Ecological security pattern identification, diagnosis and early warning comprehensive method based on machine learning

Abstract

The invention relates to a machine learning-based ecological security pattern recognition, diagnosis and early warning comprehensive method, which comprises a core system, a visual simulator interface, a database center and a learning module, and is characterized in that the core system provides ecological security pattern recognition, diagnosis and early warning functions, calculates an ecological resistance surface by using ecological source information, and calculates the ecological resistance surface; the ecological corridor is identified and diagnosed by means of the ecological resistance surface, and evolutionary simulation and early warning are carried out on ecological elements; the visual simulator interface is provided with a system user side use interface, external variable data are input in real time through a data visual interface, and diagnosis and early warning results are visually output; the database center provides database management for the core system; and the learning module provides a model basis for the core system, and is used for debugging and calibrating a machine learning model, realizing real-time and visual early warning and monitoring of an ecological security pattern, and providing scientific reference for constructing a healthier and more stable ecological security pattern.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning. Background Art

[0002] As a bridge between ecological environmental protection and the improvement of human well-being, the ecological security pattern meets the needs of regional ecological restoration, biodiversity protection, providing ecological services for social development, and promoting social sustainable development. It is of great significance to the spatial coordination of all fields and processes between ecological environmental protection and economic and social development. Therefore, building an ecological security pattern is of great significance to practicing ecological protection, promoting ecological civilization construction, and ensuring the sustainable development of national land space.

[0003] With rapid urbanization, the ecological environment has been degraded to varying degrees, putting the ecosystem under tremendous pressure. How to analyze and coordinate the overall layout in space, scientifically identify the ecological security pattern, and diagnose the key problems in the ecological security pattern is the difficulty of national space ecological protection and restoration. As an important branch of national space ecological security research, ecological security pattern identification, diagnosis and early warning are warnings of changes in the national ecological environment and the coordinated changes between the national ecological environment and the social economy. They reveal potential risks or hazards in advance, which is of great significance to maintaining regional national ecological security.

[0004] Existing ecological security pattern identification, diagnosis and early warning technologies often focus on studying the spatial characteristics of the ecosystem itself, and mostly adopt the pressure-state-response conceptual model PSR, the economic-social-environmental conceptual model EES or the method based on the ecosystem background characteristic indicator system. There is great uncertainty, and it is difficult to make real-time and intelligent predictions of future risks of the ecological security pattern.

[0005] Therefore, designing a comprehensive method for ecological security pattern identification, diagnosis, and early warning based on machine learning that can quickly complete visualization and application analysis and conduct risk early warning has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning.

[0007] The technical solution of the present invention is a comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning, including a core system, a visual simulator interface, a database center and a learning module; the core system provides ecological security pattern identification, diagnosis and early warning functions, uses ecological source information to calculate the ecological resistance surface, relies on the ecological resistance surface to identify and diagnose the ecological corridor, and performs evolution simulation and early warning on ecological elements; the visual simulator interface is provided with a system user end use interface, through the data visualization interface, external variable data is input in real time, and the diagnosis and early warning results are output visually; the database center provides database management for the core system; the learning module provides a model basis for the core system, which is used for machine learning model debugging and calibration.

[0008] As a further improvement of the present invention, the core system includes an ecological security pattern identification submodel, an ecological security pattern diagnosis submodel and an ecological security pattern early warning submodel; the ecological security pattern identification submodel provides ecological source identification, ecosystem sensitivity assessment and ecological corridor identification functions to form the current status information of the ecological security pattern; the ecological security pattern diagnosis submodule analyzes based on the current status information, discovers major risk points that may exist in the ecological security pattern, defines key areas for ecological security pattern early warning, and provides decision support for early warning; the ecological security pattern early warning submodule simulates the future development of the ecological security pattern through visual simulation, and compares it with the current status of the ecological security pattern, so as to provide real-time early warning of possible risks and problems in the ecological security pattern.

