Large-range cunninghamia lanceolata stock estimation method and system based on CR-RF hybrid model
By combining Landsat data and growth process variables with a CR-RF hybrid model, the problems of insufficient model robustness and interpretability in the estimation of Chinese fir volume were solved, and efficient and accurate estimation of Chinese fir volume over a large area was achieved.
Patent Information
- Application Number
- CN202610141865.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-02
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies for estimating Chinese fir stock volume suffer from poor model robustness and generalization ability, making it difficult to adapt to complex site environments and management practices, and their effectiveness is particularly poor when widely promoted.
A CR-RF hybrid model was adopted, combining Landsat time series data, sample plot survey data, and auxiliary data. By constructing growth curve characteristics and growth process variables, a region-specific growth equation was built. The random forest method was used for hybrid modeling to improve the robustness and interpretability of the estimation.
It significantly improves the robustness and interpretability of Chinese fir stock volume estimation, is applicable to forest resource monitoring in large-scale and complex ecological zones, and reduces data acquisition costs and model errors.
Smart Images
Figure CN122065240A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forest stock volume estimation technology, and in particular to a method and system for large-scale Chinese fir stock volume estimation based on a CR-RF hybrid model. Background Technology
[0002] Chinese fir is a widely distributed, economically valuable evergreen coniferous forest species. Due to its complex site environments and diverse management methods, accurately determining its stock volume distribution is crucial for optimizing plantation management and assessing carbon sequestration. In recent years, remote sensing technology has been widely applied to forest stock volume estimation. This primarily utilizes remote sensing data from optical, radar, and lidar sensors to extract variables and establish parametric or non-parametric regression models (such as multiple regression and random forests) with measured ground stock volumes to estimate regional stock volumes.
[0003] Existing technologies for estimating the stock volume of Chinese fir (planted forests) include the following: First, data from optical images (such as Landsat and Sentinel) and radar are used to extract spectral data, texture, and backscattering coefficients to construct a regression model for estimating stock volume. This modeling approach is constrained by multiple factors, including topography, atmospheric and phenological effects, and signal saturation. Another approach utilizes the relationship between canopy height and stock volume to construct an estimation model. Airborne LiDAR is widely recognized as the best data source for estimating stock volume or biomass, but its large data volume and high acquisition cost limit its applicability to small areas. Spaceborne LiDAR data requires combining optical data to generate comprehensive canopy height data, which carries significant uncertainty. Furthermore, using long-term optical remote sensing data to detect forest disturbance and recovery trajectories to estimate forest age and construct an age-stock volume model is also a common approach. This approach typically employs empirical growth equations or machine learning models, but it does not consider the unique management practices of Chinese fir plantations or the impact of complex site environments (such as topography and soil) on the age-stock volume relationship.
[0004] Existing technologies also have the following drawbacks: models relying solely on optical or radar intensity experience signal saturation at high biomass levels and are susceptible to interference from topography and atmospheric effects; while models relying solely on canopy height are limited by data acquisition costs or inversion accuracy, making them difficult to widely implement. When using forest age modeling, existing empirical regression models struggle to capture the nonlinear interactions between forest growth and complex site environments; while data-driven machine learning methods, although possessing strong fitting capabilities, lack biological interpretability (a black box problem), and current models do not consider the impact of frequent management practices on Chinese fir. Under diverse management practices and complex habitats for Chinese fir, the models exhibit poor robustness and generalization ability, making large-scale scalable applications difficult. Summary of the Invention
[0005] In view of this, the purpose of this invention is to provide a method and system for estimating the volume of Chinese fir trees over a wide range based on a CR-RF hybrid model, thereby improving the robustness and interpretability of volume estimation under different ecological zones and management intensities.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model, comprising the following steps:
[0007] Step 1: Collect Landsat time-series data, sample plot survey data, land type survey data, and auxiliary data, as well as Chinese fir distribution products, within the study area, and perform data preprocessing;
[0008] Step 2: Using Landsat time-series images of the Chinese fir distribution mask area, construct NBR annual median synthetic time-series data, identify potential change areas through preliminary anomaly detection, and set an adaptive disturbance threshold to achieve automatic segmentation of disturbance event groups, thereby enabling the detection of disturbance years and calculation of Chinese fir forest age;
[0009] Step 3: Set up an interference filtering strategy to post-process the age data and obtain the age distribution map of Chinese fir forests in the study area;
[0010] Step 4: Using the age of the Chinese fir forest obtained in Step 3, combined with auxiliary data, as the independent variable, and the measured stock volume of the sample plot as the dependent variable, a region-specific growth equation is constructed based on the Chapman-Richards growth equation to obtain the growth curve characteristics.
