Multi-scale-based monitoring risk assessment and prevention method for farmland non-point source pollution

Abstract

The application discloses a kind of based on multi-scale farmland non-point source pollution monitoring risk assessment and prevention and control method, comprising the following steps: under the condition of typical topography and typical crop, according to different monitoring purposes, set up farmland non-point source pollution monitoring point, monitor the farmland non-point source pollution of field scale;Water quality and quantity automatic monitoring equipment is installed in the key node of field small watershed, and the farmland non-point source pollution of watershed scale is monitored;Collect farmland non-point source pollution index data of field scale and watershed scale, carry out online transmission and data storage to index data, build farmland non-point source pollution online monitoring early warning platform;Acquire farmland non-point source pollution related data, utilize multi-model to evaluate farmland non-point source pollution, combined with hot and cold spot analysis, output target area farmland non-point source pollution load distribution characteristics, key source area and prevention and control strategy;Utilize internet of things technology to build farmland non-point source pollution space-air integration comprehensive monitoring network, realize farmland non-point source pollution monitoring evaluation early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of farmland non-point source pollution monitoring, in particular, to a farmland non-point source pollution monitoring risk assessment and prevention and control method based on multi-scale. BACKGROUND

[0002] The agricultural non-point source pollution occurs with randomness, dispersibility and universality, and the spatial and temporal heterogeneity of pollutant output is strong, which brings great difficulty to the accurate monitoring of non-point source pollution. Field monitoring is an evaluation method for representing the influence degree of agricultural non-point source pollutants on water environment, but due to the complex environmental process and geophysical and chemical transformation process of nutrients in the process of agricultural non-point source pollution, the influence of point source pollution will also be affected as the spatial scale expands, so it is difficult to effectively separate the influence of these non-agricultural activities by simply using monitoring means.

[0003] The artificial simulated rainfall method often needs power output and large transportation equipment, is inconvenient to carry, has poor mobility, is very inconvenient to assemble, disassemble, rainfall intensity and other indicators need to be calibrated on site, the test operation is complex, the test results are obtained and the post-processing period is long, the workload is large, and the use is greatly limited. In addition, the traditional monitoring methods such as field monitoring and artificial simulated rainfall method are limited to the calculation of small-scale farmland non-point source pollution load, and cannot measure the macro-scale farmland non-point source pollution status.

[0004] The existing agricultural non-point source pollution model is mostly used for calculating small-scale pollution load, and lacks identification of key source areas; since the data source is mostly based on the monitoring of field scale loss, the loss of pollutants during the migration process to the watershed outlet when the pollutants are generated and collected with runoff is not considered, and the influence of factors such as watershed topography, hydrology and climate, vegetation coverage and land use is not considered, so the model calculation result is not the actual pollution load at the outlet of the watershed, resulting in insufficient precision of the model in large-scale simulation. SUMMARY

[0005] The present application aims to overcome the problems of monitoring difficulty of farmland non-point source pollution, insufficient precision of macro-scale agricultural non-point source pollution evaluation, lack of key source area identification and prevention and control strategy in the prior art, and provides a farmland non-point source pollution monitoring risk assessment and prevention and control method based on multi-scale, realizes multi-scale evaluation of farmland non-point source pollution monitoring and evaluation "field scale-watershed scale-macro scale", accurately outputs the distribution characteristics of macro-scale farmland non-point source pollution load, key source area and prevention and control strategy, and realizes real-time monitoring and early warning of farmland non-point source pollution, and the error of the farmland non-point source pollution monitoring and evaluation result is very small.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring includes the following steps:

[0008] Under typical terrain and crop conditions, farmland non-point source pollution monitoring points were set up according to different monitoring objectives to monitor farmland non-point source pollution at the field scale, and the nitrogen and phosphorus pollution fluxes at the field scale were calculated using the LOADEST model.

[0009] Automatic water quality and quantity monitoring equipment is installed at key nodes in small watersheds to monitor farmland non-point source pollution at the watershed scale. The InVEST model is used to assess watershed-scale non-point source pollution, and the InVEST model is corrected using observation data.

[0010] Collect farmland non-point source pollution index data at the field and watershed scales, transmit and store the index data online, build an online monitoring and early warning platform for farmland non-point source pollution, and monitor the loss of nitrogen and phosphorus from farmland non-point sources online;

[0011] Acquire relevant data on farmland non-point source pollution, use the InVEST model and machine learning model to assess farmland non-point source pollution, and combine hot and cold spot analysis to output the distribution characteristics of farmland non-point source pollution load, key source areas and prevention and control strategies in the target area.

[0012] By integrating multi-source heterogeneous data from field-scale monitoring, watershed-scale monitoring, and farmland non-point source pollution assessment, and utilizing Internet of Things (IoT) technology, an integrated air-ground-space monitoring network for farmland non-point source pollution is constructed to achieve monitoring, assessment, and early warning of farmland non-point source pollution.

