the model;
[0095] For each sample (x i ,y i ), the residual is calculated by gradient descent method as:
[0096]
[0097] Among them, r m,i represents the residual of the i-th sample in the m-th regression tree; f m-1 (x i ) represents the learner obtained by training the m-1th regression tree, that is, when the independent variable is xi When calculating the residual of the ith sample of the mth regression tree, f is used. m-1 (x i ) instead of f(x i );
[0098] Using the calculated residual r of the i-th sample in the m-th regression tree m,i , we get the set {(x i ,r m,i )}, i = 1, 2, ..., N. Thus, the mth regression tree T is trained m , the partition area of its leaf node is recorded as R m,j , j=1,2,...,J.
[0099] Solve the leaf node output value, regression tree T m The calculation method of each leaf node is:
[0100]
[0101] Among them, c m,j Calculate the leaf node output value in the jth feature unit of the mth regression tree, that is, the estimated value of the tree height under the feature unit.
[0102] The regression tree T m The leaf node output value is updated, and the shrinkage parameter v is introduced as the learning rate to perform gradient improvement:
[0103]
[0104] Divide the feature space into J units {R 1 ,R 2 ,...,R J}, R m,j represents the jth feature unit of the mth regression tree, each feature unit represents a classification category; I(x∈R m,j ) is the indicator function, x is the independent variable, i.e., multispectral image data, SAR image data, and DEM data. m Determine x∈R m,j When , it means that the independent variable belongs to R m,j In the unit, the I value is 1 at this time, otherwise it is 0;
[0105] According to the process of training the mth regression tree, it is continuously iterated until M iterations have been made, and the final gradient boosting regression tree model of M regression trees superimposed is obtained as the tree height prediction model, and the formula is expressed as follows:
[0106]
[0107] in, Represents the final gradient boosted regression tree, i.e., the tree height prediction model.
[0108] In a possible implementation, the forest area image to be estimated is an image of a non-disturbed forest area. After step 104, the method further includes: estimating the forest age distribution of the disturbed forest area, which is:
[0109] Determine the pixels of disturbed forest area;
[0110] Execute for each disturbed forest area pixel separately:
[0111] Get the average initial age of trees in this pixel;
[0112] The average age of trees in a pixel is calculated by subtracting the recovery year of the disturbed pixel from the current year and adding the average initial age of trees in the pixel.
[0113] The formula for calculating the average age of trees in the disturbed forest area pixels is:
[0114] A=Y now -Y recover +a plant
[0115] Among them, A is the forest age of the pixel in the disturbed forest area; Year now The current year; Year recover is the year of recovery; plant is the average initial age of trees in the pixel.
[0116] The average initial age of trees in the pixel refers to the average age of trees in the pixel in the year of restoration of the disturbed pixel. The initial age of trees is determined by investigation, interview, data collection, or by judging the height of trees when planted. For example, the data records that in the year of restoration of the disturbed pixel, several 1-year-old saplings were planted on the ground projection of the pixel, that is, in the year of restoration of the disturbed pixel, the average age of trees in the pixel is 1, and the average initial age of trees in the pixel is 1.
[0117] In another possible implementation, the disturbed forest area pixels are determined as follows:
[0118] The Vegetation Change Tracker (VCT) method is used to track each pixel of the long-term multispectral satellite remote sensing image of the estimated area, and the IFZ time series of each pixel is obtained; if the IFZ value is increased by more than the threshold compared with the previous year, the pixel corresponding to the IFZ time series is the disturbed forest area pixel; for the disturbed forest area pixel, the IFZ value that is reduced by more than the threshold compared with the previous year is the corresponding year of the recovery of the disturbed pixel. It can be implemented on ArcGIS or GEE platform.
[0119] Here, the set of pixels in the disturbed forest area is the disturbed forest area.
[0120] In another possible implementation, the disturbed forest area pixels are determined as follows:
[0121] like Figure 3 As shown in the figure, the normalized burn rate index NBR of each pixel in the long-term multispectral satellite remote sensing image of the estimated area in each year is calculated respectively to obtain the NBR time series of each pixel. Among them, the pixel corresponding to the NBR time series with an inflection point where the NBR decreases is the pixel in the disturbed forest area; for the pixel in the disturbed forest area, the year corresponding to the inflection point where the NBR increases is the recovery year of the disturbed pixel.
