Prediction method of plant distribution affected by climate change based on artificial intelligence algorithm

Abstract

The present invention relates to a method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm, and belongs to the technical field of plant distribution prediction. Global plant distribution data is obtained from GBIF, and after cleaning, it is spatially matched with WWF ecological zone vector data, species are associated with ecological zone IDs, woody and herbaceous plants are distinguished, and species names are standardized. The number of species in each ecological zone is counted, and the proportion of woody and herbaceous plants is calculated and visualized; current and future climate data are integrated, the median of climate characteristics of each ecological zone is extracted, and a species-climate relationship model is constructed using traditional linear regression and artificial intelligence algorithms. The effects of temperature, precipitation, human footprint, and temperature seasonality are analyzed, and finally the distribution pattern of two types of plants under future climate scenarios is predicted. The present invention provides a scientific basis for biodiversity conservation and adaptive management under the background of global change through multi-scale data coupling and nonlinear modeling.

Description

Technical Field

[0001] The present invention belongs to the technical field of plant distribution prediction, and in particular relates to a method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm. Background Art

[0002] In recent years, with the intensification of global climate change and the increasingly significant impacts of human activities on ecosystems, species distribution models and related prediction techniques have become a research hotspot in ecology, conservation biology, and environmental science. Future climate models will fundamentally reshape the thermal baseline through increases in annual mean temperature, while amplifying seasonal climate extremes and interacting with increasingly severe human disturbance patterns. Currently, the stability of terrestrial ecosystems primarily depends on the functional balance between woody and herbaceous plants, but the key climate drivers that maintain this relationship between woody and herbaceous plants remain unclear. As one of the two major contrasting plant systems in terrestrial ecosystems, woody and herbaceous plants have significant differences in global biogeographic patterns and climate drivers. In this context, it is crucial to understand how climate change shapes the biogeographic patterns of woody and herbaceous plant species worldwide.

[0003] Currently, this field has made significant progress in data sources, model methods, and computing capabilities, but it still faces the following challenges:

[0004] 1. Addressing the issue of insufficient taxonomic coverage.

[0005] In the past, vascular plants were often considered a single group, ignoring the distinction between woody and herbaceous species. In this application example, a comprehensive plant species database covering both woody and herbaceous plants was established by adjusting and expanding the existing framework. Based on a global database of relevant species, extensive literature was searched to expand and improve the species database, capture underrepresented species and regions, and increase data coverage. Ultimately, the growth morphology dataset covered more than 349,000 woody and herbaceous plant species.

[0006] Second, plant species were classified based on functional traits from eight different plant trait databases, supplemented by a literature review. Species were categorized into two primary growth forms: woody and herbaceous. Species with prominent aboveground stems that persisted under diverse environmental conditions were defined as woody.

[0007] 2. Address the issue of conflicts in growth morphological classification that have not been systematically addressed.

[0008] To address the above issues, the present embodiment matches growth morphology data with species observations, which are sourced from the Global Biodiversity Information Facility (GBIF). All available plant kingdom observation data points (~498 million records) were downloaded and processed locally. Unreliable data were filtered using the R package CoordinateCleaner (v.3.0.1). After cleaning, species occurrence data were matched to the growth morphology database. Of the 349,145 species in the growth morphology database, 235,336 species were identified in the cleaned GBIF dataset. The data were analyzed using a 1-degree longitude × 1-degree latitude grid system to assess consistency across different spatial resolutions.

[0009] 3. Address the lack of quantitative integration of climate unpredictability and human factors.

[0010] To quantify climate unpredictability, we calculated interannual precipitation variability (IPV) and interannual temperature variability (ITV) as additional predictors. IPV is calculated by calculating the coefficient of variation of annual precipitation over a 30-year period (1970-2000), while ITV is calculated by calculating the standard deviation of the mean annual temperature.

[0011] We explored the associations between different climatic and anthropogenic factors and the distribution of plant growth morphological strategies. To assess the prevalence of these strategies, we calculated the proportion of species exhibiting a particular trait relative to the total number of species within each region (for example, the proportion of woody plants versus herbaceous plants relative to all plant species). Each region was then characterized using a set of climatic and anthropogenic variables.

[0012] To improve prediction accuracy and explore the complex interactions between predictors, this paper uses machine learning techniques, including random forest, extreme gradient boosting decision tree (XGBoost), and neural network. Model performance is measured by mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient ( R² ) indicators. In comparison, linear regression was used as a baseline model to evaluate the performance improvements achieved by machine learning techniques. This also revealed nonlinear relationships between climate and anthropogenic factors and the proportions of woody and herbaceous plants. Summary of the Invention

[0013] In order to solve the above technical problems, the present invention proposes a method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm, comprising:

[0014] Step S1: Obtain all plant geographical distribution data;

[0015] Step S2: Filter all plant distribution data to generate a valid new data set;

[0016] Step S3: Match the valid new dataset with the global terrestrial ecological zoning vector data according to latitude and longitude coordinates, match the ecoregion ID for each coordinate point, and finally obtain the ecoregion distribution data containing the "species" and "ECO_ID" columns;

[0017] Step S4: Obtain the species names of woody, herbaceous, and all other plants respectively, standardize the species names, and save them as csv data containing a "species" column;

[0018] Step S5: Match the data after the standardized species name with the ecological zone distribution data, obtain the ecological zone distribution data filtered by the "species" column, and save it as woody plant and herbaceous plant distribution data, which contains the "species" and "ECO_ID" columns;