[0013] Furthermore, under typical terrain conditions, field-scale monitoring of typical crops was carried out. Typical crops represent four land use types: paddy fields, dry land, irrigated land, and orchards.

[0014] Establish standardized monitoring plots for nitrogen and phosphorus loss from farmland. The monitoring indicators include runoff and leaching water volume and quality, microclimate, soil moisture, fertilizer application and irrigation water volume.

[0015] The LOADEST model was used to calculate field-scale nitrogen and phosphorus losses, and the impact of agricultural environmental information such as soil physicochemical properties, crop type, fertilization method, fertilizer ratio, fertilizer amount, and rainfall on non-point source pollution was comprehensively analyzed.

[0016] Furthermore, the principle of the LOADEST model is as follows:

[0017] Non-point source pollutant flux in farmland is the total amount of pollutants passing through a certain cross-section within a certain period of time.

[0018]

[0019] In the formula, Jt is the pollutant flux in period t; Δt is the time interval; NP is the number of discrete time intervals; Q is the period average flow; C is the instantaneous water quality; Jt is the pollutant flux in period t; Δt is the time interval; NP is the number of discrete time intervals; Q is the period average flow; C is the instantaneous water quality;

[0020] LOADEST model is based on equation (1), through the method of multiple linear regression to estimate the pollutant flux of farmland non-point source pollution

[0021]

[0022] In the formula, a0, a j is the equation coefficient; NV is the number of independent variables; X j is the independent variable;

[0023] LOADEST model provides 11 pollutant flux regression equations, and through AIC criterion and SPPC criterion to parameter calibration and optimization selection, when AIC and SPPC value is minimum, it is the optimal pollutant flux regression equation;

[0024]

[0025] In the formula, is the maximum likelihood value of data set D(X i , i = 1, n); is the maximum likelihood estimate value of equation parameters; d is the number of equation parameters; n is the number of data sets used for equation parameter estimation;

[0026] LOADEST model is based on three kinds of statistical estimation methods to establish parameter estimation method: asymptotic maximum likelihood estimation AMLE, maximum likelihood estimation MLE, least absolute deviation method LAD; when the residual error obeys normal distribution, the missing data uses AMLE to estimate the parameters, such as equation (4); non-missing data uses MLE to estimate the parameters, such as equation (5); when the residual error does not obey normal distribution, LAD is used to estimate the parameters, such as equation (6);

[0027]

[0028] In the formula, is the pollutant flux estimated by AMLE, MVUE, LAD method; H(a, b, s 2 , α, k) is the likelihood approximation function of infinite series; g m (m, s 2 , V) is Bessel function; a, b, V are independent variable functions; α, h are parameters of gamma distribution; m is degree of freedom; s 2 is the residual error variance; e k is the residual error; f is the number of missing data in data.​

[0029] Furthermore, automatic water quality and quantity monitoring equipment is installed at key nodes in small watersheds to carry out in-situ monitoring of water quality and quantity at the watershed scale. The water quantity monitoring equipment includes Parshall flume, pressure level sensor and data acquisition and transmission device, and the water quality monitoring equipment includes automatic analyzer for total nitrogen, automatic analyzer for total phosphorus, automatic analyzer for ammonia nitrogen and automatic analyzer for chemical oxygen demand.

[0030] The online transmission and storage of indicator data is achieved through an online pollutant monitoring system, which includes a sensing layer, a network layer, and an application layer.

[0031] Automatic water quality and quantity monitoring equipment is the sensing layer, installed at pollutant monitoring points and process nodes that affect pollutant discharge. It is used to monitor pollutant discharge status and process parameters and to communicate with the host computer.

[0032] The data acquisition and transmission equipment is a network layer device that collects data from automatic water quality and quantity monitoring equipment, and completes data storage and data transmission with the monitoring center using a microcontroller, industrial control computer, embedded computer, programmable automation controller, or programmable controller.

[0033] The monitoring center is an application layer device installed in the competent authorities at all levels. It connects to the automatic monitoring equipment through the network and issues query and control commands to it.

[0034] Furthermore, the InVEST model results were corrected using observational data at both the field and watershed scales. The model output error was evaluated by adjusting the Borselli K parameters and calculating the goodness of fit under different Borselli K parameters. The coefficient of determination R0 was used as the statistical measure of the goodness of fit. 2 R 2 The maximum value is 1, R 2 The closer the value is to 1, the smaller the error between the model simulation value and the observed value;

[0035]

[0036] In the formula, y 观测 These are the observed values; y is the average of the observed values; 模拟 These are simulated values; This represents the average of the simulated values.