[0009] As a further improvement of the present invention, the ecological source identification includes five ecosystem service function indicators, namely habitat quality, carbon storage, vegetation coverage, soil conservation and water conservation, and the ecosystem service importance evaluation is formed by integrating the ecosystem service function indicators; the ecosystem sensitivity evaluation is based on four aspects of ecosystem sensitivity evaluation, namely habitat, water environment, soil environment and atmospheric environment, to determine the areas with extremely important ecosystem services and extremely sensitive ecosystems, and eliminate patches with scattered distribution and small area; the ecological corridor identification function collects the exchange path information of material flow and energy flow in the ecosystem to identify the ecological source.

[0010] As a further improvement of the present invention, the habitat quality evaluation is based on land cover data, setting habitat suitability, stress factor sensitivity, stress factor impact distance and weight on the habitat, and calculating the habitat quality at the grid scale. The index range is [0,1), and the calculation formula is: , in, represents the habitat quality of the x grid in the j land class; represents the habitat suitability of land type j; represents the total threat level of grid x in land class j; K represents the half-saturation constant, which is 0.5; Select landscape factors and traffic factors that affect habitats as threat factors, and set the sensitivity of different habitats to threat factors. The landscape factors include construction land and cultivated land information, and the traffic factors include railway, national highway, provincial highway and expressway information. The carbon storage evaluation is based on land cover data and carbon density parameter tables, including four basic carbon pools: aboveground carbon pool, underground carbon pool, soil carbon pool, and litter carbon pool. The carbon storage of the basic carbon pool is determined by carbon density, and the carbon density at the grid scale is calculated using the following formula: , in, is the total carbon stock of land type j; is the carbon storage in the aboveground carbon pool of land type j; is the carbon storage in underground carbon reservoirs; is the carbon storage in the soil carbon pool; is the carbon storage in litterfall carbon pool; The vegetation coverage assessment is calculated based on Landsat remote sensing images using a pixel binary model. The calculation formula is: , in, It is the vegetation coverage within a pixel of the remote sensing image; Represents the normalized difference vegetation index NDVI of the pixel; Indicates NDVI completely covered by vegetation; Indicates the NDVI value corresponding to completely bare soil or parts without vegetation cover; The soil conservation assessment is based on the ratio of the total amount of sediment in the channel or river section to the total amount of soil erosion in the basin above the section per unit time, and the sediment transport ratio is obtained. The potential soil erosion and the actual soil erosion are calculated by combining the general soil loss equation. The soil conservation amount is obtained based on the difference between the potential soil erosion and the actual soil erosion, and the assessment result is formed. The soil erosion calculation formula is: , in, is the total amount of soil erosion in grid i; for rainfall erosivity; for soil erodibility; is the terrain factor, including slope and slope length; for vegetation cover and management; for soil and water conservation measures; Said The calculation formula is {K}_{i}=\left \{{0.2+0.3exp\left [ {-0.0256SAN\frac {1-SIL} {100}} \right ]} \right \} , Among them, SAN, SIL and CLA represent the contents of sand, silt and clay in the soil, respectively; C is the soil organic carbon content; SNI is calculated as (1-SAN) × 0.01; Said The calculation formula is , in, Indicates vegetation coverage; The water conservation evaluation is based on the water balance, calculates the grid-scale runoff, combines it with the vegetation available water content parameter PAWC, and further uses the surface runoff to build an extended model to identify areas with water conservation functions. The calculation formula is: , in, is the water conservation capacity of grid i in land type j; is the water production; is the surface runoff; is the actual evapotranspiration; is the precipitation; is the average surface runoff coefficient corresponding to the land use type; The calculation formula for vegetation available water content PAWC is: , Among them, SAN, SIL, CLA and OM are the contents of soil sand, silt, clay and organic matter, respectively.

[0011] As a further improvement of the present invention, the ecosystem sensitivity evaluation characterizes the habitat sensitivity by the spatial distribution of nature reserves, land use types and vegetation coverage; characterizes the water environment sensitivity by the distance to water bodies and key water environment polluting enterprises; characterizes the soil environment sensitivity by the actual soil erosion amount calculated by soil conservation assessment; and characterizes the atmospheric environment sensitivity by the distance of 4400m from key atmospheric environment polluting enterprises and the annual average concentration of PM2.5 according to the functional zoning classification standard for environmental air sensitive areas GB3095-2012.

