A method for inversion of subcloud / in-cloud clearance based on daily-scale nitrate wet deposition

Abstract

The present invention provides a method for inverting cloud / in-cloud clearance based on daily-scale nitrate wet deposition, which belongs to the field of remote sensing atmospheric environment applications. Based on satellite data combined with a backward trajectory model, pollutant concentration parameters in the cloud before a precipitation event are calculated to quantify the long-range pollution transport effect; a random forest machine learning model is constructed based on multi-source data such as satellite observations and ground observations, integrating meteorological factors, pollutant concentration parameters, cloud attribute parameters, socioeconomic factors, and topographic factors to estimate daily-scale deposition; a cross-validation optimization model is used and bias correction is used to improve spatial generalization ability; the contribution of each factor is analyzed, and daily-scale wet deposition flux gridded data is output. The present invention adopts the above-mentioned inverting method for cloud / in-cloud clearance based on daily-scale nitrate wet deposition, and for the first time achieves daily-resolution wet deposition estimation, accurately captures the short-term deposition dynamics of heavy precipitation events in summer, and quantifies the contribution of cloud clearance effects to clean areas.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing atmospheric environment application, and in particular relates to a sub-cloud / in-cloud clearance inversion method based on daily-scale nitrate wet deposition. Background Art

[0002] Nitrate ( ) Wet deposition removes nitrogen compounds from the atmosphere through subcloud and intracloud removal processes, and is an important pathway for nitrogen input to terrestrial ecosystems. Excessive nitrogen deposition can lead to soil acidification, water eutrophication, and biodiversity loss, posing a serious threat to ecological security and human health. To address the negative impacts of nitrogen deposition, multiple acid deposition monitoring networks (such as EANET and NADP) have been established worldwide, but their data coverage is limited, making it difficult to characterize regional-scale heterogeneity. Although chemical transport models (CTMs) can provide spatially continuous simulations, their accuracy is significantly limited by the uncertainty of emission inventories and the simplification of chemical reaction mechanisms. Therefore, the development of high-precision, high-temporal and high-spatial resolution nitrogen deposition monitoring technology has become an urgent need for environmental management.

[0003] Satellite remote sensing technology provides a new solution for estimating nitrogen deposition by inverting the vertical column concentrations of atmospheric pollutants. Existing studies often construct regression models based on monthly satellite data, using the subcloud clearance process to establish a linear relationship between local pollutant concentrations and precipitation. However, such methods have significant drawbacks: First, monthly data cannot capture the dynamic coupling between short-term precipitation events and pollutant removal, resulting in the smoothing of extreme deposition events; second, the contribution of in-cloud clearance processes to long-range pollution transport is neglected, which can lead to systematic underestimation, especially in low-pollution areas; and third, traditional statistical models have difficulty in interpreting the nonlinear interactions between meteorological, geographical, and socioeconomic factors, limiting the models' interpretability and generalization capabilities.

[0004] Although some studies have introduced machine learning methods to improve modeling accuracy, existing research still faces two major bottlenecks: 1) Model input parameters often rely on local subcloud pollutant concentrations and fail to integrate air trajectory models to quantify the contribution of cloud-based removal pathways; 2) Training data lack spatial representation, with a high proportion of urban sites leading to biased predictions in remote areas. Furthermore, existing technologies generally lack daily-scale validation, making them inadequate for the dynamic assessments required by modern air quality models. Therefore, there is an urgent need to develop high-temporal and spatial resolution nitrogen deposition monitoring methods that integrate multi-source data and account for both subcloud and cloud-based removal mechanisms to provide technical support for the accurate assessment of regional nitrogen cycles and ecological risks. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for inverting the cloud / in-cloud clearance of nitrate wet deposition based on daily scale, so as to realize high-resolution and high-precision dynamic simulation of wet deposition on a national scale. It can solve the problem that the wet deposition data-driven model ignores the cloud clearance process, improve the temporal resolution of the estimation model, effectively improve the estimation accuracy of the model, and improve the reliability and interpretability of the wet deposition results, providing a scientific basis and theoretical support for nitrogen cycle assessment, pollution control and ecological risk management.