[0037] Furthermore, the InVEST model principle is as follows:

[0038] First, the nitrogen and phosphorus loss loads are distributed proportionally between the surface and subsurface, calculated using the following formula:

[0039] load surf,p = (1-proportion)subsurface ) x load βp

[0040] load sub,p = proportion subsurface x load βp

[0041] where β is the nutrient; p is a single pixel; load surf,p is the surface nutrient load; load sub,p is the subsurface nutrient load;

[0042] The surface nutrient transport rate is calculated as:

[0043]

[0044] where NDR 0,p is the nutrient transport rate retained by downstream pixels; IC j is the topographic index; IC0 and k are calibration parameters;

[0045] The subsurface nutrient transport rate is calculated as:

[0046]

[0047] where NDR sub,p is the subsurface nutrient transport rate; eff subs is the maximum nutrient interception efficiency that can be achieved by subsurface flow; l subs is the interception length of subsurface flow; i.e. the distance over which the soil is assumed to hold its maximum capacity of nutrient, l is the distance from the pixel to the flow;

[0048] The nutrient output is calculated as:

[0049] U exp,p = load surf,p x NDR surf,p + load sub,p x NDR sub,p

[0050]

[0051] where U exp,p is the nutrient output load for each grid cell; NDR surf,p is the surface nutrient transport rate; NDR sub,p is the subsurface nutrient transport rate; is the total nutrient output load for the subcatchment.

[0052] Furthermore, the monitoring and early warning platform includes a remote intelligent control platform, an early warning platform, and a nitrogen and phosphorus non-point source runoff monitoring platform, enabling online monitoring and early warning of nitrogen and phosphorus runoff from farmland.

[0053] The system primarily uses an IoT hardware monitoring system for nitrogen and phosphorus non-point source pollution in farmland at the watershed scale. It is supplemented by field experiments to obtain basic data on nitrogen and phosphorus loss from different crops. InVEST parameter correction and optimization verification are performed to predict and assess the water, nitrogen, and phosphorus cycles in the regional farmland ecosystem. A monitoring and early warning platform is built to achieve remote online intelligent monitoring of farmland non-point source pollution. Real-time early warnings are issued for indicators exceeding the standards. The data from the constructed network points are cross-referenced and gradually extended to other regions.

[0054] Furthermore, through nitrogen and phosphorus loss assessment, remote sensing, UAVs and meteorological stations were used to obtain farmland distribution, terrain slope and meteorological data within the watershed. The NDR module of the InVEST model was used to combine field-scale and watershed-scale observation data to perform InVEST parameter correction and optimization verification, and finally simulated the output of nitrogen and phosphorus nutrients at the watershed scale.

[0055] Furthermore, through regional farmland non-point source assessment, machine learning models are used and the professional external survey software "External Survey Assistant" is introduced to obtain basic information on the target area, including planting area, soil type, farming method, soil nutrients, fertilization method, land use method, topographic slope and historical meteorological data.

[0056] After collecting data on various types of agricultural non-point source pollution, we used the ArcGIS platform and Geodatabase as the database to establish agricultural non-point source pollution data databases, meteorological data databases, and geospatial databases. We corrected and revised the InVEST model using measured data at the field and watershed scales, and used a machine learning model to convert the watershed-scale simulation into a macro-scale simulation to establish a farmland non-point source pollution model for the target area. Combined with ArcGIS hotspot analysis, we output the distribution characteristics of farmland non-point source pollution load, key source areas, and prevention and control strategies for the target area.

[0057] Furthermore, the principle of ArcGIS hot and cold spot analysis is as follows:

[0058] ArcGIS hotspot analysis is based on Getis-OrdG. i * The index is categorized into hot and cold sectors; Getis-OrdG i * An index is a statistical method used to identify high-value and low-value clusters. In geographical research, it is used to identify and interpret spatial distribution patterns, and highly significant hotspots at a 99% confidence level are considered key pollution source areas.

[0059] The calculation formula is:

[0060]

[0061] In the formula, G i * For the Getis-Ord index; V δ Let δ represent the nitrogen and phosphorus emission intensity of the δth research unit; W represents the average nitrogen emission intensity. oδ N represents the spatial weighting coefficients for regions o and δ; total This represents the total number of research units.

[0062] Compared with existing technologies, this invention utilizes the "1234" model for monitoring farmland non-point source pollution. This model involves using four crops, three major platforms, and two sets of models to construct a comprehensive, integrated air-ground monitoring network for macro-regional farmland non-point source pollution using Internet of Things (IoT) technology. This breaks through the limitations of traditional monitoring methods and addresses the difficulties in monitoring and assessing farmland non-point source pollution, the inability to accurately measure the macro-scale agricultural non-point source pollution status, and the lack of key source area identification and control strategies. It achieves multi-scale assessment of farmland non-point source pollution monitoring and assessment at the "field scale - watershed scale - macro scale," accurately outputting the macro-scale distribution characteristics of farmland non-point source pollution load, key source areas, and control strategies. This enables real-time monitoring and early warning of farmland non-point source pollution, with very small errors in the monitoring and assessment results. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating a multi-scale approach to monitoring, assessing, and controlling farmland non-point source pollution.