[0012] As a further improvement of the present invention, the ecological corridor identification includes constructing an ecological resistance surface and identifying an ecological corridor; the ecological resistance surface sets a higher ecological resistance value based on human activity concentration and complex terrain areas, and corrects the landscape ecological resistance in combination with night lights and surface curvature. The calculation formula is: , Among them, R is the corrected ecological resistance value, is the initial ecological resistance value, is the night light value, is the average nighttime light value of each landscape, is the surface curvature value, is the average surface curvature value of each landscape; The ecological corridor identification method uses the geometric center point of the ecological source as the ecological node, integrates the ecological resistance and the distance between the sources, and generates the minimum resistance path. The calculation formula is: , Among them, MCR is the minimum cumulative resistance value, is the positive correlation coefficient between ecological process and minimum cumulative resistance, is the distance between the sources, is the resistance coefficient for species migration.

[0013] As a further improvement of the present invention, the ecological security pattern diagnosis submodel includes calculating the source landscape connectivity index and the corridor interaction force; the source landscape connectivity index is calculated by the overall connectivity index dIIC and the possible connectivity index dPC, and the relative importance index dI is calculated, and the equal spacing method is used for division, and the calculation formula is: , Among them, dIIC is the overall connectivity index, n is the number of sources, a is the area of ​​the source, I is the number of corridors between sources, dPC is the possible connectivity index, P is the maximum possibility of direct flow of organisms between sources, and dI is the relative importance index; The corridor interaction force is calculated based on the gravity model to calculate the interaction force between ecological sources, and the ecological importance and sensitivity values ​​are introduced to modify the gravity model to obtain the modified interaction force. The calculation formula is: , in, is the interaction force between source corridors, is the maximum value of the corridor resistance, and is the area of ​​source areas a and b connected by the corridor, is the corridor resistance value, and is the resistance value of sources a and b connected by the corridor; is the modified interaction force of the corridor between source areas, and are the standard values ​​of ecological importance and sensitivity of source areas a and b connected by the corridor.

[0014] As a further improvement of the present invention, the ecological security pattern early warning sub-model is established based on the system dynamics model and the cellular automaton model, and integrates the artificial neural network algorithm and the roulette wheel selection mechanism; the ecological security pattern early warning sub-model collects land use data, inputs the Markov model to predict the future demand for different land types, superimposes it with the current status of the ecological security pattern, sets the ecological security pattern early warning threshold, and obtains the formula , in, and is the ground state matrix at t and t+1, Represents the transition probability matrix from type a to type b; Then set the neighborhood weight parameter, based on the land expansion capacity as construction land > unused land > water area > grassland > forest land > cultivated land, and express it in the threshold range [0,1]. The closer to 1, the stronger the expansion capacity; Finally, the suitability probability of cell s, the competition relationship between cells and the land use development goals are combined to obtain the combined improved CA model, in which the suitability probability of cell s is based on the natural condition information including soil and climate, the competition relationship between cells is based on the human activity information including GDP and population, and the land use development goal is based on the driving factors of land use change. The overall development probability of the cell is obtained by comprehensive calculation, and the calculation formula is: , in, It refers to the overall conversion probability of p cell units from the original land use type to k type of land at time t; represents the suitability probability of cell p being converted to type k of land; represents the adaptive inertia coefficient; represents the conversion cost of converting class c land into class k land; represents the neighborhood impact factor, and the calculation formula is: , in, is the number of Moore neighborhood windows; is the variable weight for different land uses; The overall development probability of the cell is input into the roulette selection mechanism to obtain the land type that each cell unit is converted into, and a global visual early warning of the ecological security pattern is carried out.

