A soil moisture data fusion method and device based on bayesian correction

Abstract

This application provides a method and apparatus for soil moisture data fusion based on Bayesian correction. In the process of fusing multiple soil moisture data products, a Bayesian model with outstanding interpretability is introduced to solve the bias correction problem. On the one hand, it can clearly reveal the interaction mechanism between variables, which helps to understand the causes of systematic bias in multi-source soil moisture products. On the other hand, it can also provide the quantification of the uncertainty of the correction results, which greatly improves the reliability of the data products. In this way, the problems of spatiotemporal inconsistency and physical bias in multi-source remote sensing soil moisture products can be effectively solved, and high-quality soil moisture data products can be obtained.

Description

Technical Field

[0001] This application relates to the field of soil moisture observation, specifically to a method and apparatus for soil moisture data fusion based on Bayesian correction. Background Technology

[0002] Soil moisture is a key variable in the terrestrial water cycle and energy exchange process, and has important scientific research value in fields such as global climate change prediction, agricultural irrigation estimation, ecosystem carbon cycle and natural disaster monitoring.

[0003] In the field of microwave remote sensing, the L-band in the 1-2 GHz range (e.g., directly locking onto 1.4 GHz) has become a key band for soil moisture retrieval due to its unique physical characteristics. Compared to higher frequency bands such as the C-band, the L-band has two significant advantages: First, its longer wavelength provides stronger vegetation penetration, enabling it to penetrate medium-density vegetation canopies; second, this band exhibits higher sensitivity to changes in soil dielectric constant, allowing for precise capture of subtle changes in soil moisture content. Based on these advantages, L-band microwave remote sensing has been widely recognized as the optimal choice for current soil moisture retrieval, playing an irreplaceable role in global and regional scale monitoring of soil moisture dynamics.

[0004] However, existing L-band soil moisture products still face two major challenges in practical applications: (1) Insufficient time series length. The existing satellites have been in operation since 2015, and their observation time series are relatively short, which is difficult to meet the needs of long-term soil moisture dynamic monitoring in climate change research, especially in analyzing long-term evolution trends and the impact of extreme climate events; (2) Data quality issues. Although the relevant existing satellites have a longer observation record (from 2009 to present), their observation data are severely affected by radio frequency interference (RFI), resulting in significant uncertainty in data quality.

[0005] To address the aforementioned challenges, it is crucial to conduct research on the correction and data fusion of soil moisture products from different satellites. Such research can not only effectively integrate the advantages of both, significantly extend the time series of high-precision soil moisture products (covering from 2009 to the present), and achieve long-term continuous observation, but also, to a certain extent, compensate for the observation gaps caused by radio frequency interference, and improve the spatial coverage and integrity of the data.

[0006] In the research of data fusion methods, machine learning algorithms have shown great potential. Machine learning methods can achieve accurate prediction of soil moisture in a specific spatiotemporal range by establishing a nonlinear mapping relationship between multiple variables and target parameters.

[0007] However, the inventors of this application have discovered that traditional machine learning methods such as neural networks and random forests are "black box" in nature, and their specific processing logic is not transparent enough, making it difficult to meet the needs of higher quality soil moisture data processing in real-world situations. Summary of the Invention

[0008] This application provides a method and apparatus for soil moisture data fusion based on Bayesian correction. In the process of fusing multiple soil moisture data products, a Bayesian model with outstanding interpretability is introduced to solve the bias correction problem. On the one hand, it can clearly reveal the interaction mechanism between variables, which helps to understand the causes of systematic bias in multi-source soil moisture products. On the other hand, it can also provide the quantification of the uncertainty of the correction results, which greatly improves the reliability of the data products. In this way, the problems of spatiotemporal inconsistency and physical bias in multi-source remote sensing soil moisture products can be effectively solved, and high-quality soil moisture data products can be obtained.

