Carbon storage calculation method based on evidence theory fusion of multi-source remote sensing data

Abstract

The present application relates to a carbon storage calculation method based on evidence theory fusion of multi-source remote sensing data, extracts band data of Sentinel and Landsat optical remote sensing image data, calculates NDVI value; uses classification samples and is based on band data and NDVI value, utilizes support vector machine model to carry out target area land classification, obtains target area Sentinel optical remote sensing image data land classification result and Landsat optical remote sensing image data land classification result; cloud is removed, and the land classification result of one kind of optical remote sensing image data classified as cloud is replaced with the land classification result of another kind of non-cloud optical remote sensing image data similar in time; evidence theory is used to fuse Sentinel and Landsat optical remote sensing image data land classification result; based on InVEST model, the obtained target area land classification result and the collected carbon density data are used to calculate the carbon storage of the target area. The present application can obtain accurate land classification result data through cloud removal, so as to more accurately calculate the carbon storage of the target area.

Description

Technical Field

[0001] This invention relates to the field of carbon storage calculation methods, and in particular to a carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data. Background Technology

[0002] Ecosystems such as forests, wetlands, and land are important carbon sinks, absorbing large amounts of carbon dioxide and storing it in vegetation and soil, thus becoming significant sources of carbon storage in their respective regions. Increases or decreases in regional carbon storage directly affect a region's greenhouse gas absorption capacity; therefore, changes in regional carbon storage are a crucial indicator for assessing climate change. By calculating carbon storage and monitoring changes in regional carbon storage, we can assess the impacts of climate change on the carbon cycle, the health of ecosystems, biodiversity, and sustainable land use, providing guidance for the rational management and protection of natural resources.

[0003] Early methods for calculating carbon storage included direct measurement, indirect estimation, and model algorithms. Direct measurement determines carbon storage by collecting samples and performing chemical analysis. This method typically requires sampling at different locations and laboratory analysis. Advantages include high accuracy and the ability to provide precise carbon storage data; however, it requires significant time and resources, making it impractical for large-scale studies or long-term monitoring. Indirect estimation infers carbon storage based on statistical models and ground observation data. This method typically uses ground observation data such as vegetation parameters and soil characteristics, combined with statistical models for estimation. It is relatively low-cost and efficient, but its accuracy is limited by model selection and the reliability of ground observation data, introducing some uncertainty. Model algorithms simulate the carbon cycle process through mathematical models to infer carbon storage. These models can estimate carbon storage at different scales and time periods and predict future changes in carbon storage. However, the accuracy of model algorithms depends on the model's structure and parameter selection, requiring calibration and validation for different regions and ecosystems. All three methods require substantial human and material resources and are mostly limited to relatively small areas.

[0004] Compared to the limitations of the methods mentioned above, the current monitoring method, which uses optical remote sensing imagery to acquire land classification data (land classification typically includes categories such as arable land, forest land, grassland, water areas, construction land, and unused land) and calculates carbon storage using the InVEST model, can cover a wide range of areas, including relatively remote and inaccessible regions. This monitoring method has the capability to assess carbon storage over large areas. In this method, the optical remote sensing imagery data is usually single-source remote sensing data provided by a single satellite such as Sentinel or Landsat. Satellite-acquired optical remote sensing images can capture surface details, thus more accurately distinguishing land features for land classification. A complete optical remote sensing image contains spectral information across multiple bands, from visible light to near-infrared, and even thermal infrared. Different land features exhibit different spectral characteristics in different bands. By capturing the spectral characteristics of each type of land feature, high-precision classification of land features can be achieved, resulting in accurate land classification data. The InVEST model (Integrated Valuation of Ecosystem Services and Tradeoffs) is a comprehensive assessment model for the evaluation of ecosystem service functions. Its carbon storage module assesses the carbon storage and sequestration functions of ecosystems. This module calculates the carbon storage or carbon sequestration over a specific time period using land classification data and carbon density data from carbon pools. The module uses four basic carbon pools: aboveground biomass, belowground biomass, soil, and dead organic matter. By inputting the carbon densities of these pools into the model, the carbon storage of each pool can be calculated. The model uses the smallest grid cell of land in each target area as the evaluation unit, and calculates the total carbon storage for each evaluation unit and the entire target area by summing the carbon storage of each pool. The InVEST model's carbon storage module utilizes land classification data and carbon density data from carbon pools to calculate the total carbon storage of the target area, providing a reliable basis for assessing the carbon storage and sequestration functions of ecosystems.

