Comparison of stratified and non-stratified most similar neighbour imputation for estimating stand tables

Many growth and yield simulators require a stand table or tree-list to set the initial condition for projections in time. Most similar neighbour (MSN) approaches can be used for estimating stand tables from information commonly available on forest cover maps (e.g. height, volume, per cent canopy cover and species composition). Simulations were used to compare MSN (using an entire database) with two stratified MSN approaches. The first stratified MSN approach used species composition to partition the population into two inventory type strata, while the second stratified MSN approach used average stand age to partition the data into two stand development stages (strata). The MSN approach was used within the whole population and within each stratum to select a reference stand and to impute the ground variables of the reference stand to each target stand. Observed vs estimated stand tables were then compared for the stratified and non-stratified simulations. The imputation within a stratum did not result in better estimates than using the MSN approach within the whole population. Possible reasons for poor performance of stratified MSN are provided.