Species distribution prediction method fused with habitat tolerance limit

Abstract

The invention discloses a gull species distribution prediction method fusing habitat tolerance limit, and relates to the technical field of ecological protection and species distribution prediction, in particular to a gull species distribution prediction method. The invention aims to solve the problem that the existing method is difficult to effectively integrate the known species ecology law into the modeling process. The method comprises the following steps: 1, quantifying the environmental tolerance limit of species and generating a spatial constraint layer; 2, predicting initial species distribution of machine learning based on tolerance limit constraint; and 3, verifying and outputting the model. According to the method, the chronic disease of'ecological unreasonable prediction 'of a traditional model is fundamentally solved, and the prediction result has more biological significance.

Description

Technical Field

[0001] This invention relates to the field of ecological conservation and species distribution prediction technology, specifically to a method for predicting species distribution. Background Technology

[0002] Species distribution models are core tools for biodiversity conservation and ecosystem management. Traditional prediction methods typically predict potential distribution areas by statistically analyzing the correlation between species distribution points and environmental variables. However, these methods have significant drawbacks: (1) Ecologically unreasonable predictions: Models may predict areas where species cannot physiologically survive based solely on statistical characteristics. For example, a waterbird that depends on low- to mid-altitude wetland ecosystems may be found in plateaus above 4,000 meters or in extremely arid deserts. (2) Sensitive to data bias: For endangered species such as the Relict Gull, whose habitat distribution points are scarce and data acquisition is difficult, the model is prone to overfitting, resulting in unstable predictions and low reliability, which may lead to the misallocation of valuable conservation resources. (3) Lack of mechanistic constraints: Existing methods are mostly "black box" models, making it difficult to effectively integrate known species ecological laws (such as physiological tolerance limits) into the modeling process.

[0003] Although ecological theory clearly states that each species has its environmental tolerance limits, such as maximum survival altitude and suitable climate zone, current technology lacks a standardized method to systematically integrate such a priori, quantifiable mechanistic knowledge into a data-driven machine learning prediction process, thereby constraining the model to generate ecologically reasonable and highly accurate prediction results. Summary of the Invention

[0004] The purpose of this invention is to solve the technical problem that existing methods cannot effectively integrate known species ecological laws into the modeling process, and to provide a species distribution prediction method that integrates "habitat tolerance limit".

[0005] The species distribution prediction method that integrates "habitat tolerance limit" is as follows:

[0006] I. Quantification of Species' Environmental Tolerance Limits and Generation of Spatial Constraint Layers

[0007] For the target species, based on its historical distribution data, its environmental tolerance limit parameters are quantitatively extracted and defined, including the upper limit of the suitable habitat altitude and the hydro-climatic ecological niche, to generate a nationwide binary raster layer, namely the elevation and regional constraint layer.

[0008] The quantitative extraction method involves extracting the corresponding elevation values ​​from all historical distribution points within the suitable habitat range of the species, determining the upper limit threshold of its distribution altitude through statistical analysis, analyzing its distribution and boundaries, and defining its suitable hydro-climatic ecological niche.

[0009] Spatial constraint layer generation: The upper limit of altitude and the boundary are generated as the spatial constraint layer in the geographic information system. The spatial constraint layer is an elevation constraint layer and a hydro-climate constraint layer. In the elevation constraint layer, areas with an altitude of less than or equal to 2000 meters are assigned a value of 1, which is allowed to pass, and areas with an altitude greater than 2000 meters are assigned a value of 0, which needs to be blocked. In the hydro-climate constraint layer, the area located in the intersection area defined by the 400 mm isohyet and the monsoon boundary is assigned a value of 1, and the area outside is assigned a value of 0.

[0010] II. Prediction of Initial Species Distribution in Machine Learning Based on Tolerance Limit Constraints

[0011] Acquire distribution point data and multi-dimensional environmental variable data of the target species; use recursive features to screen core environmental driving factors; use random forest algorithm and XGBoost algorithm to train species distribution model and generate initial, unconstrained species habitat suitability prediction layer;

[0012] The two or more spatial constraint layers generated in step one are used as spatial filters to perform pixel-by-pixel logical operations on the species habitat suitability prediction layer generated in step two.

