After three years of work, our colleague Léa Pautrel defended her doctoral thesis in ecology at the University of Montpellier last week. She conducted her research at TerrOïko in partnership with the Center for Functional and Evolutionary Ecology (CEFE).
Want to watch or rewatch her defense? You can do so here https://lnkd.in/dVAAb2XF.
Léa's research focused on the use of continuous-time hierarchical models for monitoring biodiversity with sensors. Here's a summary.
The starting point is that we are using more and more sensors, particularly camera traps, to collect data on species, which can then be used to analyze their presence and abundance, for example.
To process this data, statistical models known as “hierarchical” models must be used, which take into account the fact that the detection of animals in a given area (by a sensor or a human) is imperfect.
These models were originally designed to analyze data collected on an ad hoc basis, usually by human operators who visit the field from time to time.
However, new models have emerged more recently that are designed to process data collected continuously, such as sensor data.
Do these new models improve biodiversity monitoring?
To answer this question, Léa tested the “performance” of five models (two ‘old’ and three “new”): she fed them the same input data and compared the analysis results to see if the newer ones (those designed for continuous data) performed better than the others.
She then looked at the influence of detection covariates, i.e., factors that could influence the detection results for a species (e.g., weather conditions). She incorporated these covariates into the analyses and compared the models again.
What was the verdict of this work? We'll let you find out by watching the thesis defense.
What is certain is that, beyond simply answering the initial question, this thesis also provided an opportunity to reflect on the evolution of statistical ecology in the face of technological advances, and in particular on the operational accessibility of these methods.
This is a key issue, as the use of sensors now makes it possible to envisage automated processing chains that can be transferred to industrial sectors which, until now, have made little use of ecological modeling, via tools such as digital twins.