[Title in English: Statistical models of cumulative impactrs of human activities on wild reindeer areas: Identifying important grazing areas and scenario analyses for impact assessment and spatial planing]
Summary: Abstract
Wild reindeer are threatened both nationally and internationally, and several populations have declined sharply or become extinct in recent decades. Several factors are responsible for these trends, but it is well documented that infrastructure development and disturbance by humans can cause habitat degradation and fragmentation and can have a major negative impact on reindeer populations. This led to a large amount of research aimed at understanding and quantifying anthropogenic disturbance on reindeer and supporting sustainable land planning.
In the last decade we have developed statistical methods and a software (ConScape) to quantify the functionality of reindeer habitats, to identify movement corridors, and to measure the degree to which these are affected by the piecemeal development and "cumulative impact" of different infrastructure and human activities. The models are based on more than 3 million GPS-positions from the largest wild reindeer management areas, and on ca. 350 environmental layers describing the landscape (topography, vegetation, roads, hydropower, trails, tourist volume, private cabins, tourist cabins, etc.), climate, and, in some areas, local knowledge.
Based on these models, several maps have been produced to describe statistically how wild reindeer perceive resources and barriers on a local scale, and which areas and corridors are
most functional in the entire landscape. The latter is particularly important, as reindeer perceive the landscape as a continuous network of functional areas that they can access through
movement corridors. On the other side, reindeer also perceive the cumulative impact of different infrastructures and human activities and respond by avoiding or decreasing the use of some areas due to disturbance and/or barriers to movements. All maps, a description of methods, and the reference list to scientific and popular publications are available in the Web App:
https://www.nina.no/Naturmangfold/Hjortedyr/reindeermapsnorway.
In addition, we developed a simulation tool to guide sustainable land planning and impact assessment through scenario analyses. The tool has already been used to predict the expected
effect of 80 mitigation measures, suggested by boards of local experts to minimize cumulative impacts on reindeer habitat functionality and movement corridors in several areas in Norway
(see Appendix 1 and Web-App). The scenarios that can be tested to include a combination of the removal/closure/relocation of existing infrastructures, changes in their intensity of use, the
construction of new infrastructure (e.g. cottage, road, hiking and skiing trial, bridge over water magazines), and climate change.
The statistical approach, software, maps, web-app and scenario analyses were developed within the Norwegian Research Council projects led by NINA “RenewableReindeer”, «ProdChange» and “OneImpact”, and in the related project "Scenario analyse Øyulvsbu og formidling". Comparison between statistical and expert-based maps was funded by the project from the Norwegian Environment Agency «OneImpact og kvalitetsnorm for villrein». These projects were or are supported by several partners and funding sources including the Norwegian Environment Agency, the Directorate of Water and Energy, the Hydropower company Sira Kvina, the wild reindeer project in Setesdal, the Wild reindeer centre / wild reindeer council, Siri Bøthun nature management Norwegian University of Life Sciences NMBU, and several international collaboration partners (Universite Catholique Louvain, Julia Computing Inc, Sveriges landbruksuniversitet SLU, University of Guelp, Canada, University of Alberta, Canada, The Nature Conservancy, Universidad Politecnica de Madrid, University of Glasgow).
Parallel to this analytical work, management measures have been undertaken in Norway to counter the degradation of wild reindeer areas. In 2020 the government adopted the “Quality
Standards for wild reindeer in Norway” (Lovdata 2020), to meet both international obligations and national objectives for the conservation of viable populations within ecologically functioning
habitats. Every fourth year, each of the 23 Norwegian sub-populations is classified into good (green), medium (yellow) or poor (red) quality, based on 3 sub-standards: 1) population
conditions; 2) lichens; 3) human impact on habitats.
Sub-standard 3 is therefore an important tool to ensure sustainable management of wild reindeer habitats. Sub-standard 3 aims to identify critical declines in habitat quality and connectivity in each wild reindeer area, so that the cause of the decline can be addressed, and the status of the area restored to acceptable levels (yellow) as soon as possible. In the longer term, the aim is that all national wild reindeer areas should be of good quality (green). The procedure thus needs to robustly identify areas where habitat loss and fragmentation increased above a critical threshold, and to provide information on the relative contribution of the responsible infrastructure and human activities.
