An enhanced method and system for improving the resolution of water quality monitoring parameters

Abstract

The present invention proposes an enhanced method and system for improving the resolution of water quality monitoring parameters, which relates to the technical field of water quality monitoring. First, a preprocessing data set of multi-source remote sensing data of a water environment is obtained; then, the optimal characteristic factor set affecting the water quality monitoring parameters is screened from the preprocessing data set, and the preset water quality monitoring parameter resolution improvement model is trained using the optimal characteristic factor set to obtain a trained water quality monitoring parameter resolution improvement model; then, the multi-source remote sensing data to be processed is input into the water quality monitoring parameter resolution improvement model, and a high-resolution prediction result of the water quality monitoring parameter is output. The present invention effectively improves the prediction accuracy of water quality monitoring parameters in a complex water environment to maintain the nutritional status of the complex water environment and the health of the ecosystem.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality monitoring, and in particular to an enhanced method and system for improving the resolution of water quality monitoring parameters. Background Art

[0002] With the continuous increase of human activities and the impact of climate change, water quality monitoring of water environment has become particularly important. As an important water resource, water environment is not only directly related to the survival and development of human society, but also plays a key role in the global biogeochemical cycle and regional ecosystem. At present, the water quality of water environment is increasingly threatened by eutrophication and other problems. Among them, chlorophyll a (Chl-a) concentration, as one of the water quality monitoring parameters, is a direct indicator for assessing the degree of eutrophication of water environment, and is of great significance for the prediction of water quality monitoring parameters in water environment.

[0003] Traditional water quality monitoring parameter monitoring methods rely on ship sampling and laboratory spectrophotometry. This method is not only time-consuming and labor-intensive, but also costly and difficult to achieve continuous monitoring of large areas of water environment. Moreover, due to the rapid growth of phytoplankton and the complexity and variability of inland environment, it is difficult to obtain representative distribution of water quality monitoring parameter concentrations through sparse in-situ sampling. The development of remote sensing technology has provided a new way to solve the above problems. Satellite sensors can indirectly estimate the concentration of water quality monitoring parameters by measuring the reflectivity of the water surface, thereby achieving continuous monitoring of water quality. However, the spatial resolution of these satellite sensors is generally low. For example, the satellite MODIS-Aqua has a spatial resolution of 1 km, which is too coarse for fine monitoring of inland lakes. Although some satellite sensors such as Sentinel-3 OLCI provide a higher spatial resolution of 300m, their 300m spatial resolution still limits their wide application in small and medium-sized water environments. In order to improve the spatial resolution of concentration monitoring of water quality monitoring parameters, researchers began to explore spatial downscaling techniques for water quality monitoring parameters. For example, the methods for downscaling Chl-a concentration disclosed in the prior art mainly include methods based on image fusion and methods based on statistical regression. The image fusion-based method predicts high-resolution images of the target date by establishing statistical relationships between pairs of high- and low-resolution images at different time points. However, the accuracy of such methods is often limited when dealing with inland water environments with complex optical properties. The statistical regression-based method predicts high-resolution Chl-a concentrations using high-resolution remote sensing data by establishing a statistical model between Chl-a concentrations and auxiliary feature variables. With the development of machine learning technology, nonlinear data regression methods such as random forests and support vector regression have been widely used in downscaling research.

[0004] Therefore, the existing concentration downscaling methods for water quality monitoring parameters are mostly aimed at wide coastal water environments. For inland water environments, due to their relatively closed geographical location, complexity, and the great influence of natural and human activities, the causes of water quality monitoring parameters in inland water environments are complex and highly spatially heterogeneous, which affects the prediction accuracy of water quality monitoring parameters in complex water environments. Summary of the invention

[0005] In order to solve the problem of low prediction accuracy of water quality monitoring parameters in complex water environments in the above-mentioned prior art, the present invention proposes an enhanced method and system for improving the resolution of water quality monitoring parameters, which effectively improves the prediction accuracy of water quality monitoring parameters in complex water environments, so as to maintain the nutritional status of complex water environments and the health of the ecosystem.

[0006] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0007] An enhanced method for improving the resolution of water quality monitoring parameters comprises the following steps:

[0008] S1. Obtain preprocessed datasets of multi-source remote sensing data of water environment;

[0009] S2. Filter the optimal characteristic factor set that affects the water quality monitoring parameters from the preprocessed data set, and use the optimal characteristic factor set to train the preset water quality monitoring parameter resolution improvement model to obtain a trained water quality monitoring parameter resolution improvement model;

[0010] S3. Input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model, and output high-resolution prediction results of the water quality monitoring parameters.

[0011] Preferably, the multi-source remote sensing data includes satellite remote sensing image data and water quality monitoring parameter influencing factors, and the satellite remote sensing image data includes a first auxiliary feature data set, a second auxiliary feature data set and a third auxiliary feature data set.

[0012] Preferably, the multi-source remote sensing data is standardized and unified to obtain the preprocessed data set.

[0013] Preferably, screening the water quality monitoring parameter influencing factors from the preprocessed data set includes:

[0014] S201. Selecting the influencing factors of water quality monitoring parameters reflecting the actual situation of the water environment and having high influence from the preprocessed data set;

[0015] S202. Combining the water quality monitoring parameter influencing factors to obtain a water quality monitoring parameter influencing factor combination;

[0016] S203. Among the combinations of water quality monitoring parameter influencing factors, determine the most representative and influential combination of water quality monitoring parameter influencing factors as the optimal characteristic factor set.

