Water body pollution remote sensing parameter quality improvement method based on multi-source heterogeneous data

Abstract

The invention provides a water pollution remote sensing parameter quality improvement method based on multi-source heterogeneous data, relates to the technical field of water quality monitoring, and solves the problem of low prediction precision of chlorophyll a concentration. The method comprises the following steps: acquiring multi-source heterogeneous data; preprocessing the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes; inputting the multi-source heterogeneous data set into a pre-trained water body pollution remote sensing parameter quality improvement model, and outputting a water body pollution remote sensing parameter quality initial prediction result; and carrying out downscaling correction on the water body pollution remote sensing parameter quality initial prediction result to obtain a water body pollution remote sensing parameter quality improvement result. The water body pollution remote sensing parameter quality can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, and in particular to a method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data. Background Art

[0002] Chlorophyll-a (Chl-a), a major remote sensing parameter for water pollution, is a photosynthetic pigment present in all algal species and a key biogeochemical indicator in water. Its concentration directly reflects the biomass of phytoplankton in the water and is closely correlated with nutrient concentrations, making it a direct and appropriate representative indicator for assessing the degree of eutrophication and water quality in lakes. Therefore, regular monitoring of changes in the distribution of remote sensing parameters for water pollution not only reveals the nutrient status of water bodies, providing key data support for lake governance and fishery resource management, but also plays an important role in the prediction of remote sensing parameters for water pollution in aquatic environments.

[0003] The prior art discloses a method for reconstructing water pollution remote sensing parameters, which comprises establishing a numerical simulation data set and a satellite remote sensing data set; performing spatiotemporal matching between the numerical simulation data set and the satellite remote sensing data set, and dividing the two types of spatiotemporal matching data sets into a training set, a validation set, and a test set; based on a convolutional neural network algorithm, a deep learning model is established by adopting a sample construction method based on spatial window scanning, continuous convolution, and a maximum pooling method; the numerical simulation training set is used as the input value of the deep learning model, and the water pollution remote sensing parameters of the satellite remote sensing training set are used as the output value corresponding to the data set, and continuous training is performed; based on the trained deep learning model, all numerical simulation data sets are calculated to obtain the satellite remote sensing parameters without missing data. The satellite water pollution remote sensing parameter reconstruction process was completed using the numerical simulation dataset with time-space matching. On the one hand, the noise cannot be eliminated by the numerical simulation dataset, and the variables representing the same physical quantity or phenomenon in the numerical simulation dataset are numerically mismatched, that is, the numerical simulation dataset has inconsistent attributes. The deep learning model trained with datasets with different attributes is used to predict water pollution remote sensing parameters, and the prediction results of water pollution remote sensing parameters are low in accuracy. On the other hand, the deep learning model is used to complete the water pollution remote sensing parameter reconstruction process. Although it can achieve the quality downscaling of water pollution remote sensing parameters, it does not consider the impact of scale effect, resulting in low downscaling accuracy of water pollution remote sensing parameters, which further affects the quality of water pollution remote sensing parameters. Summary of the Invention

[0004] In order to solve the problem of poor quality of water pollution remote sensing parameters in the above-mentioned existing technologies, the present invention proposes a method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data, which can effectively improve the quality of water pollution remote sensing parameters.

[0005] In order to achieve the above technical effects, the technical solutions of the present invention are as follows:

[0006] A method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data includes the following steps:

[0007] S1. Acquire multi-source heterogeneous data;

[0008] S2. Preprocessing the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes;

[0009] S3. Input the multi-source heterogeneous dataset into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality;

[0010] S4. Downscaling the initial prediction results of the water pollution remote sensing parameter quality to obtain a water pollution remote sensing parameter quality improvement result.

[0011] Preferably, the multi-source heterogeneous data is multi-source remote sensing data, and preprocessing the multi-source remote sensing data to obtain a multi-source heterogeneous data set with consistent attributes includes:

[0012] S21. Geo-registering the multi-source remote sensing data to obtain spatially consistent multi-source remote sensing data;

[0013] S22. geometrically correcting the spatially consistent multi-source remote sensing data to obtain multi-source remote sensing geometrically corrected data;

[0014] S23. Defining a research scope for the multi-source remote sensing geometric correction data, and extracting high-precision boundary data from the multi-source remote sensing geometric correction data based on the research scope;

[0015] S24. Resample the high-precision boundary data to obtain the multi-source heterogeneous data set.

[0016] Preferably, the multi-source remote sensing data is used as input to an image registration tool, and the multi-source remote sensing data is geo-registered using the image registration tool to output spatially consistent multi-source remote sensing data.

[0017] Preferably, the geometric correction of the spatially consistent multi-source remote sensing data includes: using a denoising and smoothing algorithm to remove noise data and outliers in the spatially consistent multi-source remote sensing data to obtain the multi-source remote sensing geometrically corrected data.

[0018] Preferably, the research scope of multi-source remote sensing geometric correction data is limited, including:

[0019] S231. Using the modified normalized difference water index as a calculation tool for identifying the research range, inputting the multi-source remote sensing geometric correction data as a data source into the calculation tool, and outputting an initial range threshold;

[0020] S232. Iterate the initial range threshold until the iteration step meets the maximum number of iterations and output the optimal range threshold;

[0021] S233. Apply the optimal range threshold to the multi-source remote sensing geometric correction data to extract the high-precision boundary data.

[0022] Preferably, the water pollution remote sensing parameter quality improvement model is a spatial downscaling model, and the spatial downscaling model is constructed using a downscaling regression method.

[0023] Preferably, the calculation expression for outputting the initial prediction result of the quality of the water pollution remote sensing parameter is as follows:

[0024] CHL f =CHL c +(F c (S f )-F c (S c ))

[0025] Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f Represents the output of the initial prediction results of the quality of water pollution remote sensing parameters with high resolution, CHL c Represents the original low-resolution water pollution remote sensing parameters, F c The calculation function that represents the statistical relationship between the low-resolution water pollution remote sensing parameters and the influencing characteristic factor set of water pollution remote sensing parameters; S c represents the set of influencing feature factors at high resolution, S f Represents the set of influencing feature factors at low resolution.

[0026] Preferably, the specific expression for downscaling the initial prediction result of the water pollution remote sensing parameter quality is as follows:

[0027] CHL f '=CHL c +(F f (S f )-F c (S c ))

[0028] Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f ′ represents the improvement result of water pollution remote sensing parameter quality, CHLc Represents the original low-resolution water pollution remote sensing parameters, F f The calculation function that represents the statistical relationship between the high-resolution water pollution remote sensing parameters and the influencing characteristic factor set of the water pollution remote sensing parameter concentration; S c represents the set of influencing feature factors at high resolution, S f Represents the set of influencing feature factors at low resolution.

