A method for weakening cloud interference for reconstruction of sunlight-induced chlorophyll fluorescence SIF

Abstract

The application discloses a cloud interference weakening method for sunlight-induced chlorophyll fluorescence SIF reconstruction, and comprises the following steps: 1) constructing sunlight-induced chlorophyll fluorescence SIF product data, satellite microwave vegetation index data, vegetation data and meteorological data set; 2) pre-processing the data set; 3) constructing a SIF reconstruction model based on the microwave vegetation index; and 4) training the model and further adjusting parameters to obtain an optimal SIF reconstruction model. The application uses satellite microwave observation data, vegetation and meteorological data and photosynthetically active radiation data to realize reconstruction of daily sunlight-induced chlorophyll fluorescence SIF, thereby being directly applicable to clear sky and cloudy conditions and being conducive to reconstruction of sunlight-induced chlorophyll fluorescence SIF of vegetation in a cloudy area by the satellite microwave.

Description

Technical Field

[0001] This invention relates to the field of sunlight-induced chlorophyll fluorescence (SIF) reconstruction, and specifically to a cloud interference reduction method for sunlight-induced chlorophyll fluorescence SIF reconstruction that can directly estimate the SIF under different cloud conditions. Background Technology

[0002] SIF (Spectral Intensity Reflectance) is considered a medium for vegetation photosynthesis and helps assess its photosynthetic intensity and carbon sequestration capacity. Some satellites provide SIF data retrieved from vegetation spectral reflectance. However, SIF data retrieved directly from satellite observations suffers from coarse spatial resolution, narrow swath width, large orbital intervals, and a lack of long-term time-series data accumulation.

[0003] Scholars both domestically and internationally have attempted to construct SIF models using various explanatory factors to reconstruct SIF, thus addressing the aforementioned problems in satellite SIF products to some extent. However, existing models often use optical satellite band reflectance and meteorological information as explanatory factors to reconstruct SIF. As is well known, clouds are very common in optical satellite remote sensing imagery, and optical satellites are highly sensitive to clouds, leading to a significant reduction in the reconstruction accuracy of vegetation SIF in existing models under cloud interference.

[0004] Satellite microwave remote sensing has a stronger ability to penetrate clouds, making it valuable for all-weather, multi-temporal monitoring of surface vegetation. The C-band (6.925 GHz) and X-band (10.65 GHz) of microwave vegetation optical depth (VOD) can effectively reflect vegetation information. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies by proposing a cloud interference reduction method for sunlight-induced chlorophyll fluorescence (SIF) reconstruction. The goal is to improve the accuracy of vegetation SIF reconstruction under cloud interference in complex backgrounds, achieving effective reconstruction of vegetation SIF in complex environments, thereby enhancing the application value of SIF in regulating the global carbon cycle.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0007] The cloud interference reduction method for sunlight-induced chlorophyll fluorescence SIF reconstruction of this invention is characterized by the following steps:

[0008] Step 1: Construct a dataset of sunlight-induced chlorophyll fluorescence (SIF) products, a satellite microwave vegetation index dataset, a vegetation dataset, and a meteorological dataset;

[0009] Step 1.1: Extract the regional vector of land cover type F worldwide;

[0010] Obtain land cover imagery from the IGBP dataset and determine the boundaries of global land cover type F regions from it, thereby drawing a global vector set of land cover type F regions {ROI}. f |f=0,2,...,F};where ROI f Represents the region vector for land cover type f;

[0011] Step 1.2: Construct the SIF product dataset, satellite microwave vegetation index dataset, vegetation dataset, and meteorological dataset;

[0012] The surface reflectance dataset NBAR and the surface temperature dataset LST were acquired using the MODIS remote sensing satellite.

[0013] Data set of photosynthetically active radiation (PAR) was obtained using the CERSE remote sensing satellite.

[0014] The C-band microwave vegetation optical depth dataset CVOD and the X-band microwave vegetation optical depth dataset XVOD were obtained from the Global Long-Term Microwave Vegetation Optical Depth Climate Archive (VODCA).

[0015] SIF product datasets were obtained via the OCO-2 satellite.

[0016] The global region vector set {ROI} of land cover type F is used to define the region. f |f=0,2,...,F} are superimposed on the global surface reflectance dataset NBAR, surface temperature dataset LST, photosynthetically active radiation dataset PAR, C-band microwave vegetation optical depth dataset CVOD, and X-band microwave vegetation optical depth dataset XVOD to obtain the sample dataset. in, and These are the surface reflectance, surface temperature, photosynthetically active radiation, vegetation optical depth in the C-band, vegetation optical depth in the X-band, and sunlight-induced chlorophyll fluorescence (SIF) value of the nth sample in the f-th land cover type region of the global region, respectively, where N is the number of samples in the sample dataset.

