SATELLITE MONITORING OF DISTURBANCES IN ARCTIC ECOSYSTEMS

ABSTRACT This study explores the capability of satellite remote sensing to detect temporal changes in northern Fennoscandian regions through the application of a temporal model of surface bidirectional reflectance. Remote sensing offers the potential to monitor changes over large areas and at locations of difficult access. Specifically in remote Arctic locations, where ground surveys and aircraft observations are constrained by weather conditions and logistics, remote sensing provides a unique capability for repetitive and frequent sampling. One of the main disturbances in mountain birch forests typical of northern Sweden and Finland is caused by outbreaks of defoliating insects such as the autumn moth (Epirrita autumnata) and the winter moth (Operophtera brumata). These outbreaks occur more or less cyclically every 9-10 years and attack mainly birch (Betula spp.) leaving a mosaic of open woodland within the forest Other disturbance detected in Arctic regions are extreme and sudden winter warming events in which temperatures increase rapidly to above freezing, often causing snow melt across whole landscapes and exposure of ecosystems to warm temperatures The algorithm must detect sudden changes in the surface bidirectional reflectance distribution function (BRDF) which are indicative of a loss of vegetation, in our case the loss of foliar biomass caused by the moth caterpillar activity. In the case of the warming extreme events we look for winter snowmelt events followed by a decrease in vegetation index values during the growing season. Ideally, to successfully apply these algorithms, the surface state should remain static prior to any disturbance which is being sought. Dealing with deciduous forest this condition can only be accomplished working over a sliding window of a couple of weeks (unless some concept of phenology is included in the model). A further complication is the potential scarcity and sparseness of data available due to cloud cover. To tackle these problems we use data of both Terra and Aqua satellites in order to have as many samples as possible. The incorporation of BRDF effects in the algorithm [4] allows us to use reflectance data obtained on different days and under different geometries. Data can be unavailable for several days due to cloud cover, and natural phenological changes occurring during these periods would be detected as disturbances by the algorithm. To prevent this the algorithm must be able to account for underlying changes that can be due to vegetation phenology, as done i