[0009] Firstly, this application provides a soil moisture data fusion method based on Bayesian correction, the method comprising: The acquisition unit is used to acquire four soil moisture data products to be fused. Among the four soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing, the third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation, and the fourth soil moisture data product is a data product obtained by ground observation. The preprocessing unit is used to perform resampling and time alignment operations on the four soil moisture data products; The training unit is used to train a soil moisture data product bias correction model based on four soil moisture data products, using a first soil moisture data product (which has higher accuracy than the second soil moisture data product) constructed based on a multi-channel collaborative inversion algorithm as a benchmark. The soil moisture data product bias correction model is used to perform bias correction processing on the second soil moisture data product input into the model. Specifically, the soil moisture data product bias correction model is a Bayesian model configured under the Bayesian framework.

[0010] Secondly, this application provides a soil moisture data fusion device based on Bayesian correction, the device comprising: The acquisition unit is used to acquire four soil moisture data products to be fused. Among the four soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing, the third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation, and the fourth soil moisture data product is a data product obtained by ground observation. The preprocessing unit is used to perform resampling and time alignment operations on the four soil moisture data products; The training unit is used to train a soil moisture data product bias correction model based on four soil moisture data products, using a first soil moisture data product (which has higher accuracy than the second soil moisture data product) constructed based on a multi-channel collaborative inversion algorithm as a benchmark. The soil moisture data product bias correction model is used to perform bias correction processing on the second soil moisture data product input into the model. Specifically, the soil moisture data product bias correction model is a Bayesian model configured under the Bayesian framework.

[0011] From the above, it can be concluded that this application has the following beneficial effects: To address the goal of integrating soil moisture data products, a Bayesian model with outstanding interpretability was introduced to correct biases during the integration of multiple soil moisture data products. On the one hand, it can clearly reveal the interaction mechanism between variables, which helps to understand the causes of systematic biases in multi-source soil moisture products. On the other hand, it can also provide quantification of the uncertainty of the correction results, which greatly improves the reliability of the data products. This can effectively solve the problems of spatiotemporal inconsistency and physical bias in multi-source remote sensing soil moisture products and obtain high-quality soil moisture data products. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating a soil moisture data fusion method based on Bayesian correction, as described in this application. Figure 2 This is a schematic diagram of a soil moisture data fusion device based on Bayesian correction according to this application. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices. The naming or numbering of steps appearing in this application does not imply that the steps in the method flow must be performed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical purpose, as long as the same or similar technical effect is achieved.

[0016] The module division described in this application is a logical division. In practical applications, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual coupling, direct coupling, or communication connections may be through interfaces, and the indirect coupling or communication connections between modules may be electrical or other similar forms, none of which are limited in this application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed across multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in this application.

[0017] First, refer to Figure 1 , Figure 1 This paper illustrates a flowchart of a soil moisture data fusion method based on Bayesian correction according to this application. The soil moisture data fusion method based on Bayesian correction provided in this application may specifically include the following steps S101 to S103: Step S101: Obtain four soil moisture data products to be fused. Among the four soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing. The third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation. The fourth soil moisture data product is a data product obtained by ground observation. Understandably, the multi-soil moisture data product fusion architecture constructed in this application involves four aspects / types of soil moisture data products, so as to enable the constructed soil moisture data product deviation correction model to output high-quality soil moisture data products.

[0018] In this context, it is easy to understand that, in practice, the four soil moisture data products obtained in this application are ready-made / pre-collected soil moisture data products, such as data products provided by certain satellites / platforms under open-source / subscription services.

[0019] Of course, in specific applications, this application scheme may also be used to perform real-time autonomous data collection or trigger external systems to perform real-time data collection in accordance with the four soil moisture data product collection mechanisms.

[0020] The first soil moisture data product is a satellite soil moisture data product obtained through L-band passive microwave remote sensing. The incident angle of the L-band microwave radiometer carried by the satellite can be fixed at 40 degrees, including the ascent (6:00 pm local time) orbit and the descent (6:00 am local time) orbit, with a time resolution of 2-3 days and a spatial resolution of 36 km.