[0006] To achieve the above objectives, the present invention provides a method for inverting the cloud clearance of nitrate wet deposition on a daily scale, comprising the following steps:

[0007] Obtain and preprocess OMI NO2 column concentrations, MODIS cloud properties, meteorological data, socioeconomic data, and topographic data to obtain a raster dataset of input factors with uniform resolution;

[0008] Obtain NO3 from the ground monitoring network for each precipitation event - wet deposition data;

[0009] The HYSPLIT backward trajectory model was used to calculate the in-cloud clearance parameters, and the below-cloud clearance parameters were constructed by combining the OMI NO2 column concentration data. The pollutant concentration parameters were obtained by analyzing the in-cloud clearance parameters and the below-cloud clearance parameters.

[0010] The pollutant concentration parameters and cloud properties, meteorological data, socio-economic data, and terrain data are combined into a characteristic factor dataset. Build an optimal random forest estimation model based on wet deposition data;

[0011] The Z-score method is used to correct the system deviation and obtain the final optimal random forest estimation model;

[0012] Apply SHAP value to analyze factor contribution and output the contribution degree of each factor in different regions;

[0013] Input the input factor raster dataset to the final optimal random forest estimation model to estimate the daily scale Wet deposition grid data.

[0014] Preferably, the specific implementation method of the pretreatment of OMI NO2 column concentration is:

[0015] Use the cloud cover parameter as a mask to remove grid points with cloud cover greater than 30%;

[0016] The data of the current day is combined with the data of the previous two days to obtain three-day composite data, which is used as the pollutant concentration indicator when precipitation occurs;

[0017] The three-day composite data were interpolated using Kriging to fill in the missing grids.

[0018] Preferably, Wet deposition data were collected using the conventional acid deposition monitoring procedure HJ / T165-2004. After each precipitation event, the onset and end times of precipitation and the amount of precipitation were recorded and measured using ion chromatography. Ion concentration.

[0019] Preferably, the steps for obtaining the clearing parameters in the cloud are:

[0020] The initial height is set to 1000m and 1500m, the start time is set to the beginning of the precipitation event, and the duration parameter is set to -12h;

[0021] The following formula is used to calculate the NO2 concentration in the cloud, where the NO2 concentration in the cloud is the cloud removal parameter, and the calculation formula is:

[0022] ;

[0023] Where, represents the NO2 concentration in the cloud, N is the total number of backward trajectories, T is the time integration time length, is the coordinate of the i-th trajectory at time t, is the NO2 column concentration at the trajectory point.

[0024] Preferably, the pollutant concentration parameters and cloud properties, meteorological data, socio-economic data, and terrain data are combined into a characteristic factor data set. The optimal random forest estimation model for wet deposition data is constructed, which includes the following steps:

[0025] The random forest model was trained using the characteristic factor dataset and wet deposition ground monitoring data;

[0026] Sort by feature importance, remove the least important features, and retrain the random forest model;

[0027] Using R 2 and RMSE to evaluate the model accuracy. When RMSE increases or R 2 When it decreases, the feature combination and model evaluation score are output, otherwise the random forest model is retrained;

[0028] The hyperparameter matrix of the random forest model uses the network search method to optimize the hyperparameters to obtain the optimal hyperparameter combination.

[0029] The optimal feature combination among the feature combinations is screened, and the optimal feature combination and the optimal hyperparameter combination are used to construct the optimal random forest model.

[0030] Preferably, the deviation correction formula is:

[0031] ;

[0032] ;

[0033] ;

[0034] Where Y represents the wet deposition flux, represents the scale transformation coefficient, b represents the translation transformation coefficient, represents the expected value operator; represents the variance of the corresponding data set; Wet deposition flux estimated for the random forest model; is the observed wet deposition flux.

[0035] Preferably, the input factor grid dataset is input into the final optimal random forest estimation model to estimate the daily scale Wet deposition grid data, including

[0036] The precipitation data in the meteorological data was used as a mask to remove grids with precipitation less than or equal to 0.01 mm. Only the areas where precipitation occurred on that day were retained for wet deposition estimation, and finally the daily-scale wet deposition grid data of the study area was obtained.

[0037] Therefore, the present invention adopts the above-mentioned inversion method based on daily-scale nitrate wet deposition below / in-cloud clearance, and the technical effects are as follows:

[0038] Daily scale based on satellite remote sensing and machine learning The wet deposition estimation method solves the defects of previous monthly-scale models in smoothing extreme events and linear regression in ignoring nonlinear interactions, and can well capture the extreme deposition peaks on the daily scale.

