Acorn production is yearly variable and depend on so many ecological factors. For Oak species, Rose et al, 2012 showed that there is a weak relationship between DBH and acorn production. Also, mast year is another problem in estimations. So, do you think there would be a way to estimate this attribute so accurate that could be used for management purposes.
As you said acorn production is very variable. However you can define factors that can increase or decrease your acorn production like: size and age, growing season conditions, life history traits. Maybe if you explain better what you need the estimates for, we can contribute more.
Dear Matteo, Thank you for the response. We are going to estimate Oak acorn production at national level for some possible uses. As I mentioned before, size (which partially represents age), is the basic predictor for our models. But the relationship is so week that we can not make a reliable estimation. Even for a given site which has a similar ecological conditions, finding a proper relationship is not possible.
Aloha Hormoz. Have you tried non-linear relationship to size/age? In many trees in Hawaii there is a peak production with size and a relatively linear correlation, but at some point there is a decrease of productivity with size as the tree ages past its peak. If you are trying just a single linear correlation this inflection point could make the linear relationship very weak, but if partitioned into the growth years and the decline years you might get a much better relationship.
Iʻm not sure, but I would think that if you talk to some knowledgable indigenous people you might get some suggestions of what constitutes the productive and non-productive years. These type of things may be linked to very specific events that arenʻt obvious at all, and could take a long time to try and determine through statistical analysis (because there are so many options). Sometimes longtime observers, especially those that rely of the resource, are able to hone in on what those might be. For some crops here in Hawaii elders have told me that high rain during the first week after flowering to be a very important indicator....something I never would have been able to isolate on my own but now is showing to be a strong predictor.
Frank Lake might have some leads on who to talk to, and studies very specific ecological timings around resources (specifically how fire, the intensity and timing of fire, strongly effects the quality of forest resources).
Dear Hormoz - I have attached two files for your use. One is a reprint of a review paper I gave at a mast management conference on the effects of gypsy moth on acorn production but in it I cite a paper by Downs 1944 where he showed a relationship between tree dbh and production. I have subsequently plotted his data into graph that I use to teach foresters, wildlife biologists, and land managers about acorn production and management. I have attached some of the slides I use for that purpose. There are several systems in use in the US by wildlife managers to predict crops across large areas. If you want more detail on those, I can provide papers or if you have questions about any of the slides, let me know. . Cheers, Kurt
Hormoz -
This appears to be an interesting application, in which you have been given answers that are also intriguing, which are apparently based on experience applicable to your problem. Forestry is not my area, and I found this instructive.
You mention that the model you have is weak, and I wonder how you measure that. Is there a standard method used in this field? Do you have real test data from which you omit an observation randomly, predict for that case and then compare the prediction with the actually observed case? If you did, you could find a distribution for these 'errors' (including graphics would be good) and consider your distribution of obtained prediction errors against the theoretically estimated variance of the prediction errors for your models. Nonsampling errors such as measurement error would be included. A measure of variance, that is impacted by nonsampling error and bias which it is not designed to measure, is not perfect, but the variance of the prediction error is likely to still be a very helpful measure.
Are you using estimated variance of the prediction errors to compare models? May I assume those are found in closed form solution for your models, or can they at least be simulated? The above noted type of study could be used to test both your results and that measure of accuracy.
Many use R, but that may be problematic - though not as problematic as p-values. Willett and Singer wrote articles in The American Statistician noting that R cannot be compared directly between various models, but the problem seems to be worse than that in practice, I think. The variance of the prediction error is not a complete measure, but may be much better than R-square, which seems often volatile.
At any rate, it is important to have a basis for comparing the competing models that you might have, to decide which might perform better.
The last response above was in mid-March. Have you had occasion since then to try any of these models and compare performances? That would be interesting.
Thank you. - Jim
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis