Treating gaps and biases in biodiversity data as a missing data problem.

Area of Science:

• Ecology
• Biodiversity Science
• Data Science

Background:

• Large biodiversity datasets are crucial for monitoring species' populations and distributions over time.
• Spatial and temporal data gaps can limit the representativeness of biodiversity data, hindering large-scale inferences and conservation efforts.

Purpose of the Study:

• To conceptualize biodiversity data gaps as a missing data problem, providing a unifying framework for analysis.
• To explore the implications of data gaps for inferring species' trends and factors influencing species occurrences and abundances.

Main Methods:

• Characterizing data gaps as different classes of missing data.
• Applying missing data theory to analyze bias in species trend models.
• Reviewing empirical studies and conducting simulation studies to compare methods for handling data gaps (subsampling, weighting, imputation).

Main Results:

• Bias can arise when factors influencing data sampling/availability overlap with factors affecting species.
• Standard species trend models often fail to account for factors driving data missingness, increasing susceptibility to bias.
• Subsampling, weighting, and imputation can reduce bias but may increase uncertainty; weighting methods show potential for reducing both bias and variance.

Conclusions:

• Effective bias reduction critically depends on understanding and having data on the factors causing data gaps.
• Weighting techniques are underutilized in ecology but offer significant potential for improving biodiversity data analysis.
• Considerations for data gaps must be integrated throughout the data collection and analysis workflow for robust ecological monitoring.