Identifying climate-ready plant for urban environment: Integrating machine learning with traditional plant selection tools

Abstract

Climate change has intensified the urban heat island effect and increased extreme weather conditions, posing risks to public health and urban vegetation. To address these challenges, selecting climate-ready urban plant species is crucial. Traditional climate niche-based methods often fall short in urban contexts due to neglecting anthropogenic factors. Our study addresses this research gap by introducing an innovative urban plant selection method that integrates vulnerability metrics, expert consensus, and plant introduction records through machine learning.We identified eight climatic variables essential for the survival of urban plants in Beijing, China, and established safety margins for eight climate variables of 1,070 urban plant species across two periods: the baseline (1981-2010) and the future (2041-2070). Based on existing assessment data, expert consensus and plant introduction records, 247 plant species were classified into three levels of adaptability: backbone (highly adaptable, prevalent in Beijing), general (moderately adaptable, requiring specific care), and maladapted (poorly adaptable). Subsequently, we investigated the dynamic relationship between safety margins for eight climate variables across two time periods and the adaptability levels of plants by constructing an optimal machine learning model to predict urban plant adaptability levels, enhancing its accuracy through model comparisons and hyperparameter tuning.Our findings indicate that nearly half (49.0%) of the plant species in Beijing may face reduced adaptability to future climate conditions. However, a majority (75.9%) perform well under baseline climate conditions and are expected to adapt to future climate conditions. The results reaffirm that the species can grow well out of the niche limit, suggesting that the traditional climate niche-based method may be limited in urban contexts. Our approach overcomes the limitations of binary classification of traditional niche-based methods and the neglect of anthropogenic factors by incorporating an urban plant adaptability classification schema and ground truth derived from expert consensus and records of plant introductions through machine learning methods. This study provides a method for selecting climate-ready plant species for urban environments and supports evidence-based urban forestry management amidst climate change.