[0029] The present invention also proposes a system for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data. The system is implemented based on the method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data, and includes:

[0030] Acquisition module, used to acquire multi-source heterogeneous data;

[0031] A preprocessing module, configured to preprocess the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes;

[0032] An output module is used to input the multi-source heterogeneous data set into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality;

[0033] The improvement module is used to downscale the initial prediction results of the water pollution remote sensing parameter quality to obtain the water pollution remote sensing parameter quality improvement results.

[0034] The present invention also provides a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0035] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations of the method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data.

[0036] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0037] The present invention proposes a method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data. First, the acquired multi-source heterogeneous data are preprocessed to obtain a multi-source heterogeneous data set with consistent attributes, which solves the problems of data differentiation and lack of uniformity, lays a solid and reliable foundation for the modeling and analysis of water pollution remote sensing parameter quality improvement models, and also provides a reliable data basis and method support for the spatial downscaling research of water pollution remote sensing parameter quality; secondly, the multi-source heterogeneous data set is used as the input of a pre-trained water pollution remote sensing parameter quality improvement model, with the aim of providing accurate input data for the water pollution remote sensing parameter quality improvement model, so as to ensure that the water pollution remote sensing parameter quality improvement model outputs the initial predicted water pollution remote sensing parameter quality. The measurement results are accurate and reliable; then, the influence of scale effect is taken into account in the downscaling process of the water pollution remote sensing parameter quality improvement model, and the initial prediction results of water pollution remote sensing parameter quality are downscaled and corrected, which can accurately capture the influence of scale effect of the water pollution remote sensing parameter quality improvement model in the output of the water pollution remote sensing parameter quality initial prediction results. By performing downscaling correction in the downscaling process, the initial prediction results of water pollution remote sensing parameter quality can be further optimized and improved, and the prediction accuracy of water pollution remote sensing parameter quality can be further improved, ensuring that the water pollution remote sensing parameter quality improvement results adapt to environmental changes at different scales, and improving the reliability and accuracy of the water pollution remote sensing parameter quality improvement results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart showing a method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data proposed in an embodiment of the present invention;

[0039] Figure 2 R represents the eight combinations proposed in the embodiment of the present invention. 2 Block diagram;

[0040] Figure 3 A block diagram showing the RMSE of eight combinations proposed in an embodiment of the present invention;

[0041] Figure 4 Graphs showing the root mean square error statistics of the MVA-GBDT method proposed in an embodiment of the present invention without and after scale effect correction;

[0042] Figure 5 A structural diagram of a water pollution remote sensing parameter quality improvement system based on multi-source heterogeneous data proposed in an embodiment of the present invention;

[0043] Figure 6 A diagram showing the structure of a computer device proposed in an embodiment of the present invention;

[0044] 601. Processor; 602. Memory; 603. Communication interface; 604. Communication bus; 605. Executable instructions. DETAILED DESCRIPTION

[0045] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting the present invention;

[0046] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0047] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0048] Example 1

[0049] like Figure 1 As shown, this embodiment proposes a method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data. This method uses chlorophyll-a (Chl-a) as an example for explanation. In actual application, the water pollution remote sensing parameter is not limited to Chl-a. The method includes the following steps:

[0050] S1. Acquire multi-source heterogeneous data;

[0051] In S1, the multi-source heterogeneous data are multi-source remote sensing data. The research area for the multi-source remote sensing data is 12 inland lakes, including Qinghai Lake, Poyang Lake, Dongting Lake, Taihu Lake, Hulun Lake, Hongze Lake, Namtso Lake, Selin Co, Bosten Lake, Gaoyou Lake, Chaohu Lake, and Weishan Lake. The multi-source remote sensing data include the satellite dataset Sentinel-1, the satellite dataset Sentinel-2, and the satellite dataset Landsat-8.

[0052] S2. Preprocessing the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes;

[0053] In S2, due to the mismatch in geometry and attributes between the Sentinel-1, Sentinel-2, and Landsat-8 satellite datasets, data processing poses certain challenges. To address this problem, the multi-source heterogeneous data is selected as multi-source remote sensing data, and the multi-source remote sensing data is preprocessed for attribute consistency to obtain a multi-source heterogeneous dataset with consistent attributes, including:

[0054] S21. Geo-registering the multi-source remote sensing data to obtain spatially consistent multi-source remote sensing data;

[0055] In S21, an image registration tool based on ENVI software is used, and the multi-source remote sensing data is used as the input of the image registration tool. The multi-source remote sensing data is geographically registered using the image registration tool, and spatially consistent multi-source remote sensing data is output, thereby ensuring the accuracy and reliability of subsequent processing.

[0056] S22. geometrically correcting the spatially consistent multi-source remote sensing data to obtain multi-source remote sensing geometrically corrected data;

[0057] In S22, geometric correction of the spatially consistent multi-source remote sensing data includes: using a denoising and smoothing algorithm to remove noise data and outliers from the spatially consistent multi-source remote sensing data to obtain the multi-source remote sensing geometrically corrected data. For the surface reflectance data obtained from Sentinel-1 and Sentinel-2, the denoising and smoothing algorithm selected is a Savitzky-Golay filter that generates gaps to fill missing values ​​in the reflectance data. This step is very important for data preprocessing and can effectively reduce noise and outliers in the data, thereby improving data quality and usability. For the water surface temperature (WST) data obtained from Landsat-8, the denoising and smoothing algorithm selected is a reconstruction using the time series harmonic analysis (HANTS) method. The HANTS method is a technique widely used to reconstruct land surface temperature (LST) and normalized vegetation index (NDVI) time series. It has excellent data reconstruction capabilities and can effectively remove random noise and cloud / snow cover in the data, thereby improving data accuracy and reliability. Through the above preprocessing steps, corrected and filled data were obtained, which laid a solid foundation for subsequent research and analysis. These processed data can more accurately reflect the characteristics and changes of the water environment and provide important data support for in-depth research on the spatial distribution and changes of Chl-a concentrations.

