An aerospace-terrestrial integration small watershed water pollution monitoring system

Abstract

The present application belongs to the technical field of water environment monitoring, basin pollution simulation and intelligent water management, and specifically provides a space-air-ground integrated small watershed water pollution monitoring system, comprising: an air-based monitoring module for collecting global areal water quality spatial distribution data of a small watershed; a ground-based monitoring module for collecting time-series continuous water quality and hydrology benchmark data of the small watershed; a multi-source data management module for data processing; a model engine module for building and running a small watershed distributed hydrology and water quality mechanism model; an adaptive parameter optimization module for automatically calibrating and trigger-type dynamically updating parameters of the small watershed distributed hydrology and water quality mechanism model; an early warning and tracing module for obtaining early warning and tracing results; a long-term prediction module for obtaining water quality prediction results; and a visual decision module for displaying the early warning and tracing results and the water quality prediction results. The present application realizes monitoring, early warning, tracing and long-term prediction of small watershed water pollution.

Description

Technical Field

[0001] This invention belongs to the fields of water environment monitoring, watershed pollution simulation and smart water technology, specifically involving an integrated air-space-ground small watershed water pollution monitoring system. Background Technology

[0002] Currently, small watersheds are characterized by dispersed pollution sources, a high proportion of non-point source pollution, strong spatial heterogeneity, and significant influence of the underlying surface on hydrological and pollution processes. Accurate monitoring, rapid early warning, and long-term trend prediction are crucial for refined water environment management. Existing technologies mainly rely on three types of techniques: ground-based fixed-point monitoring, independent remote sensing monitoring by unmanned aerial vehicles (UAVs), and traditional SWAT models (Soil and Water Assessment Tool, the mainstream tool for simulating distributed mechanisms in small watersheds). However, a deeply integrated and adaptively iterative unified system has not yet been formed, and its specific shortcomings are as follows: (1) The ground monitoring space coverage is seriously insufficient and can only obtain time series data of a limited number of points. It cannot reflect the spatial distribution of pollution in areas such as tributaries, reservoir bays, shorelines, and hidden sewage outlets, making it difficult to support the judgment of the pollution situation in the whole area and accurate source tracing.

[0003] (2) The monitoring of UAVs is completely disconnected from ground monitoring and model simulation. UAVs are only used as independent inspection tools. The inversion results are not calibrated together with high-precision ground data, nor are they used as spatial constraints in the model calibration process. The advantages of its areal data have not been effectively utilized.

[0004] (3) Traditional SWAT models rely solely on single-point calibration, resulting in low accuracy and poor reliability. They generally use time series data from a single cross-section at the watershed outlet for parameter calibration, lacking spatial distribution verification and prone to “different parameters with the same effect”. The model parameters are manually set once and remain static, making it impossible to adapt to dynamic changes such as land use, meteorology, and pollution sources.

[0005] (4) Without adaptive optimization and closed-loop update mechanism, the monitoring data and model simulation are independent of each other. New UAV or ground monitoring data cannot automatically trigger model parameter optimization. The system cannot evolve itself, and the simulation error continues to increase after long-term operation.

[0006] (5) The functions of early warning, source tracing and long-term prediction are scattered, and the decision support capability is weak. Pollution early warning relies on single-point threshold alarms and cannot combine spatial information to quickly simulate diffusion; the accuracy of source tracing and load accounting is low; and there is a lack of a unified visualization platform, which makes it difficult to support long-term governance decisions. Summary of the Invention

[0007] To address the shortcomings of existing small watershed water pollution monitoring and simulation prediction technologies, such as incomplete spatial coverage, low model calibration accuracy, inability to dynamically update parameters, inaccurate early warning and source tracing, and low long-term operation and maintenance value, this invention provides an integrated air-space-ground small watershed water pollution monitoring system.

[0008] To achieve the above objectives, the present invention provides the following solution: A space-air-ground integrated small watershed water pollution monitoring system includes: The airborne monitoring module is used to collect spatial distribution data of areal water quality across the entire small watershed in the study area. The ground-based monitoring module is used to collect continuous time-series water quality and hydrological baseline data for small watersheds; The multi-source data management module is used to uniformly store, clean, spatiotemporally register and fuse the water quality spatial distribution data and the water quality hydrological benchmark data to obtain pre-processed multi-source data. The model engine module is used to build and run distributed hydrological and water quality mechanism models for small watersheds based on preprocessed multi-source data. The adaptive parameter optimization module is used to automatically calibrate and dynamically update the parameters of the small watershed distributed hydrological and water quality mechanism model with water quality spatial distribution data and water quality and hydrological benchmark data as dual objective constraints. The early warning and source tracing module is used to identify pollution anomalies, simulate pollution diffusion processes, and trace pollution sources based on the small watershed distributed hydrological and water quality mechanism model, thereby obtaining early warning and tracing results. The long-term prediction module is used to simulate future water quality change trends under multiple scenarios based on the small watershed distributed hydrological and water quality mechanism model, and obtain water quality prediction results. The visualization decision-making module is used to display the early warning and tracing results, as well as the water quality prediction results.

[0009] Preferably, the space-based monitoring module includes: The airborne data acquisition unit includes an unmanned aerial vehicle (UAV) platform, a multispectral sensor, a thermal infrared sensor, a visible light imaging device, an airborne data unit, and a ground control station, and is used to acquire airborne data. The UAV platform is used to perform periodic or emergency patrol flights. The multispectral sensor, thermal infrared sensor, and visible light imaging device are deployed in the UAV payload bay to acquire raw images. The airborne data unit and the ground control station communicate wirelessly. The image preprocessing unit is used to perform radiometric calibration, atmospheric correction, and geometric correction on the acquired raw images; The water quality parameter remote sensing inversion unit is used to invert four core water quality parameters—chlorophyll a, suspended matter, turbidity, and surface water temperature—based on preprocessed original images, and generate a global spatial raster map to characterize the spatial distribution of water quality across the entire area. The data standardization output unit is used to output the global spatial raster map containing water quality spatial distribution information obtained by inversion to the multi-source data management module through a standardized interface.

[0010] Preferably, the multi-source data management module includes: The data access unit is used to access a global spatial raster map containing water quality spatial distribution information, the water quality and hydrological baseline data, DEM topographic data, soil / land use data, daily meteorological data, and artificial sampling laboratory data; The data cleaning unit is used to perform customized cleaning based on the different data types accessed by the data access unit, and obtain cleaned multi-source data. The spatiotemporal registration unit is used to perform spatiotemporal registration on the cleaned multi-source data to obtain registered multi-source data. The data fusion unit is used to fuse registered multi-source data based on least squares support vector machine to obtain a continuous spatiotemporal water quality dataset. The data storage unit is used to store the water quality dataset, metadata, and system configuration using a hybrid architecture based on time-series databases, spatial databases, and relational databases.

