Integrating multiple data sources in species distribution modeling: a framework for data fusion.

Area of Science:

• Ecology
• Computational Biology
• Conservation Science

Background:

• Species distribution models (SDMs) are increasingly used to understand species occurrence and abundance.
• Parameterizing SDMs often involves a trade-off between data quality and quantity.
• Integrating diverse data types offers a promising solution to this challenge.

Purpose of the Study:

• To develop and evaluate novel methods for jointly modeling high-quality and lower-quality data in SDMs.
• To incorporate spatial autocorrelation in occurrence and detection error within these integrated models.
• To assess the performance of new approaches under varying data quality scenarios.

Main Methods:

• Developed three novel information-sharing models: "Shared," "Correlation," and "Covariates."
• Extended existing joint modeling approaches by incorporating Multivariate Conditional Autoregressive (MVCAR) models.
• Evaluated model performance using a case study of the Brown-headed Nuthatch and through simulations.

Main Results:

• All three novel approaches improved out-of-sample predictions compared to using a single data source.
• The "Shared" model performed best when auxiliary data was of high quality.
• The "Correlation" and "Covariates" models demonstrated robustness and superior performance with lower-quality auxiliary data.

Conclusions:

• Jointly modeling multiple data sources significantly enhances SDM accuracy.
• The developed "Correlation" and "Covariates" models provide robust alternatives for utilizing citizen science or opportunistic data.
• These methods maximize the utility of available information for more accurate species distribution estimation.