Open bobmuscarella opened 8 years ago
There is precidence for dealing with this from Condit et al. 2004 (Tropical forest dynamics across a rainfall gradient...):
Before estimating growth, we discarded negative increments where the second dbh was 4 SD1 below the first, since these were most likely due to the second type of error [measurement on wrong tree]. The same correction cannot be applied to positive outliers, since trees grow. The fastest growing species at BCI, Trema micrantha and Ochroma pyrimidale, grew by as much as 30-50 mm y-1, with a few valid records as high as 70 mm y-1 (valid because successive dbh measures from 1982 to 2000 showed consistently high growth). We thus exclude any record > 75 mm y-1 as an error.
Condit, R., Aguilar, S., Hernandez, A., Perez, R., Lao, S., Angehr, G., Hubbell, S.P. & Foster, R.B. 2004. Tropical forest dynamics across a rainfall gradient and the impact of an El Niño dry season. Journal of Tropical Ecology 20: 51-72.
I've implemented the Condit et al. growth outlier exclusion in a column called $Growth.Include.2
. Whereas the previous method (eliminating based on a fixed SD (e.g., 5*sd(tdata$growth)
) would eliminate relatively high and low values, the Condit method tends to exclude stems with high negative growth but few observations of high growth (except for 'extreme' outliers).
For now, just leave them and see if it gives problems with convergence...
Alternatively, select a standard deviation and say it is probably measurement error? e.g.
tdata$Growth.Include <- abs(tdata$growth) < (sd(tdata$growth, na.rm=T) * 15)
tdata$Growth.Include[is.na(tdata$growth)] <- F