[0011] Preferably, the model engine module includes: Sub-basins and HRU delineation units are used to discretize the watershed into distributed computing units to characterize the spatial heterogeneity of the underlying surface. Among them, the river network is extracted and sub-basins are divided based on DEM data using the D8 unidirectional algorithm, and then hydrological response units are delineated by superimposing according to land use, soil and slope. The hydrological cycle simulation unit is used to simulate surface runoff, interflow, groundwater and evapotranspiration processes based on the distributed computing unit and using the water balance equation to obtain hydrological cycle simulation results. The sediment transport simulation unit is used to simulate soil erosion and sediment transport processes based on the hydrological cycle simulation results, using a modified general soil loss equation, and to obtain sediment transport simulation results. The pollutant migration and transformation simulation unit is used to simulate the migration and transformation process of nitrogen and phosphorus nutrients based on the hydrological cycle simulation results and the sediment transport simulation results, including the transport, degradation and adsorption-desorption of dissolved and adsorbed pollutants, and to obtain the pollutant migration and transformation simulation results. The model scheduling unit is used to construct a distributed hydrological and water quality mechanism model for small watersheds based on the hydrological cycle simulation results, the sediment transport simulation results, and the pollutant migration and transformation simulation results, and to receive parameter updates from the adaptive parameter optimization module and automatically trigger model rerun.

[0012] Preferably, the adaptive parameter optimization module includes: The sensitive parameter identification unit is used to select sensitive parameters from the parameters of the small watershed distributed hydrological and water quality mechanism model that meet the preset sensitivity for their impact on water quality and hydrological simulation using a global sensitivity analysis method. The dual-objective joint objective function construction unit is used to construct a joint objective function that simultaneously constrains the fitting accuracy of time series and the similarity of spatial distribution. The intelligent optimization algorithm unit is used to iteratively optimize and solve the joint objective function based on the sensitive parameters using a hybrid complex evolutionary algorithm; The iterative convergence verification unit is used to determine whether the optimization process meets the preset convergence conditions. The parameter update triggering unit is used to automatically trigger model parameter fine-tuning when new monitoring data is added to the database or when preset conditions are met. Triggering conditions include: new UAV patrol data being added to the database, ground monitoring data reaching a preset time period, monitoring data triggering pollution anomaly warnings, or watershed land use data updates.

[0013] Preferably, the joint objective function includes a time-dimensional objective function and a spatial-dimensional objective function; wherein, the time-dimensional objective function is based on water quality and hydrological benchmark data and is constructed using the Nash-Sutcliffe efficiency coefficient; the spatial-dimensional objective function is based on water quality spatial distribution data and is constructed using structural similarity and average relative error.

[0014] Preferably, in the intelligent optimization algorithm unit, the process of iteratively optimizing and solving the joint objective function using a hybrid complex evolutionary algorithm includes: Determine the physical upper and lower limits of each sensitive parameter, and use the Latin hypercube sampling method to generate multiple sample points in the parameter space. Each sample point corresponds to a set of model parameter combinations. The sample points are sorted according to the joint objective function value, and the sorted sample points are divided into multiple complexes, each containing the same number of sample points. An evolutionary operation is performed independently for each complex, the evolutionary operation including reflection, expansion, contraction and compression steps; All sample points evolved from the complex shapes are remixed, reordered according to the objective function value, and then divided into multiple new complex shapes. Repeat the complex evolution and shuffling steps until the preset convergence condition of the iterative convergence verification unit is met, and output the optimal parameter set.

[0015] Preferably, the early warning and tracing module includes: The pollution anomaly discrimination unit is used to identify pollution events based on a triple discrimination method of national standard threshold, temporal abrupt change and spatial anomaly. The pollution diffusion simulation unit is used to simulate the diffusion path and influence range of pollution plumes based on a coupled model of the one-dimensional Saint-Venant hydrodynamic equation and the one-dimensional convection-diffusion water quality equation, and to obtain pollution diffusion simulation results. The pollution source tracing unit is used to trace pollution sources by combining reverse confluence tracing with sub-basin contribution rate inversion. The pollution load calculation unit is used to calculate the total pollution load of the basin and the pollution contribution rate of each sub-basin based on the distributed simulation results of the distributed hydrological and water quality mechanism model of the small watershed. The early warning information release unit is used to generate standardized early warning reports and push them to the visualization decision-making module based on the pollution event identification results, pollution diffusion simulation results, pollution source tracing results, and the calculation results of the total pollution load of the watershed and the pollution contribution rate of each sub-watershed.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: Spatial monitoring coverage has been significantly improved. The use of UAV-based area monitoring has increased the effective monitoring coverage of the watershed from less than 5% at a single point to 100% of the entire basin, enabling complete identification of blind spots such as tributaries, reservoir bays, and hidden sewage outlets, fundamentally solving the problem of missing spatial information. Model calibration accuracy has been greatly improved. The use of dual-target calibration with ground-based time-series data and UAV spatial data has significantly improved the fitting coefficient of determination R0 for key water quality indicators (total phosphorus, total nitrogen, and ammonia nitrogen) in the SWAT model. 2 The efficiency has been improved from 0.55-0.65 to over 0.80, significantly reducing the "different parameters, same effect" phenomenon and making the internal process simulation more realistic. Parameter adaptive updates enhance long-term prediction stability. New monitoring data automatically triggers parameter fine-tuning, ensuring that the prediction error increase is less than 5% after more than 12 months of continuous model operation, far superior to traditional static models (which typically have an error increase of >20%), significantly improving long-term prediction reliability. Pollution early warning and source tracing efficiency has been significantly improved. Combining real-time anomaly identification by drones with rapid model simulation, the time from pollution detection to early warning output has been shortened from the traditional 2-4 hours to less than 10 minutes, and the accuracy of pollution source location has increased from below 60% to over 90%.

[0017] Pollution load calculation is more accurate. The fusion of point and area data reduces the calculation error of sub-basin pollution contribution rates from ±25% to within ±10%, accurately identifying the contribution proportions of agricultural non-point sources, domestic sources, and livestock sources. Unified fusion of multi-source data enhances the system's intelligence level. Automatic cleaning, registration, and assimilation of multi-source data reduce manual intervention workload by more than 70%, achieving adaptive intelligent operation that "learns as it monitors and becomes more accurate with use."

[0018] Reduce monitoring and maintenance costs. Periodic drone patrols replace a large number of manual inspections, reducing watershed monitoring labor costs by 40%–60% and significantly reducing long-term maintenance investment. Improve the efficiency of governance fund utilization. Precise source tracing and load accounting enhance the targeting of governance projects, reducing governance fund waste by more than 30% and avoiding extensive investment across the entire region. Extend the effective lifespan of models. Adaptive updates allow models to be built once and used for a long time without frequent recalibration, saving modeling costs and time. Enhance water environment supervision capabilities. Enable early detection, early warning, and early treatment of pollution in small watersheds, reduce the risk of sudden pollution incidents, and ensure drinking water safety and aquatic ecological health.

