CN-121983163-A - Atmospheric NOy component semi-quantitative data set construction method based on multi-source data fusion

CN121983163ACN 121983163 ACN121983163 ACN 121983163ACN-121983163-A

Abstract

The invention relates to the technical field of atmospheric environment monitoring and multidimensional data processing, and discloses a construction method of an atmospheric NOy component semi-quantitative data set based on multi-source data fusion. And secondly, constructing a priori chemical manifold based on a box model, and performing topology reconstruction through differential homoembryo mapping to obtain a posterior chemical manifold adapting to the real atmosphere. Thirdly, calculating the local texture entropy and manifold chemical entropy of the image, and establishing heterogeneous feature alignment constraint. And finally, constructing a four-dimensional tensor containing space-time and component dimensions, fusing optical flow fields, chemical manifold and alignment constraint to construct a multi-constraint objective function, and executing tensor complementation. The invention effectively solves the problem of low space coverage rate of the existing monitoring means, and generates the NOy component data set with high space-time resolution and physical-chemical consistency.

Inventors

LU WEIQING
MAO JINGJING
YANG XUE
YU JINHAI
XU ZHENG
MA QIANYA

Assignees

江苏省环境监测中心

Dates

Publication Date: 20260505
Application Date: 20260122

Claims (10)

1. The construction method of the atmospheric NOy component semi-quantitative data set based on multi-source data fusion is characterized by comprising the following steps: S100, acquiring a geostationary satellite remote sensing image sequence and ground monitoring site observation data, constructing a variable optical flow model which introduces chemical reaction rate as a source and sink item, and solving a dense optical flow field of atmospheric smoke plume; s200, constructing a low-dimensional priori chemical manifold based on box model simulation data, and performing topology reconstruction on the priori chemical manifold by using the observation data as an anchor point through differential homoembryo mapping to obtain a posterior chemical manifold adapting to a real atmospheric environment; s300, calculating local texture entropy of the satellite remote sensing image sequence and chemical entropy on the posterior chemical manifold, and establishing heterogeneous feature alignment constraint in which the image texture gradient direction is consistent with the chemical evolution direction; s400, constructing a four-dimensional tensor comprising a space-time dimension and a chemical component dimension, fusing a dense optical flow field of the atmospheric smoke plume, the posterior chemical manifold and the heterogeneous characteristic alignment constraint to construct an objective function, and executing tensor complement solving to generate an atmospheric NOy component semi-quantitative data set.
2. The method for constructing an atmospheric NOy component semi-quantitative data set based on multi-source data fusion according to claim 1, wherein in step S100, constructing the variational optical flow model includes constructing a chemical source sink corrected optical flow continuity equation defined as: The partial derivative of the image intensity with respect to time, together with the dot product of the gradient vector of the image intensity with respect to the spatial coordinates and the two-dimensional velocity vector field to be solved, is equal to the net rate function of the chemical reaction determined by the chemical state parameters; In step S100, solving the dense optical flow field of the atmospheric plume includes: constructing a global total energy functional comprising a data fidelity term and a smoothing regularization term, wherein the data fidelity term comprises a residual error defined by the chemical source sink correction optical flow continuity equation; and minimizing the global total energy functional by adopting a multi-scale pyramid iteration strategy to obtain a dense optical flow field of the atmospheric smoke plume.
3. The method for constructing a semi-quantitative data set of atmospheric NOy component based on multi-source data fusion according to claim 1, wherein in step S200, the constructing a low-dimensional a priori chemical manifold based on the simulation data of the box model comprises: Setting an initial parameter vector space covering a meteorological parameter dimension and a precursor concentration dimension, running a 0-D box model to simulate the evolution process of photochemical reaction along with time, and outputting a high-dimensional simulation data set containing chemical species concentration change; And mapping the high-dimensional simulation data set to a low-dimensional embedded space by adopting a nonlinear manifold learning algorithm based on the principle of keeping the geodesic distance between sample points unchanged, and generating the priori chemical manifold.
4. The method for constructing a semi-quantitative data set of atmospheric NOy components based on multi-source data fusion according to claim 1, wherein in step S200, the topology reconstruction of the a priori chemical manifold by differential homoembryo mapping comprises: Defining a differential homoembryo mapping function which acts on the low-dimensional manifold space and is micro and reversible; Constructing a manifold topology reconstruction optimization objective function, the manifold topology reconstruction optimization objective function comprising: euclidean norms for quantifying fitting errors between reconstructed data and ground observation data, and jacobian regularization terms for controlling manifold deformation stiffness; And solving the manifold topology reconstruction optimization objective function, determining the optimal differential stratospheric mapping function, and applying the differential stratospheric mapping function to the prior chemical manifold to generate the posterior chemical manifold.
