Satellite optical remote sensing of forest canopy cover in boreal and tropical biomes

Continuous field of tree cover, or canopy cover (CC) is an important information required in ecology and international forestry. In ecology, CC quantitatively depicts the spatial heterogeneity of tree cover, and is therefore useful for spatially-explicit characterizations of ecosystem state, changes, structure, and functioning. In international forestry, CC threshold is a main criteria for a consistent definition of forest land cover, thus facilitating a globally comparable forest area statistics. In boreal forests, CC information supports the social and ecological aspects of forest management objectives. In tropical rainforests, CC is a biophysical indicator for forest degradation, and hence supports international climate policy mechanisms. The aim of this dissertation was to investigate the capability of the freely-available medium resolution, passive optical satellite data for estimating CC in boreal and tropical forests. The boreal study area included sites that span a large latitudinal gradient in Finland, encompassing a large variation in forest structure, species composition, and site fertility. The tropical study area included sites which have experienced forest degradation and deforestation, in the Borneo mega-island, South East Asia. The results showed that, in boreal forests, large area CC prediction across multiple Landsat scenes was feasible, with accuracy comparable to single-site (single-scene) CC prediction. Beta regression model with individual red spectral band as predictor was found optimal. Physically-based analysis of the sources of variations in canopy reflectance indicated that, in red band, canopy reflectance was most sensitive to CC variations, in both the boreal and tropical biomes. The new Sentinel-2 data provided a slight improvement in CC prediction accuracy, compared to Landsat-8 data. The improvement was associated with the new 705 nm red edge spectral band. In tropical rainforests, current CC variations due to varying intensities of past selective logging, could not be estimated from Landsat data. Discriminating rainforests with different degrees of past selective logging using Landsat data was not possible, due to similarity in the present canopy structural properties that drive forest reflectance. Finally, sub-annual deforestation monitoring in the insular South East Asia was feasible, using a continuous change detection algorithm based on a consecutive anomalies criterion applied to dense Landsat time series. This dissertation concluded that, in boreal forests, CC estimation accuracy can be improved most logically by accounting for the sources behind the scatter in the relationship between canopy reflectance and CC, using a physically-based approach. It was inferred that the most important sources are the reflectance adjacency effect, variability in understory reflectance, and variability in canopy shadows and scattering. In tropical rainforests, complete or partial changes in CC can be most accurately detected if done immediately as it happens, and thus continuous monitoring with integrated Landsat and Sentinel-2 data is essential.