Mechanism model and machine learning coupled high-temporal-spatial-resolution photosynthetically active radiation absorption ratio inversion method

Abstract

The invention discloses a mechanism model and machine learning coupled high temporal-spatial resolution photosynthetically active radiation absorption ratio inversion method, which comprises the following steps: simulating FY3B MERSI atmospheric layer top reflectivity, and inverting clear sky FAPAR and Landsat corresponding wave band surface reflectance from the FY3B MERSI atmospheric layer top reflectivity; establishing a first Light GBM regression model, and inverting a 30-meter clear sky FAPAR value from the real Landsat surface reflectance; a second Light GBM regression model is established, and a Landsat 30m FAPAR value missing under the cloud is reconstructed; the first Light GBM regression model and the second Light GBM regression model are sequentially applied to the regional scale, and the high-temporal-spatial-resolution photosynthetically active radiation absorption ratio is estimated; according to the method, the FAPAR time sequence with the 30-meter resolution ratio can be accurately estimated.

Description

Technical Field

[0001] The invention relates to the technical field of quantitative remote sensing parameter inversion of vegetation, and in particular to a high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning. Background Art

[0002] Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) is a key biophysical parameter that characterizes the efficiency of vegetation light energy utilization. It reflects the degree of absorption of the canopy layer of solar radiation that can be used for photosynthesis (400-700nm). FAPAR is an important input parameter for estimating the net primary productivity (NPP) of ecosystems, and is also an important indicator for assessing the impact of global carbon cycle and climate change. Accurately obtaining FAPAR with high temporal and spatial resolution is of great significance for understanding the light energy utilization mechanism of terrestrial ecosystems, simulating vegetation productivity, and evaluating ecosystem service functions. At present, there are three main methods for inverting FAPAR based on satellite remote sensing data: 1) Physical methods based on radiation transfer models, such as the inversion method based on the PROSPECT-SAIL model. This type of method has a clear physical mechanism, but the calculation is complex, the efficiency is low, and it is difficult to deal with the inversion problem under complex terrain conditions; 2) Empirical methods based on vegetation indices, such as inversion models based on NDVI, EVI and other indices. This type of method is simple and easy to implement, but lacks physical mechanism support and has limited applicability under different surface conditions; 3) Machine learning-based methods, such as neural networks, random forests, etc. This type of method has strong nonlinear fitting capabilities, but often relies on a large number of training samples and is prone to overfitting problems.

[0003] Most of the existing FAPAR products are generated based on medium- and low-spatial resolution satellite data (such as MODIS, MERIS, etc.), with a spatial resolution between 250 meters and 1 kilometer. This coarse spatial resolution is difficult to meet the needs of refined ecosystem research, especially in areas with complex terrain and diverse vegetation types, and cannot accurately characterize the impact of surface heterogeneity on FAPAR. Although some studies have attempted to invert FAPAR using high-resolution data such as Landsat, they still face the following challenges: 1) Due to the influence of factors such as atmosphere, clouds, and shadows, the effective observation of high-resolution satellite data is severely limited, resulting in poor temporal continuity of the inversion results; 2) Most existing methods directly establish an empirical relationship between remote sensing observations and FAPAR, lack consideration of the physical process of radiation transfer, and the inversion accuracy in complex environments is difficult to guarantee; 3) There is a scale effect between FAPAR products of different spatial resolutions, and how to effectively use multi-source remote sensing data for high-precision inversion is still a challenge.