In the current Quality Standards, sub-standard 3 is implemented based on expert-based assessments of changes in available wild reindeer observations, that are assumed to reflect changes in human impact. The process involves the delineation of polygons representing reindeer seasonal ranges, corridors, and a set of focal areas where socio-ecological challenges related to human activities have been identified. The reduction in use of focal areas is assessed by the experts in the last decade compared to the previous four decades. The classification is then conducted by assessing the proportion of habitat lost within all focal areas, as compared to the habitat available within the seasonal range.
In this report we provide an overview of the statistical approach, and we compare the performance of what we for simplicity call the “statistical maps” to the “expert-based polygons” developed in sub-standard 3, using both visual and quantitative approaches. We discuss the lessons learned by comparing the two approaches, and how these can help achieving the sustainability goals for the management of wild reindeer areas. This was the main focus of the project "OneImpact and quality standard for wild reindeer" funded by the Norwegian Environment Agency. As a proof of concepts, we also perform a preliminary statistical classification of the state of the wild reindeer areas following Delnorm 3 (“Sub-standard 3”). Last, we show how statistical approaches and simulations can help identifying the most effective among 76 measures suggested to mitigate cumulative impacts from anthropogenic activities in Setesdal, Nordfjella and Snøhetta (Appendix 1).
Research on cumulative impacts of infrastructure and human activities
Reindeer areas are not threatened by a single anthropogenic factor such as a hydropower reservoir, a road, a tourist resort or a cabin village. Reindeer is a migratory species that require wide and well-connected ranges, whose functionality is threatened by the combination of all these factors together. The simultaneous presence of several sources of disturbance spread across the landscape creates a series of obstacles to their free movements that can force reindeer to use only parts of their originally large ranges, and abandon pastures used in the past. Infrastructures and human activities that, individually, may have only little effect on reindeer, can together cause a large cumulative impact. This depends on how strong the impact of each of these activities are, how often or how many of these infrastructures occur on
the landscape, how far their effect can be perceived by reindeer, and on their position with respect to important reindeer grazing areas and corridors.
Statistical models are the most efficient way to understand complex interactions involving many variables at high resolution, over large areas. In the past decade we built data infrastructure,
methods and software and we used them to quantify the effect of more than 200 data layers describing vegetation, topography, climate, infrastructures (e.g. roads, hydropower, cabins, trails) and human activities (e.g. tourist volume) on more than 400 GPS-monitored reindeer.
The approach builds upon two statistical analyses: one quantifies habitat quality or habitat loss (probability to use an area, taking into account vegetation, topography, climate, infrastructures
and human activities); the other quantifies fragmentation, or barriers to movements. These analyses offer robust estimates of the magnitude of the effect of each infrastructure and human
activity, and of the distance at which such effects can be perceived by reindeer, for each 100 m pixel in Norway. We also synthesised in different ways human impact in each 100 m pixel in
Norway (current “human footprint”; “natural potential” etc), and we ranked the main factors causing habitat loss in each reindeer area. Last, we took a “bird-eye-view” of the landscape and developed new models showing all movement corridors and the most functional areas.
Models can be used also to predict beyond the data (i.e. extrapolation) in areas where no GPS data are available, or under past or future scenario of infrastructure development (e.g. build
roads, cabins, hydropower), mitigation measures (e.g. move a cabin, reduce or re-direct tourism), or climate change (e.g. changes in snow).
Comparing statistical and expert-based approaches
Statistical maps and expert-based polygons of seasonal ranges generally agree on the location of suitable reindeer pastures, though meaningful comparisons are prevented by the size of the
polygons, that at times extend to the entire management area. However, statistical models perform significantly better than the polygons in discriminating between areas used or not used
by reindeer, according to both GPS data and to reindeer observations. This is partly because the polygons are delineated by hand, and necessarily include “obviously” unsuitable areas
such as hydropower reservoir, roads, cabins etc. We show that this risks to underestimate habitat loss in the classification of Sub-standard 3.