[0017] Step 2: Preprocess the sample dataset;

[0018] Step 2.1, and Abnormal data in the QA band below the threshold are set to zero, resulting in the processed surface reflectance dataset. and the processed surface temperature dataset This represents the surface reflectance of the nth sample in the region of the f-th land cover type after processing. This represents the surface temperature of the nth sample in the region of the f-th land cover type after processing;

[0019] Step 2.2: Analyze the C-band microwave vegetation optical depth dataset. X-band microwave vegetation optical depth dataset Normalization was performed to obtain the normalized C-band microwave vegetation optical depth dataset. and the normalized X-band microwave vegetation optical depth dataset in, This represents the C-band vegetation optical depth of the nth sample in the region of the f-th land cover type after normalization. The X-band vegetation optical depth represents the normalized vegetation optical depth of the nth sample in the region of the f-th land cover type.

[0020] Step 3: Construct a SIF reconstruction model based on microwave vegetation index;

[0021] The SIF predicted value of the nth sample in the region of land cover type f is obtained using equation (1).

[0022]

[0023] In equation (1), FS is the set of all decision tree models, f k This is the k-th independent decision tree model, where K is the total number of decision trees, and x... n Let n be the nth sample, and

[0024] Step 4: Train the optimal SIF reconstruction model;

[0025] Step 4.1: Construct the loss function Loss using equation (2):

[0026]

[0027] In equation (2), SIF n For x n The true value of SIF. For x n The SIF predicted value, where N is the total number of samples and L is the mean squared error loss function;

[0028] Step 4.2: Construct the objective function F using equation (3):

[0029]

[0030] In equation (3), Ω(f k ) represents the regularization term of the k-th decision tree;

[0031] Step 4.3: Optimize the parameters of the SIF reconstruction model by minimizing the objective function F, thereby obtaining the optimal SIF reconstruction model.

[0032] The present invention provides an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the cloud interference mitigation method, and the processor is configured to execute the program stored in the memory.

[0033] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program is executed by a processor to perform the steps of the cloud interference mitigation method.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. Existing models often struggle to effectively reconstruct vegetation SIF under cloud interference. However, this invention introduces the microwave vegetation index, which has strong penetrability to clouds, and achieves effective reconstruction of vegetation SIF under complex climatic backgrounds with cloud interference.

[0036] 2. Existing models do not consider the impact of water stress on vegetation SIF. However, this invention utilizes the vegetation water stress information contained in microwave vegetation optical depth (VOD) to construct an SIF reconstruction model that considers water stress, thereby improving the reconstruction accuracy of SIF and enhancing the reliability of the model. Attached Figure Description

[0037] Figure 1 This is a flowchart of a cloud interference reduction method for sunlight-induced chlorophyll fluorescence SIF reconstruction according to the present invention.

[0038] Figure 2 This is a schematic diagram illustrating the influence of VOD, PAR, LST, and NBAR on the SIF results in an embodiment of the present invention.

[0039] Figure 3 This is a schematic diagram illustrating the correlation performance of the SIF reconstruction model with and without VOD under different cloud cover conditions in an embodiment of the present invention.

[0040] Figure 4 This is a schematic diagram illustrating the correlation between the SIF reconstruction model with and without VOD and GPP under different cloud cover conditions in this embodiment of the invention. Detailed Implementation

[0041] In this embodiment, addressing the limitations of existing models in effectively reconstructing vegetation SIF under cloud disturbance and their lack of consideration for water stress in environmental impacts, a cloud disturbance attenuation method for sunlight-induced chlorophyll fluorescence SIF reconstruction is proposed. Specifically, as follows... Figure 1 As shown, the method is performed according to the following steps:

[0042] Step 1: Construct data sets for sunlight-induced chlorophyll fluorescence (SIF) products, satellite microwave vegetation index, vegetation data, and meteorological data.

[0043] Step 1.1: Extract the regional vector of land cover type F worldwide;

[0044] Obtain land cover imagery from the IGBP dataset and determine the boundaries of global land cover type F regions from it, thereby drawing a global vector set of land cover type F regions {ROI}. f |f=0,2,...,F};where ROI f This represents the region vector for the f-th land cover type.

[0045] Step 1.2: Construct datasets of sunlight-induced chlorophyll fluorescence (SIF) product data, satellite microwave vegetation index data, vegetation data, and meteorological data;

[0046] Using the MODIS remote sensing satellite website, with a global spatial range and a time range set from January to December 2017, daily data on 0.05° surface reflectance (NBAR) and 0.05° surface temperature (LST) were acquired. Using the CERSE remote sensing satellite website, with a global spatial range and a time range set from January to December 2017, daily data on 1° photosynthetically active radiation (PAR) were acquired. Using the Global Long-Term Microwave Vegetation Optical Depth Climate Archive (VODCA), daily C-band and X-band microwave vegetation optical depth (VOD) data at 0.25° were acquired from January to December 2017. Using the OCO-2 satellite, daily SIF product data at 0.05° were acquired from January to December 2017.