[0017] Preferably, the water quality monitoring parameter influencing factors include multispectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics.

[0018] Preferably, the combination of influencing factors of water quality monitoring parameters is any combination of different types of multispectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics.

[0019] Preferably, the using of the optimal feature factor set to train a preset water quality monitoring parameter resolution enhancement model comprises:

[0020] S211. Establishing a statistical relationship between water quality monitoring parameters and an optimal characteristic factor set at low resolution through the water quality monitoring parameter resolution enhancement model;

[0021] S212. According to the statistical relationship, use the optimal characteristic factor sets of high resolution and low resolution as input, and output high resolution and low resolution water quality monitoring parameters;

[0022] S213. Calculate the difference between high-resolution and low-resolution water quality monitoring parameters, record the difference as high-resolution texture information, add the high-resolution texture information to the original low-resolution water quality monitoring parameters as a guide to train the water quality monitoring parameter resolution improvement model until the water quality monitoring parameter resolution improvement model outputs water quality monitoring parameters with high resolution, and the training is completed.

[0023] Preferably, the calculation expression for outputting the water quality monitoring parameter with high resolution is as follows:

[0024]

[0025] Among them, the subscript f Indicates high resolution, subscript c Indicates low resolution, Represents the output of water quality monitoring parameters with high resolution, Represents the original low-resolution water quality monitoring parameters, a function representing the statistical relationship; represents the optimal feature factor set at high resolution, Represents the optimal feature factor set at low resolution.

[0026] Preferably, in the process of training the preset water quality monitoring parameter resolution improvement model, the hyperparameters of the water quality monitoring parameter resolution improvement model need to be adjusted to complete the training of the water quality monitoring parameter resolution improvement model.

[0027] The present invention also proposes an enhanced water quality monitoring parameter resolution improvement system, comprising:

[0028] An acquisition module is used to obtain preprocessed data sets of multi-source remote sensing data of water environment;

[0029] A training module, used to screen the optimal characteristic factor set affecting the water quality monitoring parameters from the preprocessed data set, and train a preset water quality monitoring parameter resolution improvement model using the optimal characteristic factor set to obtain a trained water quality monitoring parameter resolution improvement model;

[0030] The output module is used to input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model and output the high-resolution prediction results of the water quality monitoring parameters. The prediction accuracy of water quality monitoring parameters in complex water environments is low.

[0031] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0032] The present invention proposes an enhanced method and system for improving the resolution of water quality monitoring parameters, which takes into account the differences in the causes of water quality monitoring parameters in inland water environments and coastal water environments. First, a preprocessing data set of multi-source remote sensing data of the water environment is obtained, and a preset water quality monitoring parameter resolution improvement model is trained through the preprocessing data set, so as to extend the prediction of water quality monitoring parameters from the coastal water environment to the inland water environment, so as to meet the challenge of fine prediction of water quality monitoring parameters in complex water environments; then, an optimal feature factor set affecting the water quality monitoring parameters is screened from the preprocessing data set, and the preset water quality monitoring parameter resolution improvement model is trained using the optimal feature factor set, and the optimal feature factor set is integrated into the training process of the water quality monitoring parameter resolution improvement model as a high spatial resolution explanatory variable to optimize the trained water quality monitoring parameter resolution improvement model; further, the multivariate analysis of the optimal feature factor set and the high efficiency of the prediction of the water quality monitoring parameter resolution improvement model are fully combined, and the output of the water quality monitoring parameter resolution improvement model is optimized through in-depth analysis, so as to effectively improve the prediction accuracy of water quality monitoring parameters in complex water environments, so as to maintain the nutritional status of the complex water environment and the health of the ecosystem. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic flow chart showing an enhanced method for improving the resolution of water quality monitoring parameters proposed in an embodiment of the present invention;

[0034] Figure 2 A diagram showing the importance of the influencing factors of water quality monitoring parameters proposed in the embodiment of the present invention and the corresponding changes in the accuracy of the downscaling model as the added elements change;

[0035] Figure 3 A heat map showing the correlation between the influencing factors of various water quality monitoring parameters proposed in the embodiment of the present invention;

[0036] Figure 4 A graph showing the first downscaling accuracy results for lakes 1 to 12 using different methods proposed in the embodiments of the present invention;

[0037] Figure 5 A graph showing the second downscaling accuracy results for lakes 1-12 using different methods proposed in the embodiments of the present invention;

[0038] Figure 6 A third downscaling accuracy result diagram for lakes No. 1 to 12 using different methods proposed in the embodiments of the present invention is shown;

[0039] Figure 7 A fourth downscaling accuracy result diagram for lakes No. 1 to No. 12 using different methods proposed in the embodiments of the present invention is shown;

[0040] Figure 8 It represents the pixel-based Sentinel-3 Chl-a concentration density scatter plot and the downscaled Chl-a concentration density scatter plot with a resolution of 300 m drawn in lakes 1 to 6 using different methods proposed in the embodiments of the present invention;

[0041] Fig. 9 It represents the pixel-based Sentinel-3 Chl-a concentration density scatter plot and the downscaled Chl-a concentration density scatter plot with a resolution of 300 m drawn in lakes 7-12 using different methods proposed in the embodiments of the present invention;

[0042] Fig.10 A graph showing the comparative analysis results of the spatial distribution differences between the Chl-a concentration downscaling maps of Lakes 1 to 6 drawn using different methods and the original Chl-a concentration of Sentinel-3 proposed in the embodiments of the present invention;

