Farmland protection forest belt age remote sensing identification method considering growth mode
By constructing time series coverage waveform curves and growth pattern characteristics, the problems of low efficiency and large identification errors in traditional methods of forest age determination are solved, and the refined and automated identification of the age of farmland shelterbelts is achieved, which is suitable for large-scale monitoring and planning.
Patent Information
- Application Number
- CN202510896073.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-09-30
AI Technical Summary
Traditional forest age determination methods are inefficient and difficult to scale up. Existing remote sensing methods are easily affected by environmental noise and cannot accurately identify the age of farmland shelterbelts. They also lack targeted design for zonal distribution characteristics, resulting in large errors in age identification and low levels of automation.
By acquiring long-term Landsat remote sensing images, a time-series forest belt coverage waveform curve is constructed, outliers are removed and smoothed, and the initial coverage recognition threshold is set based on the growth pattern characteristics. A forest belt age discrimination algorithm is constructed to achieve automatic identification of forest belt age.
It has achieved refined identification of the age of farmland shelterbelts, has a high degree of automation, adapts to the strip forest structure, avoids the problems of insufficient accuracy of single-phase remote sensing and environmental interference, and provides reliable data support for renewal assessment and ecological engineering planning at the regional scale.
Smart Images
Figure CN120726486A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of remote sensing and forest management, and specifically to a remote sensing identification method for farmland shelterbelt age taking into account growth patterns, which is suitable for dynamic age monitoring, renewal planning and comprehensive benefit evaluation of large-scale shelterbelts. Background Art
[0002] The age of shelterbelts directly impacts their ecological functions, such as windbreak and sand fixation, and soil and water conservation. As the forest ages, its canopy structure, biomass, root development, and overall benefits undergo significant changes. Young forest belts have weak shielding capabilities and have yet to form effective windbreaks; mature forest belts, on the other hand, offer excellent protection. Overmature forest belts may experience structural decline, gaps, and disconnections, compromising the overall continuity of the project. Therefore, accurate information on the age of shelterbelts is essential for orderly shelterbelt renewal, optimizing shelterbelt planting structures, and guiding scientific management.
[0003] However, traditional forest age determination relies on field tree ring analysis or manual interviews, which is inefficient and difficult to scale. Existing remote sensing methods are mostly based on spectral modeling using single-temporal imagery, which is susceptible to environmental noise. Furthermore, no effective, in-depth research has been conducted on the zonal distribution characteristics of shelterbelts, nor have optimization algorithms been developed. For example, while single-temporal NDVI can reflect vegetation cover, it cannot distinguish between forest belts planted in similar years, resulting in significant errors in age identification.
[0004] While multi-temporal remote sensing can capture dynamic vegetation changes through time series analysis, existing research has largely focused on patchy forests (such as rubber plantations and natural forests) and lacks specific design considerations for the linear characteristics of farmland shelterbelts. Furthermore, traditional methods rely on manually setting complex parameters (such as texture features and classification thresholds), resulting in low automation and limited adaptability to the rapid monitoring of forest age at a regional scale. Summary of the Invention
[0005] (1) Technical problems solved
[0006] In view of the shortcomings of the existing technology, the present invention provides a remote sensing identification method for the age of farmland shelterbelts that takes into account the growth pattern. It has the advantages of simple data processing method, high degree of automation of processing flow, high accuracy of forest age identification, and is conducive to promotion. It effectively solves the technical problems raised in the above background technology.
[0007] (2) Technical solution
[0008] The present invention realizes the automatic identification of the age of farmland shelterbelt through the following steps:
[0009] 1) Acquire and preprocess long-term Landsat remote sensing imagery: Remote sensing images are acquired year by year for months that meet monitoring requirements. Data preprocessing, including radiometric calibration, atmospheric correction, and geometric correction, is performed. Based on this data, the Normalized Difference Vegetation Index (NDVI) is calculated annually.
[0010] 2) Constructing a time series waveform curve of forest belt coverage: Applying the pixel binary model to invert the protection forest coverage, extracting the effective coverage waveform of each forest belt, and constructing a time series waveform curve of forest belt coverage (FCFS).
[0011] 3) Outlier removal and smoothing of time series coverage waveform: By setting the standard deviation threshold, outlier removal and smoothing are performed on the forest belt coverage waveform curve to extract the time series coverage representing the characteristics of the forest belt.
[0012] 4) Determine the growth pattern of forest belts based on the changes in forest belt coverage waveforms: Analyze the coverage characteristics of forest belts, comprehensively consider the planting patterns and growth characteristics of shelterbelts, and divide farmland shelterbelts into three types of forest belt growth patterns based on the changes in coverage ( Figure 2 ): Continuously updated, newly added, and never updated.