[0021] This application also specifically uses a first soil moisture data product built based on the Multi-Channel Collaborative Algorithm (MCCA), which establishes a physical parameter coupling framework for H / V dual-polarization channels. It uses the self-constraining relationship between parameters and the theoretical transformation relationship between channels to perform surface parameter inversion. The inversion process does not depend on other auxiliary data and is applicable to a variety of different load configurations. This product enhances the inversion capability of unfrozen water during the soil freezing period and improves the global coverage of soil moisture.

[0022] The second soil moisture data product is also a satellite soil moisture data product obtained through L-band passive microwave remote sensing. Its satellite is equipped with an L-band interferometric microwave radiometer, which has the ability to observe at multiple incident angles. It is mainly used for monitoring global soil moisture and sea surface salinity. It includes two orbits: ascending (6:00 am local time) and descending (6:00 pm local time). The time resolution is 2-3 days, and it is projected onto a 25km EASE-Grid 2.0 grid. EASE-Grid 2.0 is a global grid system based on the WGS84 ellipsoid with equal area, which is designed to reduce area distortion caused by projection deformation. It is particularly suitable for data processing and analysis in polar regions and globally.

[0023] Radio frequency interference is evident in this second soil moisture data product, especially in areas with high electromagnetic density. Data loss or underestimation of soil moisture and ocean salinity retrieval values ​​may occur in areas affected by radio frequency interference.

[0024] The third soil moisture data product is a data product obtained by integrating satellite observations, ground observations, and model simulations. As a satellite-era atmospheric reanalysis product, it incorporates more types of observational data for assimilation and has comprehensively updated and improved the Goddard Earth Observation System (GEOS) model and analysis scheme. Its soil temperature product covers global land areas from 1980 to the present, with a spatial resolution of approximately 50 km and a temporal resolution of hourly and monthly averages. The data has undergone quality control and long-term consistency correction and is widely used in climate research, agro-meteorology, hydrological simulation, and extreme event analysis, providing key support for research on surface energy balance and land surface processes.

[0025] The fourth soil moisture data product is a data product obtained from ground observations. Specifically, it is obtained by collecting, standardizing, and sharing soil moisture data from different ground observation networks through a global soil moisture data integration platform, providing high-quality, unified format data support for remote sensing product verification, hydrological and climate model research, etc.

[0026] In this application, soil moisture observation values ​​from the 0 to 5 cm soil layer can be used.

[0027] In order to reduce the impact of scale differences, this application can also perform spatial averaging on all data marked as "good" quality within the same pixel (where there are quality labels for the corresponding data points) in order to reduce the impact of scale differences.

[0028] Specifically, for the four types of soil moisture data products mentioned above, the design of this application specifically uses the first soil moisture data product, which has a higher accuracy (i.e., spatiotemporal resolution) than the second soil moisture data product and is constructed based on a multi-channel collaborative inversion algorithm, as a benchmark to train a soil moisture data product deviation correction model for the second soil moisture data product used to perform deviation correction processing on the input model. In terms of design direction, while ensuring that the spatiotemporal characteristics of the model's output result, i.e., the deviation-corrected second soil moisture data product, remain consistent with those of the first soil moisture data product, its coverage and other characteristics can be further improved, or its spatiotemporal resolution can be further enhanced.

[0029] Step S102: Perform resampling and time alignment operations on the four soil moisture data products; Understandably, this application may involve further preprocessing operations for the four types of soil moisture data products obtained above in order to improve their data quality and lay a good foundation for the subsequent specific model training work.

[0030] In this regard, the focus of this application may involve resampling and time alignment operations to alleviate or even completely overcome the problem of large differences in spatial scale and transit time.

[0031] Taking the third soil moisture data product as an example, in reality, the surface soil temperature has significant diurnal dynamic variation characteristics. In extreme areas, its diurnal range can reach 30-60K. Therefore, there are obvious inconsistencies between the third soil moisture data product and the first soil moisture data product in terms of spatial scale and transit time.

[0032] In the above scenario, at the operational level, the resampling operation can specifically be performed by resampling to the EASE-Grid 2.0 grid using grid linear interpolation. On the other hand, the time alignment operation here can specifically be performed using the annual average of the gridded UTC transit time (t) extracted from the first soil moisture data product. avg Using ) as the time anchor, create a ±30-minute time window to extract data from other data products at the corresponding time.