[0005] However, the aforementioned methods acquire single-source, single-frame remote sensing imagery data from a single satellite at a specific time. When calculating carbon storage for a specific date, the satellite optical remote sensing imagery for that date may be heavily obscured by clouds. To accurately identify ground features and obtain precise land classification data for the target area, cloud removal processing is necessary. Currently, the commonly used cloud removal method is filtering and noise reduction to remove the influence of cloud and fog noise. However, this method also loses a significant amount of ground feature information and weakens the reflection signals of ground features, resulting in reduced accuracy in ground feature identification, inability to obtain accurate land classification data, and ultimately, inaccurate carbon storage calculations. Summary of the Invention

[0006] The purpose of this invention is to provide a carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data, in order to solve the technical problem that existing carbon storage calculation methods based on single-source remote sensing data are prone to cloud cover when classifying land in target areas using optical remote sensing images, which makes it impossible to obtain accurate land classification data and thus accurately calculate carbon storage.

[0007] The technical problem solved by this invention can be achieved by the following solutions:

[0008] The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data includes the following steps:

[0009] Acquire Sentinel and Landsat optical remote sensing image data of the target area;

[0010] Preprocess the Sentinel and Landsat optical remote sensing image data to extract the band data from the Sentinel and Landsat optical remote sensing image data;

[0011] Calculate the NDVI values ​​of Sentinel and Landsat optical remote sensing image data;

[0012] Classification samples were drawn, and land classification of the target area was performed using the support vector machine model based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data. The land classification results of Sentinel optical remote sensing image data and Landsat optical remote sensing image data of the target area were obtained.

[0013] For cloud removal processing, if the land classification result of the acquired Sentinel optical remote sensing image data is cloud, it is replaced with the land classification result of the non-cloud Landsat optical remote sensing image data that is close in time; if the land classification result of the acquired Landsat optical remote sensing image data is cloud, it is replaced with the land classification result of the non-cloud Sentinel optical remote sensing image data that is close in time.

[0014] The land classification results of Sentinel and Landsat optical remote sensing image data are fused using evidence theory to obtain the final land classification results for the target area.

[0015] Based on the InVEST model, the carbon storage of the target area is calculated using the obtained land classification results and collected carbon density data.

[0016] Furthermore: the preprocessing of Sentinel and Landsat optical remote sensing image data to extract band data specifically includes:

[0017] Preprocess the Sentinelt optical remote sensing image data to extract data in the blue, green, red, near-infrared, and short-wave infrared bands;

[0018] Landsat optical remote sensing image data were preprocessed to extract near-shore, blue, green, red, near-infrared, and shortwave infrared band data.

[0019] Furthermore, the classification sample categories include cultivated land, forest land, grassland, water area, construction land, unused land, and cloud.

[0020] Furthermore: the use of evidence theory to fuse Sentinel and Landsat optical remote sensing image data for land classification results to obtain the final land classification results for the target area specifically includes:

[0021] Obtain the Sentinel classification probability matrix after cloud removal processing [P] s1 P s2 P s3 P s4 P s5 [P] s6 ] and Landsat classification probability matrix [P L1 [P] L2 P L3 P L4 P L5 P L6 ],

[0022] Where P s1 P s2 P s3 P s4 P s5 P s6 P represents the classification probability of Sentinel's cultivated land, forest land, grassland, water area, built-up land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Sentinel optical remote sensing image data; L1 P L2 P L3 P L4 P L5 P L6 These represent the classification probabilities of Landsat's cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Landsat optical remote sensing image data.

[0023] Calculate the normalization constant K, and calculate the evidence theory fusion classification probability P of each type of classification sample based on the normalization constant K, to obtain the evidence theory fusion classification probability matrix [P1, P2, P3, P4, P5, P6], where each element P1, P2, P3, P4, P5, P6 represents the evidence theory fusion classification probability of cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the final land classification result of the target area.