[0019] Step S6: Calculate the number of woody, herbaceous, and all other plant species in each ecological zone;

[0020] Step S7: Based on the obtained number of species in each ecoregion, the ratio of woody plants to herbaceous plants in each ecoregion is calculated, and the “ECO_ID” column is matched with the global terrestrial ecoregion vector data to generate a species ratio distribution map based on the global terrestrial ecoregion system;

[0021] Step S8: Obtain current bioclimatic characteristics and future climate characteristics, match the current and future climate characteristics to global terrestrial ecological zones according to longitude and latitude, and take the median value of each climate characteristic in each ecological zone as the climate characteristic value of the ecological zone;

[0022] Step S9: After combining the climate characteristic values ​​in pairs, the species-climate relationship is constructed using four models: linear regression model, random forest model, extreme gradient boosting decision tree model, and neural network model, respectively, with the woody plant proportion and the herbaceous plant proportion;

[0023] Step S10: Selecting a species-climate relationship model to analyze the effects of temperature and precipitation under optimal conditions on the two plants, and simultaneously adding nonlinear data of human footprint and temperature seasonality to the optimal conditions to analyze the effects of human footprint and temperature seasonality on woody and herbaceous plants at different levels;

[0024] Step S11: The allocated future climate data is transferred to the species-climate relationship model to predict the future distribution of herbaceous plants and woody plants, and visualize the predictions.

[0025] Furthermore, step S3 includes:

[0026] S31. Input observation point data and ecological region data;

[0027] S32, converting the latitude and longitude in the observation point data into point data in geographic space, and constructing a spatial data framework based on the point data and other attribute data;

[0028] S33. Project the spatial data framework onto the ecoregion, retain the intersection point data, associate each observation point with the ecoregion where it is located, and add the corresponding ECO_ID for each observation point.

[0029] Furthermore, in step S33, an inner join mode is adopted to retain only the observation points that are successfully matched to the ecological zone, and the unmatched points will be automatically filtered out; ensure that the observation points and the ecological zone use the same coordinate system; delete the geometry column and temporary index column generated during the spatial calculation process, and only retain the original observation attributes and ecological zone number.

[0030] Furthermore, the step S4 includes:

[0031] S41. Separate woody plants and herbaceous plants from the original data based on the presence or absence of xylem and growth morphology in the functional trait database, and retain a list of all plant names;

[0032] S42, performing standardization on the plant name list, and retaining the matched standard name list;

[0033] S43: Taxonomic standardization of the standard name list was performed, and plant names were converted to standard accepted names by matching against the WCVP database.

[0034] Furthermore, in step S7, the ratio of woody plants to herbaceous plants in each ecological zone is calculated by taking the number of all plant species in the ecological zone, Plant_Rich, as the denominator and the number of woody plants, Woody_Rich, and the number of herbaceous plants, Herb_Rich, as the numerators:

[0035] Woody_Proportion:

[0036] Woody_Proportion = Woody_Rich / Plant_Rich;

[0037] Herb_Proportion:

[0038] Herb_Proportion = Herb_Rich / Plant_Rich.

[0039] Furthermore, in step S9, the model constructs a species-climate relationship model, and the multiple bioclimatic characteristic values ​​are combined in pairs and then respectively constructed with the woody plant ratio and the herbaceous plant ratio. The specific steps are as follows:

[0040] (A) Multiple bioclimatic characteristic values ​​are combined in pairs to form multiple variable pairs;

[0041] (B) Multiple variables were paired with herbaceous and woody plant proportions to construct multiple species-climate relationship models;

[0042] (C) The optimal model was selected based on the correlation coefficient (R²), root mean square error (RMSE), and mean absolute error (MAE) of multiple species-climate relationship models;

[0043] (D) Determine the climate characteristics that affect the distribution of herbaceous and woody plants based on the optimal model.

[0044] Furthermore, in step S10, the formula of the model is:

[0045] ;

[0046] in: is the intercept term, , represent the main effect coefficients of the mean temperature of the coldest quarter, BioClim_11, and the precipitation of the warmest quarter, BioClim_18; represents the interaction effect coefficient between BioClim_11 and BioClim_18; , Respectively represent the linear and quadratic coefficients of the human footprint factor Human_Footprint, is the error term that follows a normal distribution, ~N(0,σ 2 ).

[0047] Furthermore, the step S11 includes:

[0048] (a) Matching future climate data to global terrestrial ecological zones;

[0049] (b) Incorporating future climate characteristics into the optimal species-climate relationship model as dependent variables to obtain the predicted proportions of woody and herbaceous plants in each ecological zone and their distribution changes under future climate conditions;

[0050] (c) Generate a map of the future distribution ratios of woody and herbaceous plants based on the global terrestrial ecological zoning system, showing the changing trends in the ratios of woody and herbaceous plants in different ecological zones.

[0051] The beneficial effects of the present invention are:

[0052] On the one hand, this application example compiles the largest plant growth morphology dataset available, covering over 349,000 woody and herbaceous plant species, 96% of which are included in the World Forestry Organization (WFO), and geolocation data is provided for 91% of them. This massive global dataset enables in-depth exploration of the mechanisms that shape the unique biogeographic patterns of woody and herbaceous plants.

