Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Enhancing species distribution models through integrated modeling and bias mitigation in African bird studies
The increasing availability of biodiversity data provides significant opportunities to improve species distribution modeling, particularly through the integration of multiple datasets. The overarching aim of this dissertation is to construct an integrated species distribution model (ISDM) for Africa...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English English |
| Published: |
Department of Statistical Sciences
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| Summary: | The increasing availability of biodiversity data provides significant opportunities to improve species distribution modeling, particularly through the integration of multiple datasets. The overarching aim of this dissertation is to construct an integrated species distribution model (ISDM) for African bird species. A central challenge in developing ISDMs is that different datasets follow distinct sampling protocols and embody different assumptions. In particular, widely available presence-only (PO) data are prone to severe sampling bias, which can substantially distort model inference if not properly addressed. In this dissertation, we evaluate how sample size, degree of spatial bias, and species preva- lence influence the accuracy and stability of ISDMs. This is achieved through simulation experiments using the virtual ecologist approach, which allows controlled manipulation of eco- logical and sampling processes. We also examine methods for mitigating sampling bias in PO datasets, including modeling the bias using covariates and incorporating an additional spatial random field specifically designed to account for the bias component. Our simulation results show that when the volume of presence-only data greatly exceeds that of presence-absence (PA) data, the PO dataset dominates model behaviour, resulting in decreased precision and reduced predictive performance. Consequently, when applying ISDMs to real-world data, PO data must be thinned to reduce the influence of sampling bias. Guided by the insights gained from the simulation study, ISDMs were then constructed for three African bird species with differing ecological and data-related characteristics. These models were developed using eBird (PO) data and SABAP2 (PA) data. As informed by the simulations, the eBird dataset was thinned prior to model construction, with thinning intensity determined using the inhomogeneous pair correlation function to minimise residual sampling bias. |
|---|