No abstract

Characterizing seedling stands with respect to their species proportions and co-occurring vegetation is important for monitoring the desired development of the forest stand. Related inventory information has traditionally been collected with costly field surveys and national forest inventory (NFI)-based models. Here, we present a novel fusion approach to combine remote sensing (RS)-based models and NFI-based models to predict seedling stand characteristics, i.e., height, density, and tending needs. We used the best linear unbiased predictor for the fusion of the NFI- and RS-based models. The NFI-based models were derived using NFI sample plots and stand features. The RS-based models were derived using ALS and color-infrared images and separate field-measured data. NFI-based models were found to be rather unreliable (RMSE = 65–115% for stem density and 59–78% for height), but they were always available without the need for any additional RS data. RS-based models provided an RMSE of 41–92% for stem density and 26–45% for height. The fusion procedure used at the prediction stage consistently increased the accuracy of all variables of interest, but the improvements were minor. In addition, we classified the tending need in seedling stands if the height of the coniferous tree was less than 1 m compared to broadleaved trees. If we simulate the decision-making situation of tending needs, we can predict tending needs (91%, user accuracy) fairly well for a stand.