[0019] Supporting scientific and precise pollution control. Providing quantifiable, verifiable, and visualized scientific evidence for watershed planning, governance scheme evaluation, and long-term management, promoting the transformation of water environment governance from "experience-based governance" to "precision and intelligent governance." Highly scalable and with significant industry demonstration value. The solution is applicable to various small watersheds with agricultural non-point source pollution, watersheds in urban-rural fringe areas, and drinking water source watersheds, and can be rapidly replicated and promoted to improve the overall level of regional water environment management. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Example 1 like Figure 1 As shown, an integrated air-space-ground small watershed water pollution monitoring system includes: an air-based monitoring module, a ground-based monitoring module, a multi-source data management module, a model engine module, an adaptive parameter optimization module, an early warning and source tracing module, a long-term prediction module, and a visualization decision-making module.

[0025] A spatial monitoring module is used to collect spatial distribution data of areal water quality across the entire small watershed in the study area; a further implementation method is that the spatial monitoring module includes: The airborne data acquisition unit includes an unmanned aerial vehicle (UAV) platform, a multispectral sensor, a thermal infrared sensor, a visible light imaging device, an airborne data unit, and a ground control station, used to acquire airborne data. The UAV platform is used for periodic or emergency patrol flights. The multispectral sensor, thermal infrared sensor, and visible light imaging device are deployed in the UAV payload bay to acquire raw imagery. Wireless communication is used between the airborne data unit and the ground control station. For the target small watershed, a dual-mode approach is adopted: a full-area orthogonal flight path and a denser flight path in key areas. The forward overlap rate is ≥80%, the lateral overlap rate is ≥70%, and the ground resolution (GSD) is ≤10cm, ensuring 100% blind-spot-free coverage of the watershed. Simultaneous acquisition of multispectral (blue, green, red, red-edge, and near-infrared 5-10 bands), thermal infrared (8-14μm), and visible light RGB data is performed.

[0026] The image preprocessing unit performs radiometric calibration, atmospheric correction, and geometric correction on the acquired raw images. The core of preprocessing is to eliminate sensor and environmental errors to obtain the true surface reflectance and accurate spatial coordinates. The core formula is as follows: Radiometric calibration: Converting the sensor's raw DN value into entrance pupil radiance.

[0027] , In the formula, The entrance pupil radiance in the λ band of the open-base multispectral / visible light band is... The calibration gain coefficient for band λ. Here is the calibration offset coefficient for band λ. This represents the original grayscale value of band λ.

[0028] Atmospheric correction: The 6S radiative transfer model is used to eliminate the effects of atmospheric scattering / absorption, yielding the true surface reflectance. , In the formula: d_ES is the Earth-Sun astronomical unit distance. λ represents the solar irradiance at the top of the atmosphere in the λ band. The zenith angle of the sun. Atmospheric transmittance. The radiometric calibration, atmospheric correction, and geometric correction in this step are used to obtain a unified spatial reference and surface reflectance, and serve as a preprocessing step for subsequent point-area fusion and model dual-objective calibration.

[0029] Geometric correction: Based on ground control points (GCP), a quadratic polynomial correction model is adopted, with a registration error of ≤1 pixel. The coordinate system is unified with CGCS2000, achieving spatial reference unification with ground point locations and watershed DEM.

[0030] The water quality parameter remote sensing inversion unit is used to invert four core water quality parameters—chlorophyll a, suspended matter, turbidity, and surface water temperature—based on preprocessed raw images, and generate a global spatial raster map to characterize the spatial distribution of water quality across the entire area.

[0031] Specifically, suspended solids (SS) inversion (core pollution indicator): red / near-infrared band ratio model.

[0032] , In the formula: These are the suspended matter inversion regression coefficients obtained from ground synchronous sampling data calibration. Red band surface reflectance, This refers to the surface reflectance in the near-infrared band.

[0033] Chlorophyll a (Chl-a) inversion: Red-edge band difference model: , In the formula: α_Chl and β_Chl are the calibration coefficients of the chlorophyll a inversion model. The surface reflectance is represented by the two red-edge bands.

[0034] Surface water temperature inversion: Based on Planck's law in the thermal infrared band: , In the formula: , L is the calibration constant for the thermal infrared sensor. λ This represents the radiance of the thermal infrared band λ, and this sign is the same as L in the aforementioned radiometric calibration and atmospheric correction formulas. λ To maintain consistency, both represent the radiance of the corresponding band λ; in this formula, λ specifically refers to the thermal infrared band.

[0035] The data standardization output unit is used to output the 1m resolution global spatial raster map containing water quality spatial distribution information, as well as the corresponding spatial coordinates, timestamps, sensor types, inversion indicators and quality control identifiers, to the multi-source data management module through a standardized interface; the identification and boundary extraction of abnormal pollution patches are performed by the early warning and source tracing module.

[0036] The ground-based monitoring module is used to collect continuous time-series water quality and hydrological benchmark data for small watersheds. It consists of fixed automatic water quality monitoring stations, flow and water level monitoring equipment, manual sampling points, and data acquisition and transmission units. Deployed at watershed outlets, tributary confluences, downstream of key pollution sources, and sensitive water areas, the monitoring data is wirelessly transmitted in real-time to the multi-source data management module. It is used to acquire continuous, high-precision time-series data on pH, COD, ammonia nitrogen, total phosphorus, total nitrogen, flow rate, and water level, providing a time-dimensional benchmark for model calibration. The core function is to acquire continuous, high-precision time-series water quality and hydrological benchmark data to provide a time-dimensional calibration basis for the model. The process is divided into three core stages: (1) Monitoring point layout and data collection: Following the "Water Environment Monitoring Standard" SL219, sampling points were set up, and in-situ sensors collected seven core parameters: pH, COD, ammonia nitrogen, total phosphorus, total nitrogen, flow rate, and water level. The sampling frequency was once every 5 minutes for hydrological parameters and once every 15 minutes for water quality parameters. Simultaneously, manual sampling points were set up, and laboratory analysis was carried out once a month according to national standard methods for sensor calibration.

[0037] (2) Data calibration and quality control: Laboratory calibration formula: Corrects for in-situ sensor system errors.

[0038] Outlier removal: The 3σ criterion (Raida criterion) is used.

[0039] like Then determine These are outliers and should be removed.

[0040] In the formula: μ is the mean of the time series. This represents the standard deviation of the time series.

[0041] Missing value imputation: For short-term missing values ​​(≤3h), linear interpolation is used; for long-term missing values ​​(>3h), Kalman filtering combined with historical data is used for imputation. The linear interpolation formula is: , In the formula, This is an estimate of the time τ to be filled. The previous valid observation time The observed values, For the next valid observation time The observed value; τ is the time point to be filled. and These are adjacent valid observation time points.

[0042] (3) Data transmission and output.

[0043] Using 4G / 5G / NB-IoT wireless communication and the MQTT protocol, a set of standardized time-series data packets with timestamps and site numbers is uploaded to the multi-source data management module every hour, with a data qualification rate of ≥95%.