[0064] Figure 2 This is a schematic diagram of an automatic water quality and quantity monitoring device.

[0065] Figure 3 This is a schematic diagram of the structure of an online pollutant monitoring system.

[0066] Figure 4 A schematic diagram of the overall planning system for the Internet of Things.

[0067] Figure 5 This is a schematic diagram of the machine learning algorithm process.

[0068] Figure 6 This is a schematic diagram of the machine learning model process. Detailed Implementation

[0069] The following description, in conjunction with the accompanying drawings and specific embodiments, further illustrates the multi-scale farmland non-point source pollution monitoring risk assessment and control method of the present invention.

[0070] Please see Figure 1This invention discloses a method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring, comprising the following steps:

[0071] Under typical terrain and crop conditions, farmland non-point source pollution monitoring points were set up according to different monitoring objectives to monitor farmland non-point source pollution at the field scale, and the nitrogen and phosphorus pollution fluxes at the field scale were calculated using the LOADEST model.

[0072] Automatic water quality and quantity monitoring equipment is installed at key nodes in small watersheds to monitor farmland non-point source pollution at the watershed scale. The InVEST model is used to assess watershed-scale non-point source pollution, and the InVEST model is corrected using observation data.

[0073] Collect farmland non-point source pollution index data at the field and watershed scales, transmit and store the index data online, build an online monitoring and early warning platform for farmland non-point source pollution, and monitor the loss of nitrogen and phosphorus from farmland non-point sources online;

[0074] Acquire relevant data on farmland non-point source pollution, use the InVEST model and machine learning model to assess farmland non-point source pollution, and combine hot and cold spot analysis to output the distribution characteristics of farmland non-point source pollution load, key source areas and prevention and control strategies in the target area.

[0075] By integrating multi-source heterogeneous data from field-scale monitoring, watershed-scale monitoring, and farmland non-point source pollution assessment, and utilizing Internet of Things (IoT) technology, an integrated air-ground-space monitoring network for farmland non-point source pollution is constructed to achieve monitoring, assessment, and early warning of farmland non-point source pollution.

[0076] Under typical topographic conditions of dry slopes and plains, typical crops were represented by four land use types: paddy fields, dry land, irrigated land, and orchards. Standardized monitoring plots for nitrogen and phosphorus loss in farmland were established according to different monitoring objectives. Several treatments were set up for each crop, with three replicates for each treatment. The plot monitoring experiment plan is shown in Table 1. Monitoring indicators included runoff and leaching water volume and quality, microclimate, soil moisture, fertilizer application rate, and irrigation water volume. The LOADEST model was used to calculate field-scale nitrogen and phosphorus loss, and the impact of soil physicochemical properties, crop type, fertilization method, fertilizer ratio, fertilizer application rate, and rainfall on agricultural environmental information and non-point source pollution was comprehensively analyzed.

[0077] Table 1 Community Monitoring Test Plan

[0078]

[0079] Field monitoring experiments were conducted on typical crops in typical terrain to monitor non-point source pollution in farmland at the field scale. Data on water quality and quantity of non-point source pollution in farmland at the field scale were collected. The LOADEST model was used to process and analyze the farmland non-point source pollution data at the field scale to estimate the flux of non-point source pollutants in farmland at the field scale, providing observational data for calibrating the simulation results of the INVEST model.

[0080] The principle of the LOADEST model is as follows:

[0081] Non-point source pollutant flux in farmland is the total amount of pollutants passing through a certain cross-section within a certain period of time.

[0082]

[0083] In the formula, τ represents the pollutant flux during the time period; Δt represents the time interval; NP represents the number of discrete time intervals; Q represents the average flow rate during the time period; and C represents the instantaneous water quality. This represents the instantaneous pollutant flux.

[0084] The LOADEST model, based on equation (1), uses multiple linear regression to analyze the fluxes of non-point source pollutants in farmland. Make an estimate:

[0085]

[0086] In the formula, a0, a j X represents the equation coefficients; NV represents the number of independent variables; X represents the equation coefficients. j is the independent variable.

[0087] The LOADEST model provides 11 pollutant flux regression equations, as shown in Table 2. The parameters are calibrated and optimized using the AIC (Akaike information criterion) and SPPC (Schwarz posterior probability criteria). The optimal pollutant flux regression equation is the one that minimizes the values ​​of AIC and SPPC.

[0088]

[0089] In the formula, For the data set D(X) i The maximum likelihood value of (i = 1, n); d represents the maximum likelihood estimate of the equation parameters; d represents the number of equation parameters; and n represents the number of data sets used for estimating the equation parameters.

