Developing a Spatiotemporal Habitat Suitability Model to Assess Biodiversity Changes in Response to Climate Change

Introduction: Biodiversity loss is a global threat to humanity. To address these challenges, international environmental organizations have adopted specific strategic goals and plans within the Kunming-Montreal Global Biodiversity Framework to understand how biodiversity changes over time and to identify the factors influencing these changes. Biodiversity modeling tools, particularly habitat suitability models (species distribution models), play a critical role in this effort. Despite significant advances in modeling techniques and the growing availability of spatial data, current models face serious limitations in accurately predicting biodiversity status and changes, hindering the development of an effective framework for monitoring biodiversity over time. One of the most significant practical limitations of these models is the inconsistency of recorded data in terms of species presence frequency and spatial extent across different time periods. This inconsistency limits the development of spatiotemporal models necessary for understanding species distribution dynamics over time. The objective of this study is to propose a solution to overcome the inconsistencies in biodiversity data over time and to develop a computational process for spatiotemporal habitat suitability modeling. Subsequently, spatiotemporal models were employed to quantify changes in species distribution. Materials and Methods: In this study, a spatiotemporal biodiversity model was developed using presence data of the Roan antelope (Hippotragus equinus). Long-term species presence data spanning from 1901 to 2020 were sourced from the global biodiversity database GBIF to develop the spatiotemporal models. Additionally, species time series data (abundance data) from the LPI and BioTime databases were used to validate the assessment of biodiversity changes. Climatic data were extracted from the CRU TS database, which was used to generate 19 annual environmental layers. After data cleaning and preparation, and selecting appropriate climatic variables by testing for multicollinearity, the time series data were integrated into a single data table or data pool. In this approach, species presence data for each year were linked to the corresponding climatic data for that year and location. To improve model efficiency and reduce uncertainty, 10 common machine learning algorithms were selected to calibrate the spatiotemporal models. After model validation, spatial distribution predictions for each year were obtained by combining predictions from different models using weighted averaging (ensemble), resulting in a 120-year time series of species distribution predictions. Next, the Sen’s slope estimator function was used to calculate the trend of habitat suitability changes over 120 years for each pixel. Results and Discussion: The results of model validation demonstrated that all modeling approaches performed exceptionally well, with AUC values ranging from 0.926 to 0.996, indicating high predictive accuracy. Analysis of the biodiversity trend maps over time revealed a gradual decline in the probability of species presence in southern latitudes. In contrast, an increase in presence probability was observed in the central African belt, suggesting shifts in species distribution patterns. Further validation of the results was carried out using time series data on species distribution and abundance from BioTime and LPI sources. This validation showed that the model accurately matched real data in 88% and 84% of the cases where habitat suitability had decreased. These findings confirm the high accuracy of the model in predicting both species distribution and changes over time. This strong correlation between model predictions and actual data underscores the effectiveness of the proposed spatiotemporal models in capturing and reflecting real-world biodiversity trends. Conclusion: The proposed solution in this study not only enables spatiotemporal modeling for analyzing species distribution patterns and their changes but also improves the accuracy of ecological niche quantification, enhancing spatial distribution predictions and reducing uncertainty in assessments. This approach addresses temporal data inconsistency challenges by increasing sample size and coverage, allowing optimal use of all available records. This study emphasizes the importance of the temporal dimension in species distribution models, particularly in regions with significant climatic changes, and can assist managers in making conservation decisions aligned with sustainable development goals, biodiversity conservation, and the KM-GBF global framework.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.