Maximizing Nature-based Solutions using Artificial Intelligence to align global biodiversity, climate, and water targets [dataset]

Here, you will find the code and data produced for the manucript titled: "Maximizing Nature-based Solutions using Artificial Intelligence to align global biodiversity, climate, and water targets." You may download the project folder and run the code for Steps 1 and 2 in Python, Step 3 in R, and Step 4 in R and Python. The four main steps of this study are summarized in the flowchart (~/Code – NbS and RL/flowchart.pdf). The file named “code_data_explanation.xlsx” explains each code in detail. Below, we provide a general explanation of each methodological step. Step 1 simulates training data, and Step 2 trains an optimization model based on the training data in Python and the Reinforcement Learning algorithm CAPTAIN. The output trained model is found in: '~/steps1_2_Trained_model/full_monitor_protect_at_once_can.log'. Step 3 creates an empirical environment, i.e., 10X10 km Planning Units spatial fishnet grid in Canada with information about threatened biodiversity, ecological integrity, carbon, water, ecozones, provinces, and land tenure in Canada. The empirical environment is available at: ~/Code – NbS and RL/puInputs/10x10kmGridCan3347.gpkg. All the input data used to create this empirical environment were obtained from public data sources (Supplemental Information Table S1) and must be downloaded or requested (e.g., IUCN and BirdLife species data) from their original sources. However, we provide demonstration data for mammals to show how the Species Habitat Index, a proxy of ecological integrity, was calculated (Steps 3A to 3E). In Steps 4A to 4F, we prepared in R the data inputs for running the Reinforcement Learning algorithm CAPTAIN. These inputs include data displaying the presence of species across 10x10Km Planning Units (PUs) that were suitable for conservation (~/Code – NbS and RL/cpInputsOutputs/pusSpeciesDbCon_V20.csv) or restoration (~/Code – NbS and RL/ cpInputsOutputs/pusSpeciesDbRes_V20.csv) based on Step 3 processes. In turn, these inputs and data from the empirical environment are used for each conservation or restoration scenario. For example, Conservation Scenario 1 (~/Code – NbS and RL/ cpInputsOutputs/conSc1) contains a specific file for species occurrence (puvsp_...V20.csv), costs based on ecological Integrity values (pu_conSc....V20.csv), and the spatial location of 10x10Km Planning Units (Planning_Units_....V20.csv). Steps 4G to 4L use the data produced in Steps 4A to 4F to obtain Conservation and Restoration Priority Scenarios using the Reinforcement Learning agent trained in Step 2. In each scenario, the Reinforcement Agent will maximize species occurrence (puvsp_...V20.csv), while avoiding high costs (pu_conSc....V20.csv) across 10x10Km Planning Units (Planning_Units_....V20.csv'). Additionally, in each scenario, the agent is provided a total sum of costs or a budget to cost-effectively prioritize a desired number of Planning Units to either achieve a conservation or restoration area target. For example, in Conservation Scenario 1, the Reinforcement Learning agent maximizes species occurrence in areas suitable for conservation along with ecological integrity. As the agent is trained to avoid areas with high costs, this environmental variable was rescaled so that high Species Habitat Index values (i.e., a proxy for ecological integrity) represented a low cost. After ten iterations of the scenario, the summary of the areas prioritized by the Reinforcement Learning agent can be found in a file named “pusPriorityConSc1_V20.csv” inside the scenario folder (~/Code – NbS and RL/ cpInputsOutputs/conSc1). The results of this scenario and other scenarios can be visualized in Steps 4M and 4N.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.