[0004] Therefore, it is urgent to develop new methods to combine the mechanistic nature of physical models with the high efficiency of machine learning, make full use of the advantages of multi-source remote sensing data and advanced algorithms, generate FAPAR products with high spatiotemporal resolution and good temporal continuity, and provide more reliable data support for ecosystem research. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method that couples a mechanism model with machine learning. The present invention can accurately estimate the FAPAR time series with a resolution of 30 meters, and provide important data support for applications such as terrestrial ecosystem light energy utilization research, vegetation productivity simulation and carbon cycle assessment.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning, comprising the following steps:

[0007] S1: The FY3B MERSI top-of-atmosphere reflectivity is simulated using a ground-air coupled radiative transfer model, and the clear sky FAPAR and the corresponding band surface reflectivity of Landsat are inverted from the FY3B MERSI top-of-atmosphere reflectivity using the SCE optimization algorithm;

[0008] S2: Taking the inverted Landsat corresponding band surface reflectance and solar zenith angle as input and the inverted clear sky FAPAR as output, the first LightGBM regression model is established to invert the 30-meter clear sky FAPAR value from the real Landsat surface reflectance;

[0009] S3: Taking the Landsat 30-meter clear sky vegetation photosynthetically active radiation absorption coefficient FAPAR as the output, similar pixels are found in the area centered on the pixel where the FAPAR value is located. The Landsat 30-meter clear sky FAPAR of similar pixels at the same period, the latitude, annual cumulative day, elevation, GLASS FAPAR, surface type, and environmental variables corresponding to the target FAPAR pixel are used as input to establish the second LightGBM regression model to reconstruct the missing Landsat 30-meter FAPAR value under the cloud;

[0010] S4: The first LightGBM regression model and the second LightGBM regression model were applied successively at the regional scale to estimate the photosynthetically active radiation absorption ratio with high temporal and spatial resolution.

[0011] As a further improvement of the present invention, in step S1, the ground-air coupled radiative transfer model simulating the FY3B MERSI top of atmosphere reflectivity includes the Walthall & Price soil model, PROSPECT-D and PROSPECT-VISIR leaf models, the 4SAIL canopy model, the physically based terrain correction model and the LibRadtran 2.0.3 atmospheric model;

[0012] During the model simulation, globally homogeneous representative training samples were selected, and the K-means unsupervised classification method was used to cluster the GLASSFAPAR time series. The selected samples after clustering should meet the following requirements: 1) All Landsat 30-meter pixels corresponding to a 250-meter pixel belong to the same land class; 2) Samples should be evenly selected in each MODIS tile, and the number of samples belonging to each class should be equal.

[0013] As a further improvement of the present invention, step S2 is specifically as follows:

[0014] When establishing the first LightGBM regression model, the spatiotemporal cross-validation method was used to determine the optimal model; for spatial cross-validation, the global sample points were divided into 10 groups according to spatial location, and 9 groups of training models were selected each time in ten repeated experiments, and the other group was independently validated; for temporal cross-validation, the global samples were divided into 10 groups according to month, and 9 groups of training models were selected each time in ten repeated experiments, and the other group was independently validated;

[0015] When applying the first LightGBM regression model to invert the 30-meter clear-sky FAPAR value from the real Landsat surface reflectance, the Fmask4.6 algorithm is used to obtain the Landsat cloud mask, and the model is applied only to the Landsat pixels marked as clear sky in the cloud mask to invert the Landsat 30-meter clear-sky FAPAR.

[0016] As a further improvement of the present invention, in step S3, when searching for similar pixels, a 3km×3km spatial window is defined with the target pixel as the center, and the Landsat 30-meter clear sky FAPAR of all pixels in the spatial window in a three-year time window is averaged to obtain an annual average FAPAR curve. If the effective value in the FAPAR curve is not higher than 70%, the time window is expanded to five years, and so on; the correlation coefficient between the target pixel and the annual average FAPAR curves of other pixels is calculated, and the top five pixels with the highest correlation coefficients are selected as similar pixels.

[0017] As a further improvement of the present invention, in step S3, the Landsat 30-meter clear sky FAPAR of the similar pixel within three years before and after the target pixel is selected as the spatiotemporal constraint part of the input features, and the latitude, annual cumulative day, elevation, GLASS FAPAR, surface type, and environmental variables corresponding to the target FAPAR pixel are used as input to establish a second LightGBM regression model, wherein the environmental variables include temperature, water vapor pressure difference, and solar radiation.