[0039] Backward trajectories are used to quantify pollutant transport within clouds, integrating cloud removal and subcloud washout. While ensuring model interpretability, the accuracy of the model is maximized to obtain high-precision and generalizable daily-scale estimates of wet deposition dynamics. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention provides a flow chart of a method for inverting the sub-cloud / in-cloud clearance of nitrate wet deposition based on daily scale. DETAILED DESCRIPTION

[0041] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0042] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0043] Example 1

[0044] like Figure 1 As shown in FIG, a method for inverting the cloud clearance of nitrate wet deposition on a daily scale includes the following steps:

[0045] Step 1: Obtain and preprocess OMI NO2 column concentration, MODIS cloud properties, meteorological data, socioeconomic data, and topographic data to obtain a raster dataset of input factors with uniform resolution;

[0046] This study selected inland China as the study area. First, OMI NO2 vertical column concentration data from 2012 to 2016 were obtained using the TEMIS platform. Pixels with a cloud probability greater than 30% were masked to ensure data quality, resulting in a cloud-free time series dataset for the study area. Second, MODIS L3 cloud attribute data (including cloud optical thickness and cloud top temperature) with a spatial resolution of 0.25° × 0.25° were obtained using the NASA LAADS platform, and CHM_PRE daily precipitation data were obtained from the National Tibetan Plateau Data Center. CN05.1 meteorological data (daily average temperature, wind speed, relative humidity, etc.), WorldPop population density data (1 km resolution), CNLUCC land use data (1 km resolution), and DEM elevation data (SRTM 1 km) were obtained as auxiliary parameters.

[0047] Then, ARCGIS software was used to perform Kriging interpolation on the OMI NO2 data and MODIS cloud attribute data to fill in the gaps, and the data of the target day and the previous two days were synthesized to characterize the pollutant accumulation and cloud attribute parameters before precipitation.

[0048] Finally, all data were resampled to a 0.25° grid to ensure that the row and column numbers of the multi-source data were aligned, and a 279×159 grid image set was finally obtained.

[0049] Step 2: Obtain NO3 from the ground monitoring network for each precipitation event - wet deposition data;

[0050] First, the wetland rainfall data of single precipitation events from 2012 to 2016 were collected from 41 stations in China. Subsidence observation data, including precipitation start time, end time, rainfall amount, Concentration, calculated per session The wet deposition flux is as follows:

[0051] ;

[0052] Where, P i is the precipitation (mm), WN i (kg N·ha -1 ) represents the i-th session Wet deposition flux, [ ] i (mg / L) is the amount of water in the ith precipitation event The concentration of nitrogen is 14 / 62. The mass fraction conversion factor in the equation is 100, which converts the unit to kg N·ha. -1 Conversion factor.

[0053] Then, daily-scale wet deposition monitoring data from eight stations within the study area in 2015 were obtained through the EANET platform to form an independent verification dataset.

[0054] Step 3: Use the HYSPLIT backward trajectory model to calculate the cloud clearance parameters, and combine them with the OMI NO2 column concentration data to construct the cloud clearance parameters. The pollutant concentration parameters are obtained by analyzing the cloud clearance parameters and the cloud clearance parameters.

[0055] Combining satellite-derived NO2 concentrations with HYSPLIT trajectory simulations, we quantify the parameters of sub-cloud and intra-cloud removal processes. We also analyze air mass motion paths to extract pollutant transport characteristics before precipitation events.

[0056] First, the OMI NO2 tropospheric vertical column concentration is used to characterize the pollutant level in the mixed layer as a subcloud clearance parameter.

[0057] Secondly, we coupled the backward trajectory model with the NO2 tropospheric vertical column to construct cloud clearance parameters. We used the HYSPLIT model to simulate the backward trajectory of air masses 12 hours before precipitation, setting initial altitudes of 1000 m and 1500 m. We then calculated the trajectory path using GDAS meteorological data. We extracted OMI NO2 concentrations along the trajectory and calculated the trajectory average concentration using the following formula:

[0058] ;

[0059] Where, represents the NO2 concentration in the cloud, N is the total number of backward trajectories, T is the time integration time length, is the coordinate of the i-th trajectory at time t, is the NO2 column concentration at the trajectory point.

