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

Published: Oct. 21, 2023

Big biodiversity datasets have great potential for monitoring and research because of their large taxonomic, geographic temporal scope. Such become especially important assessing the change species’ populations distributions. Gaps in available data, however, often hinder drawing large-scale inferences about trends. Here, we conceptualise data gaps as a missing problem, which provides unifying framework challenges solutions across different types datasets. We characterise typical classes then use theory to explore implications questions. By using this framework, show that bias due can arise when factors affecting sampling and/or availability overlap with those biodiversity. But outcome also depends on ecological questions, determines choices around analytical approach. argue approaches long-term species trend modelling are susceptible since such models do not tend account drive missingness. To identify general solutions, review empirical studies simulation compare some most frequently employed deal gaps, including subsampling, weighting imputation. All these methods reduce but may come at cost increased uncertainty parameter estimates. Weighting arguably least used so far ecology both variance Regardless method, ability critically knowledge of, on, creating gaps. our outline necessary considerations dealing stages collection analysis workflow.

Language: Английский

NicheFlow: Towards a foundation model for Species Distribution Modelling

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 18, 2024

Abstract 1. Species distribution models (SDMs) are crucial tools for understanding and predicting biodiversity patterns, yet they often struggle with limited data, biased sampling, complex species-environment relationships. Here I present NicheFlow, a novel foundation model SDMs that leverages generative AI to address these challenges advance our ability predict species distributions across taxa environments. 2. NicheFlow employs two-stage approach, combining embeddings two chained models, one generate in environmental space, second geographic space. This architecture allows the sharing of information captures complex, non-linear relationships trained on comprehensive dataset reptile evaluated its performance using both standard SDM metrics zero-shot prediction tasks. 3. demonstrates good predictive performance, particularly rare data-deficient species. The successfully generated plausible not seen during training, showcasing potential prediction. learned captured meaningful ecological information, revealing patterns niche structure taxa, latitude range sizes. 4. As proof-of-principle model, represents significant modeling, offering powerful tool addressing pressing questions ecology, evolution, conservation biology. Its joint hypothetical niches opens new avenues exploring evolutionary questions, including ancestral reconstruction community assembly processes. approach has transform improve capacity manage face global change.

Language: Английский