[0018] As a further improvement of the present invention, step S4 is specifically as follows:

[0019] For a given study area, the first LightGBM regression model is applied to invert the clear sky FAPAR from the pixels marked as clear sky by the Fmask4.6 cloud mask, and the second LightGBM regression model is applied to reconstruct the missing FAPAR under the cloud from the pixels marked as cloud or cloud shadow by the Fmask4.6 cloud mask. Finally, the Landsat 30-meter 16-day FAPAR product with seamless space and time in the study area is obtained.

[0020] The beneficial effects of the present invention are:

[0021] 1. This invention innovatively integrates physical models and machine learning methods, which not only maintains the physical mechanism of radiation transfer theory, but also makes full use of the high efficiency and nonlinear fitting ability of machine learning algorithms.

[0022] 2. The present invention effectively improves the reliability and accuracy of FAPAR inversion through multi-level data quality control and strict model verification strategy.

[0023] 3. The present invention adopts a similarity-based spatiotemporal reconstruction method, which effectively solves the data missing problem caused by cloud pollution and realizes the spatiotemporal continuity of high-resolution FAPAR products.

[0024] 4. The 30-meter resolution FAPAR product generated by the present invention can better characterize the surface heterogeneity and provide important data support for the study of ecosystem light energy utilization and carbon cycle assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] Example 1

[0028] like Figure 1 As shown in the figure, a high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method is coupled with a mechanism model and machine learning, including:

[0029] S1: The FY3B MERSI top-of-atmosphere reflectivity is simulated using a ground-air coupled radiative transfer model, and the clear sky FAPAR and the corresponding band surface reflectivity of Landsat are inverted from the FY3B MERSI top-of-atmosphere reflectivity using the SCE optimization algorithm;

[0030] S2: Taking the inverted Landsat corresponding band surface reflectance and solar zenith angle as input and the inverted clear sky FAPAR as output, the first LightGBM regression model is established to invert the 30-meter clear sky FAPAR value from the real Landsat surface reflectance;

[0031] S3: Taking the Landsat 30-meter clear sky FAPAR as the output, similar pixels are found in a 3km×3km area centered on the pixel where the FAPAR value is located. The Landsat 30-meter clear sky FAPAR of similar pixels in the same period within 3 years, the latitude, annual cumulative day, elevation, GLASS FAPAR, surface type, and environmental variables corresponding to the target FAPAR pixel are used as input to establish the second LightGBM regression model to reconstruct the missing Landsat 30-meter FAPAR value under the cloud;

[0032] S4: Apply the first and second LightGBM regression models successively at the regional scale to estimate the photosynthetically active radiation absorption ratio with high temporal and spatial resolution.

[0033] In one embodiment, step S1 further includes:

[0034] Taking the Qinghai-Tibet Plateau as an example, the terrain in this region is complex and the ecosystem types are diverse, ranging from alpine meadows to subalpine coniferous forests. First, a complete land-atmosphere coupled radiation transfer model system is established, including the Walthall&Price soil model, PROSPECT-D and PROSPECT-VISIR leaf models, 4SAIL canopy model, physically based terrain correction model and LibRadtran 2.0.3 atmospheric model.

[0035] In the training sample selection stage, the K-means clustering method was used to analyze the GLASS FAPAR time series from 2000 to 2023. It was verified that setting the number of clusters to 12 can better reflect the characteristics of different vegetation functional types. To ensure the quality of the samples, two standards were strictly implemented: one is to ensure that all 30-meter Landsat pixels under the 250-meter pixel belong to the same land type; the other is that the sample points are evenly distributed in each MODIS tile and the number of each type is equal. Finally, about 500,000 high-quality training samples were selected in the study area.

[0036] The SCE optimization algorithm is used to perform parameter inversion, with the population size set to 100, the maximum number of iterations set to 1000, and the convergence threshold set to 10. -6 During the optimization process, multiple initialization strategies are used to avoid local optimal solutions, and parallel computing is used to improve the inversion efficiency. The optimization results can simultaneously obtain the clear sky FAPAR value and the corresponding Landsat band surface reflectivity.