[0060] Step 4: Combine pollutant concentration parameters and cloud properties, meteorological data, socioeconomic data, and terrain data into a characteristic factor dataset. Build an optimal random forest estimation model based on wet deposition data;

[0061] A random forest model was constructed and optimized based on a sample dataset of single precipitation event observations from 41 stations, using factors such as pollutant concentrations, meteorological, and socioeconomic factors. The sample dataset was divided into training and test sets, and the model was constructed using the Random Forest Regressor in the Python-sklearn library. First, a backward variable elimination method was used to identify the optimal feature set from 14 meteorological parameters (precipitation, wind speed, and relative humidity), two pollutant concentration parameters (below-cloud NO2 concentration and mean intra-cloud NO2 trajectory concentration), two socioeconomic factors (population density and built-up land area), and a geographic parameter (DEM). With the goal of minimizing the RMSE and maximizing the R² in cross-validation, the least important features were iteratively removed, ultimately retaining 12 feature factors, including precipitation, built-up land area, DEM, cloud and below-cloud NO2 concentrations, wind speed, and relative humidity.

[0062] Secondly, the grid search method was used to tune the hyperparameters to optimize the maximum tree depth (max_depth=30), number of features (max_features=8), minimum number of split samples (min_samples_split=5) and number of trees (n_estimators=200) to build the optimal random forest model.

[0063] Step 5: Correct the system bias through the Z-score method to obtain the final optimal random forest estimation model;

[0064] Use the random forest estimation model obtained in step 4 to calculate the deviation between the model estimate and the true value, and eliminate the systematic deviation based on the Z-score linear transformation. The formula is as follows:

[0065] ;

[0066] ;

[0067] ;

[0068] Where Y represents the wet deposition flux, represents the scale transformation coefficient, b represents the translation transformation coefficient, represents the expected value operator; represents the variance of the corresponding data set; Wet deposition flux estimated for the random forest model; is the observed wet deposition flux.

[0069] The performance of the wet deposition estimation model was evaluated using the Pearson correlation coefficient (R) and root mean square error (RMSE). First, during model training, the dataset was divided into 10 subsets, and training and validation were performed iteratively to obtain ten-fold cross-validation results. Second, independent validation was performed using 2015 EANET station data to examine the model's generalization ability.

[0070] Step 6: Apply SHAP value to analyze the factor contribution and output the contribution of each factor in different regions;

[0071] SHAP (SHapley Additive exPlanations) values ​​are used to analyze the nonlinear contributions of each input variable. The Python shap library is used to calculate SHAP values ​​for all training samples and generate a feature contribution distribution plot.

[0072] Step 7: Input the input factor grid dataset into the final optimal random forest estimation model to estimate the daily scale Wet deposition grid data.

[0073] The trained model was applied to the 0.25° grid to achieve spatial upscaling and generate the 2015 China daily scale humidity data. The spatial distribution raster dataset of subsidence.

[0074] Bootstrap resampling was used to generate 200 sets of predictions, and the coefficient of variation (CV) was calculated to generate a spatial distribution map of uncertainty and identify areas of high uncertainty. The coefficient of variation formula is as follows:

[0075] ;

[0076] Where, Represents the standard deviation of 200 sets of prediction results, Indicates the average value.

[0077] The application conditions of the medium precipitation mask are determined by a strict threshold: if and only if the grid precipitation in the CHM_PRE daily precipitation data is ≥ 0.01 mm, the model prediction is executed and the precipitation flux value is output; grids with precipitation < 0.01 mm are marked as having no data.

[0078] Therefore, the present invention adopts the above-mentioned cloud-based / in-cloud clearance inversion method for daily-scale nitrate wet deposition, and for the first time realizes daily-resolution wet deposition estimation, accurately captures the short-term deposition dynamics of heavy rainfall events in summer, and quantifies the contribution of the in-cloud clearance effect to the clean area.