[0015] After adopting the above method, machine learning technology can quickly warn of ecological security pattern risks, and quickly complete modeling, visualization and application analysis, effectively promoting the research and development of ecological security pattern; based on the gradual identification and diagnosis of ecological source areas, ecological resistance surfaces, and ecological corridors, a comprehensive evaluation of five ecosystem service function indicators is formed to form an ecosystem service importance evaluation, determine extremely important ecosystem service areas and extremely sensitive ecosystem areas, and eliminate scattered and small patches to accurately identify ecological sources; identify major risk points that may appear in the ecological security pattern through the ecological security pattern diagnosis sub-model, and frame the key areas for ecological security pattern warning , use the gravity model to evaluate the interaction force between ecological sources, combine the internal ecological importance and sensitivity of the source that are ignored by the traditional module, use the interaction force index to represent the corridor efficiency and its importance, so that the quality of the source connected by the corridor is positively correlated with the interaction force; through the ecological security pattern early warning sub-model, based on the system dynamics model and cellular automaton model, the current ecological security pattern is used to predict the future demand for different land types, integrate the artificial neural network algorithm and the roulette selection mechanism, visualize and simulate the future development of the ecological security pattern, and give real-time early warning of possible risks and problems in the ecological security pattern, providing a scientific reference for building a healthier and more stable ecological security pattern. DETAILED DESCRIPTION

[0016] A comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning, including a core system, a visual simulator interface, a database center and a learning module; the core system provides ecological security pattern identification, diagnosis and early warning functions, uses ecological source information to calculate the ecological resistance surface, relies on the ecological resistance surface to identify and diagnose ecological corridors, and simulates and warns of ecological elements; the visual simulator interface is provided with a system user end use interface, through the data visualization interface, external variable data is input in real time, and the diagnosis and early warning results are output visually; the database center provides database management for the core system; the learning module provides a model basis for the core system, which is used for machine learning model debugging and calibration.

[0017] The core system includes an ecological security pattern identification submodel, an ecological security pattern diagnosis submodel and an ecological security pattern early warning submodel; the ecological security pattern identification submodel provides ecological source identification, ecosystem sensitivity assessment and ecological corridor identification functions to form the current status information of the ecological security pattern; the ecological security pattern diagnosis submodule analyzes the current status information to discover major risk points that may exist in the ecological security pattern, defines the key areas for ecological security pattern early warning, and provides decision support for early warning; the ecological security pattern early warning submodule simulates the future development of the ecological security pattern through visualization, and compares it with the current status of the ecological security pattern, so as to provide real-time early warning of possible risks and problems in the ecological security pattern.

[0018] The ecological source identification includes five ecosystem service function indicators: habitat quality, carbon storage, vegetation coverage, soil conservation and water conservation. The ecosystem service function indicators are integrated to form an ecosystem service importance evaluation; the ecosystem sensitivity evaluation is based on four aspects of ecosystem sensitivity evaluation: habitat, water environment, soil environment and atmospheric environment, to determine the areas with extremely important ecosystem services and extremely sensitive ecosystems, and eliminate scattered and small patches; the ecological corridor identification function collects the exchange path information of material flow and energy flow in the ecosystem to identify the ecological source.

[0019] The habitat quality evaluation is based on land cover data, setting habitat suitability, stress factor sensitivity, impact distance and weight of stress factors on habitat, and calculating the habitat quality at the grid scale. The index range is [0,1). The calculation formula is: , in, represents the habitat quality of the x grid in the j land class; represents the habitat suitability of land type j; represents the total threat level of grid x in land class j; K represents the half-saturation constant, which is 0.5; Select landscape factors and traffic factors that affect habitats as threat factors, and set the sensitivity of different habitats to threat factors. The landscape factors include construction land and cultivated land information, and the traffic factors include railway, national highway, provincial highway and expressway information. The carbon storage evaluation is based on land cover data and carbon density parameter tables, including four basic carbon pools: aboveground carbon pool, underground carbon pool, soil carbon pool, and litter carbon pool. The carbon storage of the basic carbon pool is determined by carbon density, and the carbon density at the grid scale is calculated using the following formula: , in, is the total carbon stock of land type j; is the carbon storage in the aboveground carbon pool of land type j; is the carbon storage in underground carbon reservoirs; is the carbon storage in the soil carbon pool; is the carbon storage in litterfall carbon pool; The vegetation coverage assessment is calculated based on Landsat remote sensing images using a pixel binary model. The calculation formula is: , in, It is the vegetation coverage within a pixel of the remote sensing image; Represents the normalized difference vegetation index NDVI of the pixel; Indicates NDVI completely covered by vegetation; Indicates the NDVI value corresponding to completely bare soil or parts without vegetation cover; The soil conservation assessment is based on the ratio of the total amount of sediment in the channel and river section to the total amount of soil erosion in the basin above the section per unit time, and the sediment transport ratio is obtained. The potential soil erosion and actual soil erosion are calculated by combining the general soil loss equation. The soil conservation amount is obtained based on the difference between the potential soil erosion and the actual soil erosion, forming the assessment result. The soil erosion calculation formula is: , in, is the total amount of soil erosion in grid i; for rainfall erosivity; for soil erodibility; is the terrain factor, including slope and slope length; for vegetation cover and management; for soil and water conservation measures; Said The calculation formula is {K}_{i}=\left \{{0.2+0.3exp\left [ {-0.0256SAN\frac {1-SIL} {100}} \right ]} \right \} , Among them, SAN, SIL and CLA represent the contents of sand, silt and clay in the soil, respectively; C is the soil organic carbon content; SNI is calculated as (1-SAN) × 0.01; Said The calculation formula is ,in, Indicates vegetation coverage; The water conservation evaluation is based on the water balance, calculates the grid-scale runoff, combines it with the vegetation available water content parameter PAWC, and further uses the surface runoff to build an extended model to identify areas with water conservation functions. The calculation formula is: , in, is the water conservation capacity of grid i in land type j; is the water production; is the surface runoff; is the actual evapotranspiration; is the precipitation; is the average surface runoff coefficient corresponding to the land use type; The calculation formula for vegetation available water content PAWC is: , Among them, SAN, SIL, CLA and OM are the contents of soil sand, silt, clay and organic matter, respectively.