[0122] Here, before calculating the NBR of each pixel of the long-term multispectral satellite remote sensing image of each year, it is necessary to execute:
[0123] For any year, the mean synthesis algorithm is used to calculate the annual average data of the long-term remote sensing image series of that year, and the annual average data of the long-term remote sensing image series of that year is used to calculate the NBR of each pixel.
[0124] The long-term multispectral satellite remote sensing image data of the area to be estimated from 1985 to present is used.
[0125] In another possible implementation, the forest age distribution of the undisturbed forest area and the forest age distribution of the disturbed forest area are spatially superimposed to obtain the forest age distribution of the area to be estimated.
[0126] Here, spatial overlay can be implemented using the spatial overlay function (Union) of the Geospatial Data Abstraction Library (GDAL).
[0127] In another possible implementation, the age-tree height model for any tree species is constructed as follows:
[0128] The average aboveground biomass, average tree height and average age of trees in various plots with the tree species as the dominant tree species on the shady slope are obtained respectively; the average aboveground biomass and average age are fitted to obtain the age-aboveground biomass function of the tree species on the shady slope; the average aboveground biomass and average tree height are input into the power function regression model to obtain the aboveground biomass-tree height model of the tree species on the shady slope; the aboveground biomass-tree height model of the tree species on the shady slope is substituted into its age-aboveground biomass function on the shady slope to obtain the age-tree height model of the tree species on the shady slope;
[0129] The average aboveground biomass, average tree height and average age of trees in various plots with the tree species as the dominant tree species on the sunny slope are obtained respectively; the obtained average aboveground biomass and average age are fitted to obtain the age-aboveground biomass function of the tree species on the sunny slope; the obtained average aboveground biomass and average tree height are input into the power function regression model to obtain the aboveground biomass-tree height model of the tree species on the sunny slope; the aboveground biomass-tree height model of the tree species on the sunny slope is substituted into its age-aboveground biomass function on the sunny slope to obtain the age-tree height model of the tree species on the sunny slope;
[0130] The sample plot is a forest land with a set size, a single dominant tree species, and a slope of only shady slope or sunny slope. Here, the set size can be 30×30m.
[0131] Here, the age-aboveground biomass function is expressed as A=f 1 (B) The expression for fitting the average aboveground biomass and average age based on the Richards growth function is:
[0132]
[0133] Among them, A is the average age; B is the average aboveground biomass; a, b, c are parameters. Specifically,
[0134] The average aboveground biomass and average age of the tree species in various plots on the shady slope / sunny slope are input into the above formula, and the Levenberg-Marquardt optimization algorithm is used to minimize the square of the absolute difference between the observed and estimated forest age. The parameters a, b and c are fitted and estimated by nonlinear regression, and the a, b, c parameters of the Richards growth function equation are calculated to obtain the age-aboveground biomass function of the tree species on the shady slope / sunny slope.
[0135] The aboveground biomass-tree height model is expressed as B=f 2 (H), its power function regression model expression is:
[0136] B=d×H e
[0137] Among them, H is the average tree height, d and e are parameters.
[0138] The aboveground biomass-tree height model for the tree species on the shady slope / sunny slope is as follows: the average aboveground biomass and average tree height of the tree species on the shady slope / sunny slope are input into the above formula to obtain the parameters d and e.
[0139] The age-tree height model of the tree species on the shady slope is:
[0140] Taking biomass B as the intermediate variable, we can get the functional relationship between age and tree height: A = f 1 (f 2 (H)), its expression is:
[0141]
[0142] The age-aboveground biomass function, aboveground biomass-tree height model, and age-tree height model of each tree species on shady slopes / sunny slopes are shown in Table 1.