Statistical models use data on infrastructure and human activities to quantify anthropogenic impacts directly and in real-time (as soon as new developments are updated in the data). Models can also be used to prevent impacts by testing the effect of land development plans, before infrastructures are built. The expert-based approach on the contrary is set to detect habitat loss or fragmentation only following a decline in reindeer observations in an area. This risks to delay or prevent the detection of human impacts, especially in fragmented reindeer areas where human impact is widespread, and reindeer have no access to refuge areas. In such areas, reindeer have to stay, and might suffer higher stress levels, with consequences for individual conditions.
Statistical models provide high-resolution estimates of cumulative impacts, for every 100 m of the landscape. This implies that all sources of disturbance are quantified, including both highintensity
ones (e.g. high-traffic roads, railway), or sources of disturbance that may be of lowerintensity, but diffuse and widespread in the landscape (e.g. network of hiking trails, private cabins). This allows avoiding the risk of failing to address cumulative impacts of low-intensity, diffuse sources of disturbance. A simulation showed that the same amount of disturbance spread across the entire wild reindeer area or concentrated within a specific location can lead to a vastly different classification following Sub-standard 3, as the procedure currently does not account for diffuse, low-intensity sources of disturbance.
Another advantage of using high-resolution model estimates is that there is no need to subjectively outline several polygons (i.e. one for each focal areas, seasonal range, corridor and influence area - for each season). Using simulations, we showed that the process of outlining the polygons can entirely determine the final classification of each wild reindeer area following Sub-standard 3. Hence, the final classification can be determined at the first stage of Sub-standard 3, when the polygons are drawn, irrespective from the following expert-based assessment of a possible decline in reindeer area use therein. This is because the classification depends on the proportion of the seasonal ranges covered by the focal area, and thus their respective size can fully determine the assessment. Indeed, most of the four areas we assessed could only be classified as “green” or “yellow” – not “red” – as the sum of their focal areas is too small to allow the area to be classified as red. Hence, deciding whether to include or exclude a hydropower reservoir or a glacier in the polygon may determine the final classification of that area.
Statistical analysis coupled with network models can simultaneously assess habitat loss and fragmentation (corridors and functional habitat) caused by multiple sources of disturbance and taking into account also a variety of natural factors. In other words, statistical models assess Substandard 3A (on habitat loss) and 3B (on fragmentation) within the same framework, in a
standardized way, within and across reindeer areas.
Can statistical models support decision-making processes?
The reliability of both data and models increased rapidly in recent decades, and the report shows that results are robust with respect to available reindeer data. Feedbacks from local experts are generally positive, and the results correspond fairly well with local knowledge, even in areas where GPS data are not available.
“All models are wrong, but some are useful”. The question is not “are the model representing the truth?” (no, they never will), but “are they useful to achieve the goals of Sub-standard 3?”, or “can they help identifying the most efficient mitigation measures?”. In their current form, the models seemed useful to understand why reindeer use a specific area, or why they no longer use it, which human activities cause the strongest impact in different seasons, or what is the expected impact in areas with little data or local knowledge. Statistical models can also be useful to identify areas, human activities and infrastructures that would need to be prioritized for conservation, mitigation or restoration both locally and across reindeer areas.
These models were not made for Sub-standard 3, but are a flexible, “living product” that can be adjusted, updated and improved, for instance by adding data on bucks, changing the seasons or, most importantly, by integrating local knowledge. The models are operational, flexible, reproducible, and could support Sub-standard 3 in a variety of ways.
The report discusses in detail strengths and weaknesses of both statistical and expert-based approaches and concludes that a robust interaction between statistical models and local
knowledge would be ideal for maximizing the probability of reaching the sustainability goals of Sub-standard 3, while minimizing the risks highlighted in this report.