[0047] The global vector set of land cover type F regions {ROI f |f=0,2,...,F} are superimposed on the global surface reflectance dataset NBAR, surface temperature dataset LST, photosynthetically active radiation dataset PAR, microwave vegetation optical depth dataset CVOD, and microwave vegetation optical depth dataset XVOD to obtain the sample dataset. in, and These represent the surface reflectance, surface temperature, photosynthetically active radiation, C-band vegetation optical depth, X-band vegetation optical depth, and sunlight-induced chlorophyll fluorescence (SIF) value of the nth sample in the f-th land cover type region of the global region, respectively, where N is the number of samples in the sample dataset.

[0048] Step 2: Preprocess the sample dataset;

[0049] Step 2.1, and Abnormal data in the QA band below the threshold are set to zero, resulting in the processed surface reflectance dataset. and the processed surface temperature dataset This represents the surface reflectance of the nth sample in the region of the f-th land cover type after processing. This represents the surface temperature of the nth sample in the region of the f-th land cover type after processing.

[0050] Step 2.2: Analyze the C-band microwave vegetation optical depth dataset. X-band microwave vegetation optical depth dataset Normalization was performed to obtain the normalized C-band microwave vegetation optical depth dataset. and the normalized X-band microwave vegetation optical depth dataset in, This represents the C-band vegetation optical depth of the nth sample in the region of the f-th land cover type after normalization. This represents the vegetation optical depth in the X-band of the nth sample within the normalized land cover type f region.

[0051] Step 2.3: Resample the dataset to the same resolution;

[0052] The dataset was resampled to the same resolution using bilinear interpolation. Using this method, 1°PAR, 0.05°NBAR, 0.05°LST, and 0.05°OCO-2SIF were resampled to 0.25°.

[0053] Step 3: Construct a SIF reconstruction model based on microwave vegetation index;

[0054] Choose the existing light energy utilization efficiency (LUE) model:

[0055] SIF = PAR × fPAR chl ×FE (1)

[0056] In equation (1), PAR represents photosynthetically active radiation; fPAR chlfPAR represents the portion of photosynthetically active radiation (PAR) absorbed by chlorophyll; FE represents fluorescence efficiency. Previous studies have shown that fPAR can be estimated from surface reflectance using radiative transfer models. chl Therefore, PAR and surface reflectance can be applied to SIF reconstruction. The high correlation between SIF and PAR is limited to non-stress conditions, while drought and other environmental stresses can affect FE. In the LUE model, surface temperature LST can be used as a representative of thermal effects, and microwave vegetation optical depth VOD can be used as a representative of vegetation moisture content. By adding microwave vegetation index VOD, surface reflectance NBAR, and surface temperature LST to the existing LUE model, equation (1) is derived.

[0057] The SIF predicted value of the nth sample in the region of land cover type f is obtained using equation (2).

[0058]

[0059] In equation (2), FS is the set of all decision tree models, f k This is the k-th independent decision tree model, where K is the total number of decision trees, and x... n Let n be the nth sample, and

[0060] Step 4: Train the optimal SIF reconstruction model;

[0061] Step 4.1: Construct the loss function Loss using equation (3):

[0062]

[0063] In equation (3), SIF n For x n The true value of SIF. For x n The SIF predicted value, where N is the total number of samples and L is the mean squared error loss function;

[0064] Step 4.2: Construct the objective function F using equation (4):

[0065]

[0066] Wherein, Ω(f k ) represents the regularization term of the k-th decision tree.

[0067] Step 4.4 optimizes the parameters of the SIF reconstruction model by minimizing the objective function F, thereby obtaining the optimal SIF reconstruction model. The optimal objective function F is used to measure the quality of the k-th tree structure and can be used as a gain value judgment function for the model. Substituting the objective function F into step 2, and continuously increasing the tree model f...k The process involves using a greedy algorithm to iterate through all split points, calculating the loss value for each, and then selecting the split point with the largest gain value. The smaller the maximum gain-loss ratio, the better the model's prediction, ultimately obtaining the optimal SIF reconstruction model. The influence of VOD, PAR, LST, and NBAR on the SIF results is shown as follows: Figure 2 As shown, from Figure 2 The image shows an example of the weights of VOD, PAR, LST, and NBAR obtained after training the XGBOOST model.