[0024] Furthermore: the expression for the normalization constant K is:

[0025]

[0026] The expression for each element in the evidence theory fusion classification probability matrix is ​​as follows:

[0027]

[0028] Furthermore: the calculation of carbon storage in the target area based on the InVEST model, using the obtained land classification results and collected carbon density data, specifically includes:

[0029] Based on the InVEST model, carbon pools of aboveground biomass, belowground biomass, soil, and dead organic matter were analyzed. Carbon density data for each carbon pool were collected according to the land classification results of the target area. The carbon storage of the target area was calculated using the following formula:

[0030] C total =C above +C below +C soil +C dead (3)

[0031] C above =S*ρ above (4)

[0032] C below =S*ρ below (5)

[0033] C soil =S*ρ soil (6)

[0034] C dead =S*ρ dead (7)

[0035] In the formula, C total C represents total carbon storage. above C below C soil C deadThese represent aboveground carbon storage, lower carbon storage, soil carbon storage, and dead organic matter carbon storage, respectively. S represents the land area of ​​the target region, and ρ represents the carbon storage of the target region. above ρ below ρ soil ρ above These represent the carbon densities of the aboveground biomass carbon pool, the underground biomass carbon pool, the soil carbon pool, and the dead organic matter carbon pool, respectively.

[0036] Further: The process of drawing classification samples involves using these samples and, based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data, employing a support vector machine model to classify land in the target area. This process yields land classification results for both the Sentinel and Landsat optical remote sensing image data. After obtaining these results, the Landsat optical remote sensing image data is resampled.

[0037] Furthermore, the sampling methods used to resample the Landsat optical remote sensing image data are nearest neighbor sampling, bilinear interpolation, and trilinear interpolation.

[0038] Furthermore: the Sentinel optical remote sensing image data is Sentinel-2A optical remote sensing image data, and the Landsat optical remote sensing image data is Landsat8 / 9 optical remote sensing image data.

[0039] The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data of the present invention acquires multi-source remote sensing data, including Sentinel and Landsat optical remote sensing imagery, for the target area when calculating carbon storage; preprocesses the Sentinel and Landsat optical remote sensing imagery to extract band data; calculates the NDVI values ​​of the Sentinel and Landsat optical remote sensing imagery; draws classification samples; and uses the classification samples and based on the band data and NDVI values ​​of the Sentinel and Landsat optical remote sensing imagery, performs land classification of the target area using a support vector machine model to obtain the land classification results of the Sentinel optical remote sensing imagery for the target area. The land classification results of Landsat optical remote sensing image data are used. During cloud removal, if the land classification result of the acquired Sentinel optical remote sensing image data is cloudy, it is replaced with the land classification result of non-cloudy Landsat optical remote sensing image data that is close in time. If the land classification result of the acquired Landsat optical remote sensing image data is cloudy, it is replaced with the land classification result of non-cloudy Sentinel optical remote sensing image data that is close in time. The land classification results of multi-source remote sensing data, including Sentinel and Landsat optical remote sensing image data, are fused using evidence theory to obtain the final land classification result of the target area. Based on the InVEST model, the carbon storage of the target area is calculated using the obtained land classification results of the target area and the collected carbon density data.

[0040] This calculation method acquires multi-source remote sensing data, including both Sentinel and Landsat optical remote sensing imagery, from multiple satellites. When cloud obstruction occurs in one type of satellite optical remote sensing imagery, cloud removal processing is performed. The land classification result of the optical remote sensing imagery data classified as cloud is replaced with the land classification result of another type of non-cloud optical remote sensing imagery data from a similar time period, thus resolving the cloud interference problem. Then, based on evidence theory, the classification results are fused to obtain the final land classification result for the target area. The carbon storage of the target area is calculated using the InVEST model based on this land classification result. Compared to filtering and denoising methods, the carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data in this invention does not lose ground feature information or weaken the reflection signal of ground features while removing clouds, thus improving the accuracy of ground feature identification and obtaining more accurate land classification result data. Using the land classification result data, the carbon storage of the target area can be calculated more accurately. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of the carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data according to the present invention;

[0043] Figures 2-5 This invention provides a carbon storage data map of the target area calculated using a carbon storage calculation method based on evidence theory and multi-source remote sensing data. Detailed Implementation

[0044] To make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0045] Example

[0046] This embodiment provides a method for calculating carbon storage based on evidence theory and fusing multi-source remote sensing data, such as... Figure 1 As shown, the calculation method includes the following steps:

[0047] S1. Acquire Sentinel and Landsat optical remote sensing image data of the target area;

[0048] S2. Preprocess the Sentinel and Landsat optical remote sensing image data and extract the band data from the Sentinel and Landsat optical remote sensing image data.

[0049] S3. Calculate the NDVI values ​​of Sentinel and Landsat optical remote sensing image data;

[0050] S4. Draw classification samples. Using the classification samples and based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data, use the support vector machine model to classify the land in the target area and obtain the land classification results of Sentinel optical remote sensing image data and Landsat optical remote sensing image data of the target area.

