Near-real-time ozone numerical prediction error evaluation method
By obtaining meteorological and air pollutant monitoring data in near real time and evaluating ozone forecast errors with multiple models, the problem of difficulty in integrating ozone forecasting systems in the existing technology is solved, and the accuracy of ozone forecasting and automated decision-making support is achieved.
Patent Information
- Application Number
- CN202510350543.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-24
- Publication Date
- 2025-07-08
AI Technical Summary
The existing ozone forecasting system cannot effectively integrate adjacent monitoring data, resulting in limited forecast accuracy and different error sources of different models, making it difficult to make full use of near-real-time monitoring data to correct model forecasts, which is easy to cause misjudgment of ozone generation and transmission situations.
The near-real-time acquisition of meteorological and air pollutant monitoring data is used, combined with the ozone proximity forecast model and zero-dimensional box model based on monitoring data and forecast data, the system error, cumulative error and chemical generation contribution error are calculated, and the ozone numerical forecast error assessment is performed through the XGBoost random tree model, LSTM neural network or GNN graph neural network training model.
It improves the accuracy of ozone forecasting, realizes the correction of short-term forecasting results, provides automated decision-making support, and improves the ozone pollution forecasting capabilities of environmental monitoring departments.
Smart Images
Figure HDA0005325910030000011
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of environmental monitoring and early warning, and particularly relates to a method for evaluating the error of near-real-time ozone numerical prediction. Background Art
[0002] In recent years, with the improvement of people's living quality, the public has paid more and more attention to the health problems caused by air pollution. Ozone is one of the main air pollutants in China, and the urban ozone concentration has shown an oscillating upward trend in recent years, increasing the risk of people suffering from respiratory and circulatory system diseases. To control and prevent ozone pollution, accurate ozone forecasting is required. Currently, ozone forecasting mainly relies on two types of technical systems: physical and chemical simulation based on air quality numerical models, and statistical forecasting using machine learning and deep neural networks.
[0003] However, the current forecasting system cannot efficiently incorporate near-monitoring data. The existing forecasting process is difficult to effectively incorporate the ground monitoring data obtained in real time into the model calibration link, resulting in limited forecasting accuracy. At the same time, there are certain errors in the current ozone numerical forecasting and statistical forecasting technologies, including errors caused by meteorological and emission input data, as well as errors caused by physical and chemical mechanisms and the statistical model training process. Since the error sources of different models are different, it is difficult to fully utilize near-real-time monitoring data to correct the model forecast when comprehensively judging the ozone pollution situation on the same day based on multi-model forecasting, thus it is relatively easy to misjudge the situation of ozone generation and transmission. Summary of the Invention
[0004] Aiming at the deficiencies of the existing technology, the present invention proposes a method for evaluating the error of near-real-time ozone numerical prediction to solve the technical problem that the error sources of different models are different in the existing technology, and it is relatively easy to misjudge the situation of ozone generation and transmission when comprehensively judging the ozone pollution situation on the same day based on multi-model forecasting. The technical solution adopted by the present invention is as follows:
[0005] A method for evaluating the error of near-real-time ozone numerical prediction includes the following steps:
[0006] Near-real-time obtain meteorological monitoring data and air pollutant monitoring data of the previous night and the current morning, and obtain meteorological forecast data and air pollutant forecast data of the previous night and the current morning;
[0007] According to the monitoring data and forecast data, fuse and use an ozone nowcasting model based on monitoring data, an ozone nowcasting model based on forecast data, and a zero-dimensional box model to calculate the systematic error of the forecast model, the cumulative error of air pollutant simulation, the ozone forecast error caused by weather forecast error, and the contribution error of local air pollutant chemical generation.
[0008] Furthermore, the meteorological monitoring data includes: ground meteorological station air temperature, air pressure, relative humidity, precipitation, and wind speed, and the time resolution of the meteorological monitoring data is 1 to 3 hours.
[0009] Furthermore, the air pollutant monitoring data includes: ozone, nitrogen dioxide, and volatile organic compound concentrations at ground pollutant monitoring stations, and the time resolution of the air pollutant monitoring data is 1 to 3 hours.