[0090] Table 2 Commonly used river load estimation models in the LOADEST model

[0091]

[0092] The LOADEST model establishes parameter estimation methods based on three statistical estimation methods: Asymptotic maximum likelihood estimation (AMLE), maximum likelihood estimation (MLE), and minimum absolute deviation method (LAD). When the residuals follow a normal distribution, censored data is estimated using AMLE, as shown in equation (4); uncensored data is estimated using MLE, as shown in equation (5); when the residuals do not follow a normal distribution, LAD is estimated using LAD, as shown in equation (6).

[0093]

[0094]

[0095] In the formula, The pollutant fluxes estimated using AMLE, MVUE, and LAD methods; H(a,b,s) 2 (α, k) is the likelihood approximation function of an infinite series; g m (m,s 2 (a, b, V) represents the Bessel function; a, b, V represent the independent variable functions; α, h represent the parameters of the gamma distribution; m represents the degrees of freedom; s 2 For residual variance; e k denoted as residual error; f represents the number of censored data points in the dataset.

[0096] Automatic water quality and quantity monitoring equipment should be installed at key points in small watersheds, such as... Figure 2 As shown, in-situ monitoring of water quality and quantity at the watershed scale was carried out. Water quantity monitoring equipment included a Parshall flume conforming to the Ministry of Water Resources industry standards, pressure and water level sensors, and data acquisition and transmission devices. Water quality monitoring equipment included automatic analyzers for total nitrogen, total phosphorus, ammonia nitrogen, and chemical oxygen demand.

[0097] Real-time monitoring, online transmission, and data storage of indicator data are achieved through an online pollutant monitoring system, such as... Figure 3 As shown, the online pollutant monitoring system comprises a sensing layer, a network layer, and an application layer. Automatic water quality and quantity monitoring equipment forms the sensing layer, installed at pollutant monitoring points and process nodes affecting pollutant emissions. It monitors pollutant discharge status and process parameters and communicates with a host computer. Data acquisition and transmission equipment forms the network layer, using microcontrollers, industrial control computers, embedded computers, programmable logic controllers (PLCs), or programmable controllers to collect data from the automatic water quality and quantity monitoring equipment, store the data, and transmit it to the monitoring center. The monitoring center forms the application layer, installed at various levels of relevant government departments. It links to the automatic monitoring equipment via a network and issues query and control commands.

[0098] Automatic water quality and quantity monitoring equipment was set up at key nodes in the small watershed where the field monitoring experiment was conducted to collect water quality and quantity data of farmland non-point source pollution at the watershed scale. Topographic, land use and meteorological data were obtained through remote sensing satellites, drones and agricultural meteorological stations. Combined with hydrological and water quality data collected at the field and watershed scales, the InVEST model was used to simulate farmland non-point source pollution load at the watershed scale.

[0099] The InVEST model results were corrected using observational data at both the field and watershed scales. The model output error was evaluated by adjusting the Borselli K parameters and calculating the goodness-of-fit under different Borselli K parameters. The coefficient of determination R0 was used as the statistical measure of the goodness-of-fit. 2 R 2 The maximum value is 1, R 2 The closer the value is to 1, the smaller the error between the model simulation value and the observed value.

[0100]

[0101] In the formula, y 观测 These are the observed values; y is the average of the observed values; 模拟 These are simulated values; This represents the average of the simulated values.

[0102] The InVEST model works as follows:

[0103] First, the nitrogen and phosphorus loss loads are distributed proportionally between the surface and subsurface, calculated using the following formula:

[0104] load surf,p = (1-proportion) subsurface )×load_ βp

[0105] load sub,p =proportion subsurface ×load_ βp

[0106] In the formula, β represents nutrients; p represents a single pixel; load surf,p For surface nutrient load; load sub,p For underground nutrient load;

[0107] The formula for calculating the surface nutrient transport rate is:

[0108]

[0109] In the formula, NDR 0,p Nutrient transfer rate retained by downstream pixels; IC jIC0 is the topographic index; IC0 and k are calibration parameters.

[0110] The formula for calculating the underground nutrient transport rate is:

[0111]

[0112] In the formula, NDR sub,p For underground nutrient transport rate; eff subs To achieve the maximum nutrient retention efficiency of groundwater flow; subs The interception length of the groundwater flow; that is, the distance at which the soil retains its maximum capacity of nutrients, and l is the distance from the pixel to the flow;

[0113] The formula for calculating nutrient output is:

[0114] U exp,p =load surf,p ×NDR surf,p +load sub,p ×NDR sub,p

[0115]

[0116] In the formula, U exp,p Nutrient output load for each grid cell; NDR surf,p For surface nutrient transport rate; NDR sub,p For underground nutrient transport rate; The total nutrient load output for the sub-basin.