[0053] Secondly, we used advanced machine learning techniques, including random forest models, extreme gradient boosting decision trees and neural networks, as well as traditional statistical methods (linear regression models), to simulate the complex interactions between environmental variables. We tested the current global biogeographic patterns of woody and herbaceous plants and the related hypotheses of 19 bioclimatic drivers, and compared precipitation with temperature, interannual changes with seasonal and even monthly changes. By evaluating the performance of each model, we found that the global biogeographic patterns of woody and herbaceous plants were closely related to the biogeographic patterns of 19 bioclimatic drivers. R² (correlation coefficient), RMSE (root mean square error), and MAE (mean absolute error) were used to select the model with the best fitting effect to predict how the two plant communities will respond to future climate change.

[0054] Third, because woody and herbaceous species are expected to respond differently to different fluctuations, this approach assesses a range of bioclimatic variables by incorporating extreme environmental factors (seasonal or monthly extremes in temperature and precipitation) and human-environment interactions. This approach advances vegetation modeling by capturing threshold responses to seasonal extremes and extends it to include future scenarios dominated by woody and herbaceous species under climate change. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Proportional biogeographic patterns of woody and herbaceous growth in ecoregions across parts of South America, Africa, and Asia.

[0056] Figure 2 Projected climate-mediated changes in the proportions of plant growth forms by 2060 for ecoregions in South America, Africa, and parts of Asia. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] The present invention provides a method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm, comprising the following steps:

[0059] Step S1: Obtain all plant geographic distribution data from GBIF.

[0060] GBIF is an open source platform that provides global biodiversity data. It stores a large amount of information on species distribution, classification, specimens, etc. All plant distribution data are obtained from GBIF using the rgbif package, the official toolkit for accessing GBIF data, according to the taxonKey, a digital code that identifies species taxonomic units in the biological classification Plantae.

[0061] Step S2: Filter all plant distribution data to generate a valid new data set.

[0062] All the acquired plant distribution data are filtered using the CoordinateCleaner package according to "taxonRank", "decimalLatitude", "decimalLongitude", "coordinateUncertaintyInMeters", "year", and "basisOfRecord" to generate a valid new dataset. The specific steps are as follows:

[0063] S21. Exclude data points without longitude and latitude coordinates;

[0064] S22, filter records with coordinate uncertainty greater than 100 km;

[0065] S23, exclude literature annotation and living specimen annotation records;

[0066] S24. Exclude records from 1945 and before;

[0067] S25. Filter non-species-level or subspecies-level records;

[0068] S26. After filtering is completed, extract "species", "decimalLatitude" and "decimalLongitude" and store them as a valid new data set.

[0069] Step S3: Match the generated valid new dataset with the global terrestrial ecological zoning vector data obtained from WWF according to longitude and latitude coordinates, spatially associate the CSV data containing longitude and latitude with the ecological region, match each coordinate point with the ecological region ID, and finally obtain the plant ecological zone distribution data containing the "species" and "ECO_ID" columns.

[0070] The new dataset is mapped to the corresponding ecoregion using decimal latitude and longitude data. The CSV data containing longitude and latitude are spatially associated with the ecoregion, and each coordinate point is matched to its ecoregion ID (ECO_ID). The core goal is to add spatial attributes to the point data for subsequent analysis. This includes:

[0071] S31. Input observation point data and ecological region data.

[0072] Observation point data: stored in CSV format, containing latitude and longitude fields, recording species distribution or other geographic observation information.

[0073] Ecological region data: stored in Shapefile format, containing the ecological region boundary (polygon data) and the corresponding unique identifier ECO_ID.

[0074] Observation point data is loaded as structured data using the table reading tool.

[0075] Ecoregion data were loaded through the geospatial data processing library, and only the ecoregion number and geometric boundary columns were retained to reduce redundant information.

[0076] S32. Convert the longitude and latitude in the observation point data into point data in geographic space, and construct a spatial data framework based on the point data and other attribute data.

[0077] Converts the latitude and longitude values ​​in a CSV file to geographic point data, where each point represents an observation location. Explicitly defining the coordinate system ensures accurate subsequent spatial calculations.

[0078] Combine point data with other attributes in CSV to form a spatial data frame, supporting geographic operations.

[0079] S33. Project the spatial data frame onto the ecoregion, retain the intersection point data, associate each observation point with the ecoregion where it is located, and add the corresponding ECO_ID for each observation point.

[0080] Based on the intersection of points and polygons, the spatial data frame is projected onto the ecoregion, retaining the intersection point data and associating each observation point with the ecoregion it is in. If a point is within the boundary of an ecoregion, the corresponding ECO_ID is added to the point.

[0081] Using the “inner join” mode, only observation points that are successfully matched to ecological zones are retained, and unmatched points will be automatically filtered out.

[0082] Ensure that the observation points and the ecoregion use the same coordinate system. If the ecoregion data is in a different coordinate system (such as a projected coordinate system), it needs to be dynamically converted to the same coordinate system as the point data to avoid spatial misalignment.

[0083] Delete the geometry columns (point coordinates, surface boundaries) and temporary index columns generated during the spatial calculation process, and only keep the original observation attributes and ecoregion numbers.

[0084] The final results are exported as a new CSV file in the form of pure tabular data, which is convenient for subsequent statistical analysis or importing into a database.

[0085] Step S4: Obtain the species names of woody plants, herbaceous plants, and all other plants respectively, and standardize the species names. Save the standardized data of woody and herbaceous plant names as CSV data containing a "species" column.

[0086] The species names of woody, herbaceous, and all other plants were obtained from multiple functional trait databases and literature reviews, and the species names were standardized using the U.Taxonstand package in R. The standardized data of woody and herbaceous plant names were saved as csv data containing a "species" column. The specific steps are as follows:

[0087] S41: Separate woody plants and herbaceous plants from the original data based on the presence or absence of xylem and growth morphology in the functional trait database, and retain a list of all plant names.