[0010] Furthermore, when calculating the systematic error of the forecasting model, using the meteorological forecasting data and air pollutant forecasting data of the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data to calculate and output the predicted ozone MDA8 concentration c1, use the ozone nowcasting model based on forecasting data to calculate and output the predicted ozone MDA8 concentration c2, and use the predicted difference c2 minus c1 to represent the systematic error of the forecasting model.
[0011] Furthermore, when calculating the cumulative error of air pollutant simulation, using the meteorological forecasting data and air pollutant monitoring data of the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data to calculate and output the predicted ozone MDA8 concentration c3, use the ozone nowcasting model based on forecasting data to calculate and output the predicted ozone MDA8 concentration c4, use the predicted difference c3 minus c1, and c4 minus c2 to represent the cumulative error of air pollutant simulation.
[0012] Furthermore, when calculating the ozone forecasting error caused by weather forecasting error, using the meteorological monitoring data and air pollutant forecasting data of the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data to calculate and output the predicted ozone MDA8 concentration c5, use the ozone nowcasting model based on forecasting data to calculate and output the predicted ozone MDA8 concentration c6, use the predicted differences c5 minus c1 and c6 minus c2 to represent the ozone forecasting error caused by weather forecasting error.
[0013] Furthermore, when calculating the contribution error of local air pollutant chemical generation, using near-real-time meteorological, O3, NO2, VOC monitoring data, and the simulated concentrations of other atmospheric components, use a zero-dimensional box model to calculate and output the predicted ozone MDA8 concentration c7, and use the predicted difference c7 - c1 to represent the contribution error of local air pollutant chemical generation to the ozone pollution on the current day; the other atmospheric components include SO2, NO, and / or PM 2.5 。
[0014] Furthermore, the ozone nowcasting model based on monitoring data and the ozone nowcasting model based on forecasting data are trained using the XGBoost random tree model, the LSTM neural network architecture, or the GNN graph neural network.
[0015] Further, it also includes: when the difference between the average value of the numerical service forecast result and -2×(the systematic error of the forecast model + the cumulative error of the air pollutant simulation) exceeds the ozone pollution standard of 10 micrograms per cubic meter, it is forecast as an ozone pollution exceeding standard event, and early warning and control are required;
[0016] When the sum of the average value of the numerical service forecast result and +2×(the systematic error of the forecast model + the air pollutant simulation) is lower than the ozone pollution standard of 10 micrograms per cubic meter, it is forecast that the ozone pollution does not exceed the standard.
[0017] Further, if the ozone forecast error caused by the weather forecast error is greater than 5 micrograms per cubic meter, the predicted concentrations c1 and c2 are used to replace the numerical forecast result to judge whether it exceeds the standard.
[0018] As can be seen from the above technical solutions, the beneficial technical effects of the present invention are as follows:
[0019] Based on the near-real-time observation data of the previous night and the morning of the forecast day, numerical forecast error analysis is carried out by predicting the ozone pollution situation on the forecast day; the meteorological and air pollutant observation data at adjacent times are fully utilized, which helps to improve the accuracy of ozone forecast error analysis; the sources of ozone forecast error and the contribution of pollutants in the local surrounding areas are reflected in a quantitative manner, and the monitoring data, forecast data and zero-dimensional box model are integrated for error assessment, realizing the correction of the short-term forecast result by the ozone nowcasting error analysis, improving the forecast accuracy, and providing new decision-making support information for the ozone pollution forecast judgment of the environmental monitoring department. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts do not necessarily draw according to the actual scale.
[0021] Figure 1 It is a schematic flow chart of the ozone numerical forecast error assessment method according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0022] The following will describe in detail the embodiments of the technical solutions of the present invention with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, so they are only examples and cannot be used to limit the protection scope of the present invention.
[0023] It should be noted that unless otherwise specified, the technical terms or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art to which the present invention belongs.