[0051] S5. Cloud removal: If the land classification result of the acquired Sentinel optical remote sensing image data is cloud, then replace it with the land classification result of the non-cloud Landsat optical remote sensing image data that is close in time; if the land classification result of the acquired Landsat optical remote sensing image data is cloud, then replace it with the land classification result of the non-cloud Sentinel optical remote sensing image data that is close in time.

[0052] S6. Use evidence theory to fuse the land classification results of Sentinel and Landsat optical remote sensing image data to obtain the final land classification results for the target area;

[0053] S7. Based on the InVEST model, calculate the carbon storage of the target area using the obtained land classification results and collected carbon density data.

[0054] This embodiment acquires multi-source remote sensing data, including both Sentinel and Landsat optical remote sensing image data, through Sentinel and Landsat multi-source satellites. When a certain type of satellite optical remote sensing image is obscured by clouds, cloud removal processing is performed. The land classification result of the optical remote sensing image data classified as cloud is replaced with the land classification result of another type of non-cloud optical remote sensing image data that is close in time, thereby solving the problem of cloud interference.

[0055] Furthermore, in step S1, the Sentinel optical remote sensing image data is Sentinel-2A optical remote sensing image data, and the Landsat optical remote sensing image data is Landsat8 / 9 optical remote sensing image data.

[0056] If only a single Sentinel and Landsat optical remote sensing image of the target area at a specific time is acquired, then only the carbon storage at that specific time can be obtained based on the single image data. This is insufficient to assess the long-term dynamic changes in carbon storage or monitor its temporal variations. To assess the long-term dynamic changes in carbon storage and monitor its temporal variations, further, in step S1, time-series Sentinel and Landsat optical remote sensing image data of the target area are acquired. Then, in step S2, the time-series Sentinel and Landsat optical remote sensing image data are preprocessed to extract the band data. Finally, in step S3, the NDVI values ​​of the time-series Sentinel and Landsat optical remote sensing image data are calculated. In step S4, classification samples are drawn. Using these samples and based on the band data and NDVI values ​​of the time-series Sentinel and Landsat optical remote sensing image data, a support vector machine model is used to classify the land in the target area, obtaining the land classification results of the time-series Sentinel and Landsat optical remote sensing image data for the target area. In step S5, if the land classification result of the obtained time-series Sentinel optical remote sensing image data is cloud at a certain moment, it is replaced with the non-cloud Landsat optical remote sensing image data land classification result at a similar time. In step S6, land classification results from temporal Sentinel and Landsat optical remote sensing image data are fused using evidence theory to obtain the final temporal land classification results for the target area. In step S7, based on the InVEST model, the temporal carbon storage of the target area is calculated using the obtained temporal land classification results and collected carbon density data.

[0057] Furthermore, in step S1, the time span of the time-series Sentinel optical remote sensing image data and the time-series Landsat optical remote sensing image data is more than three years. Sentinel and Landsat optical remote sensing image data that are close in time (difference of a few days) are grouped into a set of image data, with the time interval between each set of image data being more than 3 months. Furthermore, within the same set, the Sentinel and Landsat optical remote sensing image data differ in time by 3 to 5 days. Furthermore, the time interval between each set of image data is 3 to 6 months. This embodiment groups Sentinel and Landsat optical remote sensing image data that are close in time (difference of a few days) into a set of image data, and constructs multiple sets of image data based on time series (with the time interval between each set of image data being more than 3 months). These sets of image data are then applied to the carbon storage calculation method proposed in this embodiment, thereby obtaining the carbon storage data of the target area over time series (at each time node).

[0058] This embodiment's calculation method acquires multi-source remote sensing data, including both Sentinel and Landsat optical remote sensing imagery, from multiple satellites. When cloud obstruction occurs in one type of satellite optical remote sensing imagery, cloud removal processing is performed. The land classification result of the cloud-classified optical remote sensing imagery is replaced with the land classification result of another non-cloud optical remote sensing imagery from a similar time period (the two types of imagery differ by several days), thus resolving the cloud interference issue. Then, based on evidence theory, the classification results are fused to obtain the final land classification result at a specific time point. The land classification result at this time point is then used to calculate the carbon storage data of the target area at that time point using the InVEST model. By constructing multiple sets of imagery data in a time series (with time intervals of more than 3 months between each set), and applying each set of images to the above process, the carbon storage data of the target area in the time series (at each time point) can be obtained.