[0143] Table 1
[0144]
[0145] In another possible implementation, the method of obtaining the average aboveground biomass, average tree height, and average age of trees in any plot is:
[0146] Use backpack / handheld laser radar to measure the diameter at breast height and height of each tree in the plot; use tree growth cones to sample the number of annual rings of each tree core in the plot as its age;
[0147] For any tree, substitute its DBH and tree height into the allometric growth equation corresponding to the dominant tree species in the plot to calculate the aboveground biomass of the tree;
[0148] The means of the aboveground biomass, tree height and age of the trees in the sample plot were calculated respectively to obtain the average aboveground biomass, average tree height and average age of the trees in the sample plot.
[0149] The specific measurement of the DBH and height of each tree using backpack / handheld LiDAR is:
[0150] Use backpack / handheld LiDAR to collect 3D point cloud data of the tree;
[0151] The collected data are screened, adjusted and analyzed to obtain the tree's diameter at breast height and height.
[0152] Here, taking plot v where the single dominant tree species is Pinus tabulaeformis as an example, the average aboveground biomass of plot v is calculated. Average tree height Average age
[0153] calculate The formula is as follows:
[0154] W Ti =0.14187(D i 2 H i ) 0.8728 +0.0147083(D i 2 H i ) 0.9157 +0.066901(D i 2 H i ) 0.8017
[0155]
[0156] Among them, W Ti is the aboveground biomass of the ith Pinus tabulaeformis plant in plot v; D i is the DBH of the ith Pinus tabulaeformis tree in plot v; H i is the tree height of the i-th Pinus tabulaeformis in plot v; is the average aboveground biomass of plot v, and m is the total number of Pinus tabulaeformis trees in plot v.
[0157] Different tree species have different corresponding allometric growth equations. You can look up the allometric growth equations of national and local dominant tree species (groups) or consult other information to obtain them. The allometric growth equations corresponding to some tree species are shown in Table 2.
[0158] Table 2
[0159]
[0160] calculate The formula is as follows:
[0161]
[0162] in, is the average tree height of plot v, m is the total number of Pinus tabulaeformis in plot v, h Ti is the tree height of the i-th Pinus tabulaeformis tree in plot v.
[0163] The expression for calculating the average age of the plot is:
[0164]
[0165] in, is the average age of the forest in plot v, m is the total number of trees in plot v, age Tiis the age of the i-th Pinus tabulaeformis tree in plot v.
[0166] Furthermore, the effectiveness of the present invention is verified by the following method examples:
[0167] Taking the Huodong planned mining area as the area to be estimated, the forest age of the mine-forest complex area is estimated as follows:
[0168] 1. Obtain and preprocess data:
[0169] The long-term multispectral satellite remote sensing image data selected the Landsat TM / ETM+ / OLI surface reflectance image data of a mining area in the middle reaches of the Yellow River from 1985 to 2022. It was obtained and preprocessed through the GEE platform. The preprocessing includes: using the SimpleCloudScore algorithm for cloud removal, the SCS+C correction method based on the Lambertian reflectance model of DEM for terrain correction, and the Quality Mosaic algorithm for image clipping.
[0170] Sentinal-1, Sentinal-2, DEM, and GEDI data are for a mining area in the middle reaches of the Yellow River in 2022.
[0171] Get the B2, B3, B4, B5, B6, B7, B8, B11, B12 bands of Sentinal-2 MSI and calculate the remote sensing index.
[0172] Acquire GEDI L2A-level data, including spot-scale canopy relative height indicators and quality indicators.
[0173] Get DEM data.
[0174] Get the VV and VH bands of Sentinal-1SAR images.
[0175] Obtain forest distribution data from land use data.
[0176] 2. Obtain the image of the forest area to be estimated:
[0177] Select forest sample points based on Sentinal-2, use random forest machine learning classification method to obtain forest area; or use forest distribution data in land use data as mask, mask the B2, B3, B4, B5, B6, B7, B8, B11, B12 bands of Sentinal-2MSI, VV and VH bands of Sentinal-1SAR image, elevation and slope information of DEM digital to obtain the image of forest area to be estimated, such as Figure 4 shown.
[0178] 3. Determine the tree height of each pixel in the forest area image to be estimated
[0179] like Figure 5 As shown in Figure 1, 80% of the RH95% band data of the satellite-borne lidar is used as the training sample, and the Gradient Boosting Regression Tree (GBRT) or Random Forest Regression (Random Forest Regression) machine learning algorithm is used to obtain the tree height of each pixel in the forest area image to be estimated, as shown in Figure 1. Figure 6 shown.