[0024] Embodiment
[0025] When the environmental monitoring department judges the air pollution situation on the same day, it usually combines multiple air quality numerical forecasts and statistical forecast results, and judges the potential pollution possibility on the same day based on the actual situation of the previous night and the morning of the same day combined with experience, and makes responses. The standard ozone pollution situation assessment process includes:
[0026] 1. Complete the preparation of preliminary forecast data;
[0027] 2. Integrate the meteorological and air quality monitoring data of the previous night and the morning of the same day;
[0028] 3. Forecasters judge the possible pollution development trend on the same day based on data and experience.
[0029] However, the human judgment of the above process relies strongly. The current pollution situation judgment process relies too much on the experience accumulation of forecasters, specifically manifested as: lack of a standardized data fusion and weight allocation mechanism; inability to establish a unified model accuracy evaluation system; artificial intervention leads to limited system response speed and unstable effects.
[0030] The inventor of the present application has found through research that: the daily maximum concentration of ozone pollution usually appears in the afternoon, and its generation and transmission processes are deeply affected by the meteorological conditions in the morning of the same day and the previous night. Combining the above research results, this embodiment proposes a near-real-time ozone numerical forecast error assessment method, which estimates and analyzes the ozone pollution forecast error on the same day based on the near monitoring data of air pollutant concentration and meteorological parameters, and is beneficial to automated decision support.
[0031] In order to evaluate the error of ozone numerical forecast and provide multi-angle data for automated decision-making, it is first necessary to build a model. In this embodiment, there are 3 types of models in total:
[0032] Ozone nowcasting model based on monitoring data, which is used to simulate the ozone situation caused by real emissions under different meteorological conditions;
[0033] Ozone nowcasting model based on forecast data, which is used to simulate the physical and chemical processes of ozone under different meteorological conditions;
[0034] Zero-dimensional box model, which is used to simulate the local chemical generation process of ozone under different meteorological conditions.
[0035] The following describes the construction process of the above 3 types of models:
[0036] 1. Collect historical data and construct a training set
[0037] The historical data collected provides a dataset for the prediction analysis model during training in subsequent steps. The historical data includes meteorological monitoring data, air pollutant monitoring data, meteorological forecast data, and air pollutant forecast data.
[0038] Among them, the meteorological monitoring data includes: ground meteorological station air temperature, air pressure, relative humidity, precipitation, wind speed, and the above monitoring data are measured values. In some embodiments, the time resolution of the meteorological monitoring data is 1 to 3 hours.
[0039] The air pollutant monitoring data includes: ozone, nitrogen dioxide, volatile organic compound (VOC) concentration, etc. at ground pollutant monitoring stations, and the above monitoring data are measured values. In some embodiments, the time resolution of the air pollutant monitoring data is 1 to 3 hours.
[0040] The meteorological forecast data includes: operational meteorological forecast simulation data implemented based on the WRF mesoscale meteorological model, with a spatial resolution of 4 kilometers.
[0041] The air pollutant forecast data includes: operational air pollutant numerical forecast simulation data implemented based on mesoscale air quality models such as WRF-CMAQ and WRF-Chem, with a spatial resolution of 4 kilometers.
[0042] In some embodiments, the acquisition time period of the above meteorological monitoring data, air pollutant monitoring data, meteorological forecast data, and air pollutant forecast data is the recent 5 years.
[0043] In some embodiments, cases where the actual ozone MDA8 is less than or equal to 100 micrograms per cubic meter are excluded from the training set, and only near-exceedance events with MDA8 greater than 100 micrograms per cubic meter are included. Subsequently, when continuing with actual judgment, the short-term forecast model is only used when near-exceedance occurs to increase the actual forecasting ability and usage efficiency of the model.