[0013] 5) Determine the initial identification year of forest belts based on the characteristics of different growth patterns and construct a forest age calculation method: Set the initial identification threshold for coverage, combine the initial identifiable years using remote sensing, and construct a forest belt age discrimination algorithm taking into account different growth patterns to calculate the forest belt age.
[0014] 6) Forest belt growth pattern and stand age mapping: Output the growth pattern type of each shelterbelt and its corresponding stand age or stand age range.
[0015] Preferably, the radiometric calibration in step 1) converts the sensor brightness value (DN) into reflectance by utilizing the gain and offset parameters in the metadata, and the atmospheric correction uses the FLAASH algorithm to convert the top of the atmosphere reflectance (TOA) into the surface reflectance (SR), and geometrically corrects the satellite images that have spatial position offsets between different years.
[0016] Preferably, the normalized difference vegetation index calculation formula in step 1) is: NDVI=(NIR–R) / (NIR+R), where NIR is the reflectance value of the near-infrared band and R is the reflectance value of the red light band.
[0017] Preferably, the formula for calculating the coverage (FC) of each remote sensing image based on NDVI year by year using the pixel dichotomy model in step 2) is: FC=(NDVI veg –NDVI cropland ) / (NDVI mixed –NDVI cropland ), where NDVIcropland NDVI is the end member value of the cultivated land area that has not changed for a long time. veg is the NDVI end member value of the patchy poplar forest.
[0018] Preferably, in step 2), the maximum coverage waveform in the neighborhood perpendicular to the forest belt direction should be obtained for each forest belt, and valid pixels should be screened based on the standard deviation (SD) of the forest belt waveform. If the standard deviation SD of the waveform sequence is less than 0.05, it means that the forest belt structure is complete and all coverage values of the forest belt are valid; otherwise, only the coverage values higher than the average value are retained, and the mean of the effective coverage values is calculated to obtain the coverage value (FCFS) representing the characteristics of the forest belt, and a coverage waveform curve of the time series is constructed.
[0019] Preferably, the smoothing rule of the FCFS waveform curve in step 3) is: for abnormal peaks (FCFS ij >FCFS ij–1 AND FCFS ij > FCFS ij+1 ) or valley value (FCFS ij < FCFS ij–1 AND FCFS ij < FCFS ij+1 ), where i is the forest belt number and j is the year. If the fluctuation range is less than the fluctuation threshold, it is replaced by the mean of the adjacent years (j–1, j+1) (FCFS ij–1 +FCFS ij+1 ) / 2), the fluctuation threshold can be determined by statistically analyzing the distribution of coverage changes of forest belts of different ages at one-year intervals ( Figure 3 ).
[0020] Preferably, the FCFS waveform curve characteristics corresponding to the growth pattern 1 in step 4) are: a sudden drop occurs in a certain year, then it is lower than the detectable threshold for several years, and then shows an upward trend; the FCFS waveform curve characteristics corresponding to the growth pattern 2 are: it is continuously lower than the detectable threshold for a certain period of time, and then shows a slow upward trend in the later period without a sudden drop; the FCFS waveform curve characteristics corresponding to the growth pattern 3 are: since the remote sensing image can be detected, it has always been higher than the detectable threshold without a sudden drop. The method for determining the year of sudden drop is: performing first-order difference calculation on the coverage curve, the formula is: ΔFCFS j = FCFS j –FCFS j+1 , if ΔFCFS j > 0.1, it is determined that a deforestation event has occurred in the forest belt.
[0021] Preferably, the identification threshold in step 5) can be determined by the distribution of the coverage value of the cultivated land at the same period when the forest belt is 2-4 years old and first appears on the remote sensing image ( Figure 4 ) was used to determine the coverage threshold, and the final coverage threshold was set to 0.15 as the criterion for whether the farmland shelterbelt is identifiable in remote sensing images.