[0044] A multi-source data management module is used to uniformly store, clean, spatiotemporally register, and fuse water quality spatial distribution data and water quality hydrological benchmark data to obtain preprocessed multi-source data; a further implementation method is that the multi-source data management module includes: The data access unit is used to access a wide-area spatial raster map containing water quality spatial distribution information, water quality and hydrological baseline data, DEM topographic data, soil / land use data, daily meteorological data, and artificial sampling laboratory data via interfaces such as HTTP, MQTT, FTP, and direct database connection, compatible with multiple formats such as tif, shp, csv, and nc.

[0045] The data cleaning unit is used to perform customized cleaning based on the different data types accessed by the data access unit, obtaining cleaned multi-source data; and to execute customized cleaning rules for different data types. Spatial raster data (global spatial raster map): Images with cloud cover ≥10% were removed, and salt-and-pepper noise was removed using a 3×3 median filter. Formula: , In the formula: f(x, y) is the original pixel value, G(x, y) is the filtered pixel value, and u and v are the row and column offsets relative to the center pixel within the 3×3 neighborhood window.

[0046] Ground-based temporal data: For ground-based monitoring modules that have completed calibration, outlier removal, and missing value filling, we perform data entry consistency verification, timestamp verification, unit unification, and cross-source conflict verification, without repeating sensor calibration; for DEM topographic data, we perform projection unification, depression filling, and outlier elevation removal; for soil / land use data, we perform classification coding unification and boundary topology checks; for daily meteorological data, we perform missing measurement filling and unit verification; and for manual sampling laboratory data, we standardize sampling numbers, time, and indicator units to ensure that all types of access data meet the requirements for subsequent spatiotemporal registration and fusion.

[0047] The spatiotemporal registration unit is used to perform spatiotemporal registration on the cleaned multi-source data to obtain registered multi-source data. Spatial registration: CGCS2000 Gauss-Kruger projection is uniformly adopted, and raster resampling is achieved through nearest neighbor interpolation. All spatial data have uniform resolution and coordinate alignment. Temporal registration: All data are uniformly set to a daily time step, hourly ground-based data are aggregated into daily averages, UAV data are matched with daily ground-based and meteorological data, and a timestamp index is established to achieve multi-source data alignment at the same time section.

[0048] The data fusion unit is used to fuse registered multi-source data based on least squares support vector machine (LSSVM) to obtain a continuous spatiotemporal water quality dataset. Specifically, it uses LSSVM to achieve point-area data fusion, fusing high-precision single-point ground data with spatial areal data to generate a continuous spatiotemporal water quality dataset. The core optimization objective function is:

[0049] In the formula: ω is the weight vector, γ_LSSVM is the regularization parameter, e_i is the error term of the i-th fused sample, φ(x_i) is the kernel function mapping, β0_LSSVM is the bias term, y_i is the ground-based measured value, and x_i is the air-based inversion feature vector.

[0050] The data storage unit employs a hybrid architecture based on time-series databases, spatial databases, and relational databases to store water quality datasets, metadata, and system configurations. InfluxDB stores time-series data, PostgreSQL+PostGIS stores spatial data, and MySQL stores metadata and system configurations. It also includes an interface layer that provides standardized RESTful API interfaces to enable bidirectional data exchange with upstream and downstream modules, supporting batch data read / write, conditional queries, and incremental push.

[0051] The model engine module is used to build and run a distributed hydrological and water quality mechanism model for small watersheds based on preprocessed multi-source data; a further implementation method is that the model engine module includes: Sub-basins and HRU delineation units are used to discretize the watershed into distributed computing units, characterizing the spatial heterogeneity of the underlying surface. Specifically, based on DEM data, the river network is extracted using the D8 unidirectional flow algorithm and sub-basins are delineated. Then, hydrological response units are delineated by overlaying land use, soil, and slope classifications, with thresholds set at ≥5% for land use and ≥5% for soil, ensuring accurate characterization of the underlying surface features. Local slope calculation formula: , In the formula: S_slope is the local slope, z_center is the elevation of the center cell, z_neighbor is the elevation of the adjacent cell, and l_cell is the distance between the centers of two cells.

[0052] The hydrological cycle simulation unit, based on a distributed computing unit, uses water balance equations to simulate surface runoff, interflow, groundwater, and evapotranspiration processes, obtaining hydrological cycle simulation results; core formula: The governing equation for watershed water balance is: , In the formula: Let t be the soil moisture content on day t. This represents the initial soil moisture content. This refers to daily precipitation. Surface runoff, This is the actual evaporation rate. This refers to the amount of soil infiltration. This refers to underground runoff.

[0053] Surface runoff calculation (SCS-CN model): , In the formula: I a Initial loss (taken as 0.2S) ret ), S ret This represents the maximum water storage capacity in the basin. CN represents the number of runoff curves (a core sensitive parameter of the model).

[0054] The sediment transport simulation unit is used to simulate soil erosion and sediment transport processes based on hydrological cycle simulation results, employing a modified general soil loss equation to obtain sediment transport simulation results; the specific calculation formulas are as follows: , In the formula: For HRU sediment yield, Surface runoff, For the peak flow, For the HRU area, As a soil erodibility factor, For vegetation cover factor, For soil and water conservation measures, is the slope length and slope factor, and CFRG is the coarse debris factor.

[0055] The pollutant migration and transformation simulation unit is used to simulate the migration and transformation processes of nitrogen and phosphorus nutrients based on hydrological cycle simulation results and sediment transport simulation results. This includes the transport, degradation, and adsorption / desorption of dissolved and adsorbed pollutants, obtaining simulation results for pollutant migration and transformation. Core formula: Calculation of dissolved pollutant load: , In the formula: For dissolved pollutant load, This represents the concentration of pollutants in the runoff.

[0056] Calculation of adsorbed pollutant load: , In the formula: For adsorbed pollutant load, This represents the concentration of pollutants in the sediment.

[0057] The model scheduling unit is used to construct a distributed hydrological and water quality mechanism model for small watersheds based on the simulation results of hydrological cycle, sediment transport, and pollutant migration and transformation. It can perform model initialization, parameter assignment, step-by-step calculation, result output and iterative calling, and receive parameter updates from the adaptive parameter optimization module and automatically trigger model rerun.

[0058] The adaptive parameter optimization module is used to automatically calibrate and dynamically update the parameters of a small watershed distributed hydrological and water quality mechanism model, with spatial distribution data and hydrological baseline data as dual objectives. Its core functions are to achieve automatic optimization of model parameters through point-area dual-objective calibration and to achieve adaptive dynamic evolution of the model through triggered updates, thus addressing the core shortcomings of traditional models such as "different parameters having the same effect" and the accumulation of static parameter errors.

[0059] A further implementation method includes an adaptive parameter optimization module comprising: The sensitive parameter identification unit is used to select sensitive parameters from the parameters of the small watershed distributed hydrological and water quality mechanism model that meet the preset sensitivity for their impact on water quality and hydrological simulation using a global sensitivity analysis method. Specifically, it employs the LH-OAT (Latin Hypercubic Sampling-Single Perturbation) global sensitivity analysis method to select sensitive parameters from hundreds of parameters in the SWAT model, reducing the optimization dimensionality. The core sensitivity calculation formula is as follows: , In the formula: For parameters The model output value corresponding to the initial value. For parameters The model output value corresponding to the perturbation value; SI>1 is extremely sensitive, 0.2<SI≤1 is highly sensitive, and 0.05<SI≤0.2 is moderately sensitive. Only the above three types of parameters are retained in the optimization process.