[0180] 4. Calculate the average aboveground biomass, average tree height, and average age of the sample plots through the sample plot data to prepare for subsequent modeling.
[0181] 5. Obtain regional tree species distribution based on Sentinal-2 data, such as Figure 7 Based on the DEM data, the shady slope and sunny slope are distinguished, and the results are as follows Figure 8 shown.
[0182] 6. Based on the data from step 4 and the Richard growth equation, the age-biomass function and biomass-tree height function were calculated by distinguishing tree species and shady slopes and sunny slopes, and the age-tree height model was constructed through the two functions, as shown in Table 3.
[0183] Table 3
[0184]
[0185] 7. Spatially associate the tree height of each pixel with the tree species and slope aspect, using the age-tree height model A=f 1 (f 2 (H)), the forest age distribution of the undisturbed forest area was obtained, e.g. Fig. 9 shown.
[0186] 8. Construct evaluation indicators for the inversion results of the machine learning model, and use the least squares algorithm as the regression test algorithm to calculate the explained variance, mean absolute error, mean square error and determination coefficient based on machine learning to verify and evaluate the model and accuracy.
[0187] 9. Calculate the annual average NBR index of the images year by year and search for pixels with NBR index mutations, such as Fig.10 As shown, the forest pixel was disturbed in 1992 and began to recover in 1994. It is determined that the planted trees are 1-year-old saplings, that is, the initial age of the trees is 1, so the forest age of this pixel is 2023-1994+1=20 years.
[0188] Thus, the forest age of the disturbed area in a mining area in the middle reaches of the Yellow River in 2023 was obtained, such as Fig.11 shown.
[0189] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.
Claims
1. A method for estimating regional forest age, It is characterized in that include: Obtain the image of the forest area to be estimated, and perform the following steps on each pixel of the image of the forest area to be estimated to obtain the forest age distribution of the area to be estimated: From the age-tree height model set, find the age-tree height model of the tree species in the pixel at the slope of the pixel; substitute the tree height of the pixel into the found age-tree height model to obtain the average age of the trees in the pixel; Among them, the construction of the age-tree height model set is: for any tree species, take the parameters of the trees growing on the shady slope to construct the age-tree height model of the tree species on the shady slope; take the parameters of the trees growing on the sunny slope to construct the age-tree height model of the tree species on the sunny slope; the construction of the age-tree height model for any tree species is: respectively obtain the average aboveground biomass, average tree height, and average age of the trees in various plots on the shady slope where the tree species is the dominant tree species; fit the taken average aboveground biomass and average age to obtain the age-aboveground biomass function of the tree species on the shady slope; input the taken average aboveground biomass and average tree height into the power function regression model to obtain the aboveground biomass-tree height model of the tree species on the shady slope; substitute the aboveground biomass-tree height model of the tree species on the shady slope into its The age-above-ground biomass function on the shady slope is used to obtain the age-tree height model of the tree species on the shady slope; the average above-ground biomass, average tree height and average age of the trees in each sample plot with the tree species as the dominant tree species on the sunny slope are obtained respectively; the average above-ground biomass and average age are fitted to obtain the age-above-ground biomass function of the tree species on the sunny slope; the average above-ground biomass and average tree height are input into the power function regression model to obtain the above-ground biomass-tree height model of the tree species on the sunny slope; the above-ground biomass-tree height model of the tree species on the sunny slope is substituted into its age-above-ground biomass function on the sunny slope to obtain the age-tree height model of the tree species on the sunny slope; wherein the sample plot is: a set size, with only a single dominant tree species within the range, and the slope direction is only shady slope or sunny slope.
2. The method according to claim 1, It is characterized in that The forest area image to be estimated is an image of an undisturbed forest area. The method further includes: estimating the forest age distribution of the disturbed forest area as follows: Determine the pixels of disturbed forest area; Execute for each disturbed forest area pixel separately: Get the average initial age of trees in this pixel; The average age of trees in a pixel is the current year minus the recovery year of the disturbed pixel plus the average initial age of trees in the pixel. The average initial age of trees in the pixel refers to the average age of trees in the pixel in the recovery year of the disturbed pixel.