[0022] Preferably, the forest belt age discrimination algorithm taking into account different growth patterns as described in step 5) is constructed as follows: the initial year of time series image monitoring is defined as A, the monitoring end year is B, and the year when the coverage is first greater than or equal to the recognition threshold is H. In order to avoid inaccurate positioning of the H year due to abnormal coverage curve (such as interference from the reflectivity of adjacent objects), a restriction condition should be imposed on H (H>j, j is the mutation point year), the year of forest belt planting is H-3 (FCFS is significant only 3 years after planting, so it is necessary to push forward 3 years to obtain the planting year), and the calculation method of forest belt age T corresponding to different growth patterns is: Growth pattern 1: T= B–(H–3), for the forest belt coverage curve of this growth model, if multiple mutation points are detected, the age T is calculated based on the year H corresponding to the mutation point closest to the monitoring cutoff year; Growth model two: The calculation formula for the forest belt age T of growth model two is the same as that of growth model one. The difference is that for the forest belt coverage curve of growth model two, there is no mutation point, so the year H of the recognition threshold can be directly located from low years to high years according to the coverage curve; Growth model three: For the forest belt coverage curve of growth model three, the forest belt coverage value is always greater than the recognition threshold, so the forest age T is directly determined to be greater than B–A+3.
[0023] (3) Beneficial effects
[0024] Compared with the existing technology, the present invention provides a remote sensing identification method for farmland shelterbelt age taking into account growth patterns, which has the following beneficial effects:
[0025] This remote sensing-based method for identifying the age of shelterbelts uses time series fractional coverage (FCFS) curves, combined with differential analysis and remote sensing identifiability thresholds, to identify belt growth patterns and automatically extract the planting year. This method achieves refined age identification for belts at different growth stages. This method leverages the temporal information of multi-temporal Landsat imagery, effectively avoiding the time series anomalies caused by the lack of precision of single-temporal remote sensing, cloud cover, and missing data. It eliminates the need for manual parameter setting, exhibits a high degree of automation, and is highly adaptable to belt-like plantation structures. It provides reliable data support for regional-scale shelterbelt regeneration assessment, biomass estimation, and ecological engineering planning, and has significant practical application value and promotional implications. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1: A technical flow chart of a remote sensing identification method for farmland shelterbelt age taking into account growth patterns;
[0027] Figure 2 : Schematic diagram of FCFS curves of three growth modes;
[0028] Figure 3 : Distribution of coverage change values of different forest belts after one year;
[0029] Figure 4 : The distribution of forest belts aged 2-4 years that first appeared on remote sensing images and the coverage values of cultivated land at the same time;
[0030] Figure 5 : Spatial distribution map of forest age and growth pattern map of the study area in 2021;
[0031] Figure 6 : The trend of age recognition accuracy changing with the allowable error. DETAILED DESCRIPTION
[0032] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0033] Example: A remote sensing identification method for farmland shelterbelt age taking into account growth patterns comprises the following steps:
[0034] Step 1: Obtain long-term Landsat remote sensing images and preprocess them:
[0035] This example selected a typical farmland shelterbelt area located in northern Changchun City, Jilin Province, as the study area, covering Dehui City, Nong'an County, and Jiutai District, covering an area of approximately 3,600 km². A total of 32 Landsat 5 TM, Landsat 7ETM+, and Landsat 8 OLI images were acquired from 1984 to 2021, concentrated between mid-May and mid-June each year. The best images with cloud cover below 50% were selected to construct a yearly image sequence with a spatial resolution of 30 m. Age data for 357 forest belts was obtained from tree core growth ring analysis, historical image interpretation, and farmer interviews. The original Landsat images were radiometrically calibrated and atmospherically corrected, converted to surface reflectance, and NDVI was calculated. The study area images were cropped based on the study area boundary vector.
[0036] Step 2: Construct a time series waveform curve of forest belt coverage:
[0037] The coverage of farmland shelterbelt is inverted year by year using the following formula:
[0038] FC = (NDVI veg –NDVI cropland ) / (NDVI mixed –NDVI cropland )
[0039] Where is the mean NDVI value for unchanged cultivated land areas, and is the mean NDVI value for poplar forest patches. 78 sample points were selected to compare the actual coverage results obtained on the ground with the coverage values (FC) retrieved from Landsat. The coefficient of determination (R²) was 0.766, indicating that the coverage data retrieved by the model are reliable. The interpreted 2021 forest belt vector line data were superimposed with the inverted annual coverage. The maximum waveform was extracted within a 3×3 sliding window perpendicular to the forest belt. For waveforms with a standard deviation less than 0.05, the average value of all pixels was retained as the representative coverage value. Otherwise, only valid pixels with a waveform value greater than the average value were retained, and the average value was calculated as the FCFS for that year. After processing, a complete time series FCFS curve for each forest belt was obtained.
[0040] Step 3: Eliminate and smooth outliers in the time series coverage waveform:
[0041] Abnormal peaks and valleys in the FCFS sequence are detected: if the FCFS of a certain year is significantly higher or lower than the previous and next years, and the change is less than 0.1 (empirical threshold Z), it is replaced by the average of the two adjacent years; if the change is too large, smoothing is not performed to avoid misjudging important node year information.
