Dear all,
I have gathered samples of Dreissenid mussels at different locations in a lake. After counting and measuring the mussels, histograms featuring the number of individuals per shell length ranging from ≤0.4cm to 3.5cm have been created, most of which are showing multimodal distributions but some are unimodal, too.
What is being tried to do, apart from visually determining the size-frequency distributions, is to apply a statistical mathematical tool in R that goes through all of these data and attempts to classify recurring groups that then ideally represent age-groups (cohorts, populations) of the mussels.
It seems like packages in R that might be helpful are 'mclust' and 'Rmixmod' both of which have already been tried. However, in the end, always a dead-end has been reached so it got me wondering whether the arrangement of my data may be the problem or there is another underlying cause.
Has anyone already encountered and maybe solved equal problems and might possibly be up to having a look into the organisation of my size-frequency data?
Similar methods to what I am trying to do have been exercised for example in:
Comtet, Desbruyeres (1998) Population structure and recruitment in mytilid bivalves from the Lucky Strike and Menez Gwen hydrothermal vent fields (37'17'N and 37"501N on the Mid-Atlantic Ridge)
and
Taylor et al (2009) Using length-frequency data to elucidate the population dynamics of Argulus foliaceus (Crustacea: Branchiura)
Thanks to everyone wanting to help!
Benjamin Wegner
I suppose you want to estimate the individual growth, is it right??? Have you use TropFish???
ELEFAN is another option
https://www.rdocumentation.org/packages/TropFishR/versions/1.2.1/topics/ELEFAN
Hello Benjamin.
Well, since I don't know the variables in your data set, I'm not sure about the details of your problem. However, it seems to me that you have a classical classification problem.
It remembers me the Iris* data set. The mussels' measurements can be related to the measurements of the flowers (petal and sepal length).
Now, you can look to some solutions involving Random forests (a good classification algorithm) or a simpler classifier like CART*2.
Depending on your variables you may want to do clustering, mainly a density-based clustering. In this case, you may try k-means*3 or DBSCAN*4.
These are some very useful tools. You can go far and fast using R.
Best regards, Joaquim.
* https://archive.ics.uci.edu/ml/datasets/iris
*2 https://rpubs.com/minma/cart_with_rpart
*3 https://datascienceplus.com/k-means-clustering-in-r/
*4 https://cran.r-project.org/web/packages/dbscan/dbscan.pdf
Hi Benjamin,
as it appears you want clusters of mussels, according to their observed parameters.
Clusetring is worth of nothing if you don't have the good metrics and the right linkage.
Considering the multivariate data you have, I would suggest using Multiple Factor Analysis or factorial analysis of Mixed or even PCA (if all datas are numeric lengths).
The package is called FactoMineR, is very handy, and gives highly valuable insights, contrary to raw clustering.
This will enable you to find out the highest variations among your variables and thus, probably cluster by age the mussels (with the built-in function HCPC of FactoMineR).
Don't hesitate to ask me for precisions or even, send me a header of your data to see if it will suit.
Benoit
Thank you so much for your responses so far!
@ René, I am not looking at the individual growth of the mussels but at the state of sizes they had when I sampled them. I just measured the individual shell lengths and counted the frequency of lengths. Let's say one sample contained 400 mussels, then 10 had size
- Badges
- Science topic
- Similar topics
- Biological Science
- Zoology
- Invertebrate Zoology
- Arthropods
- Crustacea