Title: Optimising Occurrence Data in Species Distribution Models: Sample Size, Positional Uncertainty, and Sampling Bias Matter
Authors:
Vítězslav Moudrý, Manuele Bazzichetto, Ruben Remelgado, Rodolphe Devillers, Jonathan Lenoir, Rubén G. Mateo, Jonas J. Lembrechts, Neftalí Sillero, Vincent Lecours, Anna F. Cord, Vojtěch Barták, Petr Balej, Duccio Rocchini, Michele Torresani, Salvador Arenas-Castro, Matěj Man, Dominika Prajzlerová, Kateřina Gdulová, Jiří Prošek, Elisa Marchetto, Alejandra Zarzo-Arias, Lukáš Gábor, François Leroy, Matilde Martini, Marco Malavasi, Roberto Cazzolla Gatti, Jan Wild, Petra Šímová
Species distribution models (SDMs) have emerged as powerful tools in ecology, conservation biology, and environmental management. By predicting species occurrences and their associations with environmental variables, SDMs help to fill crucial knowledge gaps about species distributions, especially in regions or periods where data is sparse. Despite their broad applicability, however, SDMs face significant challenges due to limitations in the available species occurrence data, which can lead to inaccuracies and uncertainty in model predictions.
This article explores the key limitations of species occurrence data that impact the performance of SDMs, including sample size, positional uncertainty, and sampling bias. We also examine how species ecology plays a critical role in determining the effectiveness of SDMs and outline recommendations for improving model performance through better data practices.
Limitations in Species Occurrence Data
- Sample Size
SDMs rely heavily on the availability of species occurrence records to generate accurate predictions. When occurrence data is limited, it can result in overfitting or underfitting of the model, reducing its reliability. Small sample sizes restrict the model’s ability to capture the full range of environmental conditions a species might inhabit, leading to erroneous predictions of species-environment relationships. This is especially problematic for rare or elusive species, where data is often scarce. - Positional Uncertainty
Another significant limitation in SDMs is the accuracy of the geographic location data. Occurrence records are often collected using GPS technology, but the resolution of these points can vary greatly. Positional uncertainty arises when the recorded locations of species are imprecise, either due to inaccuracies in the data collection process or because of limited resolution in the mapping technology. This spatial inaccuracy can skew the relationship between species occurrences and environmental predictors, leading to misleading predictions of suitable habitats. - Sampling Bias
Sampling bias occurs when species occurrence data is unevenly distributed across geographic or environmental gradients. In many cases, species records are more frequently collected in areas that are easily accessible, such as near roads or urban centers, while remote or difficult-to-access areas are underrepresented. This geographic bias can lead to skewed SDM outputs, as models may fail to account for the full range of suitable environments for the species. Addressing sampling bias is crucial to improving the robustness of SDM predictions.
The Role of Species Ecology in SDMs
While the data limitations mentioned above are critical, the influence of species ecology on SDM performance cannot be understated. Species exhibit varying ecological traits, such as habitat specialization, mobility, and sensitivity to environmental changes, all of which can affect how well SDMs predict their distributions. For example, generalist species that thrive in a wide range of habitats may be easier to model compared to specialist species that require specific environmental conditions. Thus, the effectiveness of SDMs is closely tied to the ecology of the species being modeled, and this relationship must be accounted for when interpreting model outputs.
Synthesizing Research on Data Limitations and SDM Performance
Numerous studies have investigated the effects of data limitations on the accuracy and performance of SDMs. These studies provide valuable insights into the influence of sample size, positional uncertainty, sampling bias, and species ecology on SDM outcomes. However, a comprehensive synthesis of these findings is lacking, hindering our ability to fully understand their individual and combined effects. Without this synthesis, it remains difficult to predict how data quality issues impact the accuracy of SDMs, limiting the utility of these models for practical conservation and management purposes.
Recommendations for Critical Assessment of Species Data
To improve the quality of SDMs, it is essential to critically assess the species occurrence data before incorporating it into models. Based on the findings of previous studies, we provide the following recommendations:
- Increase Sample Size Where Possible: Efforts should be made to gather additional occurrence data, especially for underrepresented species or regions. Citizen science initiatives, targeted surveys, and the use of remote sensing technology can help improve sample sizes.
- Improve Positional Accuracy: Whenever possible, use high-precision GPS devices to collect occurrence data. Incorporating positional uncertainty into the modeling process, through methods such as uncertainty buffers, can help mitigate the effects of imprecise location data.
- Address Sampling Bias: Techniques such as spatial filtering, background data resampling, and bias correction methods can help reduce the impact of sampling bias on SDMs. Additionally, ensuring that occurrence data is collected from a wide range of environments and regions can improve model performance.
- Consider Species Ecology: The ecological characteristics of species should be carefully considered when building SDMs. Models for specialist species or those with narrow ecological niches may require different approaches than those for generalist species.
Conclusion
While SDMs offer valuable insights into species distributions, the limitations of species occurrence data present challenges that must be addressed to improve model accuracy. By synthesizing the results of studies on sample size, positional uncertainty, sampling bias, and species ecology, we can enhance our understanding of how these factors affect SDM performance. Implementing strategies to mitigate these limitations will lead to more reliable predictions, ultimately benefiting biodiversity conservation and management efforts.
SDMs remain a key tool for understanding species-environment relationships, but critical assessment of the data used in these models is essential for maximizing their utility and accuracy in an ever-changing world.
See the complete article: https://nsojournals.onlinelibrary.wiley.com/doi/10.1111/ecog.07294