[0059] Furthermore, in step S2, the Sentinel and Landsat optical remote sensing image data are preprocessed to extract the band data of the Sentinel and Landsat optical remote sensing image data, specifically including:

[0060] Preprocess the Sentinelt optical remote sensing image data to extract data in the blue, green, red, near-infrared, and short-wave infrared bands;

[0061] Landsat optical remote sensing image data were preprocessed to extract near-shore, blue, green, red, near-infrared, and shortwave infrared band data.

[0062] Furthermore, when the Sentinel optical remote sensing image data is Sentinel-2A, it is L2A level data and does not require radiometric calibration or atmospheric correction. When the Landsat optical remote sensing image data is Landsat 8 / 9, it is L1TP level data and requires radiometric calibration and atmospheric correction.

[0063] Furthermore, the preprocessing of Sentinel-2A optical remote sensing image data includes extracting data from the blue (B2), green (B3), red (B4), near-infrared (B8) bands and short-wave infrared (B11 and B12) bands, followed by cropping the study area. Furthermore, the accuracy of the blue (B2), green (B3), red (B4), and near-infrared (B8) band data is 10m, and the accuracy of the short-wave infrared (B11 and B12) band data is 20m. Furthermore, the short-wave infrared (B11 and B12) band data is resampled to an accuracy of 10m.

[0064] Furthermore, the preprocessing of Landsat 8 / 9 optical remote sensing image data includes radiometric calibration, atmospheric correction, extraction of near-shore (coastal aerosol), blue, green, red, near infrared, and shortwave infrared (swir1 and swir2) band data, and cropping of the study area. Furthermore, the accuracy of the near-shore (coastal aerosol), blue, green, red, near infrared, and shortwave infrared (swir1 and swir2) band data is 30m.

[0065] Furthermore, in step S3, after calculating the NDVI values ​​of the Sentinel and Landsat optical remote sensing image data, the NDVI values ​​of the Sentinel and Landsat optical remote sensing image data are combined with their respective band data, so that the NDVI values ​​also become the data basis for subsequent land classification of the target area.

[0066] Specifically, the NDVI value is calculated using the formula: NDVI = (NIR - R) / (NIR + R), where NIR represents near-infrared band data and R represents red band data. NDVI is an index used to assess vegetation cover, calculated using the reflectance of the red and near-infrared bands. Combining NDVI values ​​with band data directly extracted from optical remote sensing imagery can provide richer spectral information. Vegetation in optical remote sensing images often possesses unique spectral characteristics; combining NDVI values ​​allows for better capture and utilization of these characteristics, helping to distinguish different land cover categories and thus improving land classification accuracy and enabling more precise land classification of target areas.

[0067] Furthermore, in step S4, classification samples are drawn, and using the classification samples and based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data, a support vector machine model is used to classify the land in the target area, and the land classification results of Sentinel optical remote sensing image data and Landsat optical remote sensing image data of the target area are obtained.

[0068] Furthermore, the classification sample categories include cultivated land, forest land, grassland, water area, construction land, unused land, and cloud. Furthermore, the separability between the classification samples is greater than 1.9. Furthermore, the separability is represented by the Jeffries-Matusita distance. The Jeffries-Matusita distance is a common statistical indicator used to measure the difference between two probability density functions.

[0069] Furthermore, in step S5, the cloud removal process is performed on each group of image data from step S1. To ensure the substitutability of land classification results between Sentinel and Landsat optical remote sensing image data, furthermore, when selecting image data, the land classification results of the same group of Sentinel and Landsat optical remote sensing image data are not simultaneously classified as cloud. That is, when the land classification result of Sentinel optical remote sensing image data within the same group is cloud, the land classification result of Landsat optical remote sensing image data is not cloud; when the land classification result of Landsat optical remote sensing image data within the same group is cloud, the land classification result of Sentinel optical remote sensing image data is not cloud. When both Sentinel and Landsat optical remote sensing image data in the same group show cloud as the land classification result, the time interval between the Sentinel and Landsat optical remote sensing image data in the same group can be increased or decreased to reselect the data until the land classification result of the Sentinel optical remote sensing image data is non-cloud or the land classification result of the Landsat optical remote sensing image data is non-cloud.