[0042] Step 4: Determine the growth pattern of the forest belt based on the waveform changes of the forest belt's time series coverage:
[0043] According to the FCFS sequence morphology, forest belts are divided into three types of growth patterns:
[0044] Growth mode 1: The corresponding FCFS curve is characterized by a sudden drop in coverage followed by a re-increase trend, representing a renewal mode;
[0045] Growth mode 2: corresponds to the coverage being continuously below the threshold in the early stage and then rising significantly in the later stage, representing the late-stage new growth mode;
[0046] Growth mode 3: The coverage always remains high without mutation, indicating a non-updated mode.
[0047] Step 5: Determine the initial identification year of the forest belt based on different growth pattern characteristics and construct a forest age calculation method
[0048] For growth pattern 1, the year of sudden change was located by first-order difference analysis, and the year H with the first FCFS ≥ 0.15 after the sudden drop was found, and the forest age was 2021–(H–3);
[0049] For growth model 2, the year H that first meets FCFS ≥ 0.15 is located in FCFS, and the FCFS of the first three years is < 0.15, and the forest age is also 2021–(H–3);
[0050] For growth pattern three, there is no mutation point, and it is directly judged that the planting period is more than 40 years.
[0051] Step 6: Mapping forest belt growth patterns and stand age:
[0052] This algorithm was applied to 8422 forest belts in the region, and the forest belt ages were output to generate forest age spatial distribution maps and growth pattern maps ( Figure 5 Compared with 357 field verification samples, the overall recognition accuracy reached 82.4% when the allowable error was 3 years; among different age groups, the recognition accuracy of 1-3 year old forest belts was the highest (95%), while the accuracy of forest belts over 30 years old was relatively low ( Figure 6 In order to further evaluate the extraction effect of the proposed method on the age of shelterbelts in different age groups, the experiment used the overall accuracy, misclassification error, and omission error in the error matrix to evaluate the accuracy of age extraction. The results are shown in Table 1:
[0053] Table 1 Confusion matrix between extracted age and true age.
[0054]
Claims
1. A remote sensing identification method for the age of shelterbelts taking into account growth patterns, characterized in that: The following steps are involved: 1) Acquire and preprocess long-term Landsat remote sensing images: Obtain remote sensing images for months that meet monitoring requirements year by year, perform data preprocessing such as radiometric calibration, atmospheric correction, and geometric correction, and calculate the Normalized Difference Vegetation Index (NDVI) annually based on this data. 2) Constructing a time series forest belt coverage waveform curve: Applying a pixel binary model to invert the shelterbelt coverage, extracting the effective forest belt coverage waveform for each forest belt, and constructing a time series forest belt coverage waveform curve (FCFS); 3) Outlier removal and smoothing of time series coverage waveforms: By setting a standard deviation threshold, outliers are removed and smoothed on the forest belt coverage waveform curve to extract the time series coverage that represents the characteristics of the forest belt; 4) Determining forest belt growth patterns based on changes in forest belt cover waveforms over time: Analyzing forest belt cover characteristics and comprehensively considering the planting patterns and growth characteristics of shelterbelts, farmland shelterbelts were classified into three types of forest belt growth patterns based on changes in cover: continuously renewed, late additions, and never renewed. 5) Determine the initial identification year of forest belts based on different growth pattern characteristics and develop a forest age calculation method: Set an initial identification threshold for coverage, combine the initial identifiable years with remote sensing, and develop a forest belt age discrimination algorithm based on different growth patterns to calculate the forest belt age; 6) Forest belt growth pattern and stand age mapping: Output the growth pattern type of each shelterbelt and its corresponding stand age or stand age range.
2. The remote sensing identification method for the age of a shelterbelt taking into account growth patterns according to claim 1, characterized in that: The radiometric calibration described in step 1) converts the sensor brightness value (DN) to reflectance by using the gain and offset parameters in the metadata. The atmospheric correction uses the FLAASH algorithm to convert the top of the atmosphere reflectance (TOA) to the surface reflectance (SR). The satellite images with spatial position offsets between different years are geometrically corrected.
3. The remote sensing identification method for farmland shelterbelt age taking into account growth pattern according to claim 1, characterized in that: The calculation formula of the normalized difference vegetation index described in step 1) is: NDVI = (NIR–R) / (NIR+R), where NIR is the reflectance value of the near-infrared band and R is the reflectance value of the red light band.
