CN-121765355-B - Land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method

CN121765355BCN 121765355 BCN121765355 BCN 121765355BCN-121765355-B

Abstract

The invention belongs to the technical field of environmental data processing and pollution monitoring, and discloses a land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method; the method comprises the steps of S1, collecting and rasterizing social, economic and environmental data, S2, collecting an atmospheric nitrogen data set, integrating three types of data and matching with atmospheric nitrogen sedimentation space positions, S3, estimating land atmospheric nitrogen sedimentation data by using a CNN-LSTM model, S4, estimating marine artificial source atmospheric nitrogen sedimentation data by using a CNN-LSTM model, S5, estimating marine natural source atmospheric nitrogen sedimentation data by using an LSTM model, S6, integrating results by an inversion weighting method, S7, constructing a bay atmospheric nitrogen sedimentation training data set, S8, carrying out three-dimensional verification based on monitoring station actual measurement data, S9, carrying out lamination treatment on land and marine models, and outputting bay atmospheric nitrogen sedimentation data after training the model.

Inventors

LI SHAOBIN
Chen Jiangru
CHEN NENGWANG
Huang Zhehan
LI WEIYE
WU SHUIPING

Assignees

厦门大学

Dates

Publication Date: 20260508
Application Date: 20260302

Claims (7)

