Beyond Assumptions: Unraveling Data Limitations in Predictive Ecology

Ing. Lukáš Gábor, Ph.D.

Doctoral thesis

Abstract:

Context Species distribution models (SDMs) have proven valuable in filling gaps in our knowledge of species occurrences. However, despite their broad applicability, SDMs exhibit critical short- comings due to limitations in species occurrence data and in environmetal variables. Typical example of such limitation in species records is spatial uncertainty in their location, ranging from a few meters to tens of kilometers (e.g., positional uncertainty in the GBIF database can exceed 300 kilometers). Similarly, environmental variables may be limited by how much detailed information about the environment they provide. For example, land cover types typically indicate the amount of habitat within a spatial unit. However, it is possible that a simple binary presence/absence of suitable habitat may be the only information available or even more important aspect than area, determining species distribution. Objectives This dissertation had three main objectives. Firstly, I explored the potential of an alternative approach to incorporate environmental variables into models (i.e., binary versus continu- ous habitat information). Secondly, I investigated the influence of positional uncertainty in species records on the ecological interpretability of models. Thirdly, I evaluated whether the appropriateness of using binary or proportional (continuous) type of variables, and the influ- ence of positional error, is affected by the scale of the analyses. Specifically, I addressed the following research questions: a) Can binary land cover predictors provide models of higher accuracy than traditionally used proportional variables? b) If so, what is the role of spatial grain in determining the usability of binary land cover predictors? c) To what extent does positional uncertainty in species occurrence data affect model parameter estimation and the ecological interpretability of species distribution models? d) What are the trade-offs between analysis grain and positional uncertainty in modeling species distributions?. Results Results indicated that models’ performances were not affected by the type of the adopted habitat variable (proportional or binary but the usability of binary variables decreased with coarsening the resolution (i.e., binary representation of habitat is useful at finer grain sizes of approx. 1km2. Results confirmed that model performance decrease with increasing po- sitional error in species records, as demonstrated in prior studies. However, I have shown that coarsening the analysis grain to compensate for positional error did not improve model performance as was widely assumed. This, however, doesn’t mean we should exclude species records with high positional uncertainty from our studies, because the negative consequences of positional uncertainty on model performance did not extend as strongly to the ecological interpretability of the models. Conclusions These findings are encouraging for practitioners using SDMs to reveal relationships between species occurrences and its environmental drivers as such relationship can be to some degree estimated using positionally uncertain data and simple environmental variables describing presence or absence of a habitat. On the other hand, my findings show that positional un- certainty in species data can cause inaccurate spatial predictions leading to inaccurate maps of species distributions, especially in heterogeneous environments and when using fine res- olution environmental data. Therefore, such models are not suitable for tasks like setting up protected areas or prioritizing conservation efforts.

Keywords

positional error, binary variables, species distribution models

Language used: English

Date on which the thesis was submitted / produced: 22. 7. 2024

Thesis defence

• Supervisor: doc. Ing. Vítězslav Moudrý, Ph.D.

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Doctoral programme / specializace:
Environmental Earth Sciences / Spatial Sciences