Using model inputs from remote sensing to improve forest production forecast

The UK Forestry Commission maintains a subcompartment database of public forests which includes information regarding species, planting year, allocated management regime and habitat conditions. Currently traditional field methods are used to assess stands and to assign a yield class to predict expected productivity. These are based on field survey plots and this sampling approach aims to produce stand-level estimates which are representative of the whole. However, it is recognised that stands often demonstrate considerable heterogeneity, particularly where disturbance has occurred (e.g. management practices, wind damage, etc.). In addition, for stands which contain mixed species, the spatial distribution of species is not mapped. Where productivity potential is uncertain, a default yield class may be assigned which can differ considerably from actual potential. This study uses the Aberfoyle Research Forest in Scotland, to investigate the use of remote sensing to improve the currency of biophysical parameters estimated for forest stands and to realign forecast boundaries to better represent the heterogeneity present. eCognitionsoftware was used to re-segment forest stands into more homogeneous units whilst respecting National Forest Inventoryrequirements of minimum mapping size and form. UK-specific growth models were restructured to allow the estimation of yield class from lidar. Lidar estimates of yield class, Top Height and percentage cover were then assigned to the sub-stands. These were used as stand assessment observations at the date of the lidar flight, and were combined with species and age information in the subcompartment database to drive production forecast models. This allowed a comparison to be made between forecasts using the traditional current approach and estimates using remote sensing observations.