Rendered slides: [ Ссылка ]
Sources for the slides: [ Ссылка ]

Abstract:
Ecosystems are highly variable across time and space and the consequences of this variation have been widely studied. Interest in ecological resilience has recently increased, as we now appreciate the role environmental variability plays in maintaining biodiversity and in the identification and early detection of regime shifts. Frustratingly, many limnological time series are poorly behaved; e.g. irregular sampling and missing data. Critically, statistical evaluation of resilience is complicated in such time series, often requiring interpolation or other interventions that are difficult to account for in the analysis.

We present a novel approach to the estimation of resilience indicators based on distributional generalized additive models (GAMs), which allow the simultaneous estimation of mean, variance, and other parameters of the series. We illustrate our approach by generating continuous estimates of key resilience indicators for time series of algal pigments from a sediment core from Lake 227 (Canada), which has been experimentally manipulated to induce eutrophication. We show that our distributional GAM approach is able to reproduce previous results (Cottingham et al 2000, Ecology Letters) that identified an increased variance in algal populations post disturbance, whilst providing a continuous estimate of the variance for the entire time series.

The Lake 227 analysis is used as a case study to discuss issues with current data analysis practice applied to both big and small data in limnology and to illustrate ways forward that the community should consider if we are to make the best use of big data and data collation initiatives.

(Note this is a modified version of the abstract I submitted to the summit. In preparing the talk I decided to emphasise the issues mentioned in the last paragraph of the abstract and so present fewer details on the Lake 227 analysis itself to meet the 10 min time constraint.)

свернуть