3. The method according to claim 2, It is characterized in that The disturbed forest area pixels are determined as follows: The plant change tracking VCT method is used to track each pixel of the long-term multispectral satellite remote sensing image of the estimated area, and the comprehensive forest index IFZ time series of each pixel is obtained; If the IFZ value increases by more than the threshold compared with the previous year, the pixel corresponding to the IFZ time series is the disturbed forest area pixel; for the disturbed forest area pixel, if the IFZ value decreases by more than the threshold compared with the previous year, the corresponding year is the recovery year of the disturbed pixel; or The normalized burn rate index NBR of each pixel in the long-term multispectral satellite remote sensing image of the estimated area in each year was calculated respectively, and the NBR time series segmentation characteristics of each pixel were obtained using the LandTrendr algorithm. Among them, the pixels corresponding to the NBR time series with the inflection point of NBR decrease are the pixels in the disturbed forest area; for the pixels in the disturbed forest area, the year corresponding to the inflection point of NBR increase is the recovery year of the disturbed pixel.
4. The method according to claim 2 or 3, It is characterized in that Also includes: The forest age distribution of the undisturbed forest area and the forest age distribution of the disturbed forest area are spatially superimposed to obtain the forest age distribution of the area to be estimated.
5. The method according to claim 1, It is characterized in that The average aboveground biomass, average tree height, and average age of trees in any plot are obtained as follows: Use backpack / handheld laser radar to measure the diameter at breast height and height of each tree in the plot; use tree growth cones to sample the number of annual rings of each tree core in the plot as its age; For any tree, substitute its DBH and tree height into the allometric growth equation corresponding to the dominant tree species in the plot to calculate the aboveground biomass of the tree; The means of the aboveground biomass, tree height and age of the trees in the sample plot were calculated respectively to obtain the average aboveground biomass, average tree height and average age of the trees in the sample plot.
6. The method according to claim 1, It is characterized in that It also includes determining the tree height of each pixel in the forest area image to be estimated, which is: Construct the training set, recorded as: D={(x i ,y i )}i=1,2,…,n Wherein, D is a training set, and the independent variable sample data of D includes at least one of the multispectral image data, SAR image data and DEM data of the forest area; the dependent variable sample data is the RH95% / 98% band data of the satellite-borne lidar data of the forest area; Use D to train a random forest regression model to obtain a tree height prediction model, or use D to train a gradient boosting regression tree model to obtain a tree height prediction model; The pixel values of the image of the forest area to be estimated are respectively input into the tree height prediction model to obtain the tree height data of each pixel of the image of the forest area to be estimated.
7. The method according to claim 1, It is characterized in that It also includes determining the tree species of each pixel in the forest area image to be estimated, which is: Construct a sample set of the relationship between remote sensing index and tree species; use the constructed sample set to train a random forest model to obtain a tree species prediction model; Extract the remote sensing index of each pixel of the image of the forest area to be estimated; input the remote sensing index of each pixel into the tree species prediction model to obtain the tree species of each pixel; Among them, the independent variable data of any sample is at least one of the normalized vegetation index NDVI, enhanced vegetation index EVI, spectral band information, and texture characteristics of the multispectral image data corresponding to a sample plot; the dependent variable data of the sample is: the dominant tree species of the sample plot.
8. The method according to claim 1, It is characterized in that It also includes determining the slope direction of each pixel of the forest area image to be estimated, which is: Obtain the DEM data corresponding to the geographic coordinates of each pixel in the image of the forest area to be estimated, and determine the angle between the projection of the slope normal of each pixel on the horizontal plane and the clockwise direction of the true north direction through the obtained DEM data; determine the slope direction of the corresponding pixel based on the determined angles.
9. The method according to claim 1, It is characterized in that The method of obtaining the image of the forest area to be estimated is: Using the forest distribution data in the land use data of the area to be estimated as a mask, at least one of the multispectral image, SAR image and DEM data of the area to be estimated is masked to obtain the image of the forest area to be estimated; or, forest sample points are selected based on high-resolution images, and the random forest machine learning classification method is used to obtain the image of the forest area to be estimated.