[0033] UTC stands for Coordinated Universal Time, which is a widely used global standard time.

[0034] In this way, spatial and temporal calibration can ensure that the soil moisture data used is within the same window, providing a reliable basis for subsequent processing.

[0035] Step S103: Based on the four soil moisture data products, the first soil moisture data product, which has higher accuracy than the second soil moisture data product and is constructed based on the multi-channel collaborative inversion algorithm, is used as the benchmark to train the soil moisture data product deviation correction model. The soil moisture data product deviation correction model is used to perform deviation correction processing on the second soil moisture data product input to the model. The soil moisture data product deviation correction model is specifically a Bayesian model configured under the Bayesian framework.

[0036] It is easy to see that the main purpose of the processing of soil moisture data products in this application is to optimize the second soil moisture data product by performing bias correction processing based on a Bayesian model.

[0037] Of course, this doesn't mean directly optimizing the second soil moisture data product into the second soil moisture data product. After all, there is already a readily available second soil moisture data product, so this optimization direction is meaningless.

[0038] The more specific significance of this application in terms of deviation correction is that, in the process of deviation correction of the second soil moisture data product, the data not only possesses the spatiotemporal characteristics of the first soil moisture data product that it did not originally have, but also has a better time series length and spatial coverage than the first soil moisture data product. This also corresponds to the situation in this application of building a multi-soil moisture data product fusion architecture and introducing a variety of soil moisture data products.

[0039] Based on this, the deviation correction processing performed by the soil moisture data product deviation correction model in this application can specifically involve three major processing objectives: improving spatiotemporal resolution, improving spatial coverage, and improving temporal continuity.

[0040] Furthermore, it can be seen that the bias correction model for the soil moisture data product that performs bias correction processing, under the design of this application, specifically selected a Bayesian model.

[0041] Compared to traditional "black box" machine learning methods such as neural networks and random forests, Bayesian models have a significant advantage in interpretability, which is particularly crucial for the bias correction problem that this application aims to solve: On the one hand, Bayesian methods can clearly reveal the interaction mechanisms between variables, which helps to understand the causes of systematic biases in multi-source soil moisture products; On the other hand, the posterior distribution of the Bayesian method can provide a quantification of the uncertainty of the correction results, which can greatly improve the reliability of data products.

[0042] This also shows that the output of the Bayesian model includes, in terms of specific content, the soil moisture value after bias correction and the uncertainty of the correction result.

[0043] Therefore, the interpretability and uncertainty quantification capabilities of the Bayesian method make it an ideal choice for solving the problems of spatiotemporal inconsistency and physical bias in multi-source remote sensing soil moisture products.

[0044] More specifically, the Bayesian model, or Bayesian framework, is a reasoning paradigm based on probability theory. Its core idea is to dynamically update the understanding of unknown variables (i.e., calculate the posterior distribution) by combining prior knowledge with observed data. This framework is particularly suitable for scenarios with high uncertainty, scarce data, or significant noise. However, it is highly subjective in terms of prior knowledge, and the use of inappropriate prior knowledge (such as being too broad or significantly biased) will distort the posterior results.

[0045] Although the Markov Chain Monte Carlo (MCMC) method can obtain prior knowledge through data processing, it consumes a lot of computational resources.

[0046] Gaussian Process Regression (GPR) is one of the important implementation tools of the Bayesian framework. Its kernel function and parameters determine the prior assumptions (such as smoothness and periodicity). It can directly learn complex patterns from the data without assuming a fixed function form. By maximizing the marginal likelihood and adjusting the kernel function hyperparameters, the prior assumptions are adapted to the current data, which helps to better fit the changes and trends of the dataset.

[0047] For satellite soil moisture products, there are often problems such as interference from various errors and strong spatial heterogeneity of multi-source data. Gaussian process regression models improve the accuracy of prediction data by automatically adjusting the length scale of the kernel function to match different spatial correlation ranges, and by naturally expressing uncertainty, flexibly integrating prior knowledge and having dynamic updating capabilities. This has shown significant advantages in complex problems in the field of remote sensing.