[0060] List of core sensitive parameters: CN2 (number of runoff curves), SOL_K (soil saturated hydraulic conductivity), USLE_K (soil erodibility factor), CH_N2 (channel Manning coefficient), pollutant degradation coefficient, nitrogen / phosphorus leaching coefficient, etc.

[0061] A dual-objective joint objective function construction unit is used to construct a joint objective function that simultaneously constrains the fitting accuracy of time series and the similarity of spatial distribution. A further implementation involves the joint objective function comprising a time-dimensional objective function and a spatial-dimensional objective function. The time-dimensional objective function is constructed based on water quality and hydrological benchmark data using the Nash-Sutcliffe efficiency coefficient; the spatial-dimensional objective function is constructed based on water quality spatial distribution data using structural similarity and average relative error. The calculation formula is as follows: , In the formula: F is the joint objective function, and the smaller the value, the higher the simulation accuracy of the watershed hydrological and water quality processes; The time target weight (default 0.6) represents the weight of ground-based monitoring time-series water quality and hydrological data in the calibration process; The spatial target weight (default 0.4) represents the weight of the spatial distribution data of water quality monitored by UAVs in the calibration process. The objective function is defined by the time dimension. The objective function is the spatial dimension.

[0062] Time dimension objective function Based on ground-based time-series data, and using the Nash-Sutcliffe efficiency coefficient (NSE) as the core, the following is constructed: , , In the formula: The weighted average of flow rate and Nash-Sutcliffe efficiency coefficients for core water quality indicators (COD, ammonia nitrogen, total phosphorus, and total nitrogen) is given. The concentration of water quality indicators (such as total phosphorus mg / L, ammonia nitrogen mg / L) or flow rate value measured by the ground monitoring station at time i. ); The corresponding water quality index concentration or flow rate value output by the SWAT model at time i; NSE is the mean of the measured sequence; n is the number of samples in the time series; the closer NSE is to 1, the better the time fit. The closer it is to 0.

[0063] , , , In the formula: SSIM represents the structural similarity between the simulated and UAV-inverted raster (values ​​[0,1], the closer to 1, the higher the similarity), where: To simulate the mean values ​​of water quality indicators (suspended solids (SS), chlorophyll a) grids, This represents the mean value of the corresponding water quality index grid retrieved by the drone; , This represents the standard deviation of the corresponding raster. Let be the covariance of the two. To avoid a constant with a denominator of 0; MRE is the average relative error between the simulation and the UAV-derived water quality grid; a smaller value indicates higher spatial simulation accuracy, and m is the total number of grid cells. Let be the simulated water quality index concentration of the j-th grid; The water quality is retrieved by the UAV for the j-th grid.

[0064] The intelligent optimization algorithm unit is used to iteratively optimize and solve the joint objective function based on sensitive parameters using a hybrid complex evolutionary algorithm. A further implementation method involves the following process within the intelligent optimization algorithm unit: [Details of the iterative optimization and solution process using the hybrid complex evolutionary algorithm for the joint objective function are provided.] The physical upper and lower limits of each sensitive parameter are determined, and multiple sample points are generated in the parameter space using the Latin hypercube sampling method. Each sample point corresponds to a set of model parameter combinations. Substituting each sample point into the SWAT model, the daily watershed flow, ammonia nitrogen, total phosphorus, total nitrogen concentration, sediment load, and other results can be output, and then the corresponding joint objective function value F can be calculated.

[0065] The sample points are sorted according to the joint objective function value, and then divided into multiple composites, each containing the same number of sample points. Specifically, s sample points are sorted in ascending order of the joint objective function value F, where a smaller F indicates higher accuracy in water quality simulation. The sorted sample points are then divided into p composites, each containing m sample points, satisfying s = p × m. By performing parallel evolution of the composites, the algorithm's search efficiency is improved while ensuring full coverage of the water quality parameter space.

[0066] Evolutionary operations are performed independently on each complex, including reflection, expansion, contraction, and compression steps; specifically: Reflection: For the worst point within the complex (with the largest F value and the largest water quality simulation error), a reflection point is generated along the centroid direction of the complex. The objective function value of the reflection point is calculated. If the reflection point is better than the worst point, it is retained.

[0067] Expansion: If the reflection point is better than the optimal point within the complex shape, it is further expanded along the reflection direction to generate an expansion point. If the expansion point is better, the reflection point is replaced.

[0068] Contraction: If the reflection point is not better than the worst point, perform a contraction operation in the opposite direction to generate a contraction point.

[0069] Compression: If the shrinkage point still does not improve, compress all points in the complex except the optimal point towards the optimal point to generate a new sample set.

[0070] All sample points evolved from the composites are remixed, reordered according to the objective function value, and divided into p new composites. This breaks down the information barriers between composites, enables global sharing of excellent parameter samples, avoids local optima, and ensures that the algorithm can find a globally optimal parameter set that balances water quality time series fitting and spatial distribution simulation.

[0071] Repeat the complex evolution and shuffling steps until the preset convergence condition of the iterative convergence verification unit is met, and output the optimal parameter set.

[0072] The iterative convergence verification unit is used to determine whether the optimization process meets the preset convergence conditions. Specifically, the optimal sample result of the current iteration is input into the iterative convergence verification unit. If the following three convergence criteria bound to water quality are met simultaneously, the iteration stops; otherwise, it returns to the composite evolution and continues iterating: The objective function value changes less than 100 times in 100 consecutive iterations. .

[0073] Flow rate NSE ≥ 0.75, core water quality indicators (ammonia nitrogen, total phosphorus, total nitrogen) NSE ≥ 0.65, coefficient of determination R 2 ≥0.8.

[0074] Spatial distribution of water quality: SSIM≥0.75, MRE≤20%.

[0075] The optimal parameter set after convergence is output and written back to the SWAT model engine module to complete the calibration; at the same time, the optimal parameter set is stored as the initial value in the parameter update trigger unit for subsequent adaptive incremental fine-tuning.

[0076] Optionally, when the hybrid complex evolutionary algorithm fails to converge within a preset number of iterations or when there are multimodal solutions in the parameter space, the adaptive parameter optimization module calls the particle swarm optimization algorithm (PSO) as a secondary search strategy, using the current optimal parameter set as the initial particle center, to perform a local supplementary search for sensitive parameters; if PSO obtains a smaller joint objective function value, the optimal parameter set is updated; otherwise, the result of the hybrid complex evolutionary algorithm is retained.

[0077] The parameter update trigger unit is used to automatically trigger model parameter fine-tuning when new monitoring data is added to the database or when preset conditions are met. Trigger conditions include: adding new UAV patrol data to the database, ground monitoring data reaching a preset time period (7-day weekly update, 30-day monthly update), and monitoring data triggering pollution anomaly warnings or watershed land use data updates.

