Forest management volume inventory using lidar-based endogenous post-stratification

: Endogenous post-stratification (EPS) is a statistical method used to approximate traditional post-stratification when population-level counts are unknown. It involves assigning sample units to strata based on predicted values from a model trained on the same sample data. The population is then stratified based on fixed stratum boundaries. We evaluated the EPS method for estimating growing stock volume (GSV) for a management-level forest inventory. In such an inventory, a forest district is divided into age classes, i. e. , first-stage strata, by means of auxiliary information on the age of forest stands. Each of these age classes is then divided into three to six second-stage strata using an internal model developed using data on GSV in terrestrial sample plots and with airborne laser scanning (ALS) metrics. The first of our research questions concerned whether traditional post-stratification estimators are relevant for estimating the mean and its variance in the EPS method. The second question was whether the EPS method is more effective than the reference (EXPREF) method, in which age classes are not divided into second-stage strata. We conducted our research using a Simulated Forest District. This dataset contained simulated GSV values for individual grid cells, which were treated as “terrestrial truth”, as well as a set of original ALS metrics. We used these data to perform 1, 000 replicate sample simulations, each based on 500 sample plots. The estimated mean GSV for both the entire Forest District and the 11 individual age classes did not differ significantly from the true values. The frequency with which the true mean value fell within the confidence interval for the population mean GSV was aligned with the theoretical frequency for all inventory units. Our results show that the use of the EPS method would allow us to reduce the number of terrestrial sample plots by more than half compared with the method in which age classes are not divided into second-stage strata. Further research could focus on the use of variance estimators other than the simple expansion estimator when systematically located sample plotfs are used in the EPS method.