Read on Twitter
Sci peeps @gilromera & myself recently published an open-access piece on millennial scale ecological memory https://onlinelibrary.wiley.com/doi/full/10.1111/ecog.04772?af=R
We know, this is an odd paper, but the past is where alternatives to an uncertain future live so there we go, a thread on it
. 1/28
We know, this is an odd paper, but the past is where alternatives to an uncertain future live so there we go, a thread on it
. 1/28
Before that, aiming for #OpenScience, this paper is fully #ReproducibleScience through 3 Rmarkdown files: https://github.com/BlasBenito/EcologicalMemory
..that rely on 2 R packages I created "virtualPollen" https://blasbenito.github.io/virtualPollen/ and "memoria" https://blasbenito.github.io/memoria/ (both available at CRAN). 2/28
..that rely on 2 R packages I created "virtualPollen" https://blasbenito.github.io/virtualPollen/ and "memoria" https://blasbenito.github.io/memoria/ (both available at CRAN). 2/28
So why do even we look at ecological memory you may wonder. Well, it is a beautiful concept rooted on the idea that what we see now entails part of the past. Ogle et al (2015) summarized it as the effect of antecedent conditions on current processes https://onlinelibrary.wiley.com/doi/abs/10.1111/ele.12399 3/28
This assumes that:
1. Biotic responses are delayed to their environmental drivers,
2. Values of a biotic variable (e.g. pop. size) contain different kind of “memories” as the they depend on: 4/28
1. Biotic responses are delayed to their environmental drivers,
2. Values of a biotic variable (e.g. pop. size) contain different kind of “memories” as the they depend on: 4/28
a. past values of the biotic variable (endogenous memory)
b. past values of an environmental driver (exogenous memory)
c. concurrent values of the environmental driver (concurrent effect) 5/28
b. past values of an environmental driver (exogenous memory)
c. concurrent values of the environmental driver (concurrent effect) 5/28
For example, population size requires a lagged period before to reach a maximum after a resources input, as individuals need to gather energy before reproduction, and the new individuals need time to mature before continuing the cycle. 6/28
Ecological memory processes are out there in the wild, just looking forward to be found and explained by people like us! Look for instance at Itter et al. (2018, https://besjournals.onlinelibrary.wiley.com/doi/10.1111/1365-2745.13087) finding that tree growth shows different length response to drought & defoliation. 7/28
A great example is Liu et al (2018, https://www.nature.com/articles/s41598-018-21339-4) analyzed hydrological effects at a global scale, and found that antecedent precipitation has a stronger effect than concurrent precipitation in arid and semi-arid regions, with a memory length of 5 to 7 months. 8/28
Different ecological memory processes happen at varying time scales (min to ky). Palaeoecological data would allow to better understand memory processes at the ky scale, but palaeofolks rarely mention ecological memory as a process in action (but see https://royalsocietypublishing.org/doi/abs/10.1098/rsbl.2019.0357?af=R) 9/28
Our work answers 2 practical aspects, key to address the ecological memory concept in palaeoecology:
1. How do life traits and niche features shape ecological memory patterns?
2. How do sediment accumulation and pollen sampling intervals affect ecological memory patterns?
10/28
1. How do life traits and niche features shape ecological memory patterns?
2. How do sediment accumulation and pollen sampling intervals affect ecological memory patterns?
10/28
We hypothesize that taxa with a long life-span may show a higher endogenous memory (and are, therefore, more independent from the environmental signal and more driven by their own dynamics) than taxa with a shorter life-span. 11/28
We also test to what extent the data-sampling process in palaeo (including uneven sediment accumulation rates and a given depth between consecutive samples) may degrade the actual ecological memory patterns produced by taxa with different niche features and life traits. 12/28
However, data properties are a must to answer these questions, and this is rarely the case with actual palaeoecological data. To overcome this issue we adopted an in-silico approach based on a mechanistic model to generate our own data. 13/28
The R library virtualPollen https://blasbenito.github.io/virtualPollen generates virtual pollen curves (1yr resolution) with a mechanistic model representing the responses of synthetic taxa with known features (longevity, maturity age, niche position and breadth) to an environmental driver. 14/28
The model represents a phenomenon rarely acknowledged in the palaeoecological literature: a taxon does not "interpret" an environmental driver as is, but through its own "niche function", with a given mean (niche position) and standard deviation (specialization). 15/28
This niche function transforms the values of the environmental driver (green lines in panel b) into a "suitability index" (blue lines in the same panel), which may show varying degrees of correlation with the driver, depending on the niche features of the taxa (panel a). 16/28
If there is a low correlation between a driver and the suitability returned by the niche function of a taxon, then quantifying ecological memory directly on the driver values would lead to the wrong conclusions. We also test this idea in the paper. 17/28
virtualPollen also provides functions to apply a virtual sediment accumulation rate to the virtual pollen curve to aggregate the data unevenly, and to re-sample the data at different depth intervals between samples to simulate more realistic data conditions. 18/28
We generated 16 taxa with different niches and life traits, tracked their pollen abundances as a response to a virtual driver over 10ky, added a sediment accumulation rate, sampled the data at 1, 2, 6, and 10 cm intervals, and organized the data in lags from 20 to 240 yr. 19/28
We fitted this model with random forest:
Bt ~ Bt-1 + ... + Bt-n + Dt-1 + ... Dt-n + Dt
where:
n: lags, 20yr time-steps into the past (to 240 yr).
Bt: observed pollen
B(t-1 to t-n): endogenous memory
D(t-1 to t-n): exogenous memory
Dt: concurrent effect
20/28
Bt ~ Bt-1 + ... + Bt-n + Dt-1 + ... Dt-n + Dt
where:
n: lags, 20yr time-steps into the past (to 240 yr).
Bt: observed pollen
B(t-1 to t-n): endogenous memory
D(t-1 to t-n): exogenous memory
Dt: concurrent effect
20/28
We then computed (per taxon) the importance of the endogenous and exogenous memory over lags (memory strength, b & c), for how many lags they are important (memory length, d & e in the fig.), and for how many lags one is more important than the other (dominance, f & g) 21/28
Finally, we assessed how these ecological memory features are modified by:
- sediment accumulation rate and depth intervals between samples.
- use of "driver" instead of "suitability" as exogenous memory component.
- niche breadth, niche position, life-span, and fecundity. 22/28
- sediment accumulation rate and depth intervals between samples.
- use of "driver" instead of "suitability" as exogenous memory component.
- niche breadth, niche position, life-span, and fecundity. 22/28
The data smoothing produced by the sediment accumulation rate and increasing depth intervals between samples inflates the importance of the endogenous memory by increasing the temporal autocorrelation of the response variable. 23/28
We found that using the driver values as exogenous memory component instead of the suitability index returned by the niche function of the taxon downplays the role of every ecological memory component. This effect is exaggerated in taxa with narrow niche breadths. 24/28
Finally, our analysis shows that the dynamics of long-lived taxa is mediated by a stronger endogenous memory component, because their long life-span helps to keep their reproductive potential through periods of climate deterioration. 25/28
Our study shows that there is potential for the application of ecological memory concepts to palaeoecological data, which may help to improve our understanding on time-delayed population responses to environmental change at centennial to millennial time-scales 26/28
From here, many exciting questions are waiting for an answer: Are ecological memory patterns constant across sites and time? Do pop growth and decline show the same memory effects? May these patterns help to better understand the responses of vegetation to climate warming? 27/28
Working on this paper was great fun because we somehow bridged the gap between palaeo and neoecology. Thank you @gilromera for being my partner in crime! The end. 28/28