[0043] Fig.11 A graph showing the comparative analysis results of the spatial distribution differences between the Chl-a concentration downscaling maps of Lakes 7-12 drawn using different methods and the original Chl-a concentration of Sentinel-3 proposed in the embodiments of the present invention;

[0044] Fig.12The scatter plot and linear fitting graph of Sentinel-3 and Landsat-8 Chl-a concentrations of Qinghai Lake proposed in the embodiments of the present invention are shown;

[0045] Fig.13 The scatter plot and linear fitting graph of Sentinel-3 and Landsat-8 Chl-a concentrations of Chaohu Lake proposed in the embodiments of the present invention are shown;

[0046] Fig.14 A density scatter plot downscaling diagram showing the Chl-a concentration of the Landsat-8 Chl-a concentration of Qinghai Lake proposed in an embodiment of the present invention and the Chl-a concentration at a resolution of 30 m using a water quality monitoring parameter resolution enhancement model that incorporates multivariate analysis based on pixels;

[0047] Fig.15 A density scatter plot downscaling diagram showing the Landsat-8 Chl-a concentration of Chaohu Lake proposed in an embodiment of the present invention and the Chl-a concentration at a resolution of 30 m using a water quality monitoring parameter resolution enhancement model that incorporates multivariate analysis based on pixels;

[0048] Fig.16 A graph showing the downscaling results of Chl-a concentration in Qinghai Lake proposed in an embodiment of the present invention;

[0049] Fig.17 A graph showing the downscaling results of the Chl-a concentration of Chaohu Lake proposed in an embodiment of the present invention;

[0050] Fig.18 A structural diagram of an enhanced water quality monitoring parameter resolution improvement system proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0051] The accompanying drawings are only used for illustrative purposes and are not to be construed as limiting the present invention;

[0052] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0053] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0054] Example 1

[0055] See also Figure 1 This embodiment proposes an enhanced method for improving the resolution of water quality monitoring parameters. The water quality monitoring parameter is chlorophyll-a (Chl-a) as an example for explanation. In actual application, the water quality monitoring parameter is not limited to Chl-a. The method includes the following steps:

[0056] S1. Obtain preprocessed datasets of multi-source remote sensing data of water environment;

[0057] In S1, the water environment selected 12 inland lakes, including Qinghai Lake, Poyang Lake, Dongting Lake, Taihu Lake

[0058] Hulun Lake, Hongze Lake, Namtso Lake, Selin Co Lake, Bosten Lake, Gaoyou Lake, Chaohu Lake, Weishan Lake; the geographical location, climatic conditions and water quality of the 12 inland lakes are shown in Table 1;

[0059] Table 1 Geographical location, climate conditions and water quality of water environment

[0060]

[0061] The multi-source remote sensing data includes satellite remote sensing image data and water quality monitoring parameter influencing factors, the satellite remote sensing image data includes a first auxiliary feature data set, a second auxiliary feature data set and a third auxiliary feature data set; the first auxiliary feature data set is a satellite data set Sentinel-1, the second auxiliary feature data set is a satellite data set Sentinel-2, and the third auxiliary feature data set is a satellite data set Landsat-8, and the satellite remote sensing image data acquisition methods include:

[0062] First, Sentinel-1 and Sentinel-2 data were obtained from the open access center. The spatial resolution of Sentinel-1 and Sentinel-2 data is 10 m. Sentinel-1 data is obtained by Sentinel-1A satellite providing dual-polarization (VV+VH) C-band synthetic aperture radar data, and Sentinel-2 data is obtained by Sentinel-2A satellite equipped with 13-band multispectral data. Sentinel-2A satellite covers the range from 56 degrees south latitude to 84 degrees north latitude. Second, the secondary data of the satellite dataset Landsat-8 was downloaded from the website of the Geological Survey, with a spatial resolution of 30 m. Sentinel-1, Sentinel-2 and Landsat-8 provide rich data resources for subsequent analysis.

[0063] The 10 m resolution surface reflectance data from Sentinel-2 data was selected because of its high resolution and multi-wavelength advantages, including visible light, near infrared, shortwave infrared and unique red edge bands; the VV and VH bands of Sentinel-1 were also used to simulate the backscattering characteristics of the lake; in addition, the thermal infrared sensor suite (TIRS) on the Landsat-8 satellite was used to obtain the surface temperature data of the lake; as supplementary information, the Sentinel-2 satellite was also used to obtain the texture feature information between different lakes, including image correlation (GCLM_Cor), contrast (GCLM_Var), entropy (GCLM_Con) and variance (GCLM_Ent).

[0064] However, due to the mismatch in geometry and attributes between the Sentinel-1, Sentinel-2 and Landsat-8 datasets, this brings certain challenges to dataset processing. In order to solve this problem, the multi-source remote sensing data are standardized and unified, specifically including standardizing and unifying the Sentinel-1, Sentinel-2 and Landsat-8 datasets in turn to obtain a unified preprocessed dataset; this lays a solid and reliable foundation for subsequent modeling and analysis.

[0065] S2. Filter the optimal characteristic factor set that affects the water quality monitoring parameters from the preprocessed data set, and use the optimal characteristic factor set to train the preset water quality monitoring parameter resolution improvement model to obtain a trained water quality monitoring parameter resolution improvement model;

[0066] S3. Input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model, and output high-resolution prediction results of the water quality monitoring parameters.