1. A land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method is characterized by comprising the following steps of: s1, collecting social data, economic data and environmental data of a bay area, and performing rasterization treatment; S2, collecting an atmospheric nitrogen data set, integrating social data, economic data and environmental data, and matching with the atmospheric nitrogen sedimentation space position; S3, inputting 15-dimensional characteristics, estimating land atmospheric nitrogen sedimentation data by using a CNN-LSTM neural network model, and outputting land nitrogen sedimentation flux; s4, inputting 8-dimensional characteristics, estimating marine artificial source atmospheric nitrogen sedimentation data by using a CNN-LSTM neural network model, and outputting marine artificial source nitrogen sedimentation flux; S5, inputting 6-dimensional characteristics, estimating the atmospheric nitrogen sedimentation data of the ocean natural source by using a long-term memory network, and outputting the ocean natural source nitrogen sedimentation flux; S6, determining the weight of the marine artificial source nitrogen sedimentation flux by an inversion weight method, and obtaining the marine total nitrogen sedimentation flux by weighted summation calculation; in step S6, the calculation formula of the ocean total nitrogen sedimentation flux is: , wherein, Is ocean total nitrogen sedimentation flux; sedimentation flux for marine artificial source nitrogen; sedimentation flux for ocean natural source nitrogen; Weighting sedimentation flux of marine artificial source nitrogen; S7, constructing a bay atmospheric nitrogen sedimentation training data set according to sea Liu Zhanbi weighting in each grid of the bay by using the estimated land atmospheric nitrogen sedimentation data, the sea artificial source atmospheric nitrogen sedimentation data and the sea natural source atmospheric nitrogen sedimentation data; the specific process of step S7 includes: S71, weighting based on sea Liu Zhanbi to obtain a basic value of atmospheric nitrogen sedimentation of the bay grid, wherein the calculation formula is as follows: , wherein, A basic value of atmospheric nitrogen sedimentation for the bay grid; Sedimentation flux for terrestrial nitrogen; Is ocean total nitrogen sedimentation flux; The ratio of land area in the grid to the total area of the grid; s72, multiplying the basic value of the atmospheric nitrogen sedimentation of the bay grid with the interactive adjustment factor to obtain the final value of the atmospheric nitrogen sedimentation of the bay grid, wherein the calculation formula is as follows: , wherein, Final atmospheric nitrogen sedimentation value for bay grids; in order to interact with the adjustment factor, ∈[0.8,1.2]; S73, arranging data according to a year-month-grid ID-input feature set-training label format, and outputting a land use vector data attribute table in a CSV format and grids in a GeoTIFF format for training a laminated model; S8, adopting actual measurement data of 10-20 monitoring stations along the coast of the bay and in the sea as real values, and adopting precision, space and stability to verify the reliability of a bay atmospheric nitrogen sedimentation training data set; S9, constructing a bay nitrogen sedimentation fusion estimation model, training by using a bay atmospheric nitrogen sedimentation training data set, and carrying out bay atmospheric nitrogen sedimentation space-time dynamic estimation by using the trained bay nitrogen sedimentation fusion estimation model.
2. The method for estimating the air nitrogen settlement space-time dynamics of the gulf of land-sea cooperation according to claim 1, wherein in the step S1, the social data comprises population, domestic total production value and automobile conservation amount, the economic data comprises nitrogen fertilizer application amount, cultivation scale, industrial wastewater discharge amount related to air pollution, industrial waste gas discharge amount related to air pollution, industrial quantity related to air pollution, ship voyage mileage, tourist number and oil and gas exploitation yield, and the environmental data comprises wind speed, wind direction, precipitation amount, temperature, sea-land frequency, altitude, gradient, sea water temperature, salinity, chlorophyll a concentration, solar radiation, sea-air temperature difference, ocean current intensity and distance from coastline.
3. The method for estimating the nitrogen sedimentation space-time dynamics of the sea-land cooperation bay as claimed in claim 1, wherein in the step S1, the specific process of the rasterization treatment comprises: S11, grouping according to index types, year and month, calculating the mean value and standard deviation of each group of data, and eliminating abnormal data; s12, uniformly converting meteorological sites, land utilization and administrative district data into a WGS84 coordinate system by using ArcGIS software; s13, creating a fishing net tool by using ArcGIS software, generating a1 km multiplied by 1 km regular grid, and outputting the grid in a format of GeoTIFF; S14, setting basic weights of land types and storing the basic weights into a land utilization vector data attribute table, wherein the basic weights of the land types are set to be commercial land=1.0, industrial land=0.8, residential land=0.6, agricultural land=0.2 and woodland=0.05 by using a cutting tool of ArcGIS software to cut land utilization vector data according to the boundary of a research area and reserve land type plaques in a target area; S15, performing space superposition on the cut land utilization vector data and 1 km multiplied by 1 km regular grids by using an intersecting tool of ArcGIS software to obtain grid-land type associated plaques, calculating the area of each grid-land type associated plaque by using a calculation geometry tool of ArcGIS software, and calculating the proportion of each type of land area to the total area of the selected grids for each grid, wherein the calculation formula is as follows: , wherein, The proportion of the area of each type of land to the total area of the selected grid; Correlating areas of plaque for each grid-land type; For each grid total area, and ; S16, calculating the initial weight of the grids, namely calculating the comprehensive weight of each grid, wherein the comprehensive weight is used for representing the weighted sum of the land occupation ratio of each type and the corresponding weight, and the calculation formula is as follows: , wherein, A comprehensive weight for each grid; base weights for the type of land; And (3) total distribution of the counties, namely counting the sum of comprehensive weights of all grids in the selected counties according to the county boundaries, and calculating an initial index value of each grid, wherein the calculation formula is as follows: , , wherein, The sum of the comprehensive weights of all grids in the selected county is used; An initial index value for each grid; the total amount of the statistical indexes of the selected county is calculated; S17, calculating the total amount deviation of initial distribution values of all grids in the selected county, wherein the calculation formula is as follows: , wherein, A total amount deviation of initial allocation values for all grids in the selected county; Iterative correction, if Dev is more than 3%, repeatedly iterative adjustment of the grid value until Dev is less than or equal to 3%, obtaining the final distribution value of the grid The iterative calculation formula is: , wherein, Assigning a value to the grid of the t+1st iteration; assigning a value to the grid of the t-th iteration, wherein lambda is an adjustment coefficient and lambda=0.08; assigning an average value of the values to the grids within the selected county; for maximum function; the minimum function is obtained; s18, using a Kriging interpolation tool of ArcGIS software to interpolate the environmental data into 1 km multiplied by 1 km grids, converting the interpolation result into GeoTIFF format, and associating with the grid IDs of the social data grids and the economic data grids; And S19, integrating data, namely merging the grid data of society, economy and environment by Python Pandas software according to the index of grid ID-year-month to form the social and economic environment grid data of grid ID-index name-index value-space-time information.