[0088] Input data: A tab-delimited CSV file (woody+hurb_new.csv) containing a field for plant names (name column).

[0089] Data Reading: Use data.table::fread to efficiently read large datasets.

[0090] Encoding conversion: Convert the character encoding of the name column from latin1 to UTF-8 to resolve the garbled encoding problem of special characters (such as diacritics and non-English characters).

[0091] S42: Standardize the plant name list and retain the matched standard name list.

[0092] Call the nameMatch_WCVP function in U.Taxonstand to perform name matching against the WCVP database. Set author = FALSE to ignore author information and match only species and genus names. max.distance = 1 allows a maximum edit distance of 1, improving tolerance to spelling errors. matchFirst = TRUE causes the function to prioritize the first matching result, which is suitable for improving efficiency when dealing with large amounts of data. Append = FALSE prevents appending the original name to the result, and outputs only a list of the matched standard names.

[0093] The U.Taxonstand package was used to standardize the obtained species names using the WCVP (World Checklist of Vascular Plants) rules, including matching accepted names and deleting unmatched species names.

[0094] Step S43: taxonomically standardize the standard name list and convert the plant names into standard accepted names by matching the WCVP database.

[0095] Write the matching results to a CSV file (res_woody.csv), which contains the standard name, the matching status (such as whether it is ambiguous (TURE or FALSE)), and possible error information.

[0096] The plant name list is taxonomically standardized. By matching the WCVP (World Register of Vascular Plants) database, the original plant names are converted into standard accepted names to resolve synonyms, spelling errors or inconsistent naming, ensuring that the data conforms to authoritative classification standards.

[0097] Step S5: Match the data after the standardized species names of woody and herbaceous plants with the plant ecological zone distribution data, obtain the plant ecological zone distribution data filtered according to the "species" column, and save it as woody plant and herbaceous plant distribution data, which contains the "species" and "ECO_ID" columns.

[0098] Step S6: Calculate the number of woody, herbaceous, and all plant species in each ecological zone.

[0099] The standard for counting the number of species in each ecological zone is that each species is recorded as existing in an ecological zone only if it appears ≥5 times in the ecological zone.

[0100] S61: Initialize settings.

[0101] Input data: CSV file (herb_cell_200.csv) containing at least three columns:

[0102] ECO_ID: spatial unit identifier (such as grid number).

[0103] species: species name or code.

[0104] Other possible environment variables or observable properties (not used in the script).

[0105] Output configuration:

[0106] Define the output directory path (output_dir) and automatically create missing directories (support recursive creation).

[0107] Set the result summary file name (result_file) to save global statistical information.

[0108] S62: Data is processed step by step.

[0109] 1) Count species frequencies by spatial unit: Use data.table syntax to group by ECO_ID (spatial unit) and species (species), and count the number of times each species occurs within the unit (Freq column).

[0110] Output structure: Generates an intermediate table containing three columns: id, species, and Freq.

[0111] 2) Screening for high-frequency species: retain records of species with an occurrence frequency ≥ 5 times (Freq>= 5).

[0112] Eliminate occasional species or record noise, and focus on dominant species or species with stable distribution.

[0113] 3) Calculate and output species richness by unit: Calculate species richness: For each ECO_ID, count the number of unique species (uniqueN(species)).

[0114] The file name is dynamically generated: the format is {ECO_ID}_{num_species}.csv (for example, A12_25.csv means that the cell with ID A12 contains 25 species).

[0115] Save subfile: Write the species and their frequencies under the current ECO_ID into a CSV file and store it in the output directory.

[0116] Return metadata: record the file name and species number corresponding to each ECO_ID for subsequent summary.

[0117] 4) Generate a global statistics summary table:

[0118] Organize the metadata generated in the previous step and generate a summary table containing two columns:

[0119] ECO_ID: File name (implied spatial unit ID).

[0120] Woody_Rich: Species richness (i.e., num_species).

[0121] Write the summary table to h_cell_200.csv and store it in the output directory.

[0122] S63: Result verification.

[0123] List all CSV file paths in the output directory and print them to the console.

[0124] Quickly verify whether the file generation is successful and confirm that the number of output files is consistent with the number of spatial units.

[0125] Step S7: Based on the obtained number of species in the ecological zone, the number of all plant species in the ecological zone, Plant_Rich, is used as the denominator, and the number of woody plants, Woody_Rich, and the number of herbaceous plants, Herb_Rich, are used as the numerators to calculate the ratio of woody plants to herbaceous plants in each ecological zone: Woody_Proportion and Herb_Proportion.

[0126] Woody_Proportion = Woody_Rich / Plant_Rich;

[0127] Herb_Proportion = Herb_Rich / Plant_Rich,

[0128] According to the correspondence between the "ECO_ID" column and the global terrestrial ecological zoning vector data, the data is visualized in the form of a heat map, and finally a species proportion distribution map based on the global terrestrial ecological zoning system is generated.

