Anna Derkacheva, Gerald “JJ” Frost, Howie Epstein, and Ksenia Ermokhina, of HSE University in Russia, Alaska Biological Research, Inc., the University of Virginia, and the Russian Academy of Sciences (respectively), discuss their article: Landscape patterns of shrubification in the Siberian low arctic: A machine learning perspective
The Arctic tundra is experiencing some of the strongest effects of climate change on Earth. Changes can readily be observed within the span of a human lifetime as the reorganization of ecosystems and landscapes. Rapid expansion of tall shrubs into land previously occupied by short-statured tundra plants is an example of this and is one of the key components of “tundra greening” evident in long-term satellite records since the early 1980s. Shrub expansion includes both infilling of existing shrub canopies, and colonization of new locations. In our research, we evaluated the extent and driving local factors of “shrubification” by tall alder shrubs (genus Alnus) for three locations in northwestern Siberia, Russia.
High-resolution satellite images provide sufficient detail and coverage through time to observe and quantify recent shrub propagation in tundra landscapes. We obtained satellite image pairs acquired 10–15 years apart for three landscapes in northwestern Siberia, with a resolution that allowed for the identification of individual shrubs. We then applied artificial intelligence (AI) to create a specially trained neural network (convolutional neural network, or CNN) that recognized alder shrubs through three stages of canopy development—early colonization, open stands of mature shrubs, and closed stands with dense canopies—within our satellite imagery. This network was able to track shrub distribution and its changes hundreds of times faster than manual techniques, with comparable quality!
The second piece of our puzzle was to understand the landscape-level factors that make a habitat “attractive” for expansion of alder shrubs. Because we are operating in study areas of ~50 km2, local topographic variables are most likely to influence patterns of shrub occurrence. For the typically cold and wet Arctic, the main local influences are solar heating and ground wetness (i.e. microclimate). We evaluated these two factors, developing maps of these variables using a digital elevation model known as the ArcticDEM.
Statistical analyses confirmed some existing knowledge but also revealed other relationships that were not previously evident. For example, alder historically occupied well-drained south-facing slopes, which provided the best conditions for that species at the northern margin of its range. However, contemporary alder expansion appears to be occurring in less optimal microsites—across a range of solar radiation and wetness levels—reflecting the potential for shrubs to increase their landscape “footprint” as a result of warming temperatures. For one of our study sites, shrub occurrence increased by over 25% per decade, in contrast to another site that increased only 2.5%. While a warming climate is the main driver of shrub expansion at the circumpolar scale, patterns of shrub expansion at the landscape scale make little sense unless the availability of suitable colonization sites is considered.
While this work was conducted at only a few locations in the Siberian Arctic, our analysis pipeline using AI-driven satellite image processing, extraction of environmental factors from the ArcticDEM, and statistical cross-evaluation can be applied to larger areas with the same detail. A better understanding of Arctic shrubification goes beyond the scope of plant ecology. Many regional or global ecosystem models (e.g. for climate, carbon cycling, permafrost) rely on descriptions of actual and future vegetation cover and community composition, and the switch from largely herbaceous cover to tall shrubs is a significant environmental change that will affect Arctic wildlife, permafrost, human land-use, regional climate, and more.