[0067] In this embodiment, the overview of the water environment covers important information such as the geographical distribution range of the selected lake, the surrounding land use type and water quality conditions. These detailed information provides important background information for the research and helps to fully understand the environmental characteristics and features of the study area; these descriptions can help people better understand the research objects and provide an important reference basis for subsequent data analysis and experimental design. First, by obtaining a preprocessed data set of multi-source remote sensing data of the water environment, the sources of the data used and the classification of the acquisition paths are demonstrated in terms of data acquisition. The data acquisition process and data parameters are introduced, as well as the main functions of the selected data. These data include fine-resolution auxiliary data from Sentinel-1, Sentinel-2 and Landsat-8 satellites, covering multiple optical and radar bands, providing rich data resources for subsequent analysis; the preset water quality monitoring parameter resolution improvement model is then trained through the preprocessed data set, with the aim of extending the prediction of water quality monitoring parameters from coastal water environments to inland water environments, so as to meet the challenge of fine-tuning the prediction of water quality monitoring parameters in complex water environments; then, the optimal characteristic factor set affecting the water quality monitoring parameters is screened from the preprocessed data set, and the preset water quality monitoring parameter resolution improvement model is trained using the optimal characteristic factor set, and the optimal characteristic factor set is integrated into the training process of the water quality monitoring parameter resolution improvement model as a high spatial resolution explanatory variable to optimize the trained water quality monitoring parameter resolution improvement model; further, the multivariate analysis of the optimal characteristic factor set and the high efficiency of the prediction of the water quality monitoring parameter resolution improvement model are fully combined, and the output of the water quality monitoring parameter resolution improvement model is optimized through in-depth analysis, which effectively improves the prediction accuracy of water quality monitoring parameters in complex water environments, so as to maintain the nutrient status of complex water environments and the health of the ecosystem.

[0068] Example 2

[0069] This embodiment further illustrates the training of the water quality monitoring parameter resolution enhancement model, taking Chl-a as the water quality monitoring parameter as the research object; the screening of the water quality monitoring parameter influencing factors from the preprocessed data set includes:

[0070] S201. Selecting the influencing factors of water quality monitoring parameters reflecting the actual situation of the water environment and having high influence from the preprocessed data set;

[0071] In S201, the formation of Chl-a concentration in inland waters is complicated and diverse. It is found that Chl-a concentration is relatively sensitive to spectral reflectance, texture characteristics and water temperature. At the same time, chlorophyll and red edge index are crucial to the extraction of Chl-a concentration. Therefore, this embodiment integrates multi-source remote sensing data, selects 21 indicators, and constructs a comprehensive Chl-a indicator system. As shown in Table 2, Chl-a indicators (i.e., water quality monitoring parameter influencing factors) include six indicators: multi-spectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics; these six indicators are defined as follows:

[0072] Multispectral reflectance: It is widely considered to be an important parameter for Chl-a inversion, mainly including visible light, near infrared and short-wave infrared bands. Multispectral reflectance obtained from different wavelengths of the electromagnetic spectrum helps to detect changes in light reflection and absorption, and helps to evaluate the composition and surface characteristics of lake water bodies through the understanding of spectral response.

[0073] Chlorophyll index: including the normalized chlorophyll difference index NDCI and the blue-red-green index (BRG), is a specific indicator studied by scholars to reflect the chlorophyll in water bodies and is used to characterize the chlorophyll content in water bodies.

[0074] Red Edge Index: Calculated based on the unique red edge band of Sentinel-2 images. The red edge index derived from the vegetation red edge band is closely related to Chl-a concentration.

[0075] Texture features: including correlation, contrast, entropy and variance. Texture information can improve the accuracy of Chl-a concentration assessment. Therefore, the texture features obtained by the grey level co-occurrence matrix (GLCM) method are used as Chl-a concentration factors.

[0076] Backscatter properties: Variables related to backscatter properties, particularly from radar signals, provide insights into how these signals interact with surface and submerged features. Understanding these properties can aid in interpreting the signals and estimating the properties of lake water structures.

[0077] Thermal signature, image from Landsat-8 TIRS. Under normal conditions, the surface temperature of turbid waters is highly correlated with the Chl-a concentration, because the cooling rate is inversely proportional to the trophic state of the lake. For this reason, the WST value of two adjacent points can be expressed as:

[0078]

[0079] in, ∆WST The lake surface temperature changes. and is the Chl-a concentration of two adjacent points.

[0080] Table 2 Chl-a index system details

[0081]

[0082] S202. Combining the water quality monitoring parameter influencing factors to obtain a water quality monitoring parameter influencing factor combination;

[0083] In S202, the water quality monitoring parameter influencing factor combination is any combination of different types of multispectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics, such as multispectral reflectance, multispectral reflectance + chlorophyll index, multispectral reflectance + red edge index, multispectral reflectance + texture characteristics, multispectral reflectance + thermal characteristics, multispectral reflectance + backscattering characteristics, all water quality monitoring parameter influencing factors and other water quality monitoring parameter influencing factor combinations; by adding different types of water quality monitoring parameter influencing factor combinations and their influence on the downscaling accuracy of the water quality monitoring parameter resolution enhancement model, it is convenient to better understand and identify the water quality monitoring parameter influencing factors that are most critical to the performance improvement of the water quality monitoring parameter resolution enhancement model. This insight is of great significance for optimizing the construction of the water quality monitoring parameter resolution enhancement model and improving its prediction accuracy.

[0084] S203. Among the combinations of water quality monitoring parameter influencing factors, determine the most representative and influential combination of water quality monitoring parameter influencing factors as the optimal characteristic factor set.

