We are happy to announce that our new paper on testing theory of spatial indicators of critical slowing down and ecological transitions has appeared in the online early version of the journal Global Ecology and Biogeography. The paper was highlighted via a cover image (left) taken by Stephanie.
Congratulations to Amit Agrawal (former project assistant of our lab) and Sabiha Majumder (PhD student) for their first research publication! This project began during conversations between Stephanie Eby, Andrew P. Dobson and myself. That was back in 2010, when I was a post-doc at Princeton. They had this excellent high resolution data that made sense in the context of a theory paper from my thesis in 2009.
So what is this paper about?
Ecosystems like clear lakes or forests can abruptly collapse to ‘unhealthy’ states like toxic-turbid lakes or deserts with low vegetation. Our lab uses ideas from physics of phase transitions (e.g., water boiling to become vapour) to develop statistical tools of early warnings of such abrupt transitions. For example, two papers of my PhD thesis (Guttal and Jayaprakash 2008 and 2009) were on developing such tools to analyse time series and spatial data from ecosystems. The underlying theory is now popularly known as ‘Early warning signals of critical transitions’, “Theory of Critical slowing down”, etc.
In this paper, we were testing such tools using spatial data of large scale ecosystems. There were earlier efforts to test these quantitative tools, but mostly in simple laboratory conditions or controlled lake experiments.
Specifically, we test the prediction that the following metrics show stronger signatures of transitions ‘before a collapse of an ecosystem’
(A) Spatial variance (proposed in Guttal and Jayaprakash, 2009)
(B) Spatial skewness (proposed in Guttal and Jayaprakash, 2009)
(C) Spatial autocorrelation (proposed in Dakos et al 2010)
(D) Spatial spectra (proposed in Carpenter and Brock 2010).
This graph below shows how theoretical predictions and real data match.
There were a few subtle and insightful aspects related to analysing this dataset. First, we didn’t have an ideal dataset, so we had to make an approximation called ‘space-for-time substitution’ to compare theory with real data. We justified this approach using a model. Second, the data were discrete-state (occupied/unoccupied), unlike what the models typically assumes continuous variables (like biomass density) in each of the above papers. We developed a data preprocessing method called coarse-graining, inspired from the physics literature. We thought the method to be sufficiently important and hence the details of the method will be published separately. Analysing this dataset has motivated various thesis chapters in our lab’s PhD student Sabiha Majumder, who is from Physics department and works jointly with me and Prof. Sriram Ramaswamy.
I should add that reviewers gave detailed comments that helped the manuscript a lot. This was also our first manuscript where we used services of Axios.
All codes and some data associated with this manuscript are available on our Github page: https://github.com/tee-lab/spacetime-csd