4. The method for estimating the nitrogen sedimentation space-time dynamics of the sea-land cooperation bay as claimed in claim 1, wherein the specific process of step S2 comprises: S21, unifying grid references and index rules, namely constructing all data based on 1 km multiplied by 1 km grids, and adopting grid IDs to uniquely identify space positions, wherein the time dimension is uniformed by adopting a year-month timestamp, and finally forming a three-dimensional index of the year-month-grid IDs, so that the space-time references are kept consistent; s22, loading social and economic environment grid data and atmospheric nitrogen sedimentation grid data which all comprise grid IDs and corresponding longitude and latitude coordinates into the same space reference system through an XY data adding tool of ArcGIS software, wherein the space reference system adopts a WGS84 coordinate system; S23, a space connection tool of ArcGIS software is adopted, an atmospheric nitrogen sedimentation grid is used as a target element, a socioeconomic environment grid is used as a connection element, a complete matching rule is selected, and environment, socioeconomic fields are automatically related to a nitrogen sedimentation grid attribute table; s24, writing batch processing scripts for month time sequence data by using Python geopandas library, and one-to-one associating the batch processing scripts with the grid ID fields, wherein the specific process comprises the steps of reading an atmospheric nitrogen sedimentation grid and a socioeconomic environment grid, executing a merge function based on the grid ID fields, and outputting an associated data table; s25, screening associated data based on year-month time stamps, and eliminating invalid matching of cross-time stamps, eliminating repeated matching items through a drop_ duplicates function of a Python pandas library, and reserving unique records of year-month-grid IDs; s26, verifying matching quality, namely randomly extracting 10% of atmospheric nitrogen sedimentation grids and socioeconomic environment grids, manually checking consistency of longitude and latitude coordinates and grid IDs, counting matching success rate, and supplementing and correlating unmatched grids by adopting nearest neighbor interpolation for fault occurrence of no data.
5. The method for estimating the nitrogen sedimentation space-time dynamics of the sea-land cooperation bay as claimed in claim 1, wherein the specific process of step S3 comprises: S31, extracting land research area data from the data processed in the step S2, and dividing a training set and a verification set according to a 7:3 time sequence, wherein 12 months are taken as a time window for capturing seasonal features, and a sample pair of continuous 12 month features and 12 month nitrogen sedimentation flux is constructed to form an input feature tensor and a label tensor; S32, adopting a three-stage architecture of space feature extraction-time feature capture-output mapping, receiving dimension tensors of batch size multiplied by time window multiplied by feature number at an input layer, extracting space-related features by 2 layers of convolution at a CNN space extraction layer, adopting a ReLU as an activation function, capturing a time sequence trend in a bidirectional LSTM at an LSTM time capture layer, outputting a bidirectional feature splicing result, and linearly activating output nitrogen sedimentation flux through fitting inhibition at a fully connected output layer; S33, selecting Adam as an optimizer, setting an initial learning rate of 0.001 and a batch size of 32, adopting a mean square error adaptation regression task as a loss function, setting an early-stop strategy, and simultaneously starting L2 regularization; S34, constructing a data loader, and iteratively training for 100 rounds, calculating loss by using a verification set after each round of training, and monitoring a training curve in real time; S35, loading an optimal model to predict the verification set, calculating average absolute error, root mean square error and determining coefficient, judging that the model is qualified when the average absolute error is less than or equal to 5, the root mean square error is less than or equal to 8 and the determining coefficient is more than or equal to 0.8, outputting land nitrogen sedimentation flux of 1 km multiplied by 1 km grid month scale, and returning to adjust the model framework or parameters when the average absolute error is not more than or equal to 5, and retraining.
6. A method for estimating the air nitrogen sedimentation space time dynamics of a bay in which land and sea are cooperated according to claim 1, wherein in the step S3, the input 15-dimensional characteristics include population, domestic production total value, car reserve, nitrogen fertilizer application amount, cultivation scale, industrial wastewater discharge amount related to air pollution, industrial exhaust gas discharge amount related to air pollution, industrial quantity related to air pollution, wind speed, wind direction, precipitation amount, temperature, sea and land wind frequency, altitude and gradient, in the step S4, the input 8-dimensional characteristics include ship voyage mileage, tourist number, oil and gas exploitation yield, wind speed, wind direction, ocean current intensity and distance from coastline, and in the step S5, the input 6-dimensional characteristics include sea water temperature, salinity, chlorophyll a concentration, solar radiation, sea-air temperature difference and wind speed.
7. The method for estimating the spatial-temporal dynamics of the nitrogen sedimentation of the sea-land cooperation bay as claimed in claim 1, wherein in the step S9, the construction process of the estimation model of the nitrogen sedimentation fusion of the bay is that a three-level structure of bottom layer embedding-middle layer fusion-top layer output is adopted, in the bottom layer, CNN-LSTM neural network models pre-trained in the step S3 and the step S4 are embedded, core parameters of CNN convolution layers and LSTM layers are frozen, only 64-dimensional feature layers before full-connection output are unfrozen, in the middle layer, a transducer interaction layer has 8-head attention, a hidden layer dimension of 256 and an activation function of GELU, multidimensional integration features are input, and land-sea interaction effects are captured, and in the top layer, the full-connection layer outputs the nitrogen sedimentation flux of the sea.