[0085] In S203, in order to establish the most representative and influential combination of water quality monitoring parameter influencing factors for building a robust downscaling framework, the importance of 21 water quality monitoring parameter influencing factors was first evaluated. The gradient boosting decision tree (GBDT) method plays an important role in factor selection. This method is not only applicable to classification and regression, but also can help determine the most influential factors. Therefore, the water quality monitoring parameter resolution improvement model is selected as the downscaling model based on the gradient boosting decision tree. Then, according to the results of the importance ranking, the water quality monitoring parameter influencing factors are added one by one to the downscaling model based on the gradient boosting decision tree to conduct a downscaling experiment; finally, the combination of water quality monitoring parameter influencing factors with the highest downscaling accuracy is selected as the final optimal feature factor set. The individual importance of each water quality monitoring parameter influencing factor and the downscaling accuracy that changes with the change of the water quality monitoring parameter influencing factor are shown in Figure 2. Figure 2 As shown. Figure 2 It can be seen that the two most important characteristic factors for the water quality monitoring parameter resolution improvement model are the normalized difference vegetation index NDCI and the water surface temperature WST, with importance values ​​of 16.5% and 11.5%, respectively. The reason behind this result is that when Chl-a components (such as phytoplankton in the water body) appear in inland lakes, more near-infrared light will be reflected and more red light will be absorbed, which leads to a significant increase in NDCI. At the same time, all lakes in the study area showed a high nutrient status (see Table 1 for details), which indicates that there is a potential coupling relationship between NDCI and Chl-a concentration in the nutrient-rich lakes throughout the water area. In addition, warm temperatures provide the basic conditions for the massive reproduction of phytoplankton in the water body, so WST can significantly affect the number of phytoplankton in the water body, which in turn has a sharp impact on the changes in Chl-a concentration.

[0086] The above analysis shows that the two characteristic factors, NDCI and WST, play a vital role in the downscaling model, and there is a close correlation between them and Chl-a concentration. This conclusion not only helps to deepen our understanding of the dynamic changes of Chl-a concentration in lake ecosystems, but also provides an important reference for establishing accurate downscaling models. As key indicators reflecting the ecological environment and water quality characteristics of water bodies, the close connection between NDCI and WST and Chl-a concentration shows their importance in monitoring and evaluating the chlorophyll content of water bodies. Therefore, making full use of the information of these two characteristic factors is crucial to improving the prediction accuracy and robustness of the downscaling model. Through in-depth analysis and effective use of NDCI and WST, the spatial distribution and change trend of Chl-a concentration can be captured more accurately, providing a reliable scientific basis for lake environmental management and water quality monitoring.

[0087] In addition, through Figure 2 It can also be found that as the number of auxiliary characteristic factors included in the downscaling model gradually increases, the R 2 It has significantly increased from 32.5% in the first stage (the first 6 feature factors are added to the downscaling model) to 75.8%. This significant increase is mainly due to the relatively high importance scores of the auxiliary feature factors at this time, and the low correlation between these feature factors, with the maximum correlation coefficient being only 0.67, see Figure 3 In the medium term (as the 7th to 14th characteristic factors are added to the downscaling model), although the R 2 The improvement rate decreased significantly, but the overall accuracy continued to increase steadily, eventually reaching 75.8%. In the later stage (the 15th to 21st feature factors were added to the downscaling model), the R2 This is mainly because at this stage, the number of redundant and irrelevant feature factors increases, and there is a high correlation between the existing feature factors (such as Figure 3 As shown in the figure, the correlation between NDre2 and GCLM_Con reached -0.93, and the two characteristic factors were highly correlated), which led to the mutual influence between the data and the suppression of the independence of the data, thereby suppressing the expressiveness of the downscaling model and reducing the accuracy of the downscaling model. Therefore, considering the situation at each stage, the first 14 factors were determined as the optimal characteristic factor set. Specifically, the optimal characteristic factor set is NDCI, WST, Green, NIR, NDVIre2, Blue, VV, NDVIre3, GCLM_Ent, Red, NDre1, BRG, NDVIre1 and GCLM_Con. These optimal characteristic factor sets play a key role in the downscaling model. Their selection can not only improve the expressiveness of the downscaling model, but also reduce the redundancy between data, thereby further improving the prediction accuracy and robustness of the downscaling model.

[0088] The water quality monitoring parameter resolution enhancement model selected in step S2 is a downscaling model based on a gradient boosting decision tree, denoted as a GBDT model. The GBDT model constructs a set 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. Most downscaling algorithms are based on the scale invariance assumption, that is, the statistical relationship between Chl-a concentration and explanatory variables at low spatial resolution also holds true at high spatial resolution. Under this assumption, the preset water quality monitoring parameter resolution enhancement model is trained using the optimal feature factor set, including:

[0089] S211. Establishing a statistical relationship between water quality monitoring parameters and an optimal characteristic factor set at low resolution through the water quality monitoring parameter resolution enhancement model;

[0090] S212. According to the statistical relationship, use the optimal characteristic factor sets of high resolution and low resolution as input, and output high resolution and low resolution water quality monitoring parameters;

[0091] S213. Calculate the difference between high-resolution and low-resolution water quality monitoring parameters, record the difference as high-resolution texture information, add the high-resolution texture information to the original low-resolution water quality monitoring parameters as a guide to train the water quality monitoring parameter resolution improvement model until the water quality monitoring parameter resolution improvement model outputs water quality monitoring parameters with high resolution, and the training is completed.

