Synthesising diverse and sparse data in ecology
This source preferred by Rick Stafford
Authors: Stafford, R. and Gardner, E.
Start date: 18 August 2013
Most data in ecology, both currently and historically, are derived from one of two sources. Firstly, observations of species occurrence, interactions or animal behaviour, either through amateur naturalists and dedicated recording schemes, or through modern ‘citizen scientists’ equipped with the latest smartphone ‘apps’. Secondly, data are collected to test hypotheses, through the proper adoption of the scientific method. The first of these approaches can be very opportunistic, leading to spatially and temporally patchy data, which can prove problematic for determining actual distribution patterns or phenological changes. The second approach can be highly specific, usually optimising sample sizes for statistical power, and potentially restricted to certain species, locations and times of year.
While standardising processes such as data collection methods and sampling effort would allow for easy synthesis of such data, and its use beyond the test of a single hypothesis, such an approach may not be practical in terms of innovative methods for hypothesis testing, nor economic in terms of spatial and temporal coverage of sampling. Furthermore, such a standardisation programme would not be able to match the wealth of data which already exists, and would not be able to compare such data against existing historic records. As such, the logical conclusion is that new analysis techniques need to be developed to effectively use such data, and ideally allow it to be linked with other data from areas such as molecular biology and social sciences. Ideally, for many such techniques where data are collected by non-scientists, the outputs of such analysis techniques should be simple and initiative. In this talk I present work demonstrating how we can robustly investigate species distribution patterns and changes in animal behaviour using computationally intensive (bootstrapping), but visually intuitive processes. The processes work on standardised datasets, but also with sparse data, as collected by volunteers.
Diverse types of data all supply evidence at different levels or hierarchies of organisation (for example, cellular responses or population level responses). As such, interacting, hierarchical networks may best synthesise these diverse data types. While Bayesian belief networks can be an effective method of synthesising such data, the lack of reciprocal effects between nodes (as would be intuitive for competitive interactions) can cause problems. However, computationally intensive programing can approximately solve these kinds of issues, and I present examples of how many types of diverse evidence can be synthesised into a robust ecological decision making mechanism. This presentation is intended for all practitioners of ecology, not just computational ecologists, and will hopefully inform the debate on how computational ecology can become a useful, and user friendly, integrative tool, rather than a distant sub-disciple of ecology as a whole.