[0058] S23. Defining a research scope for the multi-source remote sensing geometric correction data, and extracting high-precision boundary data from the multi-source remote sensing geometric correction data based on the research scope;

[0059] In S23, the research scope of the multi-source remote sensing geometric correction data is limited, including:

[0060] S231. Using a modified normalized difference water index (MNDWI) as a calculation tool for identifying the study range, inputting the multi-source remote sensing geometrically corrected data as a data source into the calculation tool, determining an MNDWI threshold suitable for lake water surface identification through iterative calculation, and outputting an initial range threshold as the MNDWI threshold;

[0061] S232. Iterate the initial range threshold until the iteration step meets the maximum number of iterations and output the optimal range threshold;

[0062] S233. Apply the optimal range threshold to the multi-source remote sensing geometric correction data to extract the high-precision boundary data.

[0063] In S233, the multi-source remote sensing geometric correction data includes high spatial resolution satellite images, and an optimal range threshold is applied to the high spatial resolution satellite images to ensure that the fine scale range of the lake boundary is accurately captured, thereby extracting the high-precision boundary data.

[0064] The purpose of step S23 is to ensure the accuracy of the extracted high-precision boundary data so that the lake boundaries within the studied water environment can be fully reflected. Through this process, the lake area can be effectively located, providing a reliable spatial range for subsequent Chl-a concentration downscaling research. The use of this method of extracting boundary data within a limited research scope can ensure the accuracy and reliability of the data, provide a solid foundation for research, and facilitate a better understanding and analysis of the changes in the Chl-a concentration, a remote sensing parameter of water pollution in the water environment.

[0065] S24. Resample the high-precision boundary data to obtain the multi-source heterogeneous data set.

[0066] In S24, the high-precision boundary data is resampled using a cubic convolution interpolation resampling method, and all influencing characteristic factor data in the high-precision boundary data are resampled to 300m and 30m, so as to be used for subsequent training and prediction of water pollution remote sensing parameter quality improvement models; the application of this method helps to establish consistency between different satellite products and Chl-a concentration data, thereby providing accurate input data for the establishment and prediction of water pollution remote sensing parameter quality improvement models.

[0067] The purpose of the S2 preprocessing step is to better utilize the multi-source remote sensing dataset with consistent attributes, providing a reliable data basis and methodological support for the spatial downscaling study of Chl-a concentration. This series of data processing and preparation work provides a reliable foundation for subsequent analysis, ensures the accuracy and reliability of the water pollution remote sensing parameter quality improvement model, and provides strong technical support for in-depth research on the changes in Chl-a concentration in the water environment.

[0068] S3. Input the multi-source heterogeneous dataset into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality;

[0069] In S3, the water pollution remote sensing parameter quality improvement model is a spatial downscaling model, which is constructed using a downscaling regression method. In the process of constructing the spatial downscaling model, the influencing characteristic factors of the water pollution remote sensing parameter, chlorophyll a concentration, are screened from the multi-source remote sensing data set with consistent attributes to form an influencing characteristic factor set. The influencing characteristic factor set is used to train the spatial downscaling model constructed based on the downscaling regression method to obtain a trained spatial downscaling model. In order to construct a robust spatial downscaling model, a two-stage procedure is designed in this step, the purpose of which is to select the optimal influencing characteristic factor set when constructing the spatial downscaling model, including:

[0070] In the first stage, an importance assessment method was used, which can quantify the contribution of the influencing characteristic factors to the model and ensure that the characteristic factors with significant impact on the spatial downscaling model were selected. In this way, a series of influencing characteristic factor sets that may have a significant impact on the spatial downscaling model were preliminarily screened out.

[0071] Next, based on the importance ranking of the set of influencing factors assessed in the first phase, the second phase employed stepwise regression to refine feature selection. Stepwise regression is an effective statistical method that can gradually introduce or remove variables from a model to optimize model performance. This method is particularly suitable for addressing potential multicollinearity issues, that is, when multiple influencing factors are highly correlated, which may interfere with the model's explanatory power and stability. Through stepwise regression, it is possible to accurately eliminate the influencing factors that cause multicollinearity, retaining the factors that make a substantial contribution to the model.

[0072] Finally, through these two stages of screening and optimization, a set of optimal influencing characteristic factors was determined. The influencing characteristic factors in this set of influencing characteristic factors are not only statistically significant, but also can effectively avoid the problem of multicollinearity, providing powerful explanatory variables for the spatial downscaling model. Therefore, this carefully selected set of influencing characteristic factors was determined as the optimal set of influencing characteristic factors for constructing the spatial downscaling model, aiming to optimize the prediction performance of the spatial downscaling model, to enhance the prediction accuracy and stability of the spatial downscaling model, and to provide reliable methodological support for research and application in related fields.

[0073] Furthermore, this example distinguishes and selects six different types of influencing factors to examine their impact on the accuracy of the downscaling model. This is done to provide a more detailed understanding of their specific effects on the performance of the spatial downscaling model, particularly their contribution to improving the accuracy of Chl-a concentration downscaling predictions. This approach aims to quantitatively evaluate and identify which set of influencing factors maximizes the model's predictive power—in other words, to find the optimal set of influencing factors.

[0074] To achieve the above objectives, this embodiment designs eight different combinations of downscaling impact characteristic factor types, as shown in Table 1. Each group represents a specific configuration of impact characteristic factors, as follows:

[0075] Multispectral reflectance: Multispectral reflectance is used alone as the influencing characteristic factor of the model to explore its basic influence.

[0076] Multispectral reflectance + chlorophyll index: Add chlorophyll index on the basis of multispectral reflectance to examine the gain effect of chlorophyll index.

[0077] Multispectral reflectance + red edge index: Combines multispectral reflectance and red edge index to evaluate the contribution of red edge index to model accuracy.

[0078] Multispectral reflectance + texture features: Explore the impact of adding texture features on multispectral basis on model performance.

[0079] Multispectral reflectance + thermal features: By adding thermal features to multispectral reflectance, the improvement effect of thermal features on the model is analyzed.

[0080] Multispectral reflectance + backscattering characteristics: Evaluate the impact of combining backscattering characteristics with multispectral reflectance on model accuracy.

[0081] All influencing characterization factors: All available characterization factors are considered together to assess their overall effect.

[0082] Best influencing characteristic factors: By analyzing and comparing the effects of the first seven groups and the two-stage procedure, the combination of influencing characteristic factors that contributes most to the downscaling prediction of Chl-a concentration is selected as the best influencing characteristic factor set.