[0048] In terms of specific quantification formulas, the Bayesian framework can be expressed as follows: , in, Denotes the posterior distribution, and also indicates the distribution under a given condition. Down The probability of f is the updated understanding of f after combining the data. This represents the unknown function to be estimated. Represents observation data, This represents the likelihood function, and also represents the given... Down The probability, Denotes the prior distribution, and also represents The probability, express The probability of.

[0049] Furthermore: , , , in, Indicates input of The function (the function f here corresponds to f above). The expected value is Mean function Kernel function is Gaussian distribution process (in the formula) The wavy line that follows is the standard usage to indicate that the distribution conforms to a certain distribution. This represents the two input points of the satellite data. Indicates the noise term. This indicates a mean of 0 and a noise variance of . The normal distribution The posterior mean (representing the predicted value) is... The posterior covariance (the uncertainty of the prediction) is The Gaussian distribution.

[0050] Furthermore, regarding kernel functions within the Bayesian framework, as mentioned above... This application also has further optimized settings.

[0051] Specifically, within the Bayesian framework, kernel functions Specifically, a composite kernel function can be used, which can be represented as follows: , in, Represents the signal variance. Indicates a length measure. Used to control the smoothness of a function, i.e. The larger the value, the smoother the function. The smaller the value, the more drastic the change in the function. Indicates the distance between two points. For noise variance, Indicates when When it is 1, when The time is 0.

[0052] In Gaussian process regression models, the aforementioned composite kernel function can simultaneously model the smoothing trend of the data and observation noise. It is responsible for capturing the continuity and smoothness between input features, enabling the model to have good generalization ability. It is used to absorb random errors in the data, improve the model's robustness to noise, and automatically estimate the noise level during training.

[0053] This combination not only enhances the model's adaptability to the complexity of satellite remote sensing data, but also provides more reasonable uncertainty quantification, making it particularly suitable for practical application scenarios with potential functional relationships and measurement errors.

[0054] Meanwhile, this application will use the unknown fitting function for each grid. Treat it as a random process, assuming that the process follows a Gaussian distribution and is used as the prior distribution. The non-parametric nature of Gaussian processes means they don't make rigid assumptions about the form of the function, but rather use kernel functions to determine its form. The model implicitly models complex functional relationships, which makes it more flexible in predicting soil moisture.

[0055] Furthermore, to improve the model's processing performance, this application considers introducing color temperature as an influencing factor. In terms of details, considering that the vegetation layer absorbs, scatters, and re-emits microwave radiation, the brightness temperature received by the sensor is not only affected by the surface soil moisture state, but also closely related to the vegetation cover and water content. As the vegetation density increases, the microwave penetration ability weakens and the brightness temperature value decreases, and it becomes more sensitive to changes in the observation angle.

[0056] To this end, this application also introduces the NADI parameter, which, by calculating the ratio of the brightness temperature difference to the brightness temperature at two different incident angles under the same polarization, can highlight the characteristics of brightness temperature angle changes caused by vegetation while eliminating the influence of the absolute brightness temperature background value.

[0057] Correspondingly, the method of this application may also include: For the four soil moisture data products, corresponding first reference index data are configured. The first reference index is used as an auxiliary reference factor in the model prediction logic and is expressed as follows: , in, Indicates the primary reference indicator. Indicates brightness temperature. Indicates the polarization angle, corresponding to the same polarization conditions. and Indicates different incident angles.

[0058] At the same time, this application can also consider the fluctuations caused by the two key parameters, Vegetation Optical Depth (VOD) and Land Surface Temperature (LST). Understandably, this application believes that the attenuation effect of vegetation on microwave signals and the fluctuations in soil dielectric constant caused by dynamic changes in land surface temperature will increase the uncertainty of the soil moisture inversion process.

[0059] Correspondingly, the method of this application may also include: For the four types of soil moisture data products, corresponding second reference index data are configured. The second reference index is used as an auxiliary reference factor in the model prediction logic and includes vegetation optical thickness and surface temperature.

