|dc.description.abstract||Uncertainties in ecology are pervasive, and therefore, communicating the level of uncertainty for any inference derived from scientific research is key to sound decision-making and management of species and ecosystems. Characterising uncertainty is part of converting information into knowledge and has the added benefit of identifying fruitful avenues of further investigation. Without such care in accounting for uncertainties, we risk making misleading conclusions and inappropriate management decisions. In this thesis, it is argued strongly that rather than being something to avoid discussing, reducing uncertainty is fundamental to good ecological science. Uncertainty can come from a number of sources. Parameter estimation for demographic studies has inherently high uncertainty due to substantial variation between individuals, years, and spatial locations thus requiring considerable resources to obtain accurate estimates for survival, reproduction and growth. In some cases, certain life stages may be unseen during sampling procedures, such as seeds in the soil seed bank, or if non-breeding components of the population are not present in the selected sampling sites. While the potential sources of uncertainty are diverse, I attempted to cover a range of key areas of uncertainties relevant to ecologists over the course of this thesis. Specific areas of uncertainty were targeted using case studies to provide examples to demonstrate how these uncertainties can be addressed and how they can be used to aid inferences and provide recommendations for future data collection procedures.
First, I highlighted the prevalence of authors excluding a cryptic but important life stage, the dormant seed bank, from their data collection procedures and population models (Chapter 2). The evolution of seed banks acts as a bet hedging strategy, improving the persistence of plant populations in variable environments, thus it is crucial that we are able to address this potential knowledge gap to avoid misleading conclusions. The consequences of this exclusion on model parameters such as population growth rates and extinction risks were explored using a joint empirical and simulation approach, combining information from the published literature with Monte Carlo simulations. These simulations explored a range of assumptions that need to be considered when including a seed bank into the model, such as seed longevity, viability and germination rates. A key result of these simulations is that our perspective regarding the importance of the seed bank can differ, further depending on the species and the type of demographic year. For example, inclusion of the seed bank and demographic uncertainty in seed bank parameters were found to have little effect for stable populations with high post-seedling survival. In such cases, the seed bank can be excluded, however this should be accompanied by appropriate justification either through literature confirmation that dormancy is not existent or demonstrated via simulations that it is of little consequence. Conversely, seed banks had a more demonstrable impact on growth and extinction rates for variable populations, particularly when populations experience poor demographic years. The use of simulations and published literature can thus be an effective means to explore uncertainties resulting from the presence of cryptic life stages.
Second, I explored and demonstrated the use of multivariate auto-regressive state-space (MARSS) models as a versatile framework for capturing and addressing several sources of uncertainty including observation errors, and show how these models can be used to update and improve monitoring design (Chapter 3). MARSS models were constructed for a common, ephemeral plant using a 9 year time series dataset from multiple study sites within the Simpson Desert to explain trends over time and space. Modelling multi-dimensional time series data allowed the identification of spatial sub-population structure with respect to location and fire history, and the incorporation of population structure making use of count data for above ground plants and the seed bank. Model results suggested population dynamics to be driven primarily by geographical location possibly reflecting differences in soil conditions, local competition and local microclimate, overshadowing variation caused by fire history. The seed bank was also found to be characterised by high observation error with low environmental variability, while the converse was true for the above ground population estimates. Knowledge regarding the relative uncertainty of the above and below ground abundance estimates and the spatial distribution of population dynamics can then be used to provide guidelines for future monitoring efforts. For example, it may be more strategic to sample the seed bank less frequently as it less variable over time, and instead focus on obtaining more accurate counts when it is sampled to offset the high observation error. Additionally, the level of spatial heterogeneity in the Simpson Desert provides some justification for expanding spatial replication.
Third, the validity of using visual cover estimates as a means of monitoring vegetation and environmental changes was assessed. Visual cover estimates are particularly susceptible to observation error, and previous studies on the repeatability and reliability of such measurements have raised concerns over their value in ecological monitoring and decision making. I made use of two primary long-term monitoring datasets on spinifex grasslands, each obtained with different motivations, methods of data collection, and varying degrees of spatial and temporal coverage to assess the consistency of spatial and temporal trends between these datasets. Thus it could be determined whether the different sampling strategies and observation errors produced inconsistent and conflicting results. Observation errors were found to be quite large, often exceeding variation due to environmental changes. However, when these errors are accounted for, trends in the spatial dynamics of spinifex cover were consistent between the datasets, with population dynamics being driven primarily by time since last fire. Models also showed similar population traces over time, reflecting the effects of major temporal drivers such as rainfall and fire history. These findings vindicate visual cover estimates as a useful source of information provided that uncertainties in the measurements are appropriately addressed.
Finally, I shift the focus from single species analyses and apply dynamic factor analysis (DFA) to a large, multispecies database of abundances over time, which reduces the temporal dynamics of a large number of species to a small number of common trends. In producing these trends, interpretation of large multispecies data is greatly simplified. Furthermore, the common trends groups species with similar temporal responses, thus revealing where there is potential to borrow strength across species to supplement those that are less well sampled. Five common trends were identified for each site, and crucially, these trends were strongly associated with life form which showed distinctive signatures in the shape of their trends. Forbs and grasses for example demonstrated high levels of synchrony in their responses to rain events, although the signal for shrubs and subshrubs was weaker. These responses were also found to differ over relatively large (>20km) spatial scales. Thus plant life form is a reasonable predictor of changes in abundance over time and offers some justification for borrowing information to supplement data from poorly sampled species, provided the data are within the same locality.
The results of this thesis underpin the value of acknowledging, measuring and managing uncertainties, and that these uncertainties can be used advantageously to guide inferences, extract value from datasets thought to be unreliable, provide justification for sourcing additional sources of information or excluding others, and inform future data collection protocols. Several methods for addressing uncertainty are highlighted, such as the use of simulations when data are unavailable, powerful state-space modelling techniques to account for observation error, and identifying opportunities for supplementing data from the literature, similar sites or species with similar dynamics. There are several more options available for reducing and managing uncertainty, and it is ultimately up to the researcher to first recognise where uncertainties are likely to exist, explore their options, and decide how such uncertainties are to be addressed.||en_AU|