[0092] In S213, the calculation expression of the output water quality monitoring parameter with high resolution is as follows:

[0093]

[0094] Among them, the subscript f Indicates high resolution, subscript c Indicates low resolution, Represents the output of water quality monitoring parameters with high resolution, Represents the original low-resolution water quality monitoring parameters, a function representing the statistical relationship; represents the optimal feature factor set at high resolution, Represents the optimal feature factor set at low resolution.

[0095] In the process of training the preset water quality monitoring parameter resolution improvement model, it is necessary to adjust the hyperparameters of the water quality monitoring parameter resolution improvement model to complete the training of the water quality monitoring parameter resolution improvement model. Among them, there are three hyperparameters of the water quality monitoring parameter resolution improvement model, namely the number of decision trees in the regression program, the explanatory factors used to divide the nodes, and the loss function. The parameter adjustment of the model is crucial to avoid multicollinearity and overfitting. In this work, this embodiment adopts a two-stage procedure to achieve the hyperparameter adjustment of the water quality monitoring parameter resolution improvement model. In the first stage, the importance assessment and stepwise regression methods are 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, the grid search method and the 10-fold cross-validation method are used to determine the number of decision trees and the loss function. Finally, 100 and the minimum absolute deviation are determined as the final number and loss function of the decision tree, respectively.

[0096] For the trained water quality monitoring parameter resolution improvement model, it is necessary to evaluate its accuracy in predicting water quality monitoring parameters by verifying the indicators of the water quality monitoring parameter resolution improvement model, which include the determination coefficient R 2 , Root Mean Square Error RMSE , mean absolute error MAE and structural similarity SSIM Through statistical analysis of these four indicators, the performance and accuracy of the trained water quality monitoring parameter resolution improvement model can be comprehensively evaluated. R The value of ² is the core indicator for evaluating the prediction ability of the water quality monitoring parameter resolution improvement model. The closer its value is to 1, the higher the consistency between the prediction results of the water quality monitoring parameter resolution improvement model and the actual situation. RMSE As a key measure of forecast error, RMSEThe smaller the value, the higher the prediction accuracy of the water quality monitoring parameter resolution enhancement model. RMSE similar, MAE It is also commonly used to evaluate the accuracy of the resolution improvement model for water quality monitoring parameters. It focuses on the average of the absolute values ​​of the differences between the predicted values ​​and the actual values. SSIM It measures the structural similarity between two images. The higher the value, the higher the similarity between the downscaled image and the original image.

[0097] Coefficient of determination R 2 The calculation expression is as follows:

[0098]

[0099] Among them, the subscript f Indicates high resolution, subscript c Indicates low resolution, It represents the average value of the concentration of water quality monitoring parameter (Chl-a) at high resolution; m Indicates the number of samples;

[0100] Root mean square error RMSE The calculation expression is as follows:

[0101]

[0102] Mean absolute error MAE The calculation expression is as follows:

[0103]

[0104] Structural similarity SSIM The calculation expression is as follows:

[0105]

[0106] in, It represents the average value of the concentration of water quality monitoring parameter (Chl-a) at low resolution; represents the first constant used to maintain numerical stability; represents the second constant used to maintain numerical stability, which are set to 0.1 and 0.05 respectively; Represents the covariance of the concentration of water quality monitoring parameter (Chl-a).

[0107] Example 3

[0108] In order to further evaluate the performance of the water quality monitoring parameter resolution improvement model established by the gradient boosting decision tree downscaling method in downscaling Chl-a concentration, this embodiment introduces the multivariate analysis of the influencing factors of water quality monitoring parameters into the downscaling method, and records the water quality monitoring parameter resolution improvement model integrating the multivariate analysis as the GBDT model. The accuracy of the GBDT model in predicting Chl-a concentration is compared with that of other existing downscaling methods based on artificial neural network (ANN) to predict Chl-a concentration, based on random forest (RF) to predict Chl-a concentration, and based on multivariate polynomial regression (MPR) to predict Chl-a concentration. The four methods are all based on the same input, see Figure 4-Figure 7 , The accuracy of downscaling of multiple lakes by four different methods is demonstrated. In different lakes, the GBDT model of this embodiment always provides the best results, outperforming ANN, RF and MPR. Although the MPR method performs well in some cases, overall, the performance of the GBDT method is better.

[0109] See also Figure 8 and Fig. 9 , respectively, shows the pixel density scatter plots of four different downscaling methods. Through observation, we can have a deeper understanding of the performance differences of various methods in dealing with Chl-a concentration; first, it can be clearly observed that the data points of the GBDT method are more concentrated and show a tighter distribution. This shows that the GBDT method has higher stability and consistency in predicting Chl-a concentration. In contrast, the data points of other methods are relatively scattered.

[0110] See also Fig.10 and Fig.11 , respectively, provide an intuitive comparison of different downscaling methods in reproducing the spatial distribution of Chl-a concentration. The GBDT method shows higher accuracy and better ability to restore the original spatial features, which makes it more advantageous in dealing with the spatial downscaling task of Chl-a concentration.

[0111] In this embodiment, the downscaling results of the models established by different methods are compared. By evaluating different models such as linear regression, polynomial regression and GBDT, their effects and performance in the downscaling process are explored. The results show that compared with traditional linear regression and polynomial regression models, the GBDT model combined with multivariate analysis has higher downscaling accuracy and prediction accuracy. This shows that when dealing with the problem of Chl-a concentration downscaling, the GBDT model combined with multivariate analysis can more effectively utilize data features and complex relationships, thereby improving the prediction accuracy and robustness of the GBDT model.

