CN-122024848-A - Dynamic monitoring method for biodiversity

CN122024848ACN 122024848 ACN122024848 ACN 122024848ACN-122024848-A

Abstract

The invention provides a dynamic monitoring method of biodiversity, which relates to the technical field of ecological environment remote sensing monitoring, and comprises the steps of firstly acquiring core indexes based on a remote sensing biodiversity key variable framework, and introducing soil and situation variables; the method comprises the steps of constructing module reliability gate control to evaluate the reliability of each index module, constructing a situation self-adaptive representation model by using situation variable guidance, enabling index characteristics to adaptively change along with ecological background by using a cross attention mechanism, constructing a graph Laplacian by using a multi-relation space graph to carry out space consistency constraint on the characteristics, and finally automatically calculating weights between groups and in groups by using a double-layer self-adaptive weighting mechanism to generate a biodiversity index (SABI). The method realizes full-automatic data-driven generation of the weights, effectively improves the robustness and accuracy of the biodiversity monitoring under complex space-time situations, and provides scientific basis for regional ecological protection decisions.

Inventors

FU PINGJIE
Bu Yuankun
MENG FEI
WANG YUQIANG
ZHANG TENG
WANG HUIMENG
LI JINYU
WANG QI
WANG JIN

Assignees

山东建筑大学

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (8)

1. A method for dynamically monitoring biodiversity, comprising: Acquiring multi-source remote sensing data of a target area, and extracting core index data respectively representing species dimension, ecosystem dimension and landscape dimension from the multi-source remote sensing data based on a remote sensing biodiversity key variable framework; acquiring situation variable data of the target area, wherein the situation variable data comprises a soil factor and an environment quality factor; Resampling and spatially registering the core index data and the situation variable data based on a unified anchor grid, and carrying out normalization processing to obtain standardized index data and standardized situation data respectively; dividing the standardized index data into three modules of species, ecosystem and landscape based on the functional characteristics of the index, and respectively calculating an intra-group consistency factor, an inter-year stability factor and a data quality factor for each module by using the standardized index data in the module; Fusing the consistency factor, the annual stability factor and the data quality factor to obtain a pixel-level module reliability gating score; Mapping the standardized index data into an initial feature representation, and modulating the initial feature representation by using the standardized situation data as guiding information according to a cross attention mechanism to obtain a pixel-level multidimensional feature representation fused with situation information; constructing a multi-relation space diagram based on the geographic space adjacent relation, the physical partition adjacent relation and the ecological feature similarity relation of the target area; Obtaining a graph Laplace operator according to the multi-relation space graph, and carrying out space smoothing and consistency constraint on the pixel-level multi-dimensional feature representation by utilizing the graph Laplace operator to obtain a constrained feature representation; Based on the constrained feature characterization, the pixel level module reliability gating score and the standardized situation data, respectively calculating inter-group weights and intra-group index weights through a double-layer self-adaptive weighting mechanism; and carrying out weighted summation on the standardized index data based on the inter-group weights and the intra-group index weights to obtain the biodiversity index of the target region.
2. The method of claim 1, wherein the core index data of the species dimension comprises net primary productivity, vegetation coverage and dynamic habitat index, the core index data of the ecosystem dimension comprises precipitation, surface temperature, potential evapotranspiration, soil moisture and soil erosion, and the core index data of the landscape dimension comprises plaque shape index, shannon diversity index, simpson index and plaque density.
3. The method for dynamically monitoring biodiversity according to claim 1, wherein the soil factors comprise soil organic matters and soil textures, and the environmental factors comprise ecological environmental quality and dryness index.
4. The dynamic biodiversity monitoring method of claim 1, wherein the expression of the pixel-level module reliability gating score is: ; ; Wherein, the Representing the consistency of the group within the group, Representing the stability of the years, Representing the quality of the data and, 、、 Represents the weight coefficient of the object, Representing the composite score of the individual pieces, A temperature parameter representing the softmax of the sample, Representing all candidates Is a set of (3).
5. The dynamic monitoring method of biodiversity according to claim 4, wherein the computational expression of the pixel level multidimensional feature characterization is: ; Wherein, the Representing the scene variable(s), An index set representing the m-th set of features in the multi-modal feature matrix.
6. The method for dynamically monitoring biodiversity according to claim 1, wherein the constructing a multiple relationship space diagram based on the geospatial adjacency, the physical partition adjacency and the ecological feature similarity relationship of the target region comprises: constructing a geographic adjacency graph based on the neighborhood of the grid 8; constructing a physical adjacency graph by taking soil property partitions as boundaries; constructing an ecological similarity graph by using a K neighbor graph and a radial basis function in a space based on the pixel-level multidimensional feature representation; And carrying out weighted fusion on the adjacent matrixes of the geographic adjacent graph, the physical adjacent graph and the ecological similarity graph to obtain a final graph Laplacian.
7. The method for dynamically monitoring the biodiversity according to claim 1, wherein the expression for calculating the biodiversity index is: ; Wherein, the Represent the first A first pixel The weight to which the individual indicators are assigned, Represents the first The picture element is at the first And the value of each index.
8. The method of claim 7, wherein the value of the biodiversity index is in the range of 0-1.