[0070] Furthermore, in step S6, the land classification results of Sentinel and Landsat optical remote sensing image data are fused using evidence theory to obtain the final land classification results for the target area, specifically including:

[0071] Obtain the Sentinel classification probability P after cloud-free processing. s and the composition of P s Sentinel classification probability matrix [P] s1 P s2 P s3 P s4 P s5 P s6 [Obtain the Landsat classification probability P after cloud removal processing] L and the composition of P L Landsat classification probability matrix [P] L1 P L2 P L3 P L4 P L5 P L6 ].

[0072] Where P s1 P s2 P s3 P s4 P s5 P s6P represents the classification probability of Sentinel's cultivated land, forest land, grassland, water area, built-up land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Sentinel optical remote sensing image data; L1 P L2 P L3 P L4 P L5 P L6 These represent the classification probabilities of Landsat's cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Landsat optical remote sensing image data.

[0073] Calculate the normalization constant K, and calculate the evidence theory fusion classification probability P of each type of classification sample based on the normalization constant K. Obtain the evidence theory fusion classification probability matrix [P1, P2, P3, P4, P5, P6] that makes up the evidence theory fusion classification probability P, where each element P1, P2, P3, P4, P5, P6 represents the evidence theory fusion classification probability of cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the final land classification result of the target area.

[0074] Table 1 shows the correspondence between the classification probabilities of Sentinel classification samples, Landsat classification samples, and evidence theory fusion classification samples.

[0075] Table 1. Correspondence between classification probabilities

[0076]

[0077] The expression for the normalization constant K is:

[0078]

[0079] The expression for each element in the evidence theory fusion classification probability matrix is ​​as follows:

[0080]

[0081] To maintain consistent resolution between the Sentinel and Landsat optical remote sensing image data and facilitate the replacement of land classification results from the optical remote sensing image data, the following step is further included after step S5 and before step S6: resampling the Landsat optical remote sensing image data. Specifically, the Landsat optical remote sensing image data is sampled to the same resolution as the Sentinel optical remote sensing image data, which is 10m. Furthermore, the resampling method for the Landsat optical remote sensing image data is one of nearest neighbor sampling, bilinear interpolation, or trilinear interpolation.

[0082] Furthermore, in step S7, based on the InVEST model, the carbon storage of the target area is calculated using the obtained land classification results and collected carbon density data. Specifically, this includes:

[0083] Based on the InVEST model, carbon pools of aboveground biomass, belowground biomass, soil, and dead organic matter were analyzed. Carbon density data for each carbon pool were collected according to the land classification results of the target area. The carbon storage of the target area was calculated using the following formula:

[0084] C total =C above +C below +C soil +C dead (3)

[0085] C above =S*ρ above (4)

[0086] C below =S*ρ below (5)

[0087] C soil =S*ρ soil (6)

[0088] C above =S*ρ dead (7)

[0089] In the formula, C total C represents total carbon storage. above C below C soil C dead These represent aboveground carbon storage, lower carbon storage, soil carbon storage, and dead organic matter carbon storage, respectively. S represents the land area of ​​the target region, and ρ represents the carbon storage of the target region. above ρ below ρ soil ρ deadThese represent the carbon densities of the aboveground biomass carbon pool, the underground biomass carbon pool, the soil carbon pool, and the dead organic matter carbon pool, respectively.

[0090] The final calculated total carbon storage data for the target area is as follows: Figure 2-5 As shown, where Figure 2 To calculate the total carbon storage data for the target region as of April 28, 2020, Figure 3 To calculate the total carbon storage data for the target region as of August 2, 2021, Figure 4 To calculate the total carbon storage data for the target region as of July 7, 2022, Figure 5 The total carbon storage data for the target region as of July 17, 2023, is calculated.

[0091] This embodiment's carbon storage calculation method, based on evidence theory and fusion of multi-source remote sensing data, acquires multi-source remote sensing data including both Sentinel and Landsat optical remote sensing imagery from Sentinel and Landsat satellites. When cloud obstruction occurs in one type of satellite optical remote sensing imagery, cloud removal processing is performed. The land classification result of the cloud-classified optical remote sensing imagery is replaced with the land classification result of the non-cloud optical remote sensing imagery from a similar time period, thus resolving the cloud interference problem. Compared to filtering and denoising methods, this calculation method does not lose ground feature information or weaken the reflection signal of ground features while removing clouds, improving the accuracy of ground feature identification and obtaining more accurate land classification results. Using the land classification results, the carbon storage of the target area can be calculated more accurately.