[0129] Figure 1 Proportional biogeographic patterns of woody and herbaceous growth in ecoregions across South America, Africa, and parts of Asia. Figure 1 The longitude and latitude of the land area in Central Africa range from 4°S to 11°N, 13°E to 33°E; the longitude and latitude of the land area in Asia range from 27°N to 39°N, 93°E to 113°E; and the longitude and latitude of the land area in South America range from 3°S to 20°S, 69°W to 46°W. The geographic distribution of woody (A) and herbaceous (C) species (as a percentage of the total number of species) reveals a latitudinal gradient of decreasing woody representation, while herbaceous species exhibit the opposite trend. The analysis was based on ecoregions containing at least 10 species, encompassing 810 ecoregions. Frequency histograms of woody (B) and herbaceous (D) species across ecoregions highlight the asymmetric dominance patterns between these functionally opposing plant groups.

[0130] Step S8: Obtain the current bioclimatic characteristics of the required resolution and the future climate characteristics of the required scenario and resolution, match the current and future climate characteristics to the global terrestrial ecological zones according to longitude and latitude, and take the median value of each climate characteristic in each ecological zone as the climate characteristic value of the zone.

[0131] Obtain current bioclimatic characteristics at the desired resolution and future climate characteristics for the desired scenario and resolution from WorldClim. This example analysis uses 10-minute data from 1970 to 2000 as current climate characteristics and 10-minute data from ssp585 under the carbon emission scenario from 2041 to 2060 as future climate characteristics. Use other R packages, such as raster, to map current and future climate characteristics to global terrestrial ecological zones based on longitude and latitude. The median value of each climate characteristic in each ecological zone is used as the climate characteristic value for that zone.

[0132] Climate characteristics are obtained through WorldClim. The resolution of climate characteristics is 10 arc minutes. The median value of climate characteristics in the ecoregion is used to represent the climate characteristic value. The climate characteristic values ​​include the following 19 categories:

[0133] BioClim_1 = mean annual temperature;

[0134] BioClim_2 = mean diurnal range (monthly mean (maximum temperature - minimum temperature));

[0135] BioClim_3 = isothermality ((BIO2 / BIO7) × 100);

[0136] BioClim_4 = temperature seasonality (standard deviation × 100);

[0137] BioClim_5 = maximum temperature of the warmest month;

[0138] BioClim_6 = minimum temperature of the coldest month;

[0139] BioClim_7 = temperature annual range (BIO5-BIO6);

[0140] BioClim_8 = mean temperature of the wettest quarter;

[0141] BioClim_9 = mean temperature of the driest quarter;

[0142] BioClim_10 = mean temperature of the warmest quarter;

[0143] BioClim_11 = mean temperature of the coldest quarter;

[0144] BioClim_12 = annual precipitation;

[0145] BioClim_13 = precipitation in the wettest month;

[0146] BioClim_14 = precipitation in the driest month;

[0147] BioClim_15 = precipitation seasonality (coefficient of variation);

[0148] BioClim_16 = wettest season precipitation;

[0149] BioClim_17 = precipitation in the driest quarter;

[0150] BioClim_18 = precipitation in the warmest quarter;

[0151] BioClim_19 = precipitation in the coldest quarter;

[0152] The process is:

[0153] S81: Global configuration and initialization.

[0154] S811, Path setting:

[0155] Define the bioclimatic raster data storage path (base_path) and the ecoregion vector file path (shp_path).

[0156] The output directory is automatically created: BIO_in_eco: to store the bioclimatic statistical results within the ecological region.

[0157] BioClim: Stores the final csv file.

[0158] Exception handling: If directory creation fails, terminate immediately and prompt a permission error.

[0159] S812, Dependency Loading:

[0160] Core libraries: raster (raster), sf (vector), exactextractr (precise extraction), data.table (efficient writing), progress (progress bar).

[0161] Verification: Checks whether all packages are installed, and reports errors if they are missing.

[0162] S82: Ecoregion data loading and verification.

[0163] Use sf::st_read to load ecoregion vector data, using silent mode to avoid redundant output.

[0164] Verify the data loading results and output key metadata, including: number of spatial units and coordinate system information.

[0165] If the file is damaged or the path is incorrect, the script will be terminated immediately and a specific error message will be prompted to prevent subsequent operations from being executed ineffectively.

[0166] S83: Core processing function (variable processing).

[0167] Dynamic path construction and checking: Based on the bioclimatic variable number (e.g. BIO1 to BIO19), the raster file path (BIO{number}.tif) and the result CSV path are dynamically generated.

[0168] Pre-check whether the raster file exists. If it is missing, skip the processing and record the log to avoid interruption of the overall process due to missing files.

[0169] Raster data projection alignment:

[0170] After loading the raster data, check that its coordinate system is consistent with the ecoregion data.

[0171] If there is any inconsistency, use projectRaster for dynamic reprojection (bilinear interpolation) to ensure the accuracy of spatial overlay analysis.

[0172] Accurate raster value extraction:

[0173] Call exactextractr::exact_extract to calculate the area-weighted average of the raster values ​​within each cell using the ecoregion boundaries as a mask.

[0174] Weight selection: Based on pixel area (weights = 'area'), reflecting the actual contribution of different raster pixels to the ecoregion.

[0175] Numerical processing: The result is retained to 9 decimal places, and NaN is replaced with NA to ensure data cleanliness.

[0176] Result integration and output:

[0177] Add the extracted climate values ​​as a new column (e.g., BioClim_1) to the ecoregion attribute table and remove the geometry column to produce a pure attribute table.

[0178] After filtering out invalid values ​​(NA), use data.table::fwrite to efficiently write to the CSV file, reducing memory usage.

[0179] Memory optimization: Explicitly release temporary objects and trigger garbage collection (GC) to avoid the risk of memory overflow under large data volumes.