[0037] In one embodiment, step S2 includes:

[0038] To ensure the generalization ability of the model, a strict spatiotemporal cross-validation strategy is adopted. In the spatial dimension, the global sample points are divided into 10 groups according to their spatial locations, 9 groups are selected for model training each time, and the remaining group is used for independent validation; in the temporal dimension, the samples are divided into 10 groups according to the month, and the leave-one-out method is also used for cross-validation to ensure the applicability of the model in different seasons.

[0039] The first LightGBM model parameters were determined by grid search: the learning rate was set to 0.01, the maximum tree depth was 8, the minimum number of leaf node samples was 5, and the feature sampling ratio was 0.8. In practical applications, the Fmask4.6 algorithm was first used to detect clouds on Landsat images, and the model was applied only to pixels marked as clear sky for FAPAR inversion.

[0040] In one embodiment, step S3 includes:

[0041] The size of the spatial window is set to 3km×3km, which ensures that there are enough potential similar pixels and avoids heterogeneity caused by excessive terrain changes. For all pixels in the spatial window, the clear sky FAPAR observation values ​​within the 3-year time window are extracted to calculate the annual average FAPAR curve. If the effective observation rate of a pixel is less than 70%, the time window is extended to 5 years.

[0042] By calculating the correlation coefficient between the target pixel and the annual average FAPAR curve of other pixels in the window, the top five pixels with the highest correlation coefficient are selected as similar pixels. At the same time, various auxiliary information corresponding to these similar pixels are collected, including geographical location, topography, phenology and environment (temperature, vapor pressure difference, solar radiation) and other variables, to jointly construct the input features of the second LightGBM model.

[0043] In one embodiment, step S4 includes:

[0044] For the study area, the first LightGBM model was first applied to process the pixels marked as clear sky by Fmask4.6 to obtain the clear sky FAPAR value with a resolution of 30 meters. For the pixels marked as cloud or cloud shadow, the second LightGBM model was applied to reconstruct the FAPAR based on the historical observations of similar pixels and various auxiliary information.

[0045] In the process of model application, a partition parallel processing strategy is used to improve computational efficiency. The quality of the prediction results of each pixel is evaluated, including uncertainty analysis of model prediction and time continuity test. After strict quality control, Landsat FAPAR products with a spatial resolution of 30 meters and a temporal resolution of 16 days are finally generated for the Qinghai-Tibet Plateau region from 2000 to 2022.

[0046] The generated FAPAR products were compared with ground flux station observation data and existing low- and medium-resolution FAPAR products. The results show that the high-resolution FAPAR products generated by this method have high temporal and spatial consistency (R 2 >0.85) and good temporal continuity, which can accurately describe the spatial heterogeneity and temporal dynamic characteristics of vegetation light energy utilization under complex terrain conditions.

[0047] This embodiment can integrate the advantages of mechanism models and machine learning technology, solve the problems of insufficient physical constraints and significant cloud pollution impact when traditional methods generate high temporal and spatial resolution FAPAR products, and generate high-precision, spatiotemporally continuous 30-meter resolution FAPAR products, providing important data support for terrestrial ecosystem light energy utilization research, vegetation productivity simulation and carbon cycle assessment; integrate mechanism models and machine learning technology, make full use of the advantages of radiation transfer physical mechanisms and advanced algorithms, consider the physical constraints of FAPAR and surface heterogeneity characteristics, and generate photosynthetically active radiation absorption ratio products with high temporal and spatial resolution and a good physical basis.