[0044] 2. According to the training set, different machine learning methods are used for training, and 3 models are respectively built: (1) Ozone nowcasting model based on monitoring data
[0045] The ozone nowcasting model based on monitoring data takes the meteorological monitoring data and air pollutant monitoring data within a certain distance around the location to be predicted from the previous night to the current morning as inputs, and takes the daily maximum 8-hour ozone concentration on the current day as the output. It is a forecasting model trained using machine learning, such as the XGBoost random tree model, the LSTM neural network architecture, or the GNN graph neural network. In a specific implementation, the meteorological monitoring data and air pollutant monitoring data within 100 kilometers around the location to be predicted from 21:00 the previous night to 9:00 the current morning can be used as inputs, and the daily maximum 8-hour (MDA8) ozone concentration on the current day can be used as the output to train the model. Taking the XGBoost architecture as an example, the model training process will be described in detail:
[0046] The core principle of XGBoost is to use the gradient boosting framework for a decision tree structure, train a new decision tree through the residuals of the decision tree prediction results, and then weight and combine the new decision tree and the old decision tree; repeat the above process until finally a strong decision tree model is obtained.
[0047] Suppose there are a total of m meteorological monitoring stations within 100 kilometers around the location to be predicted, monitoring the near-surface air temperature T, relative humidity RH, east-west wind speed U, north-south wind speed V, 1-hour cumulative precipitation p, with a time resolution of 1 hour; and a total of n air pollutant monitoring stations, monitoring the near-surface ozone concentration C_O3, nitrogen dioxide concentration C_NO2, and VOC concentration C_VOC, with a time resolution of 1 hour. Define the input vector X of the XGBoost model as X = {T(i,t), RH(i,t), U(i,t), V(i,t), p(i,t), C_O3(j,t), C_NO2(j,t), C_VOC(j,t)}, where i = 1, 2, …, m; j = 1, 2, …, n; t ∈ [21, 22, 23, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9], totaling 45m + 27n features. Define the output vector Y as the daily ozone MDA8 concentration at the location to be predicted.
[0048] In a specific implementation, obtain the meteorological monitoring data for the past 5 years, collect 1825 sets of input data, and then perform normalization processing on all the data. Allocate 80% (1460) as the training set and 20% (365) as the validation set. Considering that there may be a correlation between daytime meteorology and pollutant concentrations, use the data from the first 4 years as the training set and the data from the 5th year as the validation set.
[0049] XGBoost uses the root mean square error RMSE between the predicted ozone concentration and the ozone concentration in the training set as the cost function, and after training, obtains the model M such that Y = M(X).
[0050] The hyperparameters are set as follows:
[0051] Maximum depth of tree: Controls the maximum depth of a single decision tree. Set to 3.
[0052] Minimum child weight: Controls the minimum sample weight of child nodes. Set to 2.
[0053] Subsample ratio: Controls the ratio of samples used for training each tree. Set to 0.8.
[0054] L2 regularization term: Used to control the weights of leaf nodes. Set to 0.1
[0055] Learning rate: Controls the contribution of each tree to the final model. Set to 0.1
[0056] Number of trees: The maximum number of trees in the model. Set to 1000.
[0057] Early stopping rounds: When the performance on the validation set does not improve within a certain number of rounds, stop training early.
[0058] (2) Ozone nowcasting model based on forecast data
[0059] When building the ozone nowcasting model based on forecast data, the same construction method as the ozone nowcasting model based on monitoring data is used. Machine learning is used, such as a random tree model based on XGBoost, an LSTM neural network architecture, or a GNN graph neural network, to train the obtained nowcasting model. During training, the training set is replaced with the daily meteorological forecast data and air pollutant forecast data for the past 5 years; the value-taking location of the forecast data is the model grid point corresponding to the location of each meteorological monitoring station and air pollutant monitoring station. During training, the daily meteorological forecast data and air pollutant forecast data are used as inputs, and the daily maximum 8-hour (MDA8) concentration of ozone is used as the output to train the model.