5. The method for constructing a semi-quantitative data set of atmospheric NOy components based on multi-source data fusion according to claim 1, wherein in step S300, the local texture entropy of the satellite remote sensing image sequence is calculated by a texture entropy calculation formula, the texture entropy calculation formula comprises: defining a local space neighborhood window for each pixel position in the satellite remote sensing image sequence, and counting the occurrence frequency of each gray level in the local space neighborhood window for constructing a local gray probability distribution histogram; and calculating local texture entropy based on an information theory principle, wherein the local texture entropy is a negative value of a summation result of a product of the statistical probability of each gray level occurrence and the logarithm of the statistical probability in the discretized gray level set.
6. The method for constructing a semi-quantitative data set of atmospheric NOy components based on multi-source data fusion according to claim 1, wherein in step S300, the chemical entropy on the posterior chemical manifold comprises: defining a chemical reaction path on the posterior chemical manifold, mapping the state evolution on the chemical reaction path into a marked chemical entropy index, wherein the chemical entropy index is based on a second law of thermodynamics or a Gibbs free energy minimization principle and is defined as a state function of manifold coordinates on the posterior chemical manifold, and the specific numerical value of the chemical entropy index is characterized by a negative value of the Gibbs free energy variation of a reaction system relative to the equilibrium state or by component distribution entropy.
7. The method for constructing an atmospheric NOy component semi-quantitative dataset based on multi-source data fusion according to claim 1, wherein in step S300, establishing the heterogeneous feature alignment constraint comprises constructing a heterogeneous feature alignment loss function comprising a direction alignment term and a flow field consistency regularization term; the direction alignment item comprises a cosine similarity penalty function for punishing the direction deviation between the spatial gradient vector of the local texture entropy and the spatial gradient vector of the chemical entropy corresponding to the four-dimensional tensor; The flow field consistency regularization term is used for forcing the chemical entropy to be nonnegative along the directional derivative of the dense optical flow field of the atmospheric smoke plume.
8. The method for constructing an atmospheric NOy component semi-quantitative data set based on multi-source data fusion according to claim 1, wherein in step S400, constructing the four-dimensional tensor specifically comprises: Defining the latitude of the four-dimensional tensor as longitude, latitude, time and chemical composition; and filling the four-dimensional tensor with the nitrogen dioxide column concentration in the satellite remote sensing image sequence and the total reactive nitrogen oxide concentration in the ground monitoring site observation data as known elements, and taking unobserved space-time points and components as elements to be complemented to form a highly sparse observation tensor.
9. The method for constructing a semi-quantitative data set of atmospheric NOy components based on multi-source data fusion according to claim 1, wherein in step S400, the objective function comprises: a data fidelity term for calculating a data fidelity residual; tensor kernel norm low rank constraint term for exploiting strong correlation of atmospheric chemical field over spatio-temporal distribution; an optical flow track constraint term constructed based on the dense optical flow field of the atmospheric smoke plume is used for forcing the space-time evolution of each component in the four-dimensional tensor to follow the physical transmission track of an air mass; A heterogeneous feature alignment loss function corresponding to the heterogeneous feature alignment constraint; and the manifold constraint term is used for forcing the component vector corresponding to each space-time point in the four-dimensional tensor to be positioned in the neighborhood of the posterior chemical manifold after being projected to a low-dimensional space.
10. The method of constructing an atmospheric NOy component semi-quantitative data set based on multi-source data fusion according to claim 9, wherein in step S400, the performing tensor complement solution to generate the atmospheric NOy component semi-quantitative data set comprises: iteratively solving the objective function by adopting a Riemann manifold optimization algorithm, and mapping an update step length on a tangent space to a manifold surface by utilizing a contraction operator until convergence to obtain an optimal complement tensor; extracting the concentration of each chemical component from the optimal completion tensor, and calculating the proportion coefficient of the chemical component to the total reactive nitrogen oxide concentration by utilizing a proportion formula; in combination with the calibrated total reactive nox total amount profile and the scaling factor, an atmospheric NOy component semi-quantitative dataset is generated comprising high spatial-temporal resolution.