[0079] 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 the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for inverting the cloud clearance of nitrate wet deposition on a daily scale, characterized by: The following steps are involved: Obtain and preprocess OMI NO2 column concentrations, MODIS cloud properties, meteorological data, socioeconomic data, and topographic data to obtain a raster dataset of input factors with uniform resolution; Obtain NO3 from the ground monitoring network for each precipitation event - wet deposition data; The HYSPLIT backward trajectory model was used to calculate the in-cloud clearance parameters, and the below-cloud clearance parameters were constructed by combining the OMI NO2 column concentration data. The pollutant concentration parameters were obtained by analyzing the in-cloud clearance parameters and the below-cloud clearance parameters. The pollutant concentration parameters and cloud properties, meteorological data, socioeconomic data, and terrain data are combined into a characteristic factor dataset. - Build an optimal random forest estimation model based on wet deposition data; The Z-score method is used to correct the system deviation and obtain the final optimal random forest estimation model; Apply SHAP value to analyze factor contribution and output the contribution degree of each factor in different regions; Input the input factor raster dataset into the final optimal random forest estimation model to estimate daily NO3 - Wet deposition grid data.

2. The method for inversion of sub-cloud / in-cloud clearance of nitrate wet deposition based on daily time scale according to claim 1, characterized in that: The specific implementation method of pretreatment of OMI NO2 column concentration is: Use the cloud cover parameter as a mask to remove grid points with cloud cover greater than 30%; The data of the current day is combined with the data of the previous two days to obtain three-day composite data, which is used as the pollutant concentration indicator when precipitation occurs; The three-day composite data were interpolated using Kriging to fill in the missing grids.

3. The inversion method for in-cloud / under-cloud clearance of nitrate wet deposition based on daily time scale according to claim 1, characterized in that: NO3 - Wet deposition data were collected using the conventional acid deposition monitoring procedure HJ / T165-2004. After each precipitation event, the onset and end time of precipitation and the amount of rainfall were recorded, and NO3 was measured using ion chromatography. - Ion concentration.

4. The method for inversion of sub-cloud / in-cloud clearance of nitrate wet deposition based on daily time scale according to claim 1, characterized in that: The steps to obtain the clearing parameters in the cloud are: The initial height is set to 1000m and 1500m, the start time is set to the beginning of the precipitation event, and the duration parameter is set to -12h; The following formula is used to calculate the NO2 concentration in the cloud, where the NO2 concentration in the cloud is the cloud removal parameter, and the calculation formula is: ； Where, represents the NO2 concentration in the cloud, N is the total number of backward trajectories, T is the time integration time length, is the coordinate of the i-th trajectory at time t, is the NO2 column concentration at the trajectory point.

5. The method for inversion of sub-cloud / in-cloud clearance of nitrate wet deposition based on daily time scale according to claim 1, characterized in that: The pollutant concentration parameters and cloud properties, meteorological data, socioeconomic data, and terrain data are combined into a characteristic factor dataset. - The optimal random forest estimation model for wet deposition data is constructed, which includes the following steps: The random forest model was trained using the characteristic factor dataset and wet deposition ground monitoring data; Sort by feature importance, remove the least important features, and retrain the random forest model; Using R 2 and RMSE to evaluate the model accuracy. When RMSE increases or R 2 When it decreases, the feature combination and model evaluation score are output, otherwise the random forest model is retrained; The hyperparameter matrix of the random forest model uses the network search method to optimize the hyperparameters to obtain the optimal hyperparameter combination. The optimal feature combination among the feature combinations is screened, and the optimal feature combination and the optimal hyperparameter combination are used to construct the optimal random forest model.

6. The method for inversion of sub-cloud / in-cloud clearance of nitrate wet deposition based on daily time scale according to claim 1, characterized in that: The bias correction formula is: ； ； ； Where Y represents the wet deposition flux, represents the scale transformation coefficient, b represents the translation transformation coefficient, represents the expected value operator; represents the variance of the corresponding data set; Wet deposition flux estimated for the random forest model; is the observed wet deposition flux.

7. The method for inverting cloud clearance based on daily-scale nitrate wet deposition in the cloud / in the cloud according to claim 1, characterized in that: Input the input factor raster dataset into the final optimal random forest estimation model to estimate daily NO3 - Wet deposition grid data, including The precipitation data in the meteorological data was used as a mask to remove grids with precipitation less than or equal to 0.01 mm. Only the areas where precipitation occurred on that day were retained for wet deposition estimation, and finally the daily-scale wet deposition grid data of the study area was obtained.

Images