[0083] Table 1 Detailed statistics of the composition of the influencing characteristic factors of 8 groups with different Chl-a concentrations

[0084]

[0085] These combined designs not only systematically evaluate the independent and combined effects of each influencing factor but also, through comparative analysis, intuitively reveal which influencing factors are most critical in predicting Chl-a concentrations, thereby providing a scientific basis for water monitoring and environmental assessment. This research is expected to improve the prediction accuracy of spatial downscaling models and promote the application of remote sensing technology in environmental monitoring.

[0086] In order to gain a deeper understanding and quantify the specific impact of different types of influencing characteristic factors on the accuracy of the spatial downscaling model, this example designed and performed a series of accuracy evaluation analyses. These analyses focused on comparing and quantifying the performance of various influencing characteristic factor combinations on the performance of the spatial downscaling model, aiming to reveal which characteristic factor combinations can significantly improve the accuracy of downscaling predictions. To this end, two main statistical evaluation indicators were selected: the coefficient of determination (R 2 ) and root mean square error (RMSE). These two indicators are widely considered to be the key factors in evaluating model accuracy and predictive ability, and can provide an intuitive and quantitative evaluation of the model effect. 2 This indicator measures the relative fit between the model prediction value and the actual observation value. 2 Values ​​range from 0 to 1, with values ​​closer to 1 indicating greater consistency between the model's predictions and actual observations, and thus greater explanatory power. RMSE is the square root of the sum of the squares of the differences between the observed values ​​and the model's predictions, and is used to measure the magnitude of the prediction error. Smaller values ​​indicate greater model prediction accuracy and less deviation between the predictions and the actual results.

[0087] Based on these evaluation indicators, a detailed accuracy evaluation was conducted on the eight different types of influencing characteristic factor combinations designed above.

[0088] For detailed evaluation results, see Figure 2 and Figure 3 Box plots of R² and RMSE for the seven assemblies in the studied lakes. A Wilcoxon test was used to assess whether there were significant differences in accuracy between the reference assembly and the other assemblies. MSR: multispectral reflectance; CI: chlorophyll index; REI: red edge index; TF: texture signature; TC: thermal signature; BC: backscatter signature.

[0089] from Figure 2 and Figure 3 The analysis clearly shows that when multispectral reflectance is used as the fundamental influencing feature combination for Chl-a concentration downscaling, its performance indicators are an average R² of 75.42% and an RMSE of 7.94 mg / m³. This result provides a benchmark for evaluating the specific changes in the accuracy of spatial downscaling models when other types of influencing features are added to this basis.

[0090] Further analysis shows that after introducing the chlorophyll index, red edge index, thermal characteristics and backscattering characteristics on the basis of the multispectral reflectance feature combination, the downscaling accuracy of the spatial downscaling model has shown an overall upward trend. In particular, the addition of the chlorophyll index and red edge index has a particularly significant effect on improving the accuracy of the spatial downscaling model. These two influencing characteristic factors can obviously provide important information about vegetation health and biomass, which is crucial for accurately estimating Chl-a concentrations. Specifically, when these influencing characteristic factors are integrated into the spatial downscaling model, the accuracy evaluation results show that the prediction performance of the spatial downscaling model has been significantly improved, with an average R 2 The improvements in the performance and RMSE reached a statistically significant level (p<0.01).

[0091] However, not all types of influencing feature factors will lead to an improvement in the performance of the spatial downscaling model. It is worth noting that when texture features are added to the model, the accuracy of Chl-a concentration downscaling actually decreases. Specifically, the average R 2 decreased to 73.11%, while the average RMSE increased to 8.04 mg / m 3 This phenomenon may indicate that although texture features can provide useful information for the discrimination of land cover types in some cases, they may introduce noise of non-target variables during the downscaling process used in spatial downscaling models, thereby affecting the overall prediction accuracy of the spatial downscaling models.

[0092] This shows that auxiliary type influencing characteristic factors related to the biochemical composition of phytoplankton (such as chlorophyll index, red edge index, etc.) can effectively infer the spatial variability of Chl-a concentration, while the image texture feature type, due to its non-smoothness, limits its ability to extract information on the difference in Chl-a concentration under different environments to a certain extent.

[0093] In addition, it can be found that when all types of influencing characteristic factors are added to the spatial downscaling model to participate in downscaling, the accuracy of the spatial downscaling model is not improved, but rather decreases to a certain extent. This may be because there is information redundancy caused by the mutual correlation and mutual influence of covariates among the various influencing characteristic factors. Some influencing characteristic factors are less important, and some influencing characteristic factors have strong spatial correlation. When the optimal influencing characteristic factor set in combination 8 is used to extract downscaling information from the spatial downscaling model, the average R 2 Both the RMSE and the RI were significantly improved (p < 0.001). This is primarily due to the fact that the spatial downscaling model focuses solely on the key features that contribute most to the downscaling of Chl-a concentrations, effectively avoiding the cross-redundant interference introduced by irrelevant or secondary features. Therefore, by carefully selecting and optimizing the combination of influencing features, the downscaling accuracy of the spatial downscaling model can be significantly improved, ensuring its efficiency and reliability in practical applications.

[0094] In terms of comparing different types of influencing characteristic factors, this embodiment conducted a comprehensive analysis and evaluation to determine the most effective type of influencing characteristic factors for Chl-a concentration estimation. First, multi-source remote sensing data, including remote sensing image data, hydrological and meteorological data, and geographic environment data, were collected and classified and organized. Then, statistical analysis and machine learning algorithms were used to rank the characteristic importance and evaluate the performance of different types of influencing characteristic factors, and their contributions and effects in Chl-a concentration estimation were compared. Different types of influencing characteristic factors have different roles and importance in Chl-a concentration estimation, providing more scientific and effective support for the monitoring and management of water environments.

[0095] pass Figure 2 and Figure 3 Shows the R for each combination of influencing factors 2 The RMSE values ​​are used to visually compare the performance differences of various combinations in terms of model accuracy. This comparison not only reveals the significant advantages of certain combinations of influencing feature factors in improving the prediction accuracy of spatial downscaling models, but also points out that other combinations may perform poorly due to redundant or insufficient features.

[0096] Through this precision assessment analysis, we can identify which combinations of influencing characteristic factors are most critical for improving the accuracy of the spatial downscaling model, thereby providing more precise and efficient feature selection guidance for future research, and ultimately achieving a more accurate and reliable spatial downscaling prediction model.

[0097] S4. Downscaling the initial prediction results of the water pollution remote sensing parameter quality to obtain a water pollution remote sensing parameter quality improvement result.