[0117] Collect indicator data at the field and watershed scales, transmit and store the data online, build an online monitoring and early warning platform for farmland non-point source pollution, construct an online intelligent monitoring and early warning network for farmland non-point source pollution, and form an integrated Internet of Things (IoT) overall planning system. Utilize this integrated IoT for comprehensive monitoring to achieve regional online monitoring and early warning of farmland non-point source pollution. Figure 4 As shown.

[0118] The monitoring and early warning platform includes a remote intelligent control platform, an early warning platform, and a nitrogen and phosphorus non-point source runoff monitoring platform, enabling online monitoring and early warning of nitrogen and phosphorus runoff from farmland. It primarily utilizes an IoT hardware monitoring system for farmland nitrogen and phosphorus non-point source pollution at the watershed scale, supplemented by field trials to obtain basic data on nitrogen and phosphorus runoff from different crops. InVEST parameter calibration and optimization verification are performed to predict and assess the water, nitrogen, and phosphorus cycles in the regional farmland ecosystem. A monitoring and early warning platform is established to achieve remote online intelligent monitoring of farmland non-point source pollution, providing real-time early warnings for exceeding emission standards. The data from the constructed network of monitoring points are cross-referenced and gradually extended to other regions.

[0119] This invention employs the NDR module of the InVEST model for nitrogen and phosphorus loss assessment. A machine learning model is used to convert the corrected InVEST model simulation results into macro-scale simulation results for assessing non-point source pollution in farmland of the target area. Remote sensing, UAVs, and meteorological stations are used to acquire data on farmland distribution, topographic slope, and meteorological conditions within the watershed. By using the InVEST model's NDR module and combining field-scale and watershed-scale observation data, InVEST parameters are corrected and optimized for verification.

[0120] The machine learning model is used to transform the simulation results of the corrected and revised InVEST model into macro-scale simulation results. Simultaneously, combined with ArcGIS hotspot and coldspot analysis, the distribution characteristics of farmland non-point source pollution load, key source areas, and control strategies for the target area are output. The principle of the machine learning model is as follows: Figure 5 and Figure 6 As shown, Figure 5 For machine learning algorithm flow, Figure 6 This describes the process of machine learning modeling.

[0121] By employing machine learning models and introducing the professional field survey software "Field Survey Assistant," basic information on the target area, including planting area, soil type, farming methods, soil nutrients, fertilization methods, land use patterns, topographic slope, and historical meteorological data, was obtained. After collecting data on various types of agricultural non-point source pollution, a database of agricultural data, meteorological data, and geospatial data for agricultural non-point source pollution in agricultural areas was established using the ArcGIS platform and Geodatabase. The InVEST model was corrected and revised using measured data at both the field and watershed scales. Furthermore, the watershed-scale simulation was converted into a macro-scale simulation using machine learning models to establish a farmland non-point source pollution model for the target area. Simultaneously, combined with ArcGIS hotspot and cold spot analysis, the distribution characteristics of farmland non-point source pollution load, key source areas, and control strategies for the target area were output.

[0122] The principle of ArcGIS hot and cold spot analysis is as follows:

[0123] ArcGIS hotspot analysis is based on Getis-OrdG. i * The index is categorized into hot and cold sectors. (Getis-OrdG) i * An index is a statistical method used to identify high- and low-value clusters. In geographical research, it is often used to identify and interpret spatial distribution patterns. Areas with extremely significant hotspots at a 99% confidence level are generally considered critical pollution source areas. The calculation formula is as follows:

[0124]

[0125] In the formula, G i* For the Getis-Ord index; V δ Let δ represent the nitrogen and phosphorus emission intensity of the δth research unit; W represents the average nitrogen emission intensity. oδ N represents the spatial weighting coefficients for regions o and δ; total This represents the total number of research units.

[0126] This invention utilizes the "1234" model for monitoring farmland non-point source pollution, namely, through four crops, three major platforms, and two sets of models, it constructs a comprehensive monitoring network integrating air and ground for macro-regional farmland non-point source pollution using Internet of Things (IoT) technology. This breaks through the limitations of traditional monitoring methods and addresses the difficulties in monitoring and assessing farmland non-point source pollution, the inability to accurately measure the macro-scale agricultural non-point source pollution status, and the lack of key source area identification and control strategies. It achieves multi-scale assessment of farmland non-point source pollution monitoring and assessment at the "field scale - watershed scale - macro scale," accurately outputting the macro-scale farmland non-point source pollution load distribution characteristics, key source areas, and control strategies. This enables real-time monitoring and early warning of farmland non-point source pollution, with very small errors in the monitoring and assessment results.