[0078] Using the current optimal parameter set as the initial value, incremental iterative optimization is adopted, with only minor parameter fine-tuning. The maximum number of iterations is 500 to ensure update efficiency and achieve adaptive evolution of the model, making it more accurate with use.

[0079] The early warning and source tracing module is used to identify pollution anomalies, simulate pollution diffusion processes, and trace pollution sources based on a small watershed distributed hydrological and water quality mechanism model, thereby obtaining early warning and tracing results. A further implementation method is that the early warning and source tracing module includes: The pollution anomaly detection unit is used to identify pollution events based on a triple-discrimination method of national standard thresholds, temporal abrupt changes, and spatial anomalies; specifically: National standard threshold discrimination: Based on the "Surface Water Environmental Quality Standard" GB3838, if C i >C limit An early warning will be triggered immediately, and the warning will be divided into four levels according to the multiple of the exceedance.

[0080] Time-series abrupt change detection: The sliding window Welch test is used to identify abrupt changes in water quality concentration. The formula is as follows: , In the formula: C_pre, s_pre, and n_pre represent the mean, standard deviation, and sample size of the concentration series in the preceding window, respectively; C_cur, s_cur, and n_cur represent the mean, standard deviation, and sample size of the concentration series in the current window, respectively. C_pre and C_cur are both the mean of the concentration series, but they correspond to different time windows and have different meanings, distinguished by the subscripts pre and cur. When the statistic T_w is greater than the critical value corresponding to the significance level α=0.05... When this occurs, it is identified as a mutation and an early warning is triggered.

[0081] Spatial anomaly detection: The Local Anomaly Factor (LOF) algorithm is used to identify pollution patches in UAV imagery. The formula is:

[0082] In the formula: Let be the set of k nearest neighbors of point p, and lrd be the local reachability density; LOF(p)>1 indicates an anomalous pollution patch.

[0083] The pollution diffusion simulation unit is used to simulate the diffusion path and impact range of pollution plumes based on a coupled model of the one-dimensional Saint-Venant hydrodynamic equation and the one-dimensional convection-diffusion water quality equation, and to obtain pollution diffusion simulation results; core equation: One-dimensional Saint-Venant hydrodynamic equations: Continuity equation: , Momentum equation: , In the formula: A is the cross-sectional area of ​​the water passage, Q is the flow rate, t is the time, and x is the distance along the river channel. Let z be the acceleration due to gravity, and z be the water level. This is a frictional gradient.

[0084] One-dimensional convection-diffusion water quality equation: , In the formula: C is the pollutant concentration, D_x is the longitudinal dispersion coefficient, k is the pollutant degradation coefficient, and q_src is the pollution source term per unit river segment.

[0085] Simulation output: Outputs the location, concentration, range of influence, arrival time and peak concentration of downstream sensitive targets at time steps of 10min, 30min, 1h, 2h, 6h and 24h.

[0086] The pollution source tracing unit is used to trace pollution sources using a combination of reverse confluence tracing and sub-basin contribution rate inversion. Reverse confluence time tracing: calculates the confluence time of the pollution plume from the abnormal cross-section, tracing the emission time and upstream catchment area. Formula: T_tr = Σ(L_r / v_r), In the formula: T_tr is the confluence time, L_r is the length of the r-th segment of the river, v_r is the average flow velocity of the r-th segment of the river, and r is the segment number on the reverse tracing path.

[0087] Sub-basin pollution source location: Based on the ranking of pollution contribution rates in sub-basins, the core pollution sources are located. Contribution rate formula: CR_h = Load_h / Load_out × 100%, In the formula: CR_h is the pollution contribution rate of the h-th sub-basin, Load_h is the pollution load output of the h-th sub-basin, Load_out is the total pollution load at the basin outlet, and h is the sub-basin number; combined with land use type, the pollution source type (agricultural non-point source, domestic source, livestock and poultry source, industrial point source) is identified.

[0088] The pollution load calculation unit is used to calculate the total pollution load of the watershed and the pollution contribution rate of each sub-watershed based on the distributed simulation results of the small watershed distributed hydrological and water quality mechanism model. Specifically, it calculates the total pollution load of the watershed and the load by type and region. The core formula is as follows: , , In the formula: For total pollution load, For point source load, For area source load, , , These are agricultural planting, livestock and poultry breeding, and rural domestic non-point source loads, respectively.

[0089] The early warning information release unit is used to generate standardized early warning reports and push them to the visualization decision-making module based on pollution event identification results, pollution diffusion simulation results, pollution source tracing results, and calculation results of the total pollution load of the watershed and the pollution contribution rate of each sub-watershed. Specifically, it generates standardized early warning reports, including early warning level, abnormal indicators, exceedance multiples, pollution location, diffusion prediction results, source tracing conclusions, and emergency response suggestions, and pushes them to the visualization decision-making module and management personnel.

[0090] The long-term prediction module is used to simulate future water quality change trends under multiple scenarios based on a distributed hydrological and water quality mechanism model of small watersheds, and obtain water quality prediction results. The long-term prediction module consists of a scenario setting unit, a trend simulation unit, and a governance effect evaluation unit. It calls the adaptively updated model parameters to simulate and predict the water quality trend in the next 5-10 years under scenarios such as climate change, land use change, ecological pond construction, and fertilizer reduction. The output end is connected to the visualization decision module.

[0091] The core function is to simulate long-term water quality trends and evaluate the effectiveness of watershed management under multiple scenarios, supporting long-term watershed management decisions. The implementation methods and core formulas for each unit are as follows: (1) Scenario Setting Unit: Five typical scenarios are set up to provide input boundaries for prediction: Baseline Scenario (S0): The current weather, land use, and management measures remain unchanged as a control.

[0092] Climate change scenario (S1): Based on the CMIP6 climate model, three emission scenarios, RCP2.6, RCP4.5 and RCP8.5, are set up, and daily meteorological data for the next 5-10 years are extracted.

[0093] Land use change scenario (S2): Based on national land spatial planning, scenarios such as urbanization expansion and ecological restoration are set.

[0094] Situation for governance measures (S3): Set up governance measures such as fertilizer reduction, construction of ecological ponds, ecological ditches, and soil and water conservation, and modify the corresponding parameters of the model.

[0095] Combined Scenario (S4): A comprehensive scenario combining multiple factors.

[0096] (2) Long-term trend simulation unit: Implementation process: Import the input data corresponding to the scenario and the optimal model parameters that have been adaptively updated; perform continuous distributed simulation for the next 5-10 years with a daily time step; output core results such as annual / monthly watershed outlet water quality concentration, pollution load of each sub-basin, and water quality compliance rate.

[0097] (3) Evaluation Unit for Governance Effectiveness: The effectiveness of governance measures is quantitatively evaluated using four core indicators to select the optimal solution: Water quality improvement rate: ; In the formula: For water quality improvement rate, The annual average concentration is the baseline scenario. The annual average concentration for the treatment scenario.

[0098] Pollution load reduction rate: ; In the formula: For load reduction rate, The total load is the baseline scenario. To manage the total scenario load.