[0098] Example 2

[0099] This example further illustrates the use of a downscaling regression method to construct the spatial downscaling model to verify the prediction accuracy of the spatial downscaling model. This example selects a downscaling method based on a gradient boosted decision tree (GBDT) to construct a spatial downscaling model, and compares and analyzes it with a downscaling method based on an artificial neural network, a downscaling method based on a random forest, and a downscaling method based on a multivariate polynomial regression, thereby verifying the optimal method for downscaling Chl-a concentrations between lakes.

[0100] The downscaling method based on the gradient boosting decision tree (GBDT) is different from other ML algorithms in that the gradient boosting decision tree constructs an ensemble of decision tree learners by using boosting iterations. In each iteration, a new decision tree is used to strengthen the loss function based on the sharpest gradient. Under the assumption of scale invariance, that is, the statistical relationship between Chl-a concentration and explanatory variables at low spatial resolution also holds at high spatial resolution, the downscaling method based on the gradient boosting decision tree (GBDT) is divided into three main steps to construct a spatial downscaling model: (1) First, a regression model is used to establish a statistical relationship between Chl-a concentration and influencing characteristic factors at low resolution (referring to spatial resolution); (2) Based on the established statistical relationship, the high-resolution and low-resolution influencing characteristic factors are used to predict the high-resolution and low-resolution Chl-a concentrations respectively; (3) Then, the difference between the predicted Chl-a concentrations at high resolution and low resolution (i.e., high-resolution texture information) is added to the original low-resolution Chl-a concentration, and the calculation expression for the output of the initial prediction result of the quality of water pollution remote sensing parameters is as follows:

[0101] CHL f =CHL c +(F c (S f )-F c (S c ))

[0102] Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f Represents the output of the initial prediction result of chlorophyll a concentration with high resolution, CHL c represents the original low-resolution chlorophyll a concentration, F c The calculation function that represents the statistical relationship between the chlorophyll a concentration at low resolution and the set of characteristic factors affecting the chlorophyll a concentration; S c represents the set of influencing feature factors at high resolution, S frepresents the set of influencing feature factors at low resolution. The GBDT downscaling method used in the study has three important hyperparameters: the number of decision trees in the regression procedure, the explanatory factors used to divide the nodes, and the loss function. Parameter adjustment of the spatial downscaling model is crucial to avoid multicollinearity and overfitting. In this work, we adopted a two-stage procedure to achieve model parameterization. In the first stage, importance evaluation and stepwise regression methods were used to select explanatory factors and deal with multicollinearity, because these two methods retain the physical meaning of the factors for subsequent analysis. In the second stage, grid search and 10-fold cross-validation were used to determine the number of decision trees and loss function; ultimately, 100 and minimum absolute deviation were determined as the final number of decision trees and loss function, respectively.

[0103] Downscaling methods based on artificial neural networks (ANNs) have demonstrated advanced capabilities and flexibility in complex data analysis and model building. As a nonparametric, nonlinear model, ANNs mimic the brain's receiver and information processing mechanisms, transferring and processing information between layers to achieve complex pattern recognition and learning. In an ANN model, the input layer receives raw data, while the hidden layer processes the input information and transmits it to the output layer, which then generates the final model predictions. Network initialization, including the number of neurons and the setting of initial weights, is a critical step in model building, directly impacting learning efficiency and accuracy. While ANN models have strong learning capabilities, they also carry the risk of overlearning. Therefore, selecting an appropriate number of neurons is crucial in model design, requiring a balance between learning accuracy and generalization. Determining the number of neurons is often accomplished through repeated experimentation and verification. The nnet function provided by the "nnet" package allows for efficient testing of model performance with varying numbers of neurons, thereby determining the optimal model architecture. Specifically for Chl-a concentration downscaling, fine-tuning ANN model parameters, such as the number of neurons, significantly improves the model's predictive accuracy and application value. Furthermore, during ANN model training, the mechanism for updating error values ​​and weights is crucial for ensuring the model gradually approaches the optimal solution. Through this iterative process, a robust ANN-based downscaling model was ultimately constructed.

[0104] The random forest (RF)-based downscaling method is a powerful and popular ensemble machine learning algorithm that has been used for many regression and classification problems. The core of this algorithm is to construct a series of random, simple decision trees. Essentially, the RF concept creates a large, random subset of decorrelated regression trees, bootstrapping them from samples and features. This method consists of two key elements: randomness and ensemble learning. In the random forest algorithm, randomness is primarily achieved by sampling with replacement from a dataset of N samples to construct each decision tree. This process is called bootstrapping, or bagging. Samples that are not selected for any decision tree construction are referred to as out-of-bag data. In the ensemble learning process, after random selection, a subset of N′ samples with m feature variables is considered. To optimize the tree structure, a modified form of the classification and regression tree (CART) algorithm is used, with K iterations to generate a large number of decision trees (i.e., a forest), thereby enhancing model stability and preventing overfitting. The final result is determined by ensemble learning of all K trees, averaging the predictions of all trees to arrive at the final estimate, particularly in regression problems. When constructing a random forest regression (RFR) model, two key parameters typically need to be adjusted to optimize the model's prediction accuracy. These two parameters are ntree (the total number of decision trees grown in the forest) and mtry (the number of predictor variables randomly selected at each tree node). To find the optimal combination of ntree and mtry, the recommended ranges of these parameters were referenced, and a ten-fold cross-validation method was employed. Through this series of parameter adjustment and validation processes, a robust random forest-based Chl-a concentration downscaling model was constructed.

[0105] The downscaling method based on multivariate polynomial regression (MPR) is another key method for improving the spatial resolution of Chl-a concentration. This method is based on the polynomial regression model and uses multi-source data for modeling and prediction, thereby achieving downscaling of Chl-a concentration. Multivariate polynomial regression is one of the regression methods that uses n-degree polynomials to analyze the relationship between a dependent parameter and multiple independent variables. The general form of the polynomial equation can be expressed as the following equation, which describes the nonlinear relationship between a dependent variable and an independent variable:

[0106] Y=β0+β1X+β2X 2 +…+β m X n +ε

[0107] Among them, β1, β2, …, β m is the unknown regression coefficient, and ε is the random error.