Description

Atmospheric NOy component semi-quantitative data set construction method based on multi-source data fusion Technical Field The invention relates to the technical field of atmospheric environment monitoring and multidimensional data processing, in particular to an atmospheric NOy component semi-quantitative data set construction method based on multi-source data fusion. Background Reactive nitrogen oxides (NOy) in the atmosphere are key precursors for ozone and secondary organic aerosol formation, and the components are extremely complex and mainly comprise Nitric Oxide (NO), nitrogen dioxide (NO 2), nitric acid (HNO 3), peroxyacetyl nitrates (PANs) and the like. The accurate grasp of the fine distribution of each component of NOy in space-time has important significance for the atmospheric oxidizing analysis and the prevention and control of regional composite pollution. Currently, atmospheric environmental monitoring mainly relies on geostationary satellite remote sensing and ground air quality monitoring sites. Satellite remote sensing can provide nitrogen dioxide column concentration data with large range and high frequency, has good space coverage advantage, and ground sites can provide high-precision multicomponent concentration observation, but the sites are sparse and uneven in distribution. In order to obtain a full-area continuous distribution of contaminant components, the prior art generally attempts to combine the macroscopic coverage capability of satellites with the microscopic accuracy of observation of ground sites for data fusion or inversion. However, conventional methods have several limitations in dealing with the transport and evolution of contaminants involving complex photochemical reactions. First, in terms of atmospheric flow field tracking and transmission simulation, conventional computer vision optical flow algorithms are typically based on the assumption that the intensity of pixels of the same object between successive frames is considered to remain unchanged. However, nitrogen oxides in the atmosphere undergo rapid and complex diurnal photochemical reactions and nocturnal heterogeneous reactions during transport, resulting in non-conservative changes in contaminant concentration (i.e., image intensity). In the prior art, it is difficult to distinguish the concentration source-sink change caused by chemical reaction from the advection transmission caused by physical wind field, and the direct application of the traditional method can cause significant deviation in transmission track calculation, so that the real motion path of the air mass cannot be accurately reflected. Secondly, in the aspect of chemical component inference, although the mechanism-based atmospheric chemical transmission model can simulate the evolution process of the whole components, the model has extremely high dependence on the precision of an emission list and the initial boundary condition, the calculation load is huge, and the simulation result often has systematic deviation from the real atmospheric environment. The simple data-driven interpolation or regression method, though being close to the observed data, often ignores inherent nonlinear dynamics constraint among chemical components, so that inversion results lack physicochemical interpretability in the area lacking sites, and the rationality of component proportions is difficult to ensure. In addition, existing multi-source data fusion schemes mostly treat the visual features of satellite images and the chemical state of the atmosphere as independent variables. In fact, macroscopic morphological features of the atmospheric plume (such as diffusion texture of the edges, smoothness of the plume) are intrinsically kinetically linked to the chemical aging stage in which they are subjected. The prior art lacks an effective mechanism to correlate the macroscopic image texture gradient with the microscopic chemical evolution direction, so that the inversion of chemical components cannot be restrained by utilizing high-resolution satellite image texture information in the area lacking ground anchor points, and the spatial resolution and the accuracy of data set construction are limited. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a construction method of an atmospheric NOy component semi-quantitative data set based on multi-source data fusion, which aims to solve the problems that the prior optical flow algorithm lacks real environment suitability and chemical dynamics constraint in the inversion process due to the fact that the atmospheric transmission track tracking deviation, a simple mechanism model or a data driving method are caused by neglecting chemical reaction source and sink changes, and the data spatial resolution and precision are limited due to the fact that macroscopic satellite image texture features are difficult to effectively restrict microscopic chemical component inference. The invention aim