[0092] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data, characterized by: The steps include the following: Acquire Sentinel and Landsat optical remote sensing image data of the target area; Preprocess the Sentinel and Landsat optical remote sensing image data to extract the band data from the Sentinel and Landsat optical remote sensing image data; Calculate the NDVI values ​​of Sentinel and Landsat optical remote sensing image data; Classification samples were drawn, and land classification of the target area was performed using the support vector machine model based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data. The land classification results of Sentinel optical remote sensing image data and Landsat optical remote sensing image data of the target area were obtained. For cloud removal processing, if the land classification result of the acquired Sentinel optical remote sensing image data is cloud, it is replaced with the land classification result of the non-cloud Landsat optical remote sensing image data that is close in time; if the land classification result of the acquired Landsat optical remote sensing image data is cloud, it is replaced with the land classification result of the non-cloud Sentinel optical remote sensing image data that is close in time. The land classification results of Sentinel and Landsat optical remote sensing image data are fused using evidence theory to obtain the final land classification results for the target area. Based on the InVEST model, the carbon storage of the target area is calculated using the obtained land classification results and collected carbon density data. The classification sample categories include cultivated land, forest land, grassland, water area, construction land, unused land, and cloud; The method of using evidence theory to fuse Sentinel and Landsat optical remote sensing image data to obtain the final land classification results for the target area specifically includes: Obtain the Sentinel classification probability matrix after cloud-free processing. and the Landsat classification probability matrix , in , , , , , These represent the classification probabilities of Sentinel's cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Sentinel optical remote sensing image data. , , , , , These represent the classification probabilities of Landsat's cultivated land, forest land, grassland, water area, construction land, and unused land, respectively. The classification sample with the highest probability value is the land classification result of the Landsat optical remote sensing image data. Calculate the normalization constant According to the normalization constant Calculate the evidence theory fusion classification probability of samples of each class. Obtain the evidence theory fusion classification probability matrix Each element , , , , , The evidence theory fusion classification probability is used to represent the classification samples of cultivated land, forest land, grassland, water area, construction land and unused land respectively. The classification sample with the highest probability value is the final land classification result of the target area.

2. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data according to claim 1, characterized in that: The preprocessing of Sentinel and Landsat optical remote sensing image data to extract band data specifically includes: Preprocess the Sentinelt optical remote sensing image data to extract data in the blue, green, red, near-infrared, and short-wave infrared bands; Landsat optical remote sensing image data were preprocessed to extract near-shore, blue, green, red, near-infrared, and shortwave infrared band data.

3. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data as described in claim 1, characterized in that: The normalization constant The expression is: (1) The expression for each element in the evidence theory fusion classification probability matrix is ​​as follows: (2)。 4. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data as described in claim 1, characterized in that: The calculation of carbon storage in the target area based on the InVEST model, using the obtained land classification results and collected carbon density data, specifically includes: Based on the InVEST model, carbon pools of aboveground biomass, belowground biomass, soil, and dead organic matter were analyzed. Carbon density data for each carbon pool were collected according to the land classification results of the target area. The carbon storage of the target area was calculated using the following formula: (3) (4) (5) (6) (7) In the formula, Indicates total carbon reserves. , , , These represent aboveground carbon storage, lower carbon storage, soil carbon storage, and carbon storage from dead organic matter, respectively. Indicates the land area of ​​the target region. , , , These represent the carbon densities of the aboveground biomass carbon pool, the underground biomass carbon pool, the soil carbon pool, and the dead organic matter carbon pool, respectively.

5. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data according to claim 1, characterized in that: The process of drawing classification samples involves using these samples and, based on the band data and NDVI values ​​of Sentinel and Landsat optical remote sensing image data, employing a support vector machine model to classify land in the target area. This process yields land classification results for both Sentinel and Landsat optical remote sensing image data. After obtaining these results, the Landsat optical remote sensing image data is resampled.

6. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data as described in claim 5, characterized in that: The sampling method for resampling the Landsat optical remote sensing image data is one of the nearest neighbor sampling method, bilinear interpolation method, or trilinear interpolation method.

7. The carbon storage calculation method based on evidence theory and fusion of multi-source remote sensing data as described in claim 1, characterized in that: The Sentinel optical remote sensing image data is Sentinel-2A optical remote sensing image data, and the Landsat optical remote sensing image data is Landsat8 / 9 optical remote sensing image data.

Images