In order to get tree growth curves we estabilshed an allometry campaign during which we carefully measured, subsampled, dried and then re-constructed the biomass of 154 trees belonging to 8 species. The sampling is done to allows estimation of biomasses for different compartments in order to see if the different tree species have different growth pattern. The neighbours are as well measured to look at a possible influence of competition on tree growth and tree biomass compartimentation

Trees were randomly chosen with the condition to look healthy and to have a particular DBH: because trees evenly located along the size range are needed to produce accurate growth curve estimate.General point:Every biomass or leaf area estimation come from an independent estimation and then has is own Standard Deviation. All the biomasses have been estimated SEPARATELY for each species in order to preserve specific differences (no pooling toward the grand mean). In all the models, the individual trees were considered as a random factor.The estimates and their standard deviation come from the posteriori distribution drawn out of the Markov chain Monte Carlo (three chains here). Each model was set up with 500'000 iterations from which the first 300'000 were discarded. Among the 200'000 valid iterations, one over twenty was kept in order to avoid the correlation between two consecutive estimates. So, each biomass estimate is the MEAN of the results given by the MCMC. The following variable is the SD of this mean (= "_sd").The model fit a value for each segment or each branch considered. The sum of them is done within the model: the final estimate gets a standard deviation that is right in confront with error propagation.The errors we provide are MODEL ERRORS. Thus they come from the variability around the estimates. Small trees for which every branch was measured in the lab have NO model error: the sd = 0.Bayesian estimation allows to include prior knowledge into the model: the priors. Here, we never used any informative priors, which means that our estimates come only from our own data.Branch biomass estimates: Branches are "branches" in our definition when their diameter is equal or below 3 cm of diameter. Estimates coming from the sampled branches are the total branch biomass, the branch wood biomass and the branch leaf biomass. . For each branch, the total fresh wood weight ant the total fresh leaf weight were measured, subsamples were dried. From the dry mass of the subsamples the total dry mass of the leaves, wood and total branch were gathered.Usually, we have 6 branches / tree which have the real data measured (real branch number can be seen in the "branch_nb" variable). As descriptor variable we have their diameter and their position in the crown. We use those predictors to predict the biomass of all the branches measured in vivo on the tree.For the branch biomass estimates, the model has this form:biomass estimate[i] ~ dnorm(yhat[i],tau[i]) yhat[i] <- c[thrd[i]]*b1[ind[i]]*pow(d[i],b2[ind[i]])tau[i]<-1/pow(sd[i],2) sd[i]<-a1+a2*d[i]where "thrd" is the branch position in the crown ( 3 levels), "ind" are the individual trees and "d" is the branch diameter. This exponential model is fitted as such "thrd" interacts with the diameter. Each trees has a proper set of estimators (it's own b1 and b2) but their are constrained to enter into a normal curve --> they are the random factors. b1[ind[i]] ~ dnorm(b1_mu, b1_tau) --> b1_mu is the average b1 estimated by the model with a precision b1_tau b2[ind[i]] ~ dnorm(b2_mu, b2_tau) --> b2_mu is the average b2 estimated by the model with a precision b2_tauA graphical inspection showed an increase of variability with increasing branch diameter, that's why here the sd is modelized as a linear function of the diameter.An idea of the amount of material really measured versus how much is predicted can be seen with the following variables: "field_bra_tot_weig" gives the real fresh weight of the branches, then "br_kg" is the wheight of the branches brought back to the lab. "p.branch.samp" is the precentage of branches that were brought back to the lab; as espected, the bigger the tree the smaller this percentage. "branch_nb" is the number of branches brought back to the lab.Branch leaf area estimates: The leaf area estimates are based on the same branches than the branch biomass estimates.To get the leaf area estimates, two models were run simultaneously; one to estimate the branch leaf biomass and one to estimate the leaf area/ gr of leaf for this particular branch. Then, these two estimates are multiplied within the model (right error at the branch level) and the resulting leaf area summed for each tree (right error at the tree level).The leaf biomass estimates are done exactly as above, so we don't show it here again. A subset of leaves were scanned and dried for each branch. From this we obtain a mean area/weight at the branch level that can be modeled.We