Evaluation of remote sensing methods for continuous cover forestry

The overall aim of the project was to investigate the potential and challenges in the application of high spatial and spectral resolution remote sensing to forest stands in the UK for Continuous Cover Forestry (CCF) purposes. Within the context of CCF, a relatively new forest management strategy that has been implemented in several European countries, the usefulness of digital remote sensing techniques lie in their potential ability to retrieve parameters at sub-stand level and, in particular, in the assessment of natural regeneration and light regimes. The idea behind CCF is the support of a sustainable forest management system reducing disturbance of the forest ecosystem and encouraging the use of more natural methods, e.g. natural regeneration, for which the light environment beneath the forest canopy plays a fundamental role.

The study was carried out at a test area in central Scotland, situated within the Queen Elizabeth II Forest Park (lat. 56°10' N, long. 4° 23' W). Six plots containing three different species (Norway spruce, European larch and Sessile oak), characterized by their different light regimes, were established within the area for the measurement of forest variables using a forest inventory approach and hemispherical photography. The remote sensing data available for the study consisted of Landsat ETM+ imagery, a small footprint multi-return lidar dataset over the study area, Airborne Thematic Mapper (ATM) data, and aerial photography with same acquisition date as the lidar data.

Landsat ETM+ imagery was used for the spectral characterisation of the species under study and the evaluation of phenological change as a factor to consider for future acquisitions of remotely sensed imagery. Three approaches were used for the discrimination between species: raw data, NDVI, and Principal Component Analysis (PCA). It can be concluded that no single date is ideal for discriminating the species studied (early summer was best) and that a combination of two or three datasets covering their phenological cycles is optimal for the differentiation. Although the approaches used helped to characterize the forest species, especially to the discrimination between spruces, larch and the deciduous oak species, further work is needed in order to define an optimum approach to discriminate between spruce species (e.g. Sitka spruce and Norway spruce) for which spectral responses are very similar. In general, the useful ranges of the indices were small, so a careful and accurate preprocessing of the imagery is highly recommended.

Lidar, ATM, and aerial photographic datasets were analysed for the characterisation of vertical and horizontal forest structure. A slope-based algorithm was developed for the extraction of ground elevation and tree heights from multiple return lidar data, the production of a Digital Terrain Model (DTM) and Digital Surface Model (DSM) of the area under study, and for the comparison of the predicted lidar tree heights with the true tree heights, followed by the building of a Digital Canopy Model (DCM) for the determination of percentage canopy cover and tree crown delineation. Mean height and individual tree heights were estimated for all sample plots. The results showed that lidar underestimated tree heights by an average of 1.49 m. The standard deviation of the lidar estimates was 3.58 m and the mean standard error was 0.38 m.

This study assessed the utility of an object-oriented approach for deciduous and coniferous crown delineation, based on small-footprint, multiple return lidar data, high resolution ATM imagery, and aerial photography. Special emphasis in the analysis was made in the fusion of aerial photography and lidar data for tree crown detection and classification, as it was expected that the high vertical accuracy of lidar, combined with the high spatial resolution aerial photography would render the best results and would provide the forestry sector with an affordable and accurate means for forest management and planning. Most of the field surveyed trees could be automatically and correctly detected, especially for the spruce and larch plots, but the complexity of the deciduous plots hindered the tree recognition approach, leading to poor crown extent and gap estimations. Indicators of light availability were calculated from the lidar data by calculation of laser hit penetration rates and percentage canopy cover. These results were compared to estimates of canopy openness obtained from hemispherical pictures for the same locations.

Finally, the synergistic benefits of all datasets were evaluated and the forest structural variables determined from remote sensing and hemispherical photography were examined as indicators of light availability for regenerating seedlings.