Description

Dynamic monitoring method for biodiversity Technical Field The invention relates to the technical field of ecological environment remote sensing monitoring, in particular to a dynamic biodiversity monitoring method. Background Biodiversity is the basis for maintaining long-term stable and sustainable operation of the ecosystem, and is also a genetic information base adapting to future changes. Currently, model frameworks widely used for biodiversity assessment include stress-state-response (PSR), driving force-stress-state-influence-response (DPSIR), global sustainable development target framework (SDGs), and the like. With the development of remote sensing technology, a spectrum index (such as a remote sensing ecological index RSEI) based on a remote sensing image is widely used for quantifying the characteristics of an ecological system. In order to realize global biodiversity monitoring, a biodiversity core variable (EBVs) and a remote sensing biodiversity core variable (RS-EBVs) framework are proposed by an earth observation organization biodiversity observation network (GEO BON), and key indexes are obtained by utilizing a remote sensing means so as to support biodiversity protection. However, existing techniques for monitoring and assessing biodiversity still have significant limitations in that, first, the index system construction is subjective and undercrown. The existing evaluation often does not explicitly incorporate soil factors (such as soil erosion and soil moisture) into the evaluation system, neglects the basic supporting role of soil as an ecological system 'engineer', and easily masks subtle biodiversity changes. Second, the weight settings lack data driven support. The traditional evaluation method generally adopts a fixed weight strategy such as an Analytic Hierarchy Process (AHP) or an entropy weight method, so that dynamic changes of various indexes in different time and space are difficult to effectively describe, and differences of local biodiversity are easy to ignore. Third, it is difficult to account for multi-scale situational differences. The biodiversity shows obvious differences under different environmental backgrounds (such as drought degree of climate and landscape heterogeneity), the existing model lacks system description of the ecological background where the pixels are located, and index weights cannot be adaptively adjusted according to environmental situations, so that robustness and accuracy are insufficient in cross-regional and cross-year monitoring. Disclosure of Invention In order to overcome the defects of the prior art, the invention aims to provide a dynamic monitoring method for biodiversity, which solves the problems that the construction of an index system is subjective, soil factors are often ignored, the weight setting is lack of data driving support, so that space-time dynamic change is difficult to describe, and the multi-scale ecological situation difference is not considered effectively, so that the robustness and the accuracy of an evaluation result are caused. In order to achieve the above object, the present invention provides the following solutions: a method of dynamic monitoring of biodiversity, comprising: Acquiring multi-source remote sensing data of a target area, and extracting core index data respectively representing species dimension, ecosystem dimension and landscape dimension from the multi-source remote sensing data based on a remote sensing biodiversity key variable framework; acquiring situation variable data of the target area, wherein the situation variable data comprises a soil factor and an environment quality factor; Resampling and spatially registering the core index data and the situation variable data based on a unified anchor grid, and carrying out normalization processing to obtain standardized index data and standardized situation data respectively; dividing the standardized index data into three modules of species, ecosystem and landscape based on the functional characteristics of the index, and respectively calculating an intra-group consistency factor, an inter-year stability factor and a data quality factor for each module by using the standardized index data in the module; Fusing the consistency factor, the annual stability factor and the data quality factor to obtain a pixel-level module reliability gating score; Mapping the standardized index data into an initial feature representation, and modulating the initial feature representation by using the standardized situation data as guiding information according to a cross attention mechanism to obtain a pixel-level multidimensional feature representation fused with situation information; constructing a multi-relation space diagram based on the geographic space adjacent relation, the physical partition adjacent relation and the ecological feature similarity relation of the target area; Obtaining a graph Laplace operator according to the multi-relation space graph, and