How to make use of unlabeled observations in species distribution modeling using point process models

Area of Science:

• Ecology
• Computational Biology
• Biostatistics

Background:

• Species distribution models (SDMs) are crucial for predicting species' spatial occurrences.
• Data quality, including uncertain species identification, poses a significant challenge in SDMs.
• Existing SDM platforms often lack robust methods for handling observations with unknown identities.

Purpose of the Study:

• To develop and evaluate algorithms for classifying unknown species identities within SDMs.
• To simultaneously predict multiple species distributions using spatial point processes.
• To assess the performance of new algorithms against traditional methods using simulated and real-world data.

Main Methods:

• Development of two novel algorithms for classifying observations with unknown species identities.
• Simultaneous prediction of multiple species distributions via spatial point processes.
• Comparative analysis of algorithm performance using various initializations and datasets, including a frog species (Mixophyes) dataset.

Main Results:

• Algorithm performance is influenced by species distribution correlation, abundance, and the proportion of unknown identities.
• The developed algorithms demonstrated superior performance compared to models excluding uncertain data in certain scenarios.
• Successful application to a real-world dataset of three frog species, highlighting practical utility.

Conclusions:

• The proposed algorithms offer a valuable tool for enhancing species distribution modeling with uncertain or misidentified data.
• These methods are particularly beneficial for opportunistic surveys and taxonomic revisions.
• The study underscores the importance of accounting for data quality in ecological modeling for more reliable predictions.