[0099] Water quality compliance rate: ; In the formula: P is the water quality compliance rate. For the number of days when water quality meets standards, This represents the total number of days.

[0100] Input-output ratio: ; In the formula: ROI is the input-output ratio. To address the ecological benefits of the situation, Based on the baseline scenario benefits, Costs are incurred for governance.

[0101] (4) Prediction result output unit: Output water quality trend curves, load change bar charts, spatial distribution maps, and treatment effect evaluation tables under different scenarios, and push them to the visualization decision-making module.

[0102] The visualization decision-making module displays early warning and tracing results, as well as water quality prediction results. It adopts a WebGIS (Web Geographic Information System) architecture and includes a single-map display unit, a data query unit, and a report output unit. It aggregates results from all modules and presents monitoring data, simulation results, early warning information, source tracing conclusions, and long-term prediction scenarios in a visual interface, providing support for management decision-making.

[0103] Specifically, it adopts a B / S architecture, which is divided into four layers: WebGIS map engine layer, data visualization layer, business function layer, and user interaction layer. It brings together the results of all modules to provide visual support for management decisions.

[0104] (1) WebGIS map engine layer: It adopts the Leaflet / ArcGIS API for JavaScript map engine, loads vector layers such as CGCS2000 coordinate system base map, watershed boundaries, river networks, sub-watersheds, and monitoring points, supports raster layer overlay, and realizes basic functions such as map zooming, panning, querying, and measurement.

[0105] (2) Data visualization layer: Using ECharts and D3.js visualization libraries, we can achieve graphical display of various types of data: Time-series data: Line charts show the time-series trends of flow rate and water quality changes.

[0106] Statistical data: Bar charts and pie charts show the load of sub-basins and the contribution rate of pollution sources.

[0107] Spatial data: The hierarchical color map shows the spatial distribution of water quality and pollution hotspots.

[0108] Warning data: The alarm panel displays warning information and its spread range through highlighted markers.

[0109] (3) Business Function Layer: It integrates six core functions: data query, report output, early warning management, model scheduling, scenario analysis, and decision suggestions. It supports front-end triggering of model operation, scenario setting, parameter update and other operations, and realizes real-time linkage with back-end modules.

[0110] (4) User interaction layer: It adopts a responsive web interface and is divided into five functional modules: overview map, real-time monitoring, simulation and prediction, early warning and source tracing, and decision analysis. It supports access from multiple terminals and provides corresponding operation interfaces for users with different permissions, realizing visualized decision support for the entire process of watershed water environment management.

[0111] In summary, the system design of this invention achieves the following: Spatial monitoring coverage has been significantly improved. By adopting unmanned aerial vehicle (UAV) area monitoring, the effective monitoring coverage of the watershed has been increased from the traditional single-point <5% to 100% of the entire basin. It can completely identify blind spots such as tributaries, reservoir bays, and hidden sewage outlets, and the problem of missing spatial information has been fundamentally solved.

[0112] The model calibration accuracy has been significantly improved. Dual-target calibration using ground-based time-series data and UAV spatial data has greatly improved the fitting coefficient of determination R0 for key water quality indicators (total phosphorus, total nitrogen, and ammonia nitrogen) in the SWAT model. 2 The value was increased from 0.55-0.65 to over 0.80, significantly reducing the "different parameters have the same effect" phenomenon and making the internal process simulation more realistic.

[0113] Parameters are updated adaptively, enhancing long-term prediction stability. New monitoring data automatically triggers parameter fine-tuning, ensuring that the prediction error increase is less than 5% after more than 12 months of continuous operation, which is far superior to traditional static models (error increase is usually >20%), significantly improving the reliability of long-term predictions.

[0114] The efficiency of pollution early warning and source tracing has been significantly improved. By combining real-time anomaly identification by drones with rapid model simulation, the time from pollution detection to early warning output has been shortened from the traditional 2-4 hours to less than 10 minutes, and the accuracy of pollution source location has increased from below 60% to over 90%.

[0115] Pollution load calculation is more accurate. The integration of point and area data reduces the calculation error of pollution contribution rate of sub-basins from ±25% to within ±10%, and can accurately identify the contribution ratio of agricultural non-point sources, domestic sources and livestock sources.

[0116] Unified fusion of multi-source data enhances the system's intelligence level. Automatic cleaning, registration, and assimilation of multi-source data reduce manual intervention workload by over 70%, achieving adaptive intelligent operation that "learns through monitoring and becomes more accurate with use."

[0117] Reduce monitoring and maintenance costs. Periodic drone patrols replace a large number of manual inspections, reducing the labor costs of watershed monitoring by 40% to 60% and significantly reducing long-term maintenance investment.

[0118] Improve the efficiency of governance funds. Precise source tracing and load accounting enhance the targeting of governance projects, reduce waste of governance funds by more than 30%, and avoid extensive investment across the entire region.

[0119] Extend the effective lifespan of the model. Adaptive updates allow the model to be used for a long time after a single build, eliminating the need for frequent recalibration and saving modeling costs and time.

[0120] Enhance water environment supervision capabilities. Achieve early detection, early warning, and early response to pollution in small watersheds, reduce the risk of sudden pollution incidents, and safeguard drinking water safety and aquatic ecosystem health.

[0121] Support scientific and precise pollution control. Provide quantifiable, verifiable, and visualized scientific evidence for watershed planning, governance scheme evaluation, and long-term management, and promote the transformation of water environment governance from "experience-based governance" to "precise and intelligent governance".

[0122] It is highly scalable and has significant industry demonstration value. The solution is applicable to various small watersheds with agricultural non-point source pollution, watersheds in urban-rural fringe areas, and watersheds near drinking water sources. It can be quickly replicated and promoted to improve the overall management level of regional water environment.