[0013] The corrected formula is:

[0014]

[0015] in, The value of the spatial constraint layer is 0 or 1. Through this multiplication operation, the region that is outside any tolerance limit in the initial prediction, that is, the corresponding spatial constraint layer value is 0, is forcibly corrected to 0 in terms of suitability, and is thus excluded in space. The region that is within all tolerance limits, that is, all spatial constraint layer values ​​are 1, retains its initial predicted suitability value.

[0016] III. Model Validation and Output

[0017] Using independent species distribution validation data, the accuracy index of the prediction layers before and after correction is compared, and the final species distribution prediction map after tolerance limit correction is output. This map excludes ecologically unreasonable areas and accurately indicates the potential distribution areas and core distribution ranges that conform to the physiological and behavioral characteristics of the species.

[0018] The statistical analysis described in step one includes kernel density estimation and frequency distribution histogram.

[0019] The upper limit threshold mentioned in step one is an altitude of 2000 meters.

[0020] The suitable hydro-climatic ecological niche mentioned in step one is the intersection area defined by the 400 mm annual precipitation isohyet and the monsoon line.

[0021] The multi-dimensional environmental variable data mentioned in step two include normalized vegetation index, normalized water index, enhanced vegetation index, annual precipitation, annual average temperature, diurnal temperature range, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, precipitation of the wettest month, precipitation of the driest month, precipitation seasonality, relative humidity, surface temperature, digital elevation model, topographic humidity index, land use / land cover, population density, nighttime light index, GDP, road density, distance from water body, and distance from road.

[0022] The accuracy metrics mentioned in step three are AUC, TSS, and Kappa.

[0023] The species described is the Relict Gull.

[0024] This invention specifically relates to a method and system that integrates the theory of species' "habitat tolerance limits" to improve the accuracy of machine learning models in predicting species distribution (especially endangered species such as the Relict Gull). This invention aims to overcome the shortcomings of existing technologies and provide a species distribution prediction method that integrates "habitat tolerance limits." Its core lies in solving how to transform the mechanistic environmental tolerance knowledge of species into actionable spatial constraint rules and embed them into the machine learning modeling process to correct and optimize the initial prediction results, ultimately obtaining an ecologically reliable and spatially accurate species distribution prediction map, which is particularly suitable for endangered species with scarce data.

[0025] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0026] Significantly improves the ecological rationality of predictions: By using hard constraints to exclude areas that species cannot physiologically reach or survive in (such as areas with excessively high altitudes or climate mismatches), it fundamentally solves the persistent problem of "ecologically unreasonable predictions" in traditional models, making the prediction results more biologically meaningful.

[0027] Enhancing the robustness of the model to sparse data: For data-scarce species such as the Relict Gull, introducing prior knowledge of the tolerance limit is equivalent to providing the model with strong regularization, effectively preventing overfitting and improving the stability and reliability of the model when extrapolating and predicting unknown regions.

[0028] Achieving effective integration of mechanisms and data: This invention creatively proposes a fusion framework of "data-driven prediction followed by mechanism knowledge correction." It leverages machine learning's ability to extract complex relationships from data while introducing deterministic ecological rules through a post-processing "spatial filter," serving as an effective bridge connecting related models and mechanism models.

[0029] The output results directly support conservation decisions: the final distribution map clearly identifies potential habitats and core distribution ranges that meet the actual needs of species (such as Hongjiannao and Kangba Nuoer Lake), providing direct and reliable scientific basis for delineating ecological protection red lines, planning nature reserves, and assessing the impact of climate change. Attached Figure Description

[0030] Figure 1 The figure shows the elevation tolerance limit analysis of Relict Gulls in Experiment 1. In the figure, (A) is a histogram of all the historical data of Relict Gulls, and the vertical dashed line represents the upper limit of 2000 meters elevation used to define the terrain constraint. In the figure, (B) is a kernel density map with the key hydro-climate boundary breeding grounds (orange dots) superimposed.

[0031] Figure 2 The images are a comparison of the final prediction images after constraint correction in Experiment 1, where (A) is the initial prediction image without constraints and (B) is a schematic diagram of the tolerance limit constraint layer. Detailed Implementation

[0032] The technical solution of the present invention is not limited to the specific embodiments listed below, but also includes any combination of the specific embodiments.