[0060] (3) Zero-dimensional box model
[0061] When building the zero-dimensional box model, based on the DSMACC model or F0AM model framework, the chemical process is simulated near real-time. The simulation speed of the zero-dimensional box model is significantly faster than that of the three-dimensional air quality model, and the calculation of the evolution of the chemical components of air pollutants at a specific location on the same day can be completed within a few minutes, with timeliness. In a specific implementation manner, the DSMACC model is taken as an example for detailed description:
[0062] The DSMACC model can perform integral operations based on a specific chemical reaction mechanism under specific meteorological conditions and chemical initial conditions (which can be obtained near real-time) to simulate the atmospheric chemical process; it includes the following steps:
[0063] Use the KPP chemical kinetic process module to compile a suitable atmospheric chemical mechanism;
[0064] With a 20-minute time step for chemical calculations, pre-compile the DSMACC model;
[0065] Given 9:00 am on the current day as the model start time, given the longitude and latitude of the location to be forecast, the monitored meteorological field at the location to be forecast, the monitored air pollutants (ozone, NO2, VOC), and the initial concentrations of other atmospheric components (such as SO2, NO, PM 2.5 ) simulated by the operational forecasting system as the initial field of DSMACC;
[0066] Conduct DSMACC simulations for 18 hours (from 6:00 am on the current day to 3:00 am on the next day), and calculate the daily maximum hourly concentration and the daily maximum 8-hour average concentration of ozone based on the simulation results.
[0067] In some embodiments, for the zero-dimensional box model, machine learning methods such as XGBoost can be used to train the box model into a corresponding statistical model to further ensure the comparability between models. During training, the training data can be generated by driving the box model with the simulation results and observed results of the forecast data in the past 5 years.
[0068] Based on the three models built above, in this embodiment, the near-real-time ozone numerical forecasting error assessment method includes the following steps:
[0069] S1. Near-real-time obtain the meteorological monitoring data and air pollutant monitoring data of the previous night and the current morning, as well as the meteorological forecast data and air pollutant forecast data of the previous night and the current morning
[0070] In the following technical solutions of this embodiment, the concept of near-real-time will be involved. Near-real-time means that through advanced monitoring, data processing, and transmission technologies, meteorological information can be collected, analyzed, and distributed at the fastest speed, so as to provide users with information related to the current weather and air pollution situation that is almost synchronous.
[0071] The meteorological monitoring data and air pollutant monitoring data adopt the measured data simultaneously monitored by meteorological and ecological environment monitoring stations, and the meteorological forecast data and air pollutant forecast data adopt the forecast data simultaneously released by the air quality prediction and forecasting center.
[0072] S2. Calculate and obtain the systematic error of the forecasting model, the cumulative error of air pollutant simulation, the ozone forecasting error caused by weather forecasting error, and the contribution error of local air pollutant chemical generation
[0073] According to the four types of data obtained in step S1, calculate each error in the following manner:
[0074] The systematic error e1 of the prediction model takes the meteorological prediction data and air pollutant prediction data of the previous night and the current morning as inputs, uses the ozone nowcasting model based on monitoring data for calculation, outputs the predicted ozone MDA8 concentration c1, uses the ozone nowcasting model based on prediction data for calculation, outputs the predicted ozone MDA8 concentration c2, and uses the prediction difference c2 - c1 to represent the systematic error of the prediction model.
[0075] The cumulative error e2 of air pollutant simulation takes the meteorological prediction data and air pollutant monitoring data of the previous night and the current morning as inputs, uses the ozone nowcasting model based on monitoring data for calculation, outputs the predicted ozone MDA8 concentration c3, uses the ozone nowcasting model based on prediction data for calculation, outputs the predicted ozone MDA8 concentration c4, and uses the prediction differences c3 - c1, c4 - c2 to represent the cumulative error of air pollutant simulation.
[0076] The ozone prediction error e3 caused by weather forecast error takes the meteorological monitoring data and air pollutant prediction data of the previous night and the current morning as inputs, uses the ozone nowcasting model based on monitoring data for calculation, outputs the predicted ozone MDA8 concentration c5, uses the ozone nowcasting model based on prediction data for calculation, outputs the predicted ozone MDA8 concentration c6, and uses the prediction differences c5 - c1, c6 - c2 to represent the ozone prediction error caused by weather forecast error.