[0020] The ecosystem sensitivity assessment uses the spatial distribution of nature reserves, land use types and vegetation coverage to characterize the sensitivity of habitats; uses the distance to water bodies and key water environment polluting enterprises to characterize the sensitivity of water environments; uses the actual amount of soil erosion calculated by soil conservation assessment to characterize the sensitivity of soil environments; and uses the distance of 4,400 m from key air environment polluting enterprises and the annual average concentration of PM2.5 to characterize the sensitivity of atmospheric environments according to the functional zoning classification standard for environmental air sensitive areas GB3095-2012.

[0021] The ecological corridor identification includes constructing an ecological resistance surface and identifying ecological corridors; the ecological resistance surface sets a higher ecological resistance value based on human activity concentration and complex terrain areas, and corrects the landscape ecological resistance in combination with night lights and surface curvature. The calculation formula is: , Among them, R is the corrected ecological resistance value, is the initial ecological resistance value, is the night light value, is the average nighttime light value of each landscape, is the surface curvature value, is the average surface curvature value of each landscape; The ecological corridor identification method uses the geometric center point of the ecological source as the ecological node, integrates the ecological resistance and the distance between the sources, and generates the minimum resistance path. The calculation formula is: , Among them, MCR is the minimum cumulative resistance value, is the positive correlation coefficient between ecological process and minimum cumulative resistance, is the distance between the sources, is the resistance coefficient for species migration.

[0022] As a further improvement of the present invention, the ecological security pattern diagnosis submodel includes calculating the source landscape connectivity index and the corridor interaction force; the source landscape connectivity index is calculated by the overall connectivity index dIIC and the possible connectivity index dPC, and the relative importance index dI is calculated, and the equal spacing method is used for division, and the calculation formula is: , Among them, dIIC is the overall connectivity index, n is the number of sources, a is the area of ​​the source, I is the number of corridors between sources, dPC is the possible connectivity index, P is the maximum possibility of direct flow of organisms between sources, and dI is the relative importance index; The corridor interaction force is calculated based on the gravity model to calculate the interaction force between ecological sources, and the ecological importance and sensitivity values ​​are introduced to modify the gravity model to obtain the modified interaction force. The calculation formula is: , in, is the interaction force between source corridors, is the maximum value of the corridor resistance, and is the area of ​​source areas a and b connected by the corridor, is the corridor resistance value, and is the resistance value of sources a and b connected by the corridor; is the modified interaction force of the corridor between source areas, and are the standard values ​​of ecological importance and sensitivity of source areas a and b connected by the corridor.