[0108] For multiple independent variables, different degrees of polynomial regression can be used, i.e., first degree, second degree, ..., ninth degree. For example, a quadratic (second order, n = 2) polynomial model can be expressed as the equation:

[0109]

[0110] Three polynomial regression models of varying degrees, including second-order (n=2), third-order (n=3), and fourth-order (n=4), were tested and compared to select the most accurate model for estimating Chl-a concentrations in lakes. Ultimately, the second-order polynomial regression model was determined to be the optimal Chl-a concentration multivariate polynomial regression model.

[0111] Given the lack of high-resolution measured Chl-a concentration data, a widely used "up-scaling-down" approach was employed for validation. Specifically, the original 30-m resolution impact characteristic factors and 300-m Chl-a concentration data were upscaled to 300 and 3000 m. The original 300-m Chl-a concentrations were used as a benchmark to validate the downscaled Chl-a concentrations from 3000 m.

[0112] In addition, through in-depth analysis of the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE) and structural similarity (SSIM), four key statistical indicators are used to comprehensively evaluate the performance and accuracy of the spatial downscaling model constructed by the downscaling method based on the gradient boosting decision tree. 2 The value is the core indicator for evaluating the prediction ability of the model. The closer its value is to 1, the higher the degree of consistency between the prediction result and the actual situation. RMSE is a key measure of prediction error. The smaller the RMSE value, the higher the prediction accuracy of the model. Similar to RMSE, MAE is also commonly used to evaluate the accuracy of the model. It focuses on the average of the absolute values ​​of the differences between the predicted values ​​and the actual values. SSIM measures the structural similarity between two images. The higher the value, the higher the similarity between the downscaled image and the original image. The specific expressions of these indicators are as follows:

[0113]

[0114] Among them, μ f represents the average chlorophyll a concentration at high resolution; m represents the number of samples; μ c represents the average value of chlorophyll a concentration at low resolution; C1 represents the first constant used to maintain numerical stability; C2 represents the second constant used to maintain numerical stability, which are set to 0.1 and 0.05 respectively; σ 2 represents the covariance of chlorophyll a concentration.

[0115] To evaluate the performance of the GBDT method based on multivariate analysis in downscaling Chl-a concentrations, this example also compared the accuracy of three other methods (ANN, RF, and MPR) based on the same input. Table 2 shows the statistical results of different methods to compare the performance of the 12 lakes. According to the statistical data, the GBDT method performed best in terms of accuracy, with an average determination coefficient (R 2 ) is 0.82, and the root mean square error (RMSE) is 4.47 mg / m 3 The second is the ANN method, average R 2 is 0.79, and the RMSE is 4.99 mg / m3, followed by the RF method, with an average R 2 The mean square error was 0.77, and the RMSE was 5.11 mg / m 3 , and finally the MPR method, mean R 2 The RMSE is 7.22 mg / m 3 . It is worth noting that in some lakes, there are significant differences in the performance of the two methods. For example, in Lake No. 1 (June 2021) and Lake No. 4 (September 2021), the root mean square errors of the GBDT method and the MPR method are the largest among all lakes. For the Chl-a concentration of Lake No. 4 collected in September 2021, the downscaling accuracy of the GBDT based on uniform decomposition is higher than that of the MPR method. This difference may be due to the rapid changes in the state of phytoplankton in lakes in the late rainy season. This change may cause the sub-pixel changes produced by uniform decomposition to be unable to accurately reproduce the large-scale pixels of 300m, thereby affecting the performance of the MPR method.

[0116] Table 2 Statistics of downscaling accuracy obtained by four different methods

[0117]

[0118]

[0119] In general, the GBDT downscaling method proposed in this example shows a high consistency between the downscaling results and the original values ​​after adding different types of influencing characteristic factors. The GBDT method performs well in the task of downscaling Chl-a concentration in lakes, especially in water environments with obvious seasonal changes.

[0120] Furthermore, to evaluate whether the performance of a spatial downscaling model constructed using a gradient boosted decision tree (GBDT) downscaling method is affected by overfitting, this example adds different types of influencing characteristic factors to the GBDT downscaling method to construct a spatial downscaling model, denoted as the MVA-GBDT model, and then conducts a cross-validation experiment. The MVA-GBDT model is trained using data from a single lake and applied to downscaling predictions for other lakes. The results are also tested for the 12 lakes over different time periods.

[0121] Table 3 shows the training and testing results of the 12 lakes based on the MVA-GBDT model. Each row in the table represents the MVA-GBDT model trained with data from a single lake, and each column represents the test results of the trained MVA-GBDT model.

[0122] Table 3 Error statistics of the MVA-GBDT method after downscaling model training and cross-validation for 12 different lakes at different times

[0123]

[0124]

[0125] In Lake No. 1, Qinghai Lake, the smallest RMS error was observed to be 3.10 mg / m 3 The root mean square errors of Lake No. 2, Poyang Lake, and Lake No. 4, Taihu Lake, were relatively large, at 7.83 mg / m 3 and 8.75 mg / m 3 . These results are affected by a variety of factors, including differences in the environment surrounding the lakes, data quality, and the correlation between the training data and the reference data. For example, Lake No. 2, Poyang Lake, is located in the middle and lower reaches of the Yangtze River, and its width varies greatly, which may be affected by the size of the basin and the speed of the water flow. Lake No. 4, Taihu Lake, is located at the junction of developed provinces. The nutrient content in the area is high and may be affected by human activities and industrial emissions. This is significantly different from the environmental conditions of other lakes, and therefore produces a large root mean square error in the local downscaling. These factors lead to differences in the spatial distribution of Chl-a concentrations between different lakes, which in turn affects the predictive performance of the spatial downscaling model.

[0126] Comparing the downscaling results obtained after MVA-GBDT training on the remaining lakes revealed that the root mean square error (RMS) was minimized when training data was restricted to the lake surface area. This indicates that the model performs best in predicting Chl-a concentrations at the lake surface. With the exception of two specific study areas, the test scores for the remaining lakes exhibited the lowest RMS errors, demonstrating that the model's predictive ability is relatively stable across lakes. The best results for Lake 5, Hulun Lake, and Lake 7, Nam Co, came from the eighth and ninth trained models, respectively. This may be due to insufficient adjustment of surface temperature by the explanatory variables in the study area, while the best model test results were obtained from training on a more similar study area. This suggests that the model performs better when dealing with lakes with similar environmental conditions. Notably, the differences are not significant compared to MVA-GBDT models trained on data from the study area itself. This demonstrates that the model has a certain degree of generalization ability and can be applied to downscaling Chl-a concentrations across different lakes. In summary, the proposed MVA-GBDT method showed good stability and applicability in processing the downscaling calculation of Chl-a concentrations in various inland lakes, providing an effective tool and support for the study of lake ecosystems.