[0077] The contribution error e4 of local pollutant chemical generation takes the near-real-time meteorological, O3, NO2, VOC monitoring data, and the simulated concentrations of other atmospheric components (such as SO2, NO, PM 2.5 ) and uses a zero-dimensional box model for calculation and outputs the predicted ozone MDA8 concentration c7, and uses the prediction difference c7 - c1 to represent the contribution error of local air pollutant chemical generation to the ozone pollution on the current day.
[0078] In some embodiments, when the ozone pollution prediction error on the current day is estimated and analyzed using the technical solutions described above, it can provide support for the automated decision-making of ozone pollution early warning and control measures, as follows:
[0079] Automated decision-making support, the information used includes the ozone numerical operational prediction results of multiple parties and the error sources, and is used for ozone pollution weather early warning decision-making support. The numerical operational prediction results are traditional numerical ozone prediction results that do not include nowcasting information. The automated decision-making support includes the following decision-making processes:
[0080] When the difference between the average value of the numerical operational prediction results - 2×(e1 + e2) exceeds the ozone pollution standard of 10 micrograms per cubic meter, it is defined as Scenario A, the forecast is an ozone pollution exceeding standard event, and corresponding early warning and control measures are taken;
[0081] When the sum of the average value of the numerical business forecast result + 2×(e1 + e2) is lower than the ozone pollution standard of 10 micrograms per cubic meter, it is defined as Scenario B, and the forecast is that the ozone pollution does not exceed the standard;
[0082] For other situations, it is defined as Scenario C, and auxiliary decision-making with the help of error assessment is required, including the following steps:
[0083] If e3 > 5 micrograms per cubic meter, the ozone forecast error caused by the weather forecast error is relatively large. More attention should be paid to the prediction of c1 and c2 concentrations, and the c1 and c2 predicted concentrations are used to replace the numerical forecast result to judge whether it exceeds the standard. The rules are the same as above; at the same time, closely monitor the real-time meteorological monitoring data on the same day and monitor the change trend of ozone pollution at the same time;
[0084] Under Scenario A, when the local chemical generation contribution e4 of pollutants accounts for more than 70% of the ozone exceedance standard, relatively strict emission control measures can be considered in the local area; when it is less than 70%, it is necessary to combine the meteorological field to determine the source of pollutants or precursors and take reasonable regional coordinated emission reduction measures.
[0085] The implementation process of the technical solution of this embodiment is illustrated by the following examples:
[0086] At 9:30 every morning, obtain the meteorological monitoring data, air pollutant monitoring data from 21:00 the previous night to 9:00 that morning, and the forecast data corresponding to the observation sites;
[0087] Using the above data as input, use the ozone nowcasting model based on monitoring data, the ozone nowcasting model based on forecast data, and the zero-dimensional box model to generate multiple predicted ozone MDA8 concentrations c1 - c7 for the same day, and calculate the forecast errors e1 - e4 accordingly;
[0088] Combining the model prediction results and the forecast error analysis results, carry out the nowcasting comprehensive judgment business work on the ozone pollution situation on the same day, including early warning and control measures.
[0089] The technical solution of this embodiment is based on the near-real-time observation data of the previous night and the morning of the forecast day, and conducts numerical forecast error analysis by predicting the ozone pollution situation on the forecast day; it makes full use of the meteorological and air pollutant observation data at the near moment, which helps to improve the accuracy of ozone forecast error analysis; it quantitatively reflects the source of ozone forecast error and the contribution of pollutants in the local surrounding area, and combines monitoring data, forecast data and zero-dimensional box model for error assessment, realizing the correction of the nowcasting result of ozone by ozone nowcasting error analysis, improving the forecast accuracy, and providing new decision-making support information for the ozone pollution forecast judgment of environmental monitoring departments.
[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered within the scope of the claims and the description of the present invention.
Claims
1. A method for evaluating the error of near-real-time ozone numerical prediction, characterized in that It includes the following steps: Obtain meteorological monitoring data and air pollutant monitoring data for the previous night and the current morning in near real-time, and obtain meteorological forecast data and air pollutant forecast data for the previous night and the current morning; According to the monitoring data and forecast data, use the ozone nowcasting model based on monitoring data, the ozone nowcasting model based on forecast data, and the zero-dimensional box model in combination to calculate the systematic error of the forecast model, the cumulative error of air pollutant simulation, the ozone forecast error caused by weather forecast error, and the contribution error of local air pollutant chemical generation.