[0180] Exception capture and logging:

[0181] Encapsulate the processing logic in the tryCatch block, catch errors in single variable processing, record the failed variable number, and do not affect subsequent variable processing.

[0182] S84: Batch processing and schedule management.

[0183] Loop execution: traverse BioClim_1 to BioClim_19 and call the core processing function in sequence.

[0184] Progress feedback: Use the progress package to generate a progress bar to display the processing progress, elapsed time, and remaining time estimate in real time.

[0185] Result statistics: record the number of successful and failed variables, and output the summary to the console.

[0186] S85: Processing reports and error troubleshooting.

[0187] Summary output:

[0188] Count the percentage of successfully processed variables and list the numbers of failed variables.

[0189] Provide targeted troubleshooting suggestions (such as file existence check, spatial overlap verification, and coordinate system consistency confirmation).

[0190] Path Tips:

[0191] Clearly specify the output file storage location (BioClim directory) to facilitate users to quickly locate results.

[0192] Step S9: Multiple bioclimatic characteristic values ​​are combined in pairs and then respectively combined with the woody plant proportion and the herbaceous plant proportion to construct a species-climate relationship model.

[0193] Traditional linear models and machine learning models were used to construct species-climate relationship models. The 19 bioclimatic characteristic values ​​were combined in pairs and then respectively constructed with the proportions of woody plants and herbaceous plants. This method can not only analyze how temperature and precipitation factors affect woody and herbaceous plants, but also analyze from multiple time scales (interannual changes and seasonal changes). The model with the best fitting effect was selected as the optimal model based on the R² (correlation coefficient), RMSE (root mean square error), and MAE (mean absolute error) of the model results.

[0194] The specific steps for constructing a species-climate relationship model to analyze the relationship between the proportion of herbaceous plants, the proportion of woody plants and bioclimatic characteristic factors are as follows:

[0195] (A) Bioclimatic variables were combined in pairs to form 171 variable pairs (e.g., BIO1 + BIO2);

[0196] (B) 171 variable pairs were compared with the proportions of herbaceous and woody plants to construct species-climate relationship models according to the four basic models adopted;

[0197] (C) Based on species-climate relationship models R²(correlation coefficient, the larger the better), RMSE (root mean square error, the smaller the better), and MAE (mean absolute error, the smaller the better) were used to screen the optimal model; as shown in Table 1, this is an analysis of woody / herbaceous plant species models in global terrestrial ecological zones.

[0198] The above process is completed using R language, and its main steps are as follows (taking the linear regression model as an example):

[0199] 1. Load the library and initialize the environment

[0200] Library loading: Use libraries such as caret (modeling), Metrics (evaluation), data.table (efficient data processing), glue (string operations), ggplot2 (drawing), etc.

[0201] Constant definition: set random seed (SEED=123), output directory path and dataset path.

[0202] Environment initialization: Make sure the output directory exists (dir.create()).

[0203] 2. Data loading and preprocessing

[0204] Function load_data():

[0205] 1) Use fread() to efficiently read CSV data and remove missing values ​​(na.omit()).

[0206] 2) Convert bioclimatic variables (BioClim_*) to numeric types.

[0207] 3) Divide the training set / test set (80% training, 20% testing, createDataPartition()).

[0208] 3. Model training function

[0209] Function train_lm():

[0210] 1) Use caret::train(method="lm") to train a linear model.

[0211] 2) 10-fold cross validation (method="cv", number=10), with parallel computing enabled (allowParallel=TRUE).

[0212] 4. Performance Evaluation and Visualization

[0213] Function generate_perf_plot():

[0214] 1) Calculate MAE, RMSE, and R² from the test set predictions.

[0215] 2) Create a histogram: normalize the three indicators (Scaled = indicator value / R²) and label the original value.

[0216] 3) The title contains the current variable combination (glue("Linear Model Evaluation - {comb}")).

[0217] 5. Main process: variable combination traversal

[0218] Key operations:

[0219] 1) Extract 19 bioclimatic variables (bio_vars) from the data.

[0220] 2) Generate all pairwise combinations (combn(bio_vars, 2), a total of 171 groups).

[0221] 3) Parallelization settings: Use doParallel to enable multi-core computing acceleration.

[0222] Loop through each combination:

[0223] 1) Construct a formula (e.g., Woody_Proportion ~ BioClim_1 + BioClim_2).

[0224] 2) Train the model and catch errors (tryCatch).

[0225] 3) Calculate the test set performance indicators:

[0226]

[0227]

[0228] ,

[0229] 4) Save the results: combination name, MAE, RMSE, R².

[0230] Table 1 Model analysis of woody / herbaceous plant species in global terrestrial ecoregions

[0231]

[0232] (D) Based on the numerical values ​​and species-climate relationship model in (C), determine the climate characteristic factors that affect the distribution of herbaceous and woody plants.

[0233] When subsequently predicting the species ratio under future climate conditions, the future data of climate characteristic factors that affect the distribution of herbaceous plants and woody plants are used as dependent variables to predict the future woody-herb ratio.

[0234] Step S10: Select a species-climate relationship model to analyze the effects of temperature and precipitation under optimal conditions on the two plants. At the same time, nonlinear data of human footprint and temperature seasonality are added to the species-climate relationship model to analyze the effects of human footprint and temperature seasonality on woody and herbaceous plants at different levels.