[0023] The ecological security pattern early warning sub-model is based on the system dynamics model and the cellular automaton model, and is established by integrating the artificial neural network algorithm and the roulette wheel selection mechanism; the ecological security pattern early warning sub-model collects land use data, inputs the Markov model to predict the future demand for different land types, superimposes it with the current status of the ecological security pattern, sets the ecological security pattern early warning threshold, and obtains the formula , in, and is the ground state matrix at t and t+1, Represents the transition probability matrix from type a to type b; Then set the neighborhood weight parameter, based on the land expansion capacity as construction land > unused land > water area > grassland > forest land > cultivated land, and express it in the threshold range [0,1]. The closer to 1, the stronger the expansion capacity; Finally, the suitability probability of cell s, the competition relationship between cells and the land use development goals are combined to obtain the combined improved CA model, in which the suitability probability of cell s is based on the natural condition information including soil and climate, the competition relationship between cells is based on the human activity information including GDP and population, and the land use development goal is based on the driving factors of land use change. The overall development probability of the cell is obtained by comprehensive calculation, and the calculation formula is: , in, It refers to the overall conversion probability of p cell units from the original land use type to k type of land at time t; represents the suitability probability of cell p being converted to type k of land; represents the adaptive inertia coefficient; represents the conversion cost of converting class c land into class k land; represents the neighborhood impact factor, and the calculation formula is: , in, is the number of Moore neighborhood windows; is the variable weight for different land uses; The overall development probability of the cell is input into the roulette selection mechanism to obtain the land type that each cell unit is converted into, and a global visual early warning of the ecological security pattern is carried out.

[0024] Through machine learning technology, it is possible to quickly warn of ecological security pattern risks, and quickly complete modeling, visualization and application analysis, effectively promoting the research and development of ecological security pattern; based on the gradual identification and diagnosis of ecological source areas, ecological resistance surfaces and ecological corridors, a comprehensive evaluation of five ecosystem service function indicators is formed to form an ecosystem service importance evaluation, determine the extremely important ecosystem service and the extremely sensitive ecosystem areas, and eliminate scattered and small patches to accurately identify ecological source areas; through the ecological security pattern diagnosis sub-model, major risk points that may appear in the ecological security pattern are identified, and the key areas for ecological security pattern warning are framed. The force model evaluates the interaction force between ecological sources, combines the internal ecological importance and sensitivity of the source that are ignored by the traditional module, and uses the interaction force index to represent the corridor efficiency and its importance, so that the quality of the source connected by the corridor is positively correlated with the interaction force; through the ecological security pattern early warning sub-model, based on the system dynamics model and the cellular automaton model, the future demand for different land types is predicted based on the current ecological security pattern, and the artificial neural network algorithm and the roulette selection mechanism are integrated to visualize and simulate the future development of the ecological security pattern, and to provide real-time early warning of possible risks and problems in the ecological security pattern, providing a scientific reference for building a healthier and more stable ecological security pattern.

Claims

1. A comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning, characterized by: It includes a core system, a visual simulator interface, a database center and a learning module; the core system provides ecological security pattern identification, diagnosis and early warning functions, uses ecological source information to calculate the ecological resistance surface, relies on the ecological resistance surface to identify and diagnose the ecological corridor, and simulates and warns the evolution of ecological elements; the visual simulator interface is provided with a system user-end interface, through the data visualization interface, external variable data is input in real time, and the diagnosis and early warning results are output visually; the database center provides database management for the core system; the learning module provides a model basis for the core system, which is used for machine learning model debugging and calibration.

2. According to claim 1, a comprehensive method for identifying, diagnosing and early warning ecological security patterns based on machine learning is characterized by: The core system includes an ecological security pattern identification sub-model, an ecological security pattern diagnosis sub-model and an ecological security pattern early warning sub-model; the ecological security pattern identification sub-model provides ecological source identification, ecosystem sensitivity assessment and ecological corridor identification functions to form ecological security pattern status information; The ecological security pattern diagnosis submodule analyzes the current situation information, finds the major risk points that may exist in the ecological security pattern, defines the key areas for ecological security pattern early warning, and provides decision support for early warning; The ecological security pattern early warning submodule simulates the future development of the ecological security pattern through visualization, and compares it with the current ecological security pattern, so as to warn the possible risks and problems of the ecological security pattern in real time.