2. The near real-time ozone numerical prediction error assessment method according to claim 1, characterized in that The meteorological monitoring data includes: ground meteorological station air temperature, air pressure, relative humidity, precipitation, and wind speed, and the time resolution of the meteorological monitoring data is 1 to 3 hours.
3. The near real-time ozone numerical prediction error assessment method according to claim 1, characterized in that The air pollutant monitoring data includes: ozone, nitrogen dioxide, and volatile organic compound concentrations at ground pollutant monitoring sites, and the time resolution of the air pollutant monitoring data is 1 to 3 hours.
4. The near real-time ozone numerical prediction error assessment method according to claim 1, wherein, When calculating the systematic error of the forecast model, use the meteorological forecast data and air pollutant forecast data for the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data for calculation and output the predicted ozone MDA8 concentration c1, use the ozone nowcasting model based on forecast data for calculation and output the predicted ozone MDA8 concentration c2, and use the predicted difference c2 minus c1 to represent the systematic error of the forecast model.
5. The near real-time ozone numerical prediction error assessment method according to claim 4, characterized in that, When calculating the cumulative error of air pollutant simulation, use the meteorological forecast data and air pollutant monitoring data for the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data for calculation and output the predicted ozone MDA8 concentration c3, use the ozone nowcasting model based on forecast data for calculation and output the predicted ozone MDA8 concentration c4, use the predicted difference c3 minus c1, and c4 minus c2 to represent the cumulative error of air pollutant simulation.
6. The method for evaluating the error of near-real-time ozone numerical prediction according to claim 4, characterized in that When calculating the ozone forecast error caused by weather forecast error, use the meteorological monitoring data and air pollutant forecast data for the previous night and the current morning as inputs, use the ozone nowcasting model based on monitoring data for calculation and output the predicted ozone MDA8 concentration c5, use the ozone nowcasting model based on forecast data for calculation and output the predicted ozone MDA8 concentration c6, use the predicted difference c5 minus c1, c6 minus c2 to represent the ozone forecast error caused by weather forecast error.
7. The near real-time ozone numerical prediction error assessment method according to claim 4, characterized in that, When calculating the contribution error of local air pollutant chemical generation, using near-real-time meteorological, O3, NO2, VOC monitoring data, and simulated concentrations of other atmospheric components, a zero-dimensional box model is used for calculation and output of the predicted ozone MDA8 concentration c7, and the predicted difference c7 - c1 is used to represent the contribution error of local air pollutant chemical generation to the ozone pollution on that day; the other atmospheric components include SO2, NO, and / or PM 2.5 .
8. The near-real-time ozone numerical prediction error assessment method according to any one of claims 4-6, characterized in that, The ozone nowcasting model based on monitoring data and the ozone nowcasting model based on forecast data are trained using the XGBoost random tree model, the LSTM neural network architecture, or the GNN graph neural network.
9. The near-real-time ozone numerical prediction error assessment method according to claim 1, characterized in that It also includes: When the difference between the average value of the numerical operational forecast result - 2×(systematic error of the forecast model + cumulative error of air pollutant simulation) exceeds the ozone pollution standard of 10 micrograms per cubic meter, it is forecast as an ozone pollution exceeding standard event, and early warning and control are required; When the sum of the average value of the numerical operational forecast result + 2×(systematic error of the forecast model + air pollutant simulation) is lower than the ozone pollution standard of 10 micrograms per cubic meter, it is forecast that the ozone pollution does not exceed the standard.
10. The near real-time ozone numerical prediction error assessment method according to claim 4, wherein If the ozone forecast error caused by the weather forecast error is greater than 5 μg / m³, the predicted concentrations c1 and c2 are used to replace the numerical forecast results to determine whether it exceeds the standard.