Data quantity is more important than its spatial bias for predictive species distribution modelling.

Area of Science:

• Ecology
• Biodiversity Science
• Computational Biology

Background:

• Biological records are crucial for training species distribution models (SDMs).
• Spatial sampling bias is a common issue in biological data, potentially affecting SDM performance.
• Understanding the impact of bias on SDM predictions is essential for ecological research.

Purpose of the Study:

• To evaluate the impact of spatial sampling bias, sample size, and modeling method on species distribution model (SDM) prediction performance.
• To simulate biological recording processes with real-world spatial biases.
• To quantify the relative importance of these factors in determining SDM accuracy.

Main Methods:

• Simulated presence and absence data for virtual species.
• Incorporated realistic spatial sampling biases into simulated data.
• Assessed prediction performance across different sample sizes and various SDM methods.

Main Results:

• Spatial bias in training data was found to decrease SDM prediction performance.
• Sample size and the selection of the SDM method had a greater impact on prediction performance than spatial bias.
• The study highlights the interplay between data quality and modeling choices.

Conclusions:

• While spatial bias affects SDM performance, it is not the sole determinant of accuracy.
• Optimizing sample size and selecting appropriate modeling techniques are critical for robust species distribution modeling.
• Future research should consider both data biases and methodological choices for reliable ecological predictions.