[0060] Understandably, given that vegetation optical thickness and surface temperature are existing parameters, the focus of this solution is to apply these two indicators from their original fields to the deviation correction work of the soil moisture data products specifically involved in this application. Therefore, the complex quantitative formulas for these parameters are not further elaborated.

[0061] Furthermore, as mentioned above, the temporal resolution of the first soil moisture data product used as a benchmark is 2-3 days, while the soil moisture data product bias correction model trained in this application can be involved in the processing objective of improving spatiotemporal resolution. Therefore, in the design of the solution in this application, the output result of the soil moisture data product bias correction model can be specifically configured as a global soil moisture data product on a daily scale.

[0062] As can be seen, the daily / day-by-day resolution is already superior to the first soil moisture data product that serves as the benchmark.

[0063] Under the aforementioned model configuration, the bias correction model for the trained soil moisture data product, based on evaluation metrics including root mean square error (RMSE), unbiased root mean square error (ubRMSE), bias, and correlation coefficient (CC), has been validated to achieve excellent correction performance and effectively eliminate systematic bias.

[0064] Within an acceptable range, it is essentially unbiased compared to the first soil moisture data product, indicating that it can accurately capture the spatiotemporal variation characteristics of the first soil moisture data. Within the error range, the product has the potential to replace the first soil moisture data product, and can also effectively supplement the observations during the missing periods of the first soil moisture data, significantly improving spatial coverage and temporal continuity, increasing the average number of effective observation days by 61 days per year, and also showing excellent performance in multi-site validation.

[0065] After completing the model configuration, it is obvious that the next step is to proceed with the related model application work.

[0066] In this regard, the method of this application may also include: The target soil moisture data for the target area within a specified time period is obtained through a soil moisture data product deviation correction model. Based on the target soil moisture data, soil moisture development events are analyzed and processed in the target area. The soil moisture development events involved include evapotranspiration estimation events or precipitation inversion events.

[0067] It is easy to understand that the model application process may involve the acquisition and processing of the current soil moisture data product (i.e., a second soil moisture data product) of the target area within a specified time period, which requires bias correction. This can be done through manual input, local extraction, remote extraction, or other specific acquisition methods.

[0068] In practice, model applications are usually initiated by related processing tasks. Correspondingly, the target soil moisture data product can also be carried directly or indirectly by the task information.

[0069] After the model is used to perform bias correction on the current soil moisture data product to obtain the corresponding target soil moisture data, the corresponding data application process can then proceed.

[0070] For example, the target soil moisture data can be stored locally or remotely, the results can be displayed, a message indicating that the processing is complete can be output, the results can be forwarded, or further data analysis can be performed. Obviously, these can be flexibly adjusted according to the pre-configured and real-time data application strategies.

[0071] Taking further data analysis as an example, based on the target soil moisture data, soil moisture development events can be analyzed and processed in the target area. These soil moisture development events specifically include evapotranspiration estimation events or precipitation inversion events.

[0072] Understandably, soil moisture development event analysis and processing can provide theoretical support for water-saving irrigation and soil remediation in agriculture, as well as provide scientific basis for coping with extreme climates such as drought and freeze-thaw cycles and improving the adaptability and sustainable management capacity of ecosystems.

[0073] In summary, regarding the above solutions, for the goal of integrating soil moisture data products, a Bayesian model with outstanding interpretability is introduced to address the bias correction problem during the integration of multiple soil moisture data products. On the one hand, it can clearly reveal the interaction mechanism between variables, which helps to understand the causes of systematic bias in multi-source soil moisture products. On the other hand, it can also provide quantification of the uncertainty of the correction results, significantly improving the reliability of the data products. In this way, the problems of spatiotemporal inconsistency and physical bias in multi-source remote sensing soil moisture products can be effectively solved, resulting in high-quality soil moisture data products.

[0074] The above is an introduction to the soil moisture data fusion method based on Bayesian correction provided in this application. In order to facilitate better implementation of the soil moisture data fusion method based on Bayesian correction provided in this application, this application also provides a soil moisture data fusion device based on Bayesian correction from the perspective of functional modules.