[0033] Specific Implementation Method 1: This implementation method incorporates the species distribution prediction method based on "habitat tolerance limit" as follows:

[0034] I. Quantification of Species' Environmental Tolerance Limits and Generation of Spatial Constraint Layers

[0035] For the target species, based on its historical distribution data, its environmental tolerance limit parameters are quantitatively extracted and defined, including the upper limit of the suitable habitat altitude and the hydro-climatic ecological niche, to generate a nationwide binary raster layer, namely the elevation and regional constraint layer.

[0036] The quantitative extraction method involves extracting the corresponding elevation values ​​from all historical distribution points within the suitable habitat range of the species, determining the upper limit threshold of its distribution altitude through statistical analysis, analyzing its distribution and boundaries, and defining its suitable hydro-climatic ecological niche.

[0037] Spatial constraint layer generation: The upper limit of altitude and the boundary are generated as the spatial constraint layer in the geographic information system. The spatial constraint layer is an elevation constraint layer and a hydro-climate constraint layer. In the elevation constraint layer, areas with an altitude of less than or equal to 2000 meters are assigned a value of 1, which is allowed to pass, and areas with an altitude greater than 2000 meters are assigned a value of 0, which needs to be blocked. In the hydro-climate constraint layer, the area located in the intersection area defined by the 400 mm isohyet and the monsoon boundary is assigned a value of 1, and the area outside is assigned a value of 0.

[0038] II. Prediction of Initial Species Distribution in Machine Learning Based on Tolerance Limit Constraints

[0039] Acquire distribution point data and multi-dimensional environmental variable data of the target species; use recursive features to screen core environmental driving factors; use random forest algorithm and XGBoost algorithm to train species distribution model and generate initial, unconstrained species habitat suitability prediction layer;

[0040] The two or more spatial constraint layers generated in step one are used as spatial filters to perform pixel-by-pixel logical operations on the species habitat suitability prediction layer generated in step two.

[0041] The corrected formula is:

[0042]

[0043] in, The value of the spatial constraint layer is 0 or 1. Through this multiplication operation, the region that is outside any tolerance limit in the initial prediction, that is, the corresponding spatial constraint layer value is 0, is forcibly corrected to 0 in terms of suitability, and is thus excluded in space. The region that is within all tolerance limits, that is, all spatial constraint layer values ​​are 1, retains its initial predicted suitability value.

[0044] III. Model Validation and Output

[0045] Using independent species distribution validation data, the accuracy index of the prediction layers before and after correction is compared, and the final species distribution prediction map after tolerance limit correction is output. This map excludes ecologically unreasonable areas and accurately indicates the potential distribution areas and core distribution ranges that conform to the physiological and behavioral characteristics of the species.

[0046] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the statistical analysis described in step one is kernel density estimation and frequency distribution histogram. Everything else is the same as in Specific Implementation Method One.

[0047] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the upper limit threshold mentioned in step one is an altitude of 2000 meters. Everything else is the same as in Specific Implementation Method One or Two.

[0048] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the suitable hydroclimatic ecological niche described in step one is the intersection area defined by the 400 mm annual precipitation isohyet and the monsoon line. Everything else is the same as in Specific Implementation Methods One to Three.

[0049] Specific Implementation Method Five: This implementation method differs from Specific Implementation Methods One through Four in that the multi-dimensional environmental variable data mentioned in step two include normalized difference vegetation index, normalized difference water index, enhanced vegetation index, annual precipitation, annual average temperature, diurnal temperature range, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, precipitation of the wettest month, precipitation of the driest month, precipitation seasonality, relative humidity, surface temperature, digital elevation model, topographic humidity index, land use / land cover, population density, nighttime light index, GDP, road density, distance from water bodies, and distance from roads. Everything else is the same as in Specific Implementation Methods One through Four.

[0050] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One to Five in that the accuracy indicators mentioned in step three are AUC, TSS, and Kappa. Everything else is the same as in Specific Implementation Methods One to Five.

[0051] Specific Implementation Method Seven: This implementation method differs from Specific Implementation Methods One to Six in that the species described is the Relict Gull. Otherwise, it is the same as Specific Implementation Methods One to Six.