[0048] Example 2

[0049] A high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning includes the following steps:

[0050] Step 1: Use the ground-air coupled radiative transfer model to simulate the FY3B MERSI top-of-atmosphere reflectivity, and use the SCE optimization algorithm to invert the clear sky FAPAR and the corresponding band surface reflectivity of Landsat from the FY3B MERSI top-of-atmosphere reflectivity;

[0051] Taking the Qinghai-Tibet Plateau in my country as an example, a complete set of ground-atmosphere coupled radiation transfer model system was first constructed. Considering the unique high-altitude environment of the Qinghai-Tibet Plateau, special attention was paid to the spatial variability of parameters such as atmospheric optical thickness and aerosol characteristics when designing the model. The model system includes the Walthall&Price soil model, PROSPECT-D and PROSPECT-VISIR leaf models, the 4SAIL canopy model, the physically based terrain correction model and the LibRadtran 2.0.3 atmospheric model. Each model has been parameter optimized and locally adjusted according to the characteristics of the plateau environment to accurately describe the radiation transfer process under complex terrain and high altitude conditions.

[0052] Based on the constructed radiation transfer model system, the SCE optimization algorithm is used to invert the FAPAR and Landsat corresponding band surface reflectance from the FY3B MERSI top of atmosphere reflectance. During the optimization process, the population number is set to 100, the maximum number of iterations is set to 1000, and the convergence threshold is set to 10 -6 To improve the inversion accuracy, the spectral matching of multiple MERSI bands and the physical rationality constraints of FAPAR are considered in the optimization objective function. The local optimal solution is avoided by multiple initialization strategies, and the MPI parallel computing framework is used to improve the computational efficiency. The inversion results include both FAPAR values ​​consistent with the MERSI spatial resolution and the simulated Landsat band surface reflectance.

[0053] In the process of selecting training samples, the GLASS FAPAR time series from 2000 to 2023 was analyzed in detail. The statistical characteristics of the time series (annual mean, seasonal changes, interannual trends, etc.) and phenological indicators (start, end, length of the growing season, etc.) were extracted, and the vegetation in the study area was divided into 12 functional categories using the K-means clustering method. To ensure the quality of the samples, strict screening criteria were established: using Global Land30 data to ensure the consistency of surface coverage of Landsat pixels within 250-meter pixels; requiring the effective observation ratio of sample points from 2000 to 2023 to be no less than 80%; ensuring the spatially balanced distribution of samples in MODIS tiles; considering uniform coverage of different altitudes, slopes and aspects. Finally, about 500,000 high-quality training sample points were screened in the Qinghai-Tibet Plateau.

[0054] Step 2: Establish the first LightGBM regression model to simulate the relationship between surface reflectance and high-resolution FAPAR inversion. Use the surface reflectance and solar zenith angle of the corresponding band of Landsat inverted in step 1 as input features, and the clear sky FAPAR value inverted by the physical model as the output label. Use the machine learning algorithm to mine the nonlinear relationship between the input features and FAPAR, and finally achieve high-precision mapping from real Landsat observation data to 30-meter resolution clear sky FAPAR values;

[0055] In the Qinghai-Tibet Plateau, a strict spatiotemporal cross-validation strategy is adopted to ensure the reliability and stability of the LightGBM model. In the spatial dimension, the sample points in the study area are divided into 10 sub-areas according to the latitude and longitude grid. Samples from 9 areas are selected for model training each time, and the remaining area is used for independent verification. In the temporal dimension, the samples are divided into 10 groups according to the month, and the same cross-validation strategy is adopted to ensure that the model can adapt to the surface and atmospheric conditions in different seasons.

[0056] The optimal parameter combination of the first LightGBM model was determined by the grid search method: the learning rate was set to 0.01, the maximum tree depth was 8, the minimum number of leaf node samples was 5, and the feature sampling ratio was 0.8. During the model training process, the early stopping strategy was used to prevent overfitting, and the L1 and L2 regularization terms were introduced to control the model complexity. Feature importance analysis showed that the near-infrared band reflectance and the red band reflectance contributed the most to the FAPAR estimation, followed by the influence of the solar zenith angle.