[0011] Step 5: Extract growth process variables from Landsat long-term data and use the recursive feature elimination algorithm RFECV based on cross-validation to filter variables;
[0012] Step 6: Using the growth curve characteristics obtained in Step 4 and the growth process variables extracted in Step 5 as independent variables, and the plot volume as the dependent variable, construct a hybrid model CR-RF using the random forest RF method;
[0013] Step 7: Export the CR-RF hybrid model trained in Step 6.
[0014] In a preferred embodiment, in step 1, cloud removal and annual median synthesis are performed on Landsat time-series images; climate variables are calculated based on the WorldClim bioclimatic variable system for the monthly temperature and precipitation data in the auxiliary data, and the average values of the climate variables and soil moisture data over 20 years are synthesized to generate long-term climate statistics.
[0015] In a preferred embodiment, in step 2, the annual NBR standard deviation is set to be greater than 0.05 as the initial anomaly detection threshold, and the 10th percentile of the pixel NBR sequence is used as the adaptive interference threshold.
[0016] In a preferred embodiment, in step 3, interference events with an NBR threshold greater than 0.4 and recovering within two years are defined as thinning, and the recovery threshold is set to be greater than or equal to 80% of the NBR before the interference. If the number of NBR observations greater than 0.5 within a year exceeds 50%, and no continuous downward trend is observed in the time series, the interference event is determined to be unreliable and is removed. A 3×3 moving window is used to perform spatial smoothing of the age results to obtain the age distribution map of Chinese fir in the study area with a spatial resolution of 30m.
[0017] In a preferred embodiment, in step 4, the region-specific growth equation is constructed as follows:
[0018] (1)
[0019] (2)
[0020] (3)
[0021] (4)
[0022] Where Y(t) represents the total stand volume, t represents the stand age; parameters a, b and c represent the maximum potential stand volume, growth rate and the shape and inflection point of the growth curve under general conditions, respectively; x represents the environmental vector and n represents the dimension; , and This represents the relationship between parameters a, b, c and environment variables, where , and It is the corresponding parameter vector.
[0023] In a preferred embodiment, in step 4, a two-stage optimization strategy is adopted, using the differential evolution (DE) algorithm for global optimization and the L-BFGS-B algorithm for gradient-based local optimization to obtain the final model parameters.
[0024] In a preferred embodiment, in step 4, a parameter constraint mechanism is established, with the constraint boundary ranges of parameters a, b, and c being (0, 1000), (0, 0.1), and (0, 2).
[0025] In a preferred embodiment, in step 5, the growth process variables include:
[0026] 1) Local interference features obtained using the continuous change detection and classification algorithm CCDC; the local interference features include the time and amplitude of the most recent strongest interference, the rate and amplitude of change of the spectral index NBR / NDVI before and after the interference, and the duration of the interference;
[0027] 2) Full-time statistical characteristics, including slope, intercept, mean, standard deviation, coefficient of variation and area under the curve, which describe the long-term growth trend of Chinese fir throughout the entire time series after planting;
[0028] 3) Spatial disturbance indicators, including the number of disturbing pixels in each grid cell and the standard deviation of the disturbance time, reflect the spatial heterogeneity of forest stand dynamics.
[0029] In a preferred embodiment, in step 6, the hybrid model CR-RF is developed based on the Python language and the scikit-learn library, and the specific model is as follows:
[0030] (5)
[0031] Where y represents the measured value of the stock volume in the sample plot. Y(t, env) represents the output of equation (3), ts refers to the growth process variables extracted in step 5, and env represents the auxiliary variables. In order to minimize the collinearity among the predictor variables, the environmental variables are orthogonalized to construct a set of unrelated or weakly related environmental factors env⊥.
