Consulting
Ecological modelling and data analysis
I work as an external consultant in ecological modelling and data analysis for limu.eco (Mettmenstetten, Switzerland), a company that turns aquatic plant biomass harvested from local waterways into additives for biodegradable materials.
I am available for consulting work beyond that engagement. If you have an ecological question, a monitoring dataset, or an environmental claim that needs defensible quantitative support, get in touch.
What I do
Ecological assessment of biodiversity — designing and analysing surveys and monitoring programmes for aquatic and terrestrial communities, from species composition and traits to whole food webs.
Statistical modelling — Bayesian hierarchical models, generalised linear and non-linear models, time-series and state-space models, all fitted with uncertainty carried through to the answer rather than discarded at the end.
Causal inference — separating “these two things move together” from “changing this one changes the other”. I make assumptions explicit as directed acyclic graphs (DAGs) before fitting anything, so that a model estimates the effect you actually asked about.
Ecosystem and food-web modelling — Ecopath with Ecosim /
Rpath, empirical dynamic modelling and Gaussian-Process approaches for forecasting how communities respond to warming, nutrients and exploitation.Study design before data collection — sample-size and power analysis by simulation, so you know in advance whether a planned survey can detect the effect that matters. This is usually the cheapest thing you will ever pay a statistician for.
Reproducible analysis and reporting — every project delivered as version-controlled code and a Quarto report that reruns end to end, so your analysis can be audited, reused and defended years later.
Tools
R · Stan / brms · Rpath · rEDM · isdbayes · tidyverse · Quarto · Git / GitHub · Python
How I report results
I report findings in probabilistic language. Most statistical results are not a yes or a no; they are a statement about how strongly the data support a conclusion, and pretending otherwise hides risk from whoever has to make the decision.
Suppose we test whether a new additive speeds up the degradation of a biodegradable plastic in lake water. A conventional report says:
“The treatment effect was statistically significant (p = 0.03).”
That sentence does not tell you how much faster, nor how certain we are. My report says instead:
“The additive increased the degradation rate by 18% on average (95% credible interval: 6% to 31%). Given the data and our assumptions, there is a 97% probability that the additive speeds up degradation at all, and an 82% probability that it does so by more than 10% — the threshold your process requires.”
The second version answers the question you actually have: not “is there an effect?” but “how big is it, how sure are we, and is it big enough to matter for my decision?”
This matters most when the answer is inconvenient. A result that is genuinely uncertain gets reported as uncertain — with the range of values the data support, and what further data would narrow it. You get an honest basis for a decision, not a number engineered to look conclusive.
Working principles
Assumptions first. Before any model is fitted, I write down what is assumed about how the system works and how the data were generated. If those assumptions are wrong, the result is wrong however sophisticated the model.
Uncertainty is part of the answer, not a footnote to it.
Reproducibility by default. Code and analysis are delivered so that someone else — a regulator, a reviewer, a future employee — can rerun them and get the same numbers. See my Open Science page.
I will tell you when the data cannot answer the question. That is a legitimate and often valuable finding.
Background
I hold a Ph.D. in Ecology and work as a postdoctoral researcher at Eawag, the Swiss Federal Institute of Aquatic Science and Technology. My published work spans tropical reef food webs, Amazonian river fisheries, beaver-engineered streams and Lake Constance plankton — see Research Projects for detail, and the Learning notebooks for worked examples of the methods above.
Get in touch to discuss a project.