[0123] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A space-air-ground integrated small watershed water pollution monitoring system, characterized in that, include: The airborne monitoring module is used to collect spatial distribution data of areal water quality across the entire small watershed in the study area. The ground-based monitoring module is used to collect continuous time-series water quality and hydrological baseline data for small watersheds; The multi-source data management module is used to uniformly store, clean, spatiotemporally register and fuse the water quality spatial distribution data and the water quality hydrological benchmark data to obtain pre-processed multi-source data. The model engine module is used to build and run distributed hydrological and water quality mechanism models for small watersheds based on preprocessed multi-source data; An adaptive parameter optimization module is used to automatically calibrate and trigger-based dynamically update the parameters of the small watershed distributed hydrological and water quality mechanism model. The early warning and source tracing module is used to identify pollution anomalies, simulate pollution diffusion processes, and trace pollution sources based on the small watershed distributed hydrological and water quality mechanism model, thereby obtaining early warning and tracing results. The long-term prediction module is used to simulate future water quality change trends under multiple scenarios based on the small watershed distributed hydrological and water quality mechanism model, and obtain water quality prediction results. A visualization decision-making module is used to display the early warning and tracing results, as well as the water quality prediction results; The airborne monitoring module includes an unmanned aerial vehicle (UAV) platform, a multispectral sensor, a thermal infrared sensor, a visible light imaging device, an image preprocessing unit, a water quality parameter remote sensing inversion unit, and a data standardization output unit. The UAV platform is used to perform periodic or emergency patrol flights, and the multispectral sensor, thermal infrared sensor, and visible light imaging device are used to acquire raw images. The image preprocessing unit is used to perform radiometric calibration, atmospheric correction, and geometric correction on the raw images. The water quality parameter remote sensing inversion unit is used to invert chlorophyll a, suspended solids, turbidity, and surface water temperature based on the preprocessed raw images, generating a global spatial raster map characterizing the spatial distribution of water quality across the entire area, which is then output to the multi-source data management module by the data standardization output unit. The ground-based monitoring module includes a fixed automatic water quality monitoring station, flow and water level monitoring equipment, manual sampling points, and a data acquisition and transmission unit, which is used to acquire time-series continuous water quality and hydrological reference data including pH, COD, ammonia nitrogen, total phosphorus, total nitrogen, flow and water level, and transmit the water quality and hydrological reference data to the multi-source data management module. The multi-source data management module is also used to perform spatiotemporal registration of the global spatial grid map, the water quality and hydrological reference data and meteorological data. The spatial data adopts a unified spatial reference and is grid aligned, and the temporal data is unified to the same time step, so that the UAV data and the corresponding ground monitoring data and meteorological data form multi-source data of the same time section. The model engine module includes a model scheduling unit, which is used to receive the updated parameters output by the adaptive parameter optimization module and automatically trigger the rerun of the small watershed distributed hydrological and water quality mechanism model based on the updated parameters. The adaptive parameter optimization module includes a sensitive parameter identification unit, a dual-objective joint objective function construction unit, an intelligent optimization algorithm unit, and a parameter update triggering unit. The sensitive parameter identification unit uses a global sensitivity analysis method to select sensitive parameters from the parameters of the small watershed distributed hydrological and water quality mechanism model that meet a preset sensitivity threshold for their impact on water quality and hydrological simulation. The dual-objective joint objective function construction unit constructs a joint objective function including a time-dimensional objective function and a spatial-dimensional objective function. The time-dimensional objective function is constructed using Nash-Sutcliffe efficiency coefficients based on time-series data of flow and water quality indicators measured by ground-based monitoring stations and corresponding time-series data simulated by the small watershed distributed hydrological and water quality mechanism model. The spatial-dimensional objective function is constructed using structural similarity and average relative error based on spatial grids of suspended matter and chlorophyll a simulated by the small watershed distributed hydrological and water quality mechanism model and corresponding spatial grids obtained from UAV inversion. The intelligent optimization algorithm unit iteratively optimizes and solves the joint objective function based on the sensitive parameters. The parameter update triggering unit is used to perform incremental iterative optimization with the current optimal parameter set as the initial value when at least one of the following conditions is met: new UAV patrol data is added to the database, ground monitoring data reaches a preset time period, monitoring data triggers a pollution anomaly warning, or watershed land use data is updated; and the updated parameters are sent to the model scheduling unit. The long-term prediction module is used to simulate the long-term trend of watershed water quality under multiple scenarios based on the small watershed distributed hydrological and water quality mechanism model and by calling the model parameters adaptively updated by the adaptive parameter optimization module.

2. The integrated air-space-ground small watershed water pollution monitoring system according to claim 1, characterized in that, The multi-source data management module includes: a data access unit, used to access a global spatial raster map containing water quality spatial distribution information, the water quality and hydrological benchmark data, DEM topographic data, soil / land use data, daily meteorological data, and artificial sampling laboratory data; a data cleaning unit, used to perform customized cleaning according to the different data types accessed by the data access unit to obtain cleaned multi-source data; a data fusion unit, used to fuse the spatiotemporally registered multi-source data based on least squares support vector machine to obtain a global spatiotemporally continuous water quality dataset; and a data storage unit, used to store the water quality dataset, metadata, and system configuration using a hybrid architecture based on time-series databases, spatial databases, and relational databases.

3. The integrated air-space-ground small watershed water pollution monitoring system according to claim 1, characterized in that, The model engine module further includes: a sub-basin and HRU partitioning unit, used to discretize the watershed into distributed computing units, wherein the river network is extracted and sub-basins are divided based on DEM data using the D8 unidirectional flow algorithm, and then hydrological response units are divided by land use, soil, and slope classification; a hydrological cycle simulation unit, used to simulate surface runoff, interflow, groundwater, and evapotranspiration processes based on the distributed computing units using the water balance equation; a sediment transport simulation unit, used to simulate soil erosion and sediment transport processes based on the hydrological cycle simulation results using the modified general soil loss equation; and a pollutant migration and transformation simulation unit, used to simulate the migration and transformation processes of nitrogen and phosphorus nutrients based on the hydrological cycle simulation results and the sediment transport simulation results, including the transport, degradation, and adsorption / desorption of dissolved and adsorbed pollutants.

4. The integrated air-space-ground small watershed water pollution monitoring system according to claim 1, characterized in that, The adaptive parameter optimization module also includes an iterative convergence verification unit, used to determine whether the optimization process meets the preset convergence conditions; The intelligent optimization algorithm unit uses a hybrid composite evolutionary algorithm to iteratively optimize and solve the joint objective function. The iterative optimization and solution includes: determining the physical upper and lower limits of each sensitive parameter, and using the Latin hypercube sampling method to generate multiple sample points in the parameter space, with each sample point corresponding to a set of model parameter combinations. The sample points are sorted according to the joint objective function value, and the sorted sample points are divided into multiple complexes, each containing the same number of sample points. An evolution operation is performed independently on each complex, which includes reflection, expansion, contraction, and compression steps. The sample points after all complexes have evolved are remixed, resorted according to the joint objective function value, and divided into multiple new complexes again. The complex evolution and shuffling steps are repeated until the preset convergence condition is met, and the optimal parameter set is output.

5. The integrated air-space-ground small watershed water pollution monitoring system according to claim 4, characterized in that, The early warning and source tracing module includes: a pollution anomaly discrimination unit, used to identify pollution events based on a triple discrimination method of national standard threshold, temporal abrupt change, and spatial anomaly; a pollution diffusion simulation unit, used to simulate the diffusion path and impact range of pollution plumes based on a coupled model of one-dimensional Saint-Venant hydrodynamic equation and one-dimensional convection-diffusion water quality equation, and obtain pollution diffusion simulation results; a pollution source tracing unit, used to trace the pollution source using a combination of reverse confluence tracing and sub-basin contribution rate inversion; a pollution load calculation unit, used to calculate the total pollution load of the basin and the pollution contribution rate of each sub-basin based on the distributed simulation results of the small watershed distributed hydrological and water quality mechanism model; and an early warning information release unit, used to generate a standardized early warning report and push it to the visualization decision module based on the pollution event identification results, pollution diffusion simulation results, pollution source tracing results, and the calculation results of the total pollution load of the basin and the pollution contribution rate of each sub-basin.

Images