Description

Land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method Technical Field The invention belongs to the technical field of environmental data processing and pollution monitoring, and particularly relates to a land-sea cooperative bay atmospheric nitrogen settlement space-time dynamic estimation method. Background The method has the advantages that the method is mainly used for solving the ecological problems of atmospheric nitrogen settlement research focusing on the eutrophication of a bay and the like, and is currently used for constructing a technical basis of monitoring-simulation-source analysis by depending on numerical models of ground stations, satellite remote sensing and the like, but two major core challenges are faced, namely, the research is more limited to small-range areas such as a single bay (such as a Jiaozhou bay) and the like, the large-range research across the bay is difficult to advance due to non-uniform observation standards and insufficient model suitability, and social and economic indexes such as GDP, automobile conservation quantity and the like are quantitatively fitted with nitrogen settlement flux by few people, and the human activity driving mechanism is difficult to quantify due to the lack of multi-dimensional database support. Therefore, how to effectively integrate the socioeconomic and meteorological multisource data, not only breaks the bottleneck of non-uniform observation standards and insufficient model suitability in large-scale research across gulf, but also fills the gap of multi-dimensional database support, realizes quantitative fitting of socioeconomic indexes and nitrogen sedimentation flux, and finally accurately estimates atmospheric nitrogen sedimentation data, provides technical support for quantifying human activity driving mechanisms, and becomes a problem to be solved urgently in the current environmental data processing field. Disclosure of Invention In order to solve the problems, the invention provides a land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method, which can solve the bottleneck problems of low precision, insufficient space-time resolution and low calculation efficiency of the traditional atmospheric nitrogen sedimentation estimation method (such as single ground station interpolation and a numerical model without fusion of land Taiwan Strait Exchange Association and data), obviously improves the simulation speed on the premise of keeping high precision and high space-time resolution, efficiently generates a long-time-sequence and high-resolution hydraulic field, provides reliable and efficient power input for a bay water quality model, an ecological model and the like, and has good popularization and application prospects. In order to achieve the above purpose, the present invention adopts the following technical scheme: a land-sea cooperative bay atmospheric nitrogen sedimentation space-time dynamic estimation method comprises the following steps: s1, collecting social data, economic data and environmental data of a bay area, and performing rasterization treatment; S2, collecting an atmospheric nitrogen data set, integrating social data, economic data and environmental data, and matching with the atmospheric nitrogen sedimentation space position; S3, inputting 15-dimensional characteristics, estimating land atmospheric nitrogen sedimentation data by using a CNN-LSTM neural network model, and outputting land nitrogen sedimentation flux; s4, inputting 8-dimensional characteristics, estimating marine artificial source atmospheric nitrogen sedimentation data by using a CNN-LSTM neural network model, and outputting marine artificial source nitrogen sedimentation flux; S5, inputting 6-dimensional characteristics, estimating the atmospheric nitrogen sedimentation data of the ocean natural source by using a long-term memory network, and outputting the ocean natural source nitrogen sedimentation flux; S6, determining the weight of the marine artificial source nitrogen sedimentation flux by an inversion weight method, and obtaining the marine total nitrogen sedimentation flux by weighted summation calculation; S7, constructing a bay atmospheric nitrogen sedimentation training data set according to sea Liu Zhanbi weighting in each grid of the bay by using the estimated land atmospheric nitrogen sedimentation data, the sea artificial source atmospheric nitrogen sedimentation data and the sea natural source atmospheric nitrogen sedimentation data; S8, adopting actual measurement data of 10-20 monitoring stations along the coast of the bay and in the sea as real values, and adopting precision, space and stability to verify the reliability of a bay atmospheric nitrogen sedimentation training data set; S9, constructing a bay nitrogen sedimentation fusion estimation model, training by using a bay atmospheric nitrogen sedimentation training data set, and carrying out bay atmo