[0112] Example 4

[0113] This embodiment further verifies the reliability and effectiveness of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis by using satellite remote sensing image data. In order to verify the performance of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis, the widely used "zoom-in-zoom-out" technology is adopted. Specifically, the original 300 m Sentinel-3 Chl-a concentration is first enlarged to 3000 m, and then the 3000 m Chl-a concentration is reduced to 300 m using the water quality monitoring parameter resolution improvement model integrated with multivariate analysis, and compared with the reference 300 m Chl-a concentration; for further verification, two Landsat-8 Chl-a concentration images with a spatial resolution of 30 m (Lake No. 1 Qinghai Lake and Lake No. 11 Chaohu Lake) are collected, and the performance of the water quality monitoring parameter resolution improvement model is tested by reducing the Chl-a concentration from 300m to 30m. This verification method can not only evaluate the applicability of the water quality monitoring parameter resolution improvement model at different spatial resolutions, but also verify its generalization ability on different data sources and lakes, thereby ensuring the reliability and effectiveness of the model.

[0114] Fig.12 and Fig.13 The scatter plot and linear regression relationship between Sentinel-3 and Landsat-8 Chl-a concentration products at coarse spatial resolution for the two sub-study areas are shown. Fig.12 and Fig.13 As shown in the figure, within the 95% confidence interval, the determination coefficients of Sentinel-3 and Landsat-8 Chl-a concentrations reached 0.93 and 0.88, respectively, and the root mean square error was controlled below 3.50 mg / m³. This shows that there is a high correlation between Sentinel-3 and Landsat-8 Chl-a concentration products, and the linear regression relationship is stable. Therefore, the Landsat-8 Chl-a concentration product can be used as an effective verification data for the downscaling results of Sentinel-3 Chl-a concentration, and the reliability and accuracy of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis under different data sources are verified. This verification method not only helps to evaluate the applicability of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis on different data sources, but also verifies the generalization ability and reliability of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis, which provides strong support for the promotion of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis in practical applications.

[0115] Fig.14 and Fig.15The pixel density scatter plots of Landsat-8 Chl-a concentration and downscaled Chl-a concentration at 30 m resolution using the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis in two lakes are shown. Although these evaluation results are slightly different from those from 3000 m to 300 m, the main results are similar. For different inland lakes, the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis based on the downscaling method of gradient boosting decision tree has achieved satisfactory results, with an average RMSE lower than 4.00 mg / m³ and R² higher than 0.80. This shows that the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis is very reliable in downscaling Chl-a concentration from 300 m to 30 m.

[0116] See also Fig.16 and Fig.17 , the water quality monitoring parameter resolution improvement model integrated with multivariate analysis downscales the Chl-a concentration while maintaining the overall spatial pattern of the 300 m Chl-a concentration, adding rich Chl-a concentration details, making it very similar to the reference Chl-a concentration. This means that the water quality monitoring parameter resolution improvement model integrated with multivariate analysis can not only accurately reproduce the overall spatial distribution of Chl-a concentration, but also capture more detailed local changes, thereby improving the resolution of the spatial characteristics of Chl-a concentration. By increasing the capture of local characteristics inside the lake, the water quality monitoring parameter resolution improvement model integrated with multivariate analysis retains more detailed information during the downscaling process, making the downscaling results more refined and accurate. Further verification also shows that when the lake heterogeneity is high, the root mean square error of the water quality monitoring parameter resolution improvement model integrated with multivariate analysis for downscaling all Chl-a concentrations is within a satisfactory range. This shows that the water quality monitoring parameter resolution improvement model method integrated with multivariate analysis has strong adaptability and robustness when processing Chl-a concentration data from different lakes. Even in the case of lakes with complex topographic structures and hydrological characteristics, the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis can effectively capture the spatial distribution pattern of Chl-a concentration and produce reliable downscaling results.

[0117] In this example, these verification results provide solid support for the downscaling of Sentinel-3 Chl-a concentration from 300 m to 30 m by using the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis in practical applications. The accuracy and robustness of the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis make it a reliable tool for high-resolution spatial analysis of lake Chl-a concentration. In future research and applications, the applicability of the water quality monitoring parameter resolution enhancement model integrated with multivariate analysis under different environmental and regional conditions can be further explored, and it can be further optimized and improved to meet the needs for spatial information on Chl-a concentration in different scenarios.

[0118] The robustness of these verification results further verified the reliability and effectiveness of the water quality monitoring parameter resolution improvement model that integrates multivariate analysis. Through verification in different lakes and at different resolutions, the applicability of the water quality monitoring parameter resolution improvement model that integrates multivariate analysis under different conditions was confirmed, and the wide applicability of the water quality monitoring parameter resolution improvement model that integrates multivariate analysis in Chl-a concentration downscaling was proved. This provides a reliable basis for further using the water quality monitoring parameter resolution improvement model that integrates multivariate analysis to monitor and spatially analyze Chl-a concentrations in inland lakes, and provides important technical support for research and application in related fields.

[0119] Example 5

[0120] This embodiment also proposes an enhanced water quality monitoring parameter resolution improvement system, including:

[0121] An acquisition module is used to obtain preprocessed data sets of multi-source remote sensing data of water environment;

[0122] A training module, used to screen the optimal characteristic factor set affecting the water quality monitoring parameters from the preprocessed data set, and train a preset water quality monitoring parameter resolution improvement model using the optimal characteristic factor set to obtain a trained water quality monitoring parameter resolution improvement model;

[0123] The output module is used to input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model and output the high-resolution prediction results of the water quality monitoring parameters.