[0127] Example 3

[0128] This embodiment further explains S4. In S4, it is assumed that at different spatial resolutions, the same statistical relationship exists between Chl-a concentration and influencing characteristic factors. However, the actual situation is not always the case. The relationship between Chl-a concentration and certain influencing characteristic factors will change with the change of spatial resolution. For example, the relationship between Chl-a concentration and NDCI will change with the change of scale, and the relationship between Chl-a concentration and WST also has significant differences in scale. Therefore, this embodiment further downscales the initial prediction result of chlorophyll a concentration by considering the scale effect to obtain the downscaled result of chlorophyll a concentration.

[0129] This embodiment selects a downscaling method based on a gradient boosting decision tree (GBDT) to construct a spatial downscaling model, and adds different types of influencing characteristic factors, which is simplified in English as the MVA-GBDT method. The MVA-GBDT method uses a comparison scheme to quantify the uncertainty caused by the scale effect. The scheme is based on the high-resolution Chl-a concentration obtained by Landsat-8, which shows a high degree of consistency. The key to this scheme is to replace the statistical relationship established between the scale factor and the Chl-a concentration at low resolution with the statistical relationship at high resolution. After having the high-resolution statistical relationship, the specific expression for downscaling correction of the initial prediction results of the water pollution remote sensing parameter quality is as follows:

[0130] CHLf '=CHL c +(F f (S f )-F c (S c ))

[0131] Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f ′ represents the downscaling result of chlorophyll a concentration, CHL c represents the original low-resolution chlorophyll a concentration, F f The calculation function that represents the statistical relationship between the chlorophyll a concentration and the set of influencing characteristic factors of chlorophyll a concentration at high resolution; S c represents the set of influencing feature factors at high resolution, S f Represents the set of influencing feature factors at low resolution.

[0132] The advantage of downscaling the initial chlorophyll-a concentration prediction method is that by considering high-resolution statistical relationships in the low-resolution Chl-a concentration estimation process, the impact of scale effects can be more accurately captured and corrected during the downscaling process. In this way, the chlorophyll-a concentration downscaling results can not only accurately reflect the relationship between Chl-a concentration and influencing characteristic factors, but also better adapt to environmental changes at different scales.

[0133] This example quantifies the proportional effect of MVA-GBDT and shows the results in Figure 4 The results show that accounting for scale effects can significantly improve the accuracy of downscaling methods. In several heterogeneous regions, the root mean square error (RMS) after accounting for scale effects was significantly reduced by approximately 20% compared to when excluding scale effects. This demonstrates that the accuracy of the proposed downscaling method is significantly improved by accounting for scale effects.

[0134] Careful consideration of scale effects not only provides deeper insights into the complex relationships between Chl-a concentrations and influencing factors, but also enables more accurate capture and reflection of the subtle impacts of environmental change during downscaling. This is particularly important in heterogeneous regions with diverse geographical and environmental conditions. In these regions, scale effects are particularly pronounced due to significant differences in environmental conditions and water quality characteristics. This requires a more nuanced consideration of scale factors during downscaling to ensure a comprehensive assessment of the performance of downscaling methods and to make effective adjustments and corrections to environmental changes at different scales.

[0135] Furthermore, in-depth analysis and consideration of scale effects not only improves the accuracy and stability of downscaling methods in capturing environmental changes, but also greatly enhances their practicality and reliability. This provides a powerful tool for monitoring and analyzing Chl-a concentrations, particularly in environmental management and scientific research requiring high precision and stability. Incorporating scale effects into Chl-a concentration downscaling methods means we can more accurately predict and assess the nutrient status and ecological health of water bodies, thereby providing a scientific basis and decision-making support for the management of various water bodies, such as lakes, rivers, and oceans.

[0136] Example 4

[0137] See also Figure 5 This embodiment proposes a system for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data. The system is implemented based on the method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data described in the above embodiment, and includes:

[0138] Acquisition module, used to acquire multi-source heterogeneous data;

[0139] A preprocessing module, configured to preprocess the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes;

[0140] An output module is used to input the multi-source heterogeneous data set into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality;

[0141] The improvement module is used to downscale the initial prediction results of the water pollution remote sensing parameter quality to obtain the water pollution remote sensing parameter quality improvement results.

[0142] In this embodiment, the acquired multi-source heterogeneous data are first preprocessed to obtain a multi-source heterogeneous data set with consistent attributes, which solves the problems of data differentiation and lack of uniformity, lays a solid and reliable foundation for the modeling and analysis of the water pollution remote sensing parameter quality improvement model, and also provides a reliable data basis and method support for the spatial downscaling research of water pollution remote sensing parameter quality; secondly, the multi-source heterogeneous data set is used as the input of the pre-trained water pollution remote sensing parameter quality improvement model, with the aim of providing accurate input data for the water pollution remote sensing parameter quality improvement model, so as to ensure that the initial prediction results of water pollution remote sensing parameter quality output by the water pollution remote sensing parameter quality improvement model are accurate and reliable. ; Then, the influence of scale effect is taken into account in the downscaling process of the water pollution remote sensing parameter quality improvement model, and the initial prediction results of the water pollution remote sensing parameter quality are downscaled and corrected, which can accurately capture the influence of scale effect of the water pollution remote sensing parameter quality improvement model in the output water pollution remote sensing parameter quality initial prediction results. By performing downscaling correction in the downscaling process, the initial prediction results of the water pollution remote sensing parameter quality can be further optimized and improved, further improving the prediction accuracy of the water pollution remote sensing parameter quality, ensuring that the water pollution remote sensing parameter quality improvement results adapt to environmental changes at different scales, and improving the reliability and accuracy of the water pollution remote sensing parameter quality improvement results.

[0143] Example 5

[0144] This embodiment also proposes a computer device, see Figure 6 , comprising: a processor 601, a memory 602, a communication interface 603 and a communication bus 604, wherein the processor 601, the memory 602 and the communication interface 603 communicate with each other via the communication bus 604;

[0145] Processor 601, memory 602, and communication interface 603 communicate with each other via a communication bus 604. Communication interface 603 is used for network communication with other devices, such as a client or other servers. Processor 601 is used to execute executable instructions 605, specifically, to perform the steps described in the aforementioned embodiment of the data processing-based chlorophyll a concentration spatial downscaling method.