4. The remote sensing identification method for farmland shelterbelt age taking into account growth pattern according to claim 1, characterized in that: The formula for calculating the coverage (FC) of each remote sensing image based on NDVI year by year using the pixel binary model in step 2) is: FC=(NDVI veg –NDVI cropland ) / (NDVI mixed –NDVI cropland ), where NDVI cropland NDVI is the end member value of the cultivated land area that has not changed for a long time. veg is the NDVI end member value of the patchy poplar forest.
5. The remote sensing identification method for farmland shelterbelt age taking into account growth pattern according to claim 1, characterized in that: As described in step 2), the maximum coverage waveform in the neighborhood perpendicular to the forest belt should be obtained for each forest belt, and valid pixels should be screened based on the standard deviation (SD) of the forest belt waveform. If the standard deviation SD of the waveform sequence is less than 0.05, it means that the forest belt structure is complete and all coverage values of the forest belt are valid; otherwise, only the coverage values higher than the average value are retained, and the mean of the effective coverage values is calculated to obtain the coverage value (FCFS) representing the characteristics of the forest belt, and a time series coverage waveform curve is constructed.
6. The remote sensing identification method for shelterbelt age taking into account growth patterns according to claim 1, characterized in that: The smoothing rule of the FCFS waveform curve in step 3) is: for abnormal peaks (FCFS ij > FCFS ij–1 ANDFCFS ij > FCFS ij+1 ) or valley value (FCFS ij < FCFS ij–1 AND FCFS ij < FCFS ij+1 ), where i is the forest belt number and j is the year. If the fluctuation range is less than the fluctuation threshold, it is replaced by the mean of the adjacent years (j–1, j+1) (FCFS ij–1 +FCFS ij+1 ) / 2), the fluctuation threshold can be determined by statistically analyzing the distribution of coverage change values of forest belts of different ages at one-year intervals.
7. The remote sensing identification method for shelterbelt age taking into account growth patterns according to claim 1, characterized in that: The FCFS waveform curve characteristics corresponding to the growth pattern 1 described in step 4) are: a sudden drop occurs in a certain year, then it is lower than the detectable threshold for several years, and then it shows an upward trend; the FCFS waveform curve characteristics corresponding to the growth pattern 2 are: it is continuously lower than the detectable threshold for a certain period of time, and then it shows a slow upward trend in the later period without a sudden drop; the FCFS waveform curve characteristics corresponding to the growth pattern 3 are: since the remote sensing image can be detected, it has always been higher than the detectable threshold without a sudden drop. The method for determining the year of sudden drop is to perform first-order difference calculation on the coverage curve, and the formula is: ΔFCFS j = FCFS j –FCFS j+1 , if ΔFCFS j > 0.1, it is determined that a deforestation event has occurred in the forest belt.
8. The remote sensing identification method for farmland shelterbelt age taking into account growth pattern according to claim 1, characterized in that: The identification threshold in step 5) can be determined by the initial appearance of forest belts aged 2-4 years in remote sensing images and the distribution of coverage values of cultivated land in the same period. The final coverage threshold is set to 0.15 as the standard for whether farmland shelterbelts are identifiable in remote sensing images.
9. The remote sensing identification method for shelterbelt age taking into account growth patterns according to claim 1, characterized in that: The forest belt age discrimination algorithm described in step 5) is constructed based on different growth patterns as follows: the initial year of time series image monitoring is defined as A, the monitoring end year is defined as B, and the year when the coverage first meets the recognition threshold is defined as H. In order to avoid inaccurate positioning of the H year due to abnormal coverage curve, such as interference from the reflectivity of adjacent objects; a restriction condition should also be imposed on H, H>j, where j is the mutation point year; the planting year of the forest belt is H-3, and FCFS is significant only 3 years after planting, so it is necessary to push forward 3 years to obtain the planting year; the calculation method of the forest belt age T corresponding to different growth patterns is: Growth pattern 1: T=B-(H –3), for the forest belt coverage curve of this growth pattern, if multiple mutation points are detected, the age T is calculated based on the year H corresponding to the mutation point closest to the monitoring cutoff year; Growth pattern 2: The calculation formula for the forest belt age T of growth pattern 2 is the same as that of growth pattern 1. The difference is that for the forest belt coverage curve of growth pattern 2, there is no mutation point, so the year H of the recognition threshold can be directly located from low years to high years according to the coverage curve; Growth pattern 3: For the forest belt coverage curve of growth pattern 3, the forest belt coverage value is always greater than the recognition threshold, so the forest age T is directly determined to be greater than B–A+3.