3. According to claim 2, a comprehensive method for identifying, diagnosing and warning ecological security patterns based on machine learning is characterized by: The ecological source identification includes five ecosystem service function indicators: habitat quality, carbon storage, vegetation coverage, soil conservation and water conservation. The ecosystem service function indicators are integrated to form an ecosystem service importance evaluation; the ecosystem sensitivity evaluation is based on four aspects of ecosystem sensitivity evaluation: habitat, water environment, soil environment and atmospheric environment, to determine the areas with extremely important ecosystem services and extremely sensitive ecosystems, and eliminate scattered and small patches; the ecological corridor identification function collects the exchange path information of material flow and energy flow in the ecosystem to identify the ecological source.

4. According to claim 3, a comprehensive method for ecological security pattern identification, diagnosis and early warning based on machine learning is characterized by: The habitat quality evaluation is based on land cover data, setting habitat suitability, stress factor sensitivity, impact distance and weight of stress factors on habitat, and calculating the habitat quality at the grid scale. The index range is [0,1). The calculation formula is: ,in, represents the habitat quality of the x grid in the j land class; represents the habitat suitability of land type j; represents the total threat level of grid x in land class j; K represents the half-saturation constant, which is 0.5; Select landscape factors and traffic factors that affect habitats as threat factors, and set the sensitivity of different habitats to threat factors. The landscape factors include construction land and cultivated land information, and the traffic factors include railway, national highway, provincial highway and expressway information. The carbon storage evaluation is based on land cover data and carbon density parameter tables, including four basic carbon pools: aboveground carbon pool, underground carbon pool, soil carbon pool, and litter carbon pool. The carbon storage of the basic carbon pool is determined by carbon density, and the carbon density at the grid scale is calculated using the following formula: , in, is the total carbon stock of land type j; is the carbon storage in the aboveground carbon pool of land type j; is the carbon storage in underground carbon reservoirs; is the carbon storage in the soil carbon pool; is the carbon storage in litterfall carbon pool; The vegetation coverage assessment is calculated based on Landsat remote sensing images using a pixel binary model. The calculation formula is: , in, It is the vegetation coverage within a pixel of the remote sensing image; Represents the normalized difference vegetation index NDVI of the pixel; Indicates NDVI completely covered by vegetation; Indicates the NDVI value corresponding to completely bare soil or parts without vegetation cover; The soil conservation assessment is based on the ratio of the total amount of sediment in the channel or river section to the total amount of soil erosion in the basin above the section per unit time, and the sediment transport ratio is obtained. The potential soil erosion and the actual soil erosion are calculated by combining the general soil loss equation. The soil conservation amount is obtained based on the difference between the potential soil erosion and the actual soil erosion, and the assessment result is formed. The soil erosion calculation formula is: , in, is the total amount of soil erosion in grid i; for rainfall erosivity; for soil erodibility; is the terrain factor, including slope and slope length; for vegetation cover and management; for soil and water conservation measures; Said The calculation formula is ， ， ， Among them, SAN, SIL and CLA represent the contents of sand, silt and clay in the soil, respectively; C is the soil organic carbon content; SNI is calculated as (1-SAN) × 0.01; Said The calculation formula is ， in, Indicates vegetation coverage; The water conservation evaluation is based on the water balance, calculates the grid-scale runoff, combines it with the vegetation available water content parameter PAWC, and further uses the surface runoff to build an extended model to identify areas with water conservation functions. The calculation formula is: ， ， ， in, is the water conservation capacity of grid i in land type j; is the water production; is the surface runoff; is the actual evapotranspiration; is the precipitation; is the average surface runoff coefficient corresponding to the land use type; The calculation formula for vegetation available water content PAWC is: ， Among them, SAN, SIL, CLA and OM are the contents of soil sand, silt, clay and organic matter, respectively.

5. According to claim 4, a comprehensive method for identifying, diagnosing and warning ecological security patterns based on machine learning is characterized by: The ecosystem sensitivity assessment is based on the spatial distribution of nature reserves, land use types and vegetation coverage to characterize the sensitivity of habitats; The distance from water bodies and key water pollution-discharging enterprises are used to characterize the sensitivity of water environment. The actual soil erosion amount calculated by soil conservation assessment is used to characterize the sensitivity of soil environment. According to the functional zone classification standard for environmental air sensitive areas GB3095-2012, the distance from key air pollution-discharging enterprises to 4400m and the annual average concentration of PM2.5 are used to characterize the sensitivity of atmospheric environment.