[0127] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit disclosed in the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring, characterized in that, Includes the following steps: Under typical terrain and crop conditions, farmland non-point source pollution monitoring points were set up according to different monitoring objectives to monitor farmland non-point source pollution at the field scale, and the nitrogen and phosphorus pollution fluxes at the field scale were calculated using the LOADEST model. Automatic water quality and quantity monitoring equipment is installed at key nodes in small watersheds to monitor farmland non-point source pollution at the watershed scale. The InVEST model is used to assess watershed-scale non-point source pollution, and the InVEST model is corrected using observation data. Collect farmland non-point source pollution index data at the field and watershed scales, transmit and store the index data online, build an online monitoring and early warning platform for farmland non-point source pollution, and monitor the loss of nitrogen and phosphorus from farmland non-point sources online; Acquire relevant data on farmland non-point source pollution, use the InVEST model and machine learning model to assess farmland non-point source pollution, and combine hot and cold spot analysis to output the distribution characteristics of farmland non-point source pollution load, key source areas and prevention and control strategies in the target area. By integrating multi-source heterogeneous data from field-scale monitoring, watershed-scale monitoring, and farmland non-point source pollution assessment, and utilizing Internet of Things (IoT) technology, an integrated air-ground-space monitoring network for farmland non-point source pollution is constructed to achieve monitoring, assessment, and early warning of farmland non-point source pollution.

2. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, Typical crop field-scale monitoring was carried out under typical terrain conditions. Typical crops represent four land use types: paddy fields, dry land, irrigated land, and orchards. Establish standardized monitoring plots for nitrogen and phosphorus loss from farmland. The monitoring indicators include runoff and leaching water volume and quality, microclimate, soil moisture, fertilizer application and irrigation water volume. The LOADEST model was used to calculate field-scale nitrogen and phosphorus losses, and the impact of agricultural environmental information such as soil physicochemical properties, crop type, fertilization method, fertilizer ratio, fertilizer amount, and rainfall on non-point source pollution was comprehensively analyzed.

3. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, The principle of the LOADEST model is as follows: Non-point source pollutant flux in farmland is the total amount of pollutants passing through a certain cross-section within a certain period of time. In the formula, τ represents the pollutant flux during the time period; Δt represents the time interval; NP represents the number of discrete time intervals; Q represents the average flow rate during the time period; and C represents the instantaneous water quality. This refers to the instantaneous pollutant flux; The LOADEST model, based on equation (1), uses multiple linear regression to analyze the fluxes of non-point source pollutants in farmland. Make an estimate: In the formula, a0, a j X represents the equation coefficients; NV represents the number of independent variables; X represents the equation coefficients. j As the independent variable; The LOADEST model provides 11 pollutant flux regression equations and uses the AIC and SPPC criteria for parameter calibration and optimization. The optimal pollutant flux regression equation is the one that minimizes the AIC and SPPC values. In the formula, For data set D(a) i The maximum likelihood value of (i = 1, n); is the maximum likelihood estimate of the equation parameters; d is the number of equation parameters; n is the number of data sets used for estimating the equation parameters; The LOADEST model establishes parameter estimation methods based on three statistical estimation methods: Asymptotic maximum likelihood estimation (AMLE), maximum likelihood estimation (MLE), and minimum absolute deviation method (LAD). When the residuals follow a normal distribution, censored data is estimated using AMLE, as shown in equation (4); uncensored data is estimated using MLE, as shown in equation (5); when the residuals do not follow a normal distribution, LAD is estimated using LAD, as shown in equation (6). In the formula, The pollutant fluxes estimated using AMLE, MVUE, and LAD methods; H(a,b,s) 2 (α, k) is the likelihood approximation function of an infinite series; g m (m,s 2 (a, b, V) represents the Bessel function; a, b, V represent the independent variable functions; α, h represent the parameters of the gamma distribution; m represents the degrees of freedom; s 2 For residual variance; e k denoted as residual error; f represents the number of censored data points in the dataset.

4. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, Automatic water quality and quantity monitoring equipment is installed at key nodes in small watersheds in the field to carry out in-situ monitoring of water quality and quantity at the watershed scale. The water quantity monitoring equipment includes Parshall flume, pressure level sensor and data acquisition and transmission device. The water quality monitoring equipment includes automatic analyzer for total nitrogen, automatic analyzer for total phosphorus, automatic analyzer for ammonia nitrogen and automatic analyzer for chemical oxygen demand. The online transmission and storage of indicator data is achieved through an online pollutant monitoring system, which includes a sensing layer, a network layer, and an application layer. Automatic water quality and quantity monitoring equipment is the sensing layer, installed at pollutant monitoring points and process nodes that affect pollutant discharge. It is used to monitor pollutant discharge status and process parameters and to communicate with the host computer. The data acquisition and transmission equipment is a network layer device that collects data from automatic water quality and quantity monitoring equipment, and completes data storage and data transmission with the monitoring center using a microcontroller, industrial control computer, embedded computer, programmable automation controller, or programmable controller. The monitoring center is an application layer device installed in the competent authorities at all levels. It connects to the automatic monitoring equipment through the network and issues query and control commands to it.

5. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, The InVEST model results were corrected using observational data at both the field and watershed scales. The model output error was evaluated by adjusting the Borselli K parameters and calculating the goodness of fit under different Borselli K parameters. The coefficient of determination R0 was used as the statistical measure of the goodness of fit. 2 R 2 The maximum value is 1, R 2 The closer the value is to 1, the smaller the error between the model simulation value and the observed value; In the formula, y 观测 These are the observed values; y is the average of the observed values; 模拟 These are simulated values; This represents the average of the simulated values.

6. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 5, characterized in that, The InVEST model works as follows: First, the nitrogen and phosphorus loss loads are distributed proportionally between the surface and subsurface, calculated using the following formula: load surf,p =(1-proportion subsurface )×load_ βp load sub,p =proportion subsurface ×load_ βp In the formula, β represents nutrients; p represents a single pixel; load surf,p For surface nutrient load; load sub,p For underground nutrient load; The formula for calculating the surface nutrient transport rate is: In the formula, NDR 0,p Nutrient transfer rate retained by downstream pixels; IC j IC0 is the topographic index; IC0 and k are calibration parameters. The formula for calculating the underground nutrient transport rate is: In the formula, NDR sub,p For underground nutrient transport rate; eff subs To achieve the maximum nutrient retention efficiency of groundwater flow; subs The interception length of the groundwater flow; that is, the distance at which the soil retains its maximum capacity of nutrients, and l is the distance from the pixel to the flow; The formula for calculating nutrient output is: U exp,p =load surf,p ×NDR surf,p +load sub,p ×NDR sub,p In the formula, U exp,p Nutrient output load for each grid cell; NDR surf,p For surface nutrient transport rate; NDR sub,p For underground nutrient transport rate; The total nutrient load output for the sub-basin.

7. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, The monitoring and early warning platform includes a remote intelligent control platform, an early warning platform, and a nitrogen and phosphorus non-point source runoff monitoring platform, enabling online monitoring and early warning of nitrogen and phosphorus runoff from farmland. The system primarily uses an IoT hardware monitoring system for nitrogen and phosphorus non-point source pollution in farmland at the watershed scale. It is supplemented by field experiments to obtain basic data on nitrogen and phosphorus loss from different crops. InVEST parameter correction and optimization verification are performed to predict and assess the water, nitrogen, and phosphorus cycles in the regional farmland ecosystem. A monitoring and early warning platform is built to achieve remote online intelligent monitoring of farmland non-point source pollution. Real-time early warnings are issued for indicators exceeding the standards. The data from the constructed network points are cross-referenced and gradually extended to other regions.

8. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, By assessing nitrogen and phosphorus loss, data on farmland distribution, topographic slope, and meteorological conditions within the watershed were obtained using remote sensing, drones, and weather stations. The NDR module of the InVEST model was used, and InVEST parameters were corrected and optimized by combining field-scale and watershed-scale observation data. Finally, the output of nitrogen and phosphorus nutrients at the watershed scale was simulated.

9. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 1, characterized in that, By conducting regional farmland non-point source assessments, using machine learning models and introducing the professional external survey software "External Survey Assistant," basic information such as planting area, soil type, farming methods, soil nutrients, fertilization methods, land use patterns, topographic slope, and historical meteorological data of the target area can be obtained. After collecting data on various types of agricultural non-point source pollution, we used the ArcGIS platform and Geodatabase as the database to establish agricultural non-point source pollution data databases, meteorological data databases, and geospatial databases. We corrected and revised the InVEST model using measured data at the field and watershed scales, and used a machine learning model to convert the watershed-scale simulation into a macro-scale simulation to establish a farmland non-point source pollution model for the target area. Combined with ArcGIS hotspot analysis, we output the distribution characteristics of farmland non-point source pollution load, key source areas, and prevention and control strategies for the target area.

10. The method for risk assessment and control of farmland non-point source pollution based on multi-scale monitoring according to claim 9, characterized in that, The principle of ArcGIS hot and cold spot analysis is as follows: ArcGIS hotspot analysis is based on Getis-OrdG. i * The index is categorized into hot and cold sectors; Getis-OrdG i * An index is a statistical method used to identify high-value and low-value clusters. In geographical research, it is used to identify and interpret spatial distribution patterns, and highly significant hotspots at a 99% confidence level are considered key pollution source areas. The calculation formula is: In the formula, G i * For the Getis-Ord index; V δ Let δ represent the nitrogen and phosphorus emission intensity of the δth research unit; W represents the average nitrogen emission intensity. oδ N represents the spatial weighting coefficients for regions o and δ; total This represents the total number of research units.

Images