[0057] In the actual application stage, the Fmask4.6 algorithm is first used to detect clouds in Landsat images. The applicability of this algorithm in plateau areas has been verified, and it can effectively identify clouds, cloud shadows, and snow. Only the pixels marked as clear sky are used to perform FAPAR inversion using the trained LightGBM model. To further ensure the reliability of the inversion results, a valid value range constraint based on the physical mechanism is set to screen out possible outliers.

[0058] Step 3: Based on the acquired Landsat 30-meter clear sky FAPAR, a search window is established around the target pixel to identify similar pixels. At the same time, the second LightGBM regression model is established by combining the historical observations, geographical location, terrain conditions, phenological characteristics, and environmental variables of similar pixels to achieve high-precision reconstruction of the missing FAPAR under the cloud.

[0059] Under the complex terrain conditions of the Qinghai-Tibet Plateau, the size of the spatial window is particularly important. Through experimental analysis, it is determined that 3km×3km centered on the target pixel is the most appropriate, which can not only contain enough potential similar pixels, but also ensure the relative consistency of surface conditions. For all pixels in the spatial window, the clear sky FAPAR observation sequence within the 3-year time window is extracted to calculate the annual average FAPAR curve. For pixels with less than 70% effective observations, more observation samples are obtained by gradually expanding the time window to 5 years.

[0060] Similar pixels are identified using a method based on time series similarity. The annual average FAPAR curve correlation coefficient between the target pixel and other pixels in the window is calculated, and the top five pixels with the highest correlation coefficient are selected as references. At the same time, multi-source auxiliary information of these similar pixels is collected, including latitude, annual accumulation day, elevation, slope, aspect, GLASS FAPAR, land use type, temperature, vapor pressure difference, solar radiation and other environmental variables, as input features of the second LightGBM model.

[0061] The second LightGBM model pays special attention to the maintenance of temporal continuity during training. The model structure design adopts a multi-input and multi-output form, taking into account both the observation value sequence of similar pixels and the historical observation records of the target pixel, and ensuring the temporal smoothness of the reconstruction results through the design of the loss function. The model validation adopts the holdout method, using 20% ​​of the samples for independent testing.

[0062] Step 4: Fmask4.6 is used to detect clouds in all Landsat images in the study area. For pixels marked as clear sky, the first LightGBM model is directly used for FAPAR inversion; for pixels marked as clouds or cloud shadows, the second LightGBM model is used for FAPAR reconstruction.

[0063] The study area is divided into several sub-areas according to the UTM projection zone, each sub-area is processed separately, and finally seamless mosaicking is performed. In order to ensure the reliability of the inversion results, a multi-level quality control system is established: first, the effective value range test based on the physical mechanism; second, the time continuity test to eliminate abnormal time jumps; and finally, the spatial consistency test to ensure the smoothness of the spatial transition.

[0064] The inversion results have been rigorously verified for accuracy. By comparing with the FAPAR data from ground flux observation stations, the FAPAR products with a 30-meter resolution and a 16-day time interval generated over the Qinghai-Tibet Plateau from 2000 to 2022 have achieved good accuracy (R 2 >0.85, RMSE<0.1). The product not only maintains good temporal continuity, but also can clearly depict the spatial heterogeneity of vegetation light energy utilization under complex plateau terrain conditions.

[0065] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A high spatiotemporal resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning, characterized in that: The following steps are involved: S1: The FY3B MERSI top-of-atmosphere reflectivity is simulated using a ground-air coupled radiative transfer model, and the clear sky FAPAR and the corresponding band surface reflectivity of Landsat are inverted from the FY3B MERSI top-of-atmosphere reflectivity using the SCE optimization algorithm; S2: Taking the inverted Landsat corresponding band surface reflectance and solar zenith angle as input and the inverted clear sky FAPAR as output, the first LightGBM regression model is established to invert the 30-meter clear sky FAPAR value from the real Landsat surface reflectance; S3: Taking the Landsat 30-meter clear sky vegetation photosynthetically active radiation absorption coefficient FAPAR as the output, similar pixels are found in the area centered on the pixel where the FAPAR value is located. The Landsat 30-meter clear sky FAPAR of similar pixels at the same period, the latitude, annual cumulative day, elevation, GLASS FAPAR, surface type, and environmental variables corresponding to the target FAPAR pixel are used as input to establish the second LightGBM regression model to reconstruct the missing Landsat 30-meter FAPAR value under the cloud; S4: The first LightGBM regression model and the second LightGBM regression model were applied successively at the regional scale to estimate the photosynthetically active radiation absorption ratio with high temporal and spatial resolution.