[0235] When analyzing the impact of changes in climate characteristic factors and human footprint on herbaceous and woody plants, a linear regression model was used to add human footprint and BioClim_4 = temperature seasonality (standard deviation × 100) to the climate conditions of the species-climate relationship model, and analyze their effects on woody and herbaceous plants. For example, the linear regression model written in R language was added to the optimal climate conditions of the linear regression model to analyze the impact of human footprint on woody plants:

[0236] The model fits the woody plant proportion Woody_Proportion with BioClim_11, BioClim_18, and Human_Footprint, taking into account the interaction between BioClim_11 and BioClim_18. The relationship between the square of Human_Footprint and the woody plant proportion Woody_Proportion is then analyzed under the interaction between BioClim_11 and BioClim_18. The formula of the model is:

[0237] ;

[0238] in: is the intercept term, , represent the main effect coefficients of BioClim_11 and BioClim_18, respectively; represents the interaction effect coefficient between BioClim_11 and BioClim_18; , Respectively represent the linear and quadratic coefficients of Human_Footprint, is the error term that follows a normal distribution ( ~N(0,σ 2 )).

[0239] Coefficients in a linear regression model ~ Solved by Ordinary Least Squares (OLS), the following are the specific steps and principles:

[0240] 1. Model expression and matrix form

[0241] The model formula is:

[0242] ;

[0243] in:

[0244] y is the vector of dependent variables (Woody_Proportion).

[0245] X is the design matrix (containing the intercept term, main effects, interaction terms, and polynomial terms).

[0246] is the coefficient vector to be found.

[0247] is the error term.

[0248] 2. Construct the design matrix X

[0249] According to the model

[0250] ,

[0251] The design matrix XX contains the following columns:

[0252] Intercept term: a column with all 1s (corresponding to ).

[0253] Main effects:

[0254] BioClim_11 (corresponding )

[0255] BioClim_18 (corresponding to )

[0256] Interaction term: BioClim_11 × BioClim_18 (corresponding to )

[0257] Polynomial terms: Generate two orthogonal polynomial sequences (corresponding to ).

[0258] 3. Least Squares Method

[0259] Estimated values ​​of the coefficients Calculated by the following formula:

[0260] ;

[0261] in:

[0262] is the transposed matrix of X.

[0263] yes The inverse matrix of (if invertible).

[0264] The results of this process are shown in Table 2, which shows the linear regression analysis of the effect of temperature seasonality (BioClim 4) on the proportion of woody species, that is, the effect of BioClim 4 on woody plants in the ecoregion; Table 3 shows the linear regression analysis of the effect of temperature seasonality (BioClim 4) on the proportion of herbaceous species, that is, the effect of BioClim 4 on herbaceous plants in the ecoregion; Table 4 shows the effect of human footprint on woody species in the ecoregion through multiple linear regression analysis based on ecoregion; Table 5 shows the effect of human footprint on herbaceous species in the ecoregion through multiple linear regression analysis based on ecoregion.

[0265] Table 2 Effects of BioClim 4 on woody plants in the ecological zone

[0266]

[0267] Table 3 Effects of BioClim 4 on herbaceous plants in the ecological zone

[0268]

[0269] Table 4 Impact of human footprint on woody species in ecological regions

[0270]

[0271] Table 5 Impact of human footprint on herbaceous species in ecological regions

[0272]

[0273] Step S11: The allocated future climate data is transferred to the species-climate relationship model to predict the future distribution of herbaceous plants and woody plants, and visualized, such as Figure 2 As shown, from left to right, the ecoregions of South America, Africa, and parts of Asia are predicted to change the proportion of plant growth forms mediated by climate by 2060. Figure 2The latitude and longitude range for the land area of ​​Central Africa is 4°S–11°N, 13°E–33°E; the longitude and latitude range for the land area of ​​Asia is 27°N–39°N, 93°E–113°E; and the longitude and latitude range for the land area of ​​South America is 3°S–20°S, 69°W–46°W. Absolute changes in the proportion of woody (A) and herbaceous (B) plants by 2060 for 780 ecoregions meeting data quality thresholds. Excluded ecoregions (N = 45) are shown in gray. Color intensity indicates the magnitude of the projected change.

[0274] Specific steps for predicting and visualizing the future distribution of woody and herbaceous plants:

[0275] (a) Matching future climate data to global terrestrial ecological zones;

[0276] (b) Incorporating future climate characteristics into the optimal species-climate relationship model as dependent variables to obtain predicted values ​​for the proportions of woody and herbaceous plants in each ecological zone and their distribution changes under future climate conditions;

[0277] (c) Generate a map of the future distribution ratios of woody and herbaceous plants based on the global terrestrial ecological zoning system, showing the changing trends in the ratios of woody and herbaceous plants in different ecological zones.

[0278] Future climate data includes projections from 2041 to 2060 at a 10-minute resolution under various GCM scenarios, including SSP585 from the CMIP6 model. For example, projections based on the available future climate data indicate that woody vegetation will shrink in the tropics while expanding poleward in the temperate zones, while herbaceous species will exhibit the opposite response, increasing in the tropics and decreasing in the temperate zones. An ecoregion analysis indicates that woody plant communities may experience range reductions in 24.6% of the Earth's ecoregions, particularly in major tropical rainforest ecoregions such as the Amazon Basin, the Congo Basin, and the Southeast Asian Rainforest Complex. Conversely, 39.2% of ecoregions, primarily in boreal and temperate biomes, may see woody plant range expansions.