[0068] Step 5: Model Evaluation;

[0069] The trained model is evaluated using the test set from step 2, with R² and RMSE used as evaluation metrics to assess the model's fit and prediction accuracy, respectively. In this example, the training results from January 1, 2017 to December 31, 2017 are: RMSE = 0.25, R² = 0.92. The SIF correlation of the method of this invention and the traditional model under increased cloud cover is selected, and the impact of increased cloud cover on the SIF correlation of the SIF reconstruction model with and without VOD is shown below. Figure 3 As shown, this invention reduces cloud sensitivity by 48% compared to traditional methods, eliminating the influence of cloud pollution. When evaluating model performance, GPP and its correlation are often used to determine performance. GPP represents the overall ability of plants to absorb and utilize light energy for growth and biomass increase. In ecology and environmental science, GPP is commonly used as an important indicator of ecosystem function, crucial for understanding ecosystem productivity, carbon cycling, and ecosystem function. The correlation between the method of this invention and traditional models and GPP under increased cloud cover is shown below, revealing the correlation between the VOD-based SIF reconstruction model and the VOD-free SIF reconstruction model and GPP under increased cloud cover. Figure 4 As shown, this method reduces cloud sensitivity by 46% compared to traditional methods.

[0070] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0071] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A method for reducing cloud interference in sunlight-induced chlorophyll fluorescence SIF reconstruction, characterized in that, The procedure is as follows: Step 1: Construct a dataset of sunlight-induced chlorophyll fluorescence (SIF) products, a satellite microwave vegetation index dataset, a vegetation dataset, and a meteorological dataset; Step 1.1: Extract the regional vector of land cover type F worldwide; Obtain land cover imagery from the IGBP dataset and determine the boundaries of global land cover type F regions from it, thereby drawing a global vector set of land cover type F regions {ROI}. f |f=0,2,...,F};where ROI f Represents the region vector for land cover type f; Step 1.2: Construct the SIF product dataset, satellite microwave vegetation index dataset, vegetation dataset, and meteorological dataset; The surface reflectance dataset NBAR and the surface temperature dataset LST were acquired using the MODIS remote sensing satellite. Data set of photosynthetically active radiation (PAR) was obtained using the CERSE remote sensing satellite. The C-band microwave vegetation optical depth dataset CVOD and the X-band microwave vegetation optical depth dataset XVOD were obtained from the Global Long-Term Microwave Vegetation Optical Depth Climate Archive (VODCA). SIF product datasets were obtained via the OCO-2 satellite. The global region vector set {ROI} of land cover type F is used to define the region. f |f=0,2,...,F} are superimposed on the global surface reflectance dataset NBAR, surface temperature dataset LST, photosynthetically active radiation dataset PAR, C-band microwave vegetation optical depth dataset CVOD, and X-band microwave vegetation optical depth dataset XVOD to obtain the sample dataset. in, and These are the surface reflectance, surface temperature, photosynthetically active radiation, vegetation optical depth in the C-band, vegetation optical depth in the X-band, and sunlight-induced chlorophyll fluorescence (SIF) value of the nth sample in the f-th land cover type region of the global region, respectively, where N is the number of samples in the sample dataset. Step 2: Preprocess the sample dataset; Step 2.1, and Abnormal data in the QA band below the threshold are set to zero, resulting in the processed surface reflectance dataset. and the processed surface temperature dataset This represents the surface reflectance of the nth sample in the region of the f-th land cover type after processing. This represents the surface temperature of the nth sample in the region of the f-th land cover type after processing; Step 2.2: Analyze the C-band microwave vegetation optical depth dataset. X-band microwave vegetation optical depth dataset Normalization was performed to obtain the normalized C-band microwave vegetation optical depth dataset. and the normalized X-band microwave vegetation optical depth dataset in, This represents the C-band vegetation optical depth of the nth sample in the region of the f-th land cover type after normalization. The X-band vegetation optical depth represents the normalized vegetation optical depth of the nth sample in the region of the f-th land cover type. Step 3: Construct a SIF reconstruction model based on microwave vegetation index; The SIF predicted value of the nth sample in the region of land cover type f is obtained using equation (1). In equation (1), FS is the set of all decision tree models, f k This is the k-th independent decision tree model, where K is the total number of decision trees, and x... n Let n be the nth sample, and Step 4: Train the optimal SIF reconstruction model; Step 4.1: Construct the loss function Loss using equation (2): In equation (2), SIF n For x n The true value of SIF. For x n The SIF predicted value, where N is the total number of samples and L is the mean squared error loss function; Step 4.2: Construct the objective function F using equation (3): In equation (3), Ω(f k ) represents the regularization term of the k-th decision tree; Step 4.3: Optimize the parameters of the SIF reconstruction model by minimizing the objective function F, thereby obtaining the optimal SIF reconstruction model.

2. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports the processor in executing the cloud interference mitigation method of claim 1, wherein the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to perform the steps of the cloud interference reduction method of claim 1.

Images