2. The high spatiotemporal resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning according to claim 1 is characterized in that: In step S1, the ground-air coupled radiative transfer model for simulating the FY3B MERSI top-of-atmosphere reflectivity includes the Walthall & Price soil model, PROSPECT-D and PROSPECT-VISIR leaf models, 4SAIL canopy model, physically based terrain correction model and LibRadtran 2.0.3 atmospheric model; During the model simulation, globally homogeneous representative training samples were selected, and the K-means unsupervised classification method was used to cluster the GLASS FAPAR time series. The selected samples after clustering should meet the following requirements: 1) All Landsat 30-meter pixels corresponding to a 250-meter pixel belong to the same land class; 2) Samples should be evenly selected in each MODIS tile, and the number of samples belonging to each class should be equal.

3. The high spatiotemporal resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning according to claim 1 is characterized in that: Step S2 is specifically as follows: When establishing the first LightGBM regression model, the spatiotemporal cross-validation method was used to determine the optimal model; for spatial cross-validation, the global sample points were divided into 10 groups according to spatial location, and 9 groups of training models were selected each time in ten repeated experiments, and the other group was independently validated; for temporal cross-validation, the global samples were divided into 10 groups according to month, and 9 groups of training models were selected each time in ten repeated experiments, and the other group was independently validated; When applying the first LightGBM regression model to invert the 30-meter clear-sky FAPAR value from the real Landsat surface reflectance, the Fmask4.6 algorithm is used to obtain the Landsat cloud mask, and the model is applied only to the Landsat pixels marked as clear sky in the cloud mask to invert the Landsat 30-meter clear-sky FAPAR.

4. The high spatiotemporal resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning according to claim 1 is characterized in that: In step S3, when searching for similar pixels, a 3km×3km spatial window is defined with the target pixel as the center, and the Landsat 30-meter clear sky FAPAR of all pixels in the spatial window is averaged over a three-year time window to obtain an annual average FAPAR curve. If the effective value in the FAPAR curve is not higher than 70%, the time window is expanded to five years, and so on. The correlation coefficient between the target pixel and the annual average FAPAR curves of other pixels is calculated, and the top five pixels with the highest correlation coefficients are selected as similar pixels.

5. The high temporal and spatial resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning according to claim 4 is characterized in that: In step S3, the Landsat 30-meter clear sky FAPAR of the similar pixel within three years before and after the target pixel is selected as the spatiotemporal constraint part of the input features, and the latitude, annual cumulative day, elevation, GLASS FAPAR, surface type, and environmental variables corresponding to the target FAPAR pixel are used as input to establish the second LightGBM regression model, where the environmental variables include temperature, water vapor pressure difference, and solar radiation.

6. The high spatiotemporal resolution photosynthetically active radiation absorption ratio inversion method coupled with a mechanism model and machine learning according to claim 1 is characterized in that: Step S4 is specifically as follows: For a given study area, the first LightGBM regression model is applied to invert the clear sky FAPAR from the pixels marked as clear sky by the Fmask4.6 cloud mask, and the second LightGBM regression model is applied to reconstruct the missing FAPAR under the cloud from the pixels marked as cloud or cloud shadow by the Fmask4.6 cloud mask. Finally, the Landsat 30-meter 16-day FAPAR product with seamless space and time in the study area is obtained.

Images