[0052] The following experiments were used to verify the effectiveness of the invention:

[0053] Experiment 1:

[0054] The following is a method for predicting the species distribution of Relict Gulls that incorporates the "habitat tolerance limit":

[0055] Step 1: Quantification of the environmental tolerance limit of relict gulls and generation of spatial constraint layers

[0056] Distribution data of Relict Gulls throughout China from 2008 to 2023 were collected, and the altitude values ​​of each point were extracted to create a frequency distribution histogram. Figure 1 This is the spatial constraint condition for the Relict Gull distribution model. Figure (A) is a histogram recording elevations. The vertical dashed line represents the 2000-meter elevation limit used to define the topographic constraint. The median and mean elevations are 43 meters and 439.08 meters, respectively. Figure (B) is a kernel density map overlaid with key hydroclimatic boundary breeding grounds (orange dots). The core distribution of this species is restricted between the 400 mm isohyet (green) and the East Asian monsoon boundary (blue), defining the hydroclimatic constraint. Analysis shows that over 95% of the distribution points are below 2000 meters in elevation. Therefore, "elevation ≤ 2000 meters" is defined as the core elevation tolerance limit for the Relict Gull. In ArcGIS software, based on 30-meter resolution DEM data, a nationwide binary raster layer, namely the "elevation constraint layer," is generated, where the pixel value for elevation ≤ 2000 meters is 1, and the pixel value for > 2000 meters is 0.

[0057] Simultaneously, spatially superimpose the distribution points of all breeding grounds with the 400 mm annual precipitation line and the East Asian monsoon boundary line. Figure 1 (B) It was found that its core breeding area is strictly confined to the transition zone between these two boundaries. Based on this, the area "located within the region bounded by the 400 mm isohyet and the monsoon boundary" is defined as the hydroclimatic niche of the Relict Gull. A second binary raster layer, namely the "hydroclimatic constraint layer", is generated, with a pixel value of 1 within this region and a pixel value of 0 outside the region.

[0058] Step 2: Predicting the initial relict gull distribution using machine learning based on tolerance limit constraints

[0059] We collected multi-dimensional environmental variable data related to the Relict Gull's ecology, including Normalized Difference Vegetation Index (NDI), Normalized Difference Water Index (NDI), Enhanced Vegetation Index (EDI), annual precipitation, annual mean temperature, diurnal temperature range, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, precipitation of the wettest month, precipitation seasonality, relative humidity, surface temperature, digital elevation model (DEM), topographic humidity index, land use / land cover, population density, nighttime light index, GDP, road density, distance from water bodies, and distance from roads, totaling 23 variables. Using recursive feature elimination, we identified 15 core driving factors: distance from water bodies, enhanced vegetation index, population density, diurnal temperature range, distance from roads, relative humidity, minimum temperature of the coldest month, maximum temperature of the warmest month, precipitation of the wettest month, precipitation seasonality, road density, annual precipitation, temperature seasonality, annual mean temperature, and NDI. Random forest and XGBoost algorithms were used respectively, with the selected variables as input, to train the Relict Gull habitat suitability model, generating initial habitat suitability prediction layers for each year and the average from 2008 to 2023. Figure 2 This is a suitable habitat for Relict Gulls during their breeding season in China. Figure 2 (A) The habitat suitability distribution map without altitude masking shows that although most areas of the Qinghai-Tibet Plateau (altitude > 2,000 meters) are not suitable for the survival of Relict Gulls, they were initially predicted to be hotspots for Relict Gull releases, indicating that there is a high degree of false suitability in the Qinghai-Tibet Plateau and other areas.

[0060] Multiply the "elevation constraint layer" and "hydroclimate constraint layer" generated in step one with the initial habitat suitability prediction layer obtained in step two in the raster calculator.

[0061] The formula is: .

[0062] Based on this calculation, the suitability values ​​of all pixels in the initial habitat suitability prediction layer located above 2000 meters in altitude (such as the Qinghai-Tibet Plateau) or outside the hydro-climatic ecological niche (such as the humid South China region and the extremely arid Northwest region) were corrected to 0 (corresponding to...). Figure 2 (B). A map showing the distribution of habitat suitability under cover at elevations of 2,000 meters and below, revealing that highly suitable areas (red–orange) are concentrated in the Ordos Plateau, Ningxia, and western Inner Mongolia, coinciding with the 400 mm isohyet (green) and the monsoon–non-monsoon boundary (blue). The revised map shows that highly suitable areas are precisely defined within the inland salt lake clusters of Inner Mongolia and Hebei.