[0279] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. The embodiments should therefore be considered illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A prediction method for the impact of climate change on plant distribution based on artificial intelligence algorithms, characterized by: include: Step S1: Obtain all plant geographical distribution data; Step S2: Filter all plant distribution data to generate a valid new data set; Step S3: Match the valid new dataset with the global terrestrial ecological zoning vector data according to longitude and latitude coordinates, and match the ecoregion ID for each coordinate point to obtain the ecoregion distribution data containing the "species" and "ECO_ID" columns; Step S4: Obtain the species names of woody, herbaceous, and all other plants respectively, standardize the species names, and save them as CSV data containing a "species" column; Step S5: Match the data after the standardized species name with the ecological zone distribution data, obtain the ecological zone distribution data filtered by the "species" column, and save it as woody plant and herbaceous plant distribution data, which contains the "species" and "ECO_ID" columns; Step S6: Calculate the number of woody, herbaceous, and all other plant species in each ecological zone; Step S7: Based on the obtained number of species in each ecoregion, calculate the ratio of woody plants to herbaceous plants in each ecoregion, and generate a species ratio distribution map based on the global terrestrial ecoregion system by matching the "ECO_ID" column with the global terrestrial ecoregion vector data; Step S8: Obtain current bioclimatic characteristics and future climate characteristics, match the current and future climate characteristics to global terrestrial ecological zones according to longitude and latitude, and take the median value of each climate characteristic in each ecological zone as the climate characteristic value of the ecological zone; Step S9: After combining the climate characteristic values ​​in pairs, the species-climate relationship is constructed using four models: linear regression model, random forest model, extreme gradient boosting decision tree model, and neural network model, respectively, with the woody plant proportion and the herbaceous plant proportion; Step S10: Selecting a species-climate relationship model to analyze the effects of temperature and precipitation under optimal conditions on the two plants, and simultaneously adding nonlinear data of human footprint and temperature seasonality to the optimal conditions to analyze the effects of human footprint and temperature seasonality on woody and herbaceous plants at different levels; Step S11: The allocated future climate data is transferred to the species-climate relationship model to predict the future distribution of herbaceous plants and woody plants, and visualize the predictions.

2. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 1, characterized in that: The step S3 comprises: S31. Input observation point data and ecological region data; S32, converting the latitude and longitude in the observation point data into point data in geographic space, and constructing a spatial data framework based on the point data and other attribute data; S33. Project the spatial data framework onto the ecoregion, retain the intersection point data, associate each observation point with the ecoregion where it is located, and add the corresponding ECO_ID for each observation point.

3. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 2, characterized in that: In step S33, an inner join mode is used to retain only observation points that are successfully matched to the ecoregion, and unmatched points are automatically filtered out; it is ensured that the observation points and the ecoregion use the same coordinate system; the geometry columns and temporary index columns generated during the spatial calculation process are deleted, and only the original observation attributes and the ecoregion number are retained.

4. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 1, characterized in that: The step S4 comprises: S41. Separate woody plants and herbaceous plants from the original data based on the presence or absence of xylem and growth morphology in the functional trait database, and retain a list of all plant names; S42, performing standardization on the plant name list, and retaining the matched standard name list; S43: Taxonomic standardization of the standard name list was performed, and plant names were converted to standard accepted names by matching against the WCVP database.

5. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 1, characterized in that: In step S7, the ratio of woody plants to herbaceous plants in each ecological zone is calculated by taking the number of all plant species in the ecological zone, Plant_Rich, as the denominator and the number of woody plants, Woody_Rich, and the number of herbaceous plants, Herb_Rich, as the numerators: Woody_Proportion: Woody_Proportion = Woody_Rich / Plant_Rich; Herb_Proportion: Herb_Proportion = Herb_Rich / Plant_Rich.

6. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 1, characterized in that: In step S9, the model constructs a species-climate relationship model, and the multiple bioclimatic characteristic values ​​are combined in pairs and then respectively combined with the woody plant ratio and the herbaceous plant ratio to construct the model. The specific steps are: (A) Multiple bioclimatic characteristic values ​​are combined in pairs to form multiple variable pairs; (B) Multiple species-climate relationship models were constructed by pairing multiple variables with the proportions of herbaceous plants and woody plants; (C) The optimal model was selected based on the correlation coefficient (R²), root mean square error (RMSE), and mean absolute error (MAE) of multiple species-climate relationship models; (D) Determine the climate characteristics that affect the distribution of herbaceous and woody plants based on the optimal model.

7. The method for predicting the impact of climate change on plant distribution based on artificial intelligence algorithm according to claim 5, characterized in that: In step S10, the formula of the model is: ； in: is the intercept term, , represent the main effect coefficients of the mean temperature of the coldest quarter, BioClim_11, and the precipitation of the warmest quarter, BioClim_18; represents the interaction effect coefficient between BioClim_11 and BioClim_18; , Respectively represent the linear and quadratic coefficients of the human footprint factor Human_Footprint, is the error term that follows a normal distribution, ~N(0,σ 2 ).

8. The method for predicting the impact of climate change on plant distribution based on an artificial intelligence algorithm according to claim 1, characterized in that: The step S11 includes: (a) Matching future climate data to global terrestrial ecological zones; (b) Incorporating future climate characteristics into the optimal species-climate relationship model as dependent variables to obtain the predicted proportions of woody and herbaceous plants in each ecological zone and their distribution changes under future climate conditions; (c) Generate a map of the future distribution ratios of woody and herbaceous plants based on the global terrestrial ecological zoning system, showing the changing trends in the ratios of woody and herbaceous plants in different ecological zones.

Images