ISSUES

should linear spline for age be on log(X) or on X?
autocorr in true incr
autocorr in incr msmt error - probably hard
dbh outliers - probably handled by t dist
error in neil dbh?
biomass scaling
long-term trends (often down for smallish trees) - how deal with - linear trend? kernel-smoothed AR process on 10-year interval knots; cubic spline with few knots?
Wishart on covar of spline coefficients to induce long-term structure? but if parameterized in terms of calendar time, we have less info as go further back...
work on having alive/dead be random for periods in between censuses
data issues - confer with Neil and Dan
experiment w/ taking out dbh data except neil's and seeing if we predict dbh and total biomass increment well
can I use census data from outside plots to estimate increment variations by size and taxa? is large incr of fagr/tsca/beal consistent with these data?
deal w/ uncertain date of death - sometime before census in which recorded as D
get 1962 census data
include trees that were dead at all times in census but recorded if go back further intime - only 2 of these and both are ~13 cm dbh in 69
include dead dbh measurements with more error and an offset?
put in tree 307 as having neil's dbh in 2012 so can use incr data
could all increments be systematically off - add a bias/error term for this which deals w/ one aspect of possible autocorrelation in increments - quantify this with Neil's data from different #s of cores per tree
assess impact of excluding small trees from analysis in terms of biomass increment
dbh obs error on log scale?
incr obs error on log scale?
how estimate NPP over stand given the 3 plots?
how incorporate info from nested rings?
does increment decrease in years just before death? - not a lot of evidence in cores as few dead cored trees but dead trees were generally growing slowly before death; need to look at dbh data too
what paper had growth be a multiple of tree size? - could think about this - but maybe essentially same as incr being on log scale
do I need to model mortality as related to growth so that can model when ghost trees died?
can we do upper bound on tree size via dinterval and not dconstraint
might also want t dist for x[i,j] process to account for very low values that occur occasionally; this may help with sig_x, sig_x_obs mixing
any way to get at long-term autocorr in increment?  probably hard and boulder jct discussion suggested maybe it's ok not to address it

* t dist for dbh error
* 1991 was actually 3 years
* dbh is bgin/end of grow season? -jun/jul - have prior on how much of ring is laid down - put at .2-.8 and estimate? use Audrey's dates - estimate a linearly scaled effective growing season between 4/1 and 10/1 - try logistic transformation
* should include info that if was alive in later survey and not present in earlier, that means it was less than 5 cm observed (e.g., tree 5837) - actually there are msmts of 4.57 cm, so use 4.5 as upper bound
* spp variability - have linear spline for age have spp-specific intercept (exchangeable) and constant slope
* effect of age on increment - try linear spline - baseline constant plus slope at size of 25/30 cm dbh (originally was thinking to have flag for small/shade-intol taxa (i.e. indivs)?)try lin spline with spp-specific intercept and pooling of intercepts