[0032] This invention also provides a large-scale Chinese fir volume estimation system based on a CR-RF hybrid model, comprising a processor, a memory, and a bus. The memory stores machine-readable instructions executed by the processor. When the system is running, the processor communicates with the memory via the bus, and the machine-readable instructions executed by the processor are as described in the large-scale Chinese fir volume estimation method based on a CR-RF hybrid model.
[0033] Compared with existing technologies, this invention has the following advantages: The large-scale remote sensing estimation method for Chinese fir stock volume constructed in this invention utilizes the dynamic capture capability of long-term remote sensing data, which is of great value for stock volume estimation in the monitoring of large-scale plantation resources with short management cycles and frequent interference. Compared with data from optical and radar sources, which are constrained by multiple factors such as topography, atmosphere, phenology, and signal saturation, the hybrid modeling strategy of this invention fully integrates the region-specific growth curves constructed from forest age and auxiliary data with the growth process characteristics represented by remote sensing time series, significantly improving the robustness and interpretability of stock volume estimation under different ecological zones and management intensities. Attached Figure Description
[0034] Figure 1 This is a flowchart of a preferred embodiment of the method for estimating the volume of Chinese fir trees according to the present invention. Detailed Implementation
[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0036] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0037] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application; as used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise; furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.
[0038] The embodiments of the present invention estimate the volume of Chinese fir trees in China based on multi-source time-series data and auxiliary data, referring to... Figure 1 The technical solution and specific steps of this invention are as follows:
[0039] Step 1. Collect Landsat time-series data for the study area using the Google Earth cloud computing platform (GEE). Collect plot survey data (plot volume, forest age), land use survey data and auxiliary data (climate, soil, topography), and Chinese fir distribution products, and perform data preprocessing;
[0040] In this embodiment, cloud removal and annual median composites are performed on Landsat time-series imagery. Climate variables are calculated from the monthly temperature and precipitation data in the auxiliary data using the WorldClim bioclimatic variable system. The 20-year data for these climate variables, along with soil moisture data, are then averaged to generate long-term climate statistics. All data are uniformly set to a 30m resolution.
[0041] Step 2. Using Landsat time-series images of the Chinese fir distribution mask area, construct NBR annual median synthetic time-series data, identify potential change areas through preliminary anomaly detection, and set an adaptive disturbance threshold to achieve automatic segmentation of disturbance event groups, thereby enabling the detection of disturbance years and calculation of Chinese fir forest age;
[0042] In this embodiment, based on empirical calibration of calibration samples and the growth prior of Chinese fir, an interannual NBR standard deviation greater than 0.05 is set as the initial anomaly detection threshold, and the 10th percentile of the pixel NBR sequence is used as the adaptive interference threshold.
[0043] Step 3. Set up an interference filtering strategy to post-process the age data and obtain the age distribution map of Chinese fir forests in the study area;
[0044] In this embodiment, to distinguish between thinning and clear-cutting, interference with an NBR threshold greater than 0.4 and recovery within two years is defined as thinning, and the recovery threshold is set to be greater than or equal to 80% of the NBR before the interference. To remove noise in early time series, if the number of NBR observations greater than 0.5 within a year exceeds 50% and no continuous downward trend is observed in the time series, the interference event is deemed unreliable and removed. To eliminate spatial abrupt changes and maintain regional consistency, a 3×3 moving window is used to spatially smooth the age results, resulting in a Chinese fir age distribution map of the study area with a spatial resolution of 30m.
[0045] Step 4. Using the age of the Chinese fir forest obtained in Step 3, combined with auxiliary data (climate, topography, soil) as the independent variable, and the measured stock volume of the sample plot as the dependent variable, a region-specific growth equation is constructed based on the Chapman-Richards (CR) growth equation.
[0046] (1)
[0047] (2)
[0048] (3)
[0049] (4)
[0050] Where Y(t) represents the total stand volume, and t represents the stand age. Parameters a, b, and c represent the maximum potential stand volume, growth rate, and the shape and inflection point of the growth curve under general conditions, respectively. To incorporate environmental effects, each parameter is expressed as a linear function of the environmental vector x, with dimension n. Specifically, , and This describes the relationship between parameters a, b, c and environment variables, where , and It is the corresponding parameter vector.
[0051] A two-stage optimization strategy was adopted, using the differential evolution (DE) algorithm for global optimization and the L-BFGS-B (Limited-memory Broyden-Fletcher-Goldfarb-Shanno with Box constraints) algorithm for gradient-based local optimization to obtain the final model parameters.
[0052] In this embodiment, in order to ensure the rationality of the growth function, a parameter constraint mechanism is established, and the constraint boundary ranges of parameters (a, b, and c) are (0, 1000), (0, 0.1), and (0, 2).
[0053] Step 5. Extract growth process variables from Landsat long-term data and use the recursive feature elimination algorithm based on cross-validation (RFECV) to filter variables;
[0054] In this embodiment, the growth process variables include (1) local disturbance features obtained using the Continuous Change Detection and Classification (CCDC) algorithm. These features include the time and magnitude of the most recent strongest disturbance, the rate and magnitude of change of the spectral index (NBR / NDVI) before and after the disturbance, and the duration of the disturbance; (2) full-time statistical features, including slope, intercept, mean, standard deviation, coefficient of variation, and area under the curve, which describe the long-term growth trend of the Chinese fir throughout the entire time series after planting; and (3) spatial disturbance indices, including the number of disturbing pixels in each grid cell (1 km) and the standard deviation of the disturbance time, reflecting the spatial heterogeneity of the stand dynamics.
[0055] Step 6. Using the growth curve features obtained in Step 4 and the growth process variables extracted in Step 5 as independent variables, and the plot volume as the dependent variable, a mixed model (CR-RF) is constructed using the Random Forest (RF) method. The model is developed based on Python and the scikit-learn library. The specific model is as follows:
[0056] (5)
[0057] Where y represents the measured value of the stock volume in the sample plot. Y(t, env) represents the constructed random forest model, Y(t, env) represents the output of equation (3), ts refers to the time series variables extracted in step 7, env represents the auxiliary variables. In order to minimize the collinearity among the predictor variables, the environmental variables were orthogonalized to construct a set of uncorrelated (or weakly correlated) environmental factors env⊥.
[0058] In this embodiment, the random forest model is configured with 200 decision trees and a maximum depth of 8.
[0059] Step 7. The model trained in Step 6 is exported through the geemap.ml module so that it can be inverted on the Google Earth Engine (GEE) platform to obtain a spatial distribution map of the Chinese fir stock volume for the entire study area.
[0060] In this embodiment, the random forest model is configured with 200 decision trees and a maximum depth of 8. The model is then inverted to the entire study area to obtain a spatial distribution map of the Chinese fir stock volume with a spatial resolution of 30m.
[0061] The following is a further explanation of the algorithm involved in this invention:
[0062] It is a recursive feature elimination algorithm based on cross-validation (RFECV): it iteratively removes the least important features and retrains the model, using five-fold cross-validation at each step to ensure robust performance evaluation.
[0063] Continuous Change Detection and Classification (CCDC): Robustly identifies the timing and manner of land cover changes over time using all available Landsat data. The algorithm first fits a harmonic regression model, detects anomalies by comparing model predictions with actual observations, and uses the parameters of the fitted model as input to a land cover classifier for land cover feature analysis.
[0064] Theil-Sen trend analysis: used to estimate the slope and intercept of time series. This non-parametric method reduces the impact of outliers by calculating the median slope, thereby improving the robustness of the results.
[0065] Differential Evolution (DE) algorithm: It is a population-based, efficient global optimization algorithm that can effectively prevent the model from getting trapped in local optima.
[0066] The L-BFGS-B algorithm is a gradient-based local optimization nonlinear optimization method that allows for efficient computation under simple boundary constraints.