[0075] See Figure 2 , Figure 2 This is a schematic diagram of a soil moisture data fusion device based on Bayesian correction according to this application. In this application, the soil moisture data fusion device 200 based on Bayesian correction may specifically include the following structure: The acquisition unit 201 is used to acquire four soil moisture data products to be fused. Among the four soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing, the third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation, and the fourth soil moisture data product is a data product obtained by ground observation. Preprocessing unit 202 is used to perform resampling and time alignment operations on four types of soil moisture data products; Training unit 203 is used to train a soil moisture data product bias correction model based on four soil moisture data products, using a first soil moisture data product with higher accuracy than the second soil moisture data product and constructed based on a multi-channel collaborative inversion algorithm as a benchmark. The soil moisture data product bias correction model is used to perform bias correction processing on the second soil moisture data product input into the model. The soil moisture data product bias correction model is specifically a Bayesian model configured under the Bayesian framework.

[0076] As an exemplary embodiment, the deviation correction process performed by the soil moisture data product deviation correction model specifically involves improving spatiotemporal resolution, improving spatial coverage, and improving temporal continuity.

[0077] As another exemplary embodiment, the resampling operation specifically employs grid linear interpolation to resample to the EASE-Grid 2.0 grid; The time alignment operation specifically uses the annual average of the gridded UTC transit time extracted from the first soil moisture data product as the time anchor point, and creates a ±30-minute time window to extract data from other data products at the corresponding time.

[0078] As yet another exemplary embodiment, the Bayesian framework is represented as follows: , in, Denotes the posterior distribution, and also indicates the distribution under a given condition. Down The probability, This represents the unknown function to be estimated. Represents observation data, This represents the likelihood function, and also represents the given... Down The probability, Denotes the prior distribution, and also represents The probability, express The probability of; , , , in, Indicates input of function, The expected value is Mean function Kernel function is Gaussian distribution process, This represents the two input points of the satellite data. Indicates the noise term. This indicates a mean of 0 and a noise variance of . The normal distribution The posterior mean is... The posterior covariance is The Gaussian distribution.

[0079] As yet another exemplary embodiment, within the Bayesian framework, the kernel function Specifically, a composite kernel function is used, which is represented as follows: , in, Represents the signal variance. Indicates a length measure. Indicates the distance between two points. Indicates when When it is 1, when The time is 0.

[0080] As another exemplary embodiment, the device further includes a configuration unit 204 for: For the four soil moisture data products, corresponding first reference index data are configured. The first reference index is used as an auxiliary reference factor in the model prediction logic and is expressed as follows: , in, Indicates the primary reference indicator. Indicates brightness temperature. Indicates the polarization angle. and Indicates different incident angles.

[0081] As another exemplary embodiment, the device further includes a configuration unit 204 for: For the four types of soil moisture data products, corresponding second reference index data are configured. The second reference index is used as an auxiliary reference factor in the model prediction logic and includes vegetation optical thickness and surface temperature.

[0082] As another exemplary embodiment, the output of the soil moisture data product bias correction model is specifically a daily-scale global soil moisture data product.

[0083] As another exemplary embodiment, the device further includes an application unit 205, configured to: The target soil moisture data for the target area within a specified time period is obtained through a soil moisture data product deviation correction model. Based on the target soil moisture data, soil moisture development events are analyzed and processed in the target area. The soil moisture development events involved include evapotranspiration estimation events or precipitation inversion events.

[0084] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the Bayesian correction-based soil moisture data fusion device and its corresponding units described above can be found in [reference needed]. Figure 1 The description of the soil moisture data fusion method based on Bayesian correction in the corresponding embodiment will not be repeated here.

