Use of automated image analysis to detect changes in megafaunal densities at HAUSGARTEN (79°N west off Svalbard) between 2002 and 2004High latitudes are amongst the most sensitive ecosystems with respect to climate change, which prompted the launch of the first and only deep-sea long-term observatory beyond the polar circle, HAUSGARTEN (eastern Fram Strait), in 1999. An understanding of the abundance and spatial distribution of organisms is vital to assess the effects of global change. To map the distribution of megafaunal organisms deep-sea research relies strongly on the use of high-resolution cameras that are fitted to towed sledges, drop-down frames and remotely operated or autonomous underwater vehicles. Inevitably, such techniques generate large quantities of footage. The visual analysis of images is labour-intensive, time-consuming and subjective. Over recent years, the increase in computer power has facilitated advances to automate image analysis. Here, we present a novel approach for the automatic detection and classification of particular biological classes within image data. For this purpose, we applied machine-learning algorithms to two photographic transects from the HAUSGARTEN central station (2500m) taken by an ocean floor observation system in 2002 and 2004. Each transect contains some 700 photographs. The main goal is the automatic identification of important biological classes (e.g. sea cucumbers, sea lilies) to assess the densities of the most frequent organisms over time. So far, our system shows a promising performance in detecting sea cucumbers and sea lilies with sensitivities and positive predictive values between 75 - 80%.Results from manual analysis of 66 images taken at the central part of the transect indicate a significant decline in the mean density of sea cucumbers (Elpidia glacialis), sea lilies (Bathycrinus cf. carpenteri), burrow entrances and total megafaunal densities from 2002 to 2004 which concurs with a decrease in sea ice coverage, particulate flux to the sea floor, sediment-bound nutrients and pigments, microbial biomass and changes in meiofaunal community structure. Results from automated image analysis will increase the spatial resolution and statistical power of our analyses as it enables us to process larger quantities of images.