Claims
1. A method for estimating the volume of Chinese fir trees over a large area based on a CR-RF hybrid model, characterized in that, Includes the following steps: Step 1: Collect Landsat time-series data, sample plot survey data, land type survey data, and auxiliary data, as well as Chinese fir distribution products, within the study area, and perform data preprocessing; Step 2: Using Landsat time-series images of the Chinese fir distribution mask area, construct NBR annual median synthetic time-series data, identify potential change areas through preliminary anomaly detection, and set an adaptive disturbance threshold to achieve automatic segmentation of disturbance event groups, thereby enabling the detection of disturbance years and calculation of Chinese fir forest age; Step 3: Set up an interference filtering strategy to post-process the age data and obtain the age distribution map of Chinese fir forests in the study area; Step 4: Using the age of the Chinese fir forest obtained in Step 3, combined with auxiliary data, as the independent variable, and the measured stock volume of the sample plot as the dependent variable, a region-specific growth equation is constructed based on the Chapman-Richards growth equation to obtain the growth curve characteristics. Step 5: Extract growth process variables from Landsat long-term data and use the recursive feature elimination algorithm RFECV based on cross-validation to filter variables; Step 6: Using the growth curve characteristics obtained in Step 4 and the growth process variables extracted in Step 5 as independent variables, and the plot volume as the dependent variable, construct a hybrid model CR-RF using the random forest RF method; Step 7: Draw a spatial distribution map of the cedar wood stock volume from the hybrid model CR-RF trained in Step 6.
2. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 1, characterized in that, In step 1, cloud removal and annual median synthesis are performed on Landsat time-series images; climate variables are calculated for monthly temperature and precipitation data in auxiliary data based on the WorldClim bioclimatic variable system, and the average values of these climate variables and soil moisture data over 20 years are synthesized to generate long-term climate statistics.
3. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 1, characterized in that, In step 2, the annual NBR standard deviation is set to be greater than 0.05 as the initial anomaly detection threshold, and the 10th percentile of the pixel NBR sequence is used as the adaptive interference threshold.
4. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 1, characterized in that, In step 3, interferences with an NBR threshold greater than 0.4 and recovering within two years are considered intermittent. The recovery threshold is set to be greater than or equal to 80% of the NBR before the interference. If the number of NBR observations greater than 0.5 within a year exceeds 50% and no continuous downward trend is observed in the time series, the interference event is determined to be unreliable and is removed. Spatial smoothing of the age results was performed using a 3×3 moving window to obtain the age distribution map of Chinese fir in the study area, with a spatial resolution of 30m.
5. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 1, characterized in that, In step 4, the region-specific growth equation is constructed as follows: (1) (2) (3) (4) Where Y(t) represents the total stand volume, t represents the stand age; parameters a, b and c represent the maximum potential stand volume, growth rate and the shape and inflection point of the growth curve under general conditions, respectively; x represents the environmental vector and n represents the dimension; , and This represents the relationship between parameters a, b, c and environment variables, where , and It is the corresponding parameter vector.
6. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 5, characterized in that, In step 4, a two-stage optimization strategy is adopted, using the differential evolution (DE) algorithm for global optimization and the L-BFGS-B algorithm for gradient-based local optimization to obtain the final model parameters.
7. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 6, characterized in that, In step 4, a parameter constraint mechanism is established, with the constraint boundary ranges of parameters a, b, and c being (0, 1000), (0, 0.1), and (0, 2), respectively.
8. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 1, characterized in that, In step 5, the variables in the growth process include: 1) Local interference features obtained using the continuous change detection and classification algorithm CCDC; the local interference features include the time and amplitude of the most recent strongest interference, the rate and amplitude of change of the spectral index NBR / NDVI before and after the interference, and the duration of the interference; 2) Full-time statistical characteristics, including slope, intercept, mean, standard deviation, coefficient of variation and area under the curve, which describe the long-term growth trend of Chinese fir throughout the entire time series after planting; 3) Spatial disturbance indicators, including the number of disturbing pixels in each grid cell and the standard deviation of the disturbance time, reflect the spatial heterogeneity of forest stand dynamics.
9. The method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model according to claim 5, characterized in that, In step 6, the hybrid model CR-RF was developed using Python and the scikit-learn library. The specific model is as follows: (5) Where y represents the measured value of the stock volume in the sample plot. Y(t, env) represents the output of equation (3), ts refers to the growth process variables extracted in step 5, and env represents the auxiliary variables. In order to minimize the collinearity among the predictor variables, the environmental variables are orthogonalized to construct a set of unrelated or weakly related environmental factors env⊥.
10. A large-scale Chinese fir volume estimation system based on a CR-RF hybrid model, comprising a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executed by the processor; characterized in that, When the system is running, the processor and the memory communicate via a bus, and the machine-readable instructions are executed by the processor as described in any one of claims 1 to 9. This is a method for estimating the large-scale stock volume of Chinese fir based on a CR-RF hybrid model.