[0124] In this embodiment, firstly, a preprocessed data set of multi-source remote sensing data of a water environment is obtained, and a preset water quality monitoring parameter resolution improvement model is trained through the preprocessed data set, with the purpose of extending the prediction of water quality monitoring parameters from coastal water environments to inland water environments, so as to meet the challenge of fine prediction of water quality monitoring parameters in complex water environments; then, an optimal feature factor set affecting water quality monitoring parameters is screened from the preprocessed data set, and the preset water quality monitoring parameter resolution improvement model is trained using the optimal feature factor set, and the trained water quality monitoring parameter resolution improvement model is optimized by integrating the optimal feature factor set as a high spatial resolution explanatory variable into the training process of the water quality monitoring parameter resolution improvement model; further, the multivariate analysis of the optimal feature factor set and the high efficiency of the prediction of the water quality monitoring parameter resolution improvement model are fully combined, and the output of the water quality monitoring parameter resolution improvement model is optimized through in-depth analysis, so as to effectively improve the prediction accuracy of water quality monitoring parameters in complex water environments, so as to maintain the nutritional status of the complex water environment and the health of the ecosystem.

[0125] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the implementation methods of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all implementation methods here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. An enhanced method for improving the resolution of water quality monitoring parameters, characterized in that: The following steps are involved: S1. Obtain a preprocessed data set of multi-source remote sensing data of a water environment; the multi-source remote sensing data includes satellite remote sensing image data and water quality monitoring parameter influencing factors, and the satellite remote sensing image data includes a first auxiliary feature data set, a second auxiliary feature data set and a third auxiliary feature data set; S2. Filter the optimal characteristic factor set that affects the water quality monitoring parameters from the preprocessed data set, and use the optimal characteristic factor set to train the preset water quality monitoring parameter resolution improvement model to obtain a trained water quality monitoring parameter resolution improvement model; Screening the influencing factors of water quality monitoring parameters from the preprocessed data set includes: S201. Selecting the influencing factors of water quality monitoring parameters reflecting the actual situation of the water environment and having high influence from the preprocessed data set; S202. Combining the water quality monitoring parameter influencing factors to obtain a water quality monitoring parameter influencing factor combination; S203. Determine the most representative and influential combination of water quality monitoring parameter influencing factors among the combinations of water quality monitoring parameter influencing factors as the optimal characteristic factor set; The preset water quality monitoring parameter resolution enhancement model is trained using the optimal feature factor set, including: S211. Establishing a statistical relationship between water quality monitoring parameters and an optimal characteristic factor set at low resolution through the water quality monitoring parameter resolution enhancement model; S212. According to the statistical relationship, use the optimal characteristic factor sets of high resolution and low resolution as input, and output high resolution and low resolution water quality monitoring parameters; S213. Calculate the difference between the high-resolution and low-resolution water quality monitoring parameters, record the difference as high-resolution texture information, add the high-resolution texture information to the original low-resolution water quality monitoring parameters as a guide to train the water quality monitoring parameter resolution enhancement model until the water quality monitoring parameter resolution enhancement model outputs water quality monitoring parameters with high resolution, and the training is completed; S3. Input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model, and output high-resolution prediction results of the water quality monitoring parameters.

2. The enhanced water quality monitoring parameter resolution improvement method according to claim 1 is characterized in that: The multi-source remote sensing data is standardized and unified to obtain the preprocessed data set.

3. The enhanced water quality monitoring parameter resolution improvement method according to claim 1 is characterized in that: The water quality monitoring parameter influencing factors include multispectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics.

4. The enhanced water quality monitoring parameter resolution improvement method according to claim 3 is characterized in that: The water quality monitoring parameter influencing factor combination is any combination of different types of multi-spectral reflectance, chlorophyll index, red edge index, texture characteristics, backscattering characteristics and thermal characteristics.

5. The enhanced water quality monitoring parameter resolution improvement method according to claim 4 is characterized in that: The calculation expression of the output water quality monitoring parameter with high resolution is as follows: Among them, the subscript f Indicates high resolution, subscript c Indicates low resolution, Represents the output of water quality monitoring parameters with high resolution, Represents the original low-resolution water quality monitoring parameters, a function representing the statistical relationship; represents the optimal feature factor set at high resolution, Represents the optimal set of feature factors at low resolution.

6. The enhanced water quality monitoring parameter resolution improvement method according to claim 5 is characterized in that: In the process of training the preset water quality monitoring parameter resolution improvement model, it is necessary to adjust the hyperparameters of the water quality monitoring parameter resolution improvement model to complete the training of the water quality monitoring parameter resolution improvement model.

7. An enhanced water quality monitoring parameter resolution improvement system, the system is implemented based on the enhanced water quality monitoring parameter resolution improvement method according to any one of claims 1 to 6, characterized in that: include: An acquisition module is used to obtain preprocessed data sets of multi-source remote sensing data of water environment; A training module, used to screen the optimal characteristic factor set affecting the water quality monitoring parameters from the preprocessed data set, and train a preset water quality monitoring parameter resolution improvement model using the optimal characteristic factor set to obtain a trained water quality monitoring parameter resolution improvement model; The output module is used to input the multi-source remote sensing data to be processed into the water quality monitoring parameter resolution enhancement model and output the high-resolution prediction results of the water quality monitoring parameters.

Images