6. According to claim 4, a comprehensive method for identifying, diagnosing and early warning ecological security patterns based on machine learning is characterized by: The ecological corridor identification includes constructing ecological resistance surface and identifying ecological corridor; The ecological resistance surface is based on human activity concentration and complex terrain areas, setting a higher ecological resistance value, and combining night lights and surface curvature to correct the landscape ecological resistance. The calculation formula is: , Among them, R is the corrected ecological resistance value, is the initial ecological resistance value, is the night light value, is the average nighttime light value of each landscape, is the surface curvature value, is the average surface curvature value of each landscape; The ecological corridor identification method uses the geometric center point of the ecological source as the ecological node, integrates the ecological resistance and the distance between the sources, and generates the minimum resistance path. The calculation formula is: , Among them, MCR is the minimum cumulative resistance value, is the positive correlation coefficient between ecological process and minimum cumulative resistance, is the distance between the sources, is the resistance coefficient for species migration.

7. According to claim 2, a comprehensive method for identifying, diagnosing and early warning ecological security patterns based on machine learning is characterized by: The ecological security pattern diagnosis sub-model includes calculating the source landscape connectivity index and the corridor interaction force; the source landscape connectivity index is calculated by the overall connectivity index dIIC and the possible connectivity index dPC, and the relative importance index dI is calculated, and the equal spacing method is used for division. The calculation formula is: ， ， ， Among them, dIIC is the overall connectivity index, n is the number of sources, a is the area of ​​the source, I is the number of corridors between sources, dPC is the possible connectivity index, P is the maximum possibility of direct flow of organisms between sources, and dI is the relative importance index; The corridor interaction force is calculated based on the gravity model to calculate the interaction force between ecological sources, and the ecological importance and sensitivity values ​​are introduced to modify the gravity model to obtain the modified interaction force. The calculation formula is: , ,in, is the interaction force between source corridors, is the maximum value of the corridor resistance, and is the area of ​​source areas a and b connected by the corridor, is the corridor resistance value, and is the resistance value of sources a and b connected by the corridor; is the modified interaction force of the corridor between source areas, and are the standard values ​​of ecological importance and sensitivity of source areas a and b connected by the corridor.

8. According to claim 2, a comprehensive method for identifying, diagnosing and early warning ecological security patterns based on machine learning is characterized by: The ecological security pattern early warning sub-model is established based on the system dynamics model and the cellular automaton model, and integrates the artificial neural network algorithm and the roulette wheel selection mechanism; The ecological security pattern early warning sub-model collects land use data, inputs the Markov model to predict the future demand for different land types, superimposes it with the current ecological security pattern, sets the ecological security pattern early warning threshold, and obtains the formula , in, and is the ground state matrix at t and t+1, Represents the transition probability matrix from type a to type b; Then set the neighborhood weight parameter, based on the land expansion capacity as construction land > unused land > water area > grassland > forest land > cultivated land, and express it in the threshold range [0,1]. The closer to 1, the stronger the expansion capacity; Finally, the suitability probability of cell s, the competition relationship between cells and the land use development goals are combined to obtain the combined improved CA model, in which the suitability probability of cell s is based on the natural condition information including soil and climate, the competition relationship between cells is based on the human activity information including GDP and population, and the land use development goal is based on the driving factors of land use change. The overall development probability of the cell is obtained by comprehensive calculation, and the calculation formula is: , in, It refers to the overall conversion probability of p cell units from the original land use type to k type of land at time t; represents the suitability probability of cell p being converted to type k of land; represents the adaptive inertia coefficient; represents the conversion cost of converting class c land into class k land; represents the neighborhood impact factor, and the calculation formula is: ,in, is the number of Moore neighborhood windows; is the variable weight for different land uses; The overall development probability of the cell is input into the roulette selection mechanism to obtain the land type that each cell unit is converted into, and a global visual early warning of the ecological security pattern is carried out.

Images