[0146] Specifically, executable instructions 605 may include program code. Processor 601 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. A computer device includes one or more processors, which may be of the same type, such as one or more CPUs, or different types, such as one or more CPUs and one or more ASICs.

[0147] The memory 602 is used to store executable instructions 605. The memory 602 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0148] The executable instructions 605 may be specifically invoked by the processor 601 to cause the computer device to perform the following operations:

[0149] S1. Acquire multi-source heterogeneous data;

[0150] S2. Preprocessing the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes;

[0151] S3. Input the multi-source heterogeneous dataset into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality;

[0152] S4. Downscaling the initial prediction results of the water pollution remote sensing parameter quality to obtain a water pollution remote sensing parameter quality improvement result.

[0153] In this embodiment, the acquired multi-source heterogeneous data are first preprocessed to obtain a multi-source heterogeneous data set with consistent attributes, which solves the problems of data differentiation and lack of uniformity, lays a solid and reliable foundation for the modeling and analysis of the water pollution remote sensing parameter quality improvement model, and also provides a reliable data basis and method support for the spatial downscaling research of water pollution remote sensing parameter quality; secondly, the multi-source heterogeneous data set is used as the input of the pre-trained water pollution remote sensing parameter quality improvement model, with the aim of providing accurate input data for the water pollution remote sensing parameter quality improvement model, so as to ensure that the initial prediction results of water pollution remote sensing parameter quality output by the water pollution remote sensing parameter quality improvement model are accurate and reliable. ; Then, the influence of scale effect is taken into account in the downscaling process of the water pollution remote sensing parameter quality improvement model, and the initial prediction results of the water pollution remote sensing parameter quality are downscaled and corrected, which can accurately capture the influence of scale effect of the water pollution remote sensing parameter quality improvement model in the output water pollution remote sensing parameter quality initial prediction results. By performing downscaling correction in the downscaling process, the initial prediction results of the water pollution remote sensing parameter quality can be further optimized and improved, further improving the prediction accuracy of the water pollution remote sensing parameter quality, ensuring that the water pollution remote sensing parameter quality improvement results adapt to environmental changes at different scales, and improving the reliability and accuracy of the water pollution remote sensing parameter quality improvement results.

[0154] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. A person skilled in the art would be able to make other variations or modifications based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data, characterized in that: The following steps are involved: S1. Acquire multi-source heterogeneous data; S2. Preprocessing the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes; S3. Input the multi-source heterogeneous dataset into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality; S4. Downscaling the initial prediction results of the water pollution remote sensing parameter quality to obtain a water pollution remote sensing parameter quality improvement result.

2. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 1 is characterized in that: The multi-source heterogeneous data is multi-source remote sensing data, and the multi-source remote sensing data is preprocessed to obtain a multi-source heterogeneous data set with consistent attributes, including: S21. Geo-registering the multi-source remote sensing data to obtain spatially consistent multi-source remote sensing data; S22. geometrically correcting the spatially consistent multi-source remote sensing data to obtain multi-source remote sensing geometrically corrected data; S23. Defining a research scope for the multi-source remote sensing geometric correction data, and extracting high-precision boundary data from the multi-source remote sensing geometric correction data based on the research scope; S24. Resample the high-precision boundary data to obtain the multi-source heterogeneous data set.

3. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 2 is characterized in that: The multi-source remote sensing data is used as an input of an image registration tool, and the multi-source remote sensing data is geo-registered using the image registration tool to output spatially consistent multi-source remote sensing data.

4. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 2, characterized in that: The geometric correction of the spatially consistent multi-source remote sensing data includes: using a denoising and smoothing algorithm to remove noise data and abnormal values ​​in the spatially consistent multi-source remote sensing data to obtain the multi-source remote sensing geometrically corrected data.

5. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 2, characterized in that: The scope of research on multi-source remote sensing geometric correction data is limited to: S231. Using the modified normalized difference water index as a calculation tool for identifying the research range, inputting the multi-source remote sensing geometric correction data as a data source into the calculation tool, and outputting an initial range threshold; S232. Iterate the initial range threshold until the iteration step meets the maximum number of iterations and output the optimal range threshold; S233. Apply the optimal range threshold to the multi-source remote sensing geometric correction data to extract the high-precision boundary data.

6. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 1, characterized in that: The water pollution remote sensing parameter quality improvement model is a spatial downscaling model, which is constructed using a downscaling regression method.

7. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 1, characterized in that: The calculation expression for the output of the initial prediction result of the water pollution remote sensing parameter quality is as follows: CHL f =CHL c +(F c (S f )-F c (S c )) Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f Represents the output of the initial prediction results of the quality of water pollution remote sensing parameters with high resolution, CHL c Represents the original low-resolution water pollution remote sensing parameters, F c The calculation function that represents the statistical relationship between the low-resolution water pollution remote sensing parameters and the influencing characteristic factor set of water pollution remote sensing parameters; S c represents the set of influencing feature factors at high resolution, S f Represents the set of influencing feature factors at low resolution.

8. The chlorophyll a concentration spatial downscaling method based on data processing according to claim 1, characterized in that: The specific expression for downscaling the initial prediction result of the water pollution remote sensing parameter quality is as follows: CHL f '=CHL c +(F f (S f )-F c (S c )) Among them, the subscript f indicates high resolution, the subscript c indicates low resolution, and CHL f ′ represents the improvement result of water pollution remote sensing parameter quality, CHL c Represents the original low-resolution water pollution remote sensing parameters, F f The calculation function that represents the statistical relationship between the high-resolution water pollution remote sensing parameters and the influencing characteristic factor set of the water pollution remote sensing parameter concentration; S c represents the set of influencing feature factors at high resolution, S f Represents the set of influencing feature factors at low resolution.

9. A system for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data, the system being implemented based on the method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data according to any one of claims 1 to 8, characterized in that: include: Acquisition module, used to acquire multi-source heterogeneous data; A preprocessing module, configured to preprocess the multi-source heterogeneous data to obtain a multi-source heterogeneous data set with consistent attributes; An output module is used to input the multi-source heterogeneous data set into a pre-trained water pollution remote sensing parameter quality improvement model and output an initial prediction result of the water pollution remote sensing parameter quality; The improvement module is used to downscale the initial prediction results of the water pollution remote sensing parameter quality to obtain the water pollution remote sensing parameter quality improvement results.

10. A computer device, characterized in that: include: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform the operation of the method for improving the quality of water pollution remote sensing parameters based on multi-source heterogeneous data as described in any one of claims 1-8.

Images