[0085] The foregoing has provided a detailed description of the soil moisture data fusion method and apparatus based on Bayesian correction provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the core ideas of this application; furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for fusing soil moisture data based on Bayesian correction, characterized in that, The method includes: Four types of soil moisture data products to be fused are obtained. Among the four types of soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing, the third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation, and the fourth soil moisture data product is a data product obtained by ground observation. Resampling and time alignment operations were performed on the four soil moisture data products. Based on the four types of soil moisture data products, the first soil moisture data product, which has higher accuracy than the second soil moisture data product and is constructed based on a multi-channel collaborative inversion algorithm, is used as a benchmark to train a soil moisture data product deviation correction model. The soil moisture data product deviation correction model is used to perform deviation correction processing on the second soil moisture data product input into the model. Specifically, the soil moisture data product deviation correction model is a Bayesian model configured under the Bayesian framework.

2. The method according to claim 1, characterized in that, The deviation correction process performed by the soil moisture data product deviation correction model specifically involves improving spatiotemporal resolution, spatial coverage, and temporal continuity.

3. The method according to claim 1, characterized in that, The resampling operation specifically employs grid linear interpolation to resample to the EASE-Grid 2.0 grid; The time alignment operation specifically uses the annual average of the grid UTC transit time extracted from the first soil moisture data product as the time anchor point, and creates a ±30-minute time window to extract data from other data products at the corresponding time.

4. The method according to claim 1, characterized in that, The Bayesian framework is represented as follows: ， in, Denotes the posterior distribution, and also indicates the distribution under a given condition. Down The probability, This represents the unknown function to be estimated. Represents observation data, This represents the likelihood function, and also represents the given... Down The probability, Denotes the prior distribution, and also represents The probability, express The probability of; ， ， ， in, Indicates input of function, The expected value is Mean function Kernel function is Gaussian distribution process, This represents the two input points of the satellite data. Indicates the noise term. This indicates a mean of 0 and a noise variance of . The normal distribution The posterior mean is... The posterior covariance is The Gaussian distribution.

5. The method according to claim 4, characterized in that, Within the Bayesian framework, kernel function Specifically, a composite kernel function is used, which is represented as follows: ， in, Indicates the signal variance. Indicates length measurement. Indicates the distance between two points. Indicates when When it is 1, when The time is 0.

6. The method according to claim 1, characterized in that, The method further includes: For the four types of soil moisture data products, corresponding first reference index data are configured. The first reference index is used as an auxiliary reference factor in the model prediction logic and is expressed as follows: ， in, This refers to the first reference indicator. Indicates brightness temperature. Indicates the polarization angle. and Indicates different incident angles.

7. The method according to claim 1, characterized in that, The method further includes: The four types of soil moisture data products are configured with corresponding second reference index data. The second reference index is used as an auxiliary reference factor in the model prediction logic and includes vegetation optical thickness and surface temperature.

8. The method according to claim 1, characterized in that, The output of the soil moisture data product bias correction model is specifically a daily-scale global soil moisture data product.

9. The method according to claim 1, characterized in that, The method further includes: The target soil moisture data for the target area within a specified time period is obtained through the soil moisture data product deviation correction model. Based on the target soil moisture data, soil moisture development events are analyzed and processed in the target area. The soil moisture development events involved specifically include evapotranspiration estimation events or precipitation inversion events.

10. A soil moisture data fusion device based on Bayesian correction, characterized in that, The device includes: The acquisition unit is used to acquire four soil moisture data products to be fused. Among the four soil moisture data products, the first soil moisture data product and the second soil moisture data product are different data products obtained by L-band passive microwave remote sensing, the third soil moisture data product is a data product obtained by integrating satellite observation, ground observation and model simulation, and the fourth soil moisture data product is a data product obtained by ground observation. The preprocessing unit is used to perform resampling and time alignment operations on the four types of soil moisture data products; The training unit is used to train a soil moisture data product bias correction model based on the four soil moisture data products, using the first soil moisture data product, which has higher accuracy than the second soil moisture data product and is constructed based on a multi-channel collaborative inversion algorithm, as a benchmark. The soil moisture data product bias correction model is used to perform bias correction processing on the second soil moisture data product input into the model. Specifically, the soil moisture data product bias correction model is a Bayesian model configured under the Bayesian framework.

Images