[0063] Step 3: Model Validation and Output

[0064] Using independent observational data of Relict Gulls not involved in the modeling, the AUC, TSS, and Kappa accuracy indices of the predicted maps before and after correction were calculated. Validation results show that all accuracy indices of the corrected model are significantly improved, and the predicted core habitats highly match the key breeding grounds confirmed by field surveys. Finally, this ecologically constrained, high-precision potential distribution map of Relict Gulls is output to guide habitat protection and restoration projects.

Claims

1. A species distribution prediction method integrating "habitat tolerance limit", characterized in that... The species distribution prediction method that incorporates "habitat tolerance limit" is as follows: I. Quantification of Species' Environmental Tolerance Limits and Generation of Spatial Constraint Layers For the target species, based on its historical distribution data, its environmental tolerance limit parameters are quantitatively extracted and defined, including the upper limit of the suitable habitat altitude and the hydro-climatic ecological niche, to generate a nationwide binary raster layer, namely the elevation and regional constraint layer. The quantitative extraction method involves extracting the corresponding elevation values ​​from all historical distribution points within the suitable habitat range of the species, determining the upper limit threshold of its distribution altitude through statistical analysis, analyzing its distribution and boundaries, and defining its suitable hydro-climatic ecological niche. Spatial constraint layer generation: The upper limit of altitude and the boundary are generated as the spatial constraint layer in the geographic information system. The spatial constraint layer is an elevation constraint layer and a hydro-climate constraint layer. In the elevation constraint layer, areas with an altitude of less than or equal to 2000 meters are assigned a value of 1, which is allowed to pass, and areas with an altitude greater than 2000 meters are assigned a value of 0, which needs to be blocked. In the hydro-climate constraint layer, the area located in the intersection area defined by the 400 mm isohyet and the monsoon boundary is assigned a value of 1, and the area outside is assigned a value of 0. II. Prediction of Initial Species Distribution in Machine Learning Based on Tolerance Limit Constraints Acquire distribution point data and multi-dimensional environmental variable data of the target species; use recursive features to screen core environmental driving factors; use random forest algorithm and XGBoost algorithm to train species distribution model and generate initial, unconstrained species habitat suitability prediction layer; The two or more spatial constraint layers generated in step one are used as spatial filters to perform pixel-by-pixel logical operations on the species habitat suitability prediction layer generated in step two. The corrected formula is: Final fitness value = Initial fitness value × Spatial constraint layer 1 × Spatial constraint layer 2 × … × Spatial constraint layer N Where N≥2, the value of the spatial constraint layer is 0 or 1. Through this multiplication operation, the region that is outside any tolerance limit in the initial prediction, that is, the corresponding spatial constraint layer value is 0, is forcibly corrected to 0 in terms of suitability and thus excluded in space. The region that is within all tolerance limits, that is, all spatial constraint layer values ​​are 1, retains its initial predicted suitability value. III. Model Validation and Output Using independent species distribution validation data, the accuracy metrics of the prediction layers before and after correction are compared, and the final species distribution prediction map after tolerance limit correction is output.

2. The species distribution prediction method integrating "habitat tolerance limit" as described in claim 1, characterized in that... The statistical analysis described in step one includes kernel density estimation and frequency distribution histogram.

3. The species distribution prediction method integrating "habitat tolerance limit" according to claim 1, characterized in that... The upper limit threshold mentioned in step one is an altitude of 2000 meters.

4. The method for predicting the distribution of Relict Gull species by incorporating "habitat tolerance limit" as described in claim 1, characterized in that... The suitable hydro-climatic ecological niche mentioned in step one is the intersection area defined by the 400 mm annual precipitation isohyet and the monsoon line.

5. The species distribution prediction method incorporating "habitat tolerance limit" as described in claim 1, characterized in that... The multi-dimensional environmental variable data mentioned in step two include normalized vegetation index, normalized water index, enhanced vegetation index, annual precipitation, annual average temperature, diurnal temperature range, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, precipitation of the wettest month, precipitation of the driest month, precipitation seasonality, relative humidity, surface temperature, digital elevation model, topographic humidity index, land use / land cover, population density, nighttime light index, GDP, road density, distance from water body, and distance from road.

6. The species distribution prediction method incorporating "habitat tolerance limit" as described in claim 1, characterized in that... The accuracy metrics mentioned in step three are AUC, TSS, and Kappa.

7. The species distribution prediction method incorporating "habitat tolerance limit" as described in claim 1, characterized in that... The species described is the Relict Gull.

Images