did not find any pattern for the leaf area / gr of leave with the branch diameter, so this variable does not appear in our model. In contrast, leaves tend to have a lower surface / gr when being situated higher in the crown, so "thrd" is in our model. As before, the individuals are the random factor.area[i]~dnorm(yhatar[i],tauar) yhatar[i] <- c2[thrd[i]]*b9[ind[i]]where "thrd" is the branch position in the crown ( 3 levels) and "ind" are the individual treesSegments biomasses: The segments are the part of the wood belonging to the stem or branches with a diameter above 3 cm. The whole volume of the tree is recorded by "segment". In some of those segments, a slice was take and a real density gathered. This density is here modelled, applied to the segment, then multiplied by the segment volume, which give the segment biomass and their error (segment level) and summed within the model to have the biomasses with the right error term at the individual level.segment_density]~dnorm(yhat[i],tau[i]) yhat[i]<-b0[indiv[i]]+b1[s_or_b[i]]+b2[indiv[i]]*llength[i] + b3[indiv[i]]*larea[i] + b4[indiv[i]]*relpos[i] tau[i]<-1/pow(sd,2) biom[i]<- yhat[i]*segvol[i]Where "indiv" is the individual marker, "s_or_b" is a two level factor to indicate if the segment is from the Stem or a Branch, "llength" is the log of the total length of the stem or the branch where the segment is situated, "larea" is the log of the average sectional area of the segment, "relpos" is the relative position of the segment within the branch or the stem. Then, to get the biomass "biom" of the segment "i", we multiply it by its volume = "segvol".Here, to avoid a super complicated model but still letting a lot of individual freedom for the estimates, we don't make interactions but still having different parameter estimates for all the variables at the individual level. . As before, the individuals are the random factor.To know how much biomass was really measured versus estimated, you may look at "sl_kg" that gives the total dry weight of the slices brought back to the lab, "slice_nb" that gives the number of slices brought back to the lab and "p.seg.samp" wich is the percentage of dry biomass really measured compared to the biomass predicted.Dead Attached Material(DAM) biomasses: The DAM were less extensively sampled, so we have only one density value per individual. For some (strange) reasons that occur in the field, quite a certain number of trees don't have a proper sample. So we decided to calculate an average DAM density at the species level (so, here there is no "[indiv]" marker). The individual estimates were used to asses the "noise" around that mean.In the field, we measured the diameter and the length of the DAM. The DAM volume is estimated applying the formula of a cone (might not be the best).Within the WinBUGS model, the density is applied to each DAM which give us its biomass and the error. Then the different DAM are summed at the individual level which provide the error at the individual level. Here the error reflects our uncertainty around the specific mean.the model could not be simpler:DAM_density[i]~dnorm(mu,tau)Total living biomass (tot_bio), total biomass (tot_bio_big) and total wood biomass (total_wood_biomass): To get these estimates with the right error, we re-used the models described above and we made them run in parallel and then we summed their estimates. "tot_bio_big" is obtained by having three models running, one for the DAM, one for the total branch biomass and one for the total segment biomass. Their estimates are summed inside the model. Similarly, the total living biomass is obtain by summing the total segment biomass and the total branch biomass. The total wood biomass is obtained by summing the total segment biomass with the woody part of the branches biomass. ----- data describe among others: variables that describe the tree shape, the place where it was growing and its social situation;

According to Google Map, the site is N 29.113494, E 118.008817. The forest is an exploited plantation with about 80% of the trees being Cunninghamia lanceolata. The 20% remaining is naturally present there and comes from the surrounding natural forest. The site is situated at the end of a small valley with relatively strong slope

Time needed to sample the 154 trees

154 trees belonging to 8 different species (n, DBH size range in cm): 2 coniferous: Pinus massoniana (19, 2.9 – 23.2) and Cunninghamia lanceolata (17, 1 – 17.7), 3 deciduous: Liquidambar formosana (15, 2.6 – 37.5), Sassafras tzumu (20, 5.1 – 27.7) and Alniphyllum fortunei (21, 2.4 – 18.7), 3 evergreens: Castanopsis fargesi (25, 2.5 – 27.3), Castanopsis sclerophylla (16, 2.7 – 16.5) and Schima superba (21, 1.2 – 23.1).

Exponential curves according to species and other covariates; This data are not "raw", to get the biomasses from the raw data one needs to many steps, so here we provide the processed data and the methods required to get them