Individual tree detection using airborne laser scanning (ALS) can provide relevant data to complement forest inventory data. Local Maxima-based (LM) methods for individual tree detection are suitable for applications over large extents, but their performance depends on the type of pre-processing of the input data, as well as forest structure and composition. We developed a model that improves LM through statistical modeling using prior knowledge about forest structure. The model selects the optimal canopy height model (CHM) pre-processing filters based on forest structure variables like the dominant canopy height and degree of cover derived from ALS data, the dominant leaf type derived from Sentinel data, and terrain metrics. The model performance was evaluated by assessing tree detection errors for the canopy stem count in National Forest Inventory (NFI) plots in Switzerland (n=5254). For plots with point densities of more than 15 points per square meter and, at most, 6 years between ALS acquisition and inventory (n=2676), the results showed a mean absolute error of 61 stems per ha compared to 174 stems per ha when detecting trees using an unprocessed CHM. The model showed a stable performance for different dominant leaf types (broadleaved-dominated, mixed, coniferous-dominated) and for different degrees of cover. We consider the developed model to be suitable for applications that require data on forest structure or individual tree positions and heights over large areas.