Discussion and Conservation Implications

Due to the remoteness and restricted accessibility, there are few historical observations of the summering ground of this population (Ruokonen et al . 2004), and our current knowledge on the breeding distribution and habitat preference is limited (Fig. 1 and see Artiukhov and Syroechkovski-Jr. 1999; Egorov and Okhlopkov 2007; Solovieva and Vartanyan, 2011). In this context, rapid development of animal tracking technologies offers new insights to determine distribution range and habitat preferences (Kays et al . 2015). In this study, we combined historical records with recent tracking data to model potentially suitable areas of the east sub-population of A. erythropus across the more than 7,400,000 km2 of arctic and subarctic of north-eastern Russia. Our findings assist conservation of this threatened species by identifying the most suitable breeding grounds and assessing existing and future threats. As A. erythropus often co-occurs with other geese (e.g. Greater White-fronted goose (A. albifrons ), Bean Goose (A. fabalis ) and Brent Goose (Branta bernicla ) and other waterfowl including ducks and tundra swan (Hodges and Eldridge, 2001; Pozdnyakov, 2002; Krechmar and Kondratiev 2006), the breeding habitat map could also be used for prioritizing waterbird conservation including through identification of high-priority conservation areas.

Model accuracy and breeding range

In recent years, animal tracking point data have been used in SDM construction either through direct use for model fitting (Williamset al . 2017) or for validating the output of the model (Pintoet al . 2016). By combining three-year tracking data and historical surveys, our dataset represents the most comprehensive presence record and offers a solid basis to delineate the breeding range of the poorly-known eastern sub-population of A. erythropus . The cross-validation results showed that the training and testing AUC are both high (i.e. greater than 0.92) and comparable, suggesting that the output is highly reliable (Phillips and Dudík 2008).
The Maxent output suggested a continuous rather than patchy breeding range of the A. erythropus on the plains adjusted to the Laptev, East-Siberian and Chukchi Seas and in the Anadyr Lowland. Within this over 4,000 km area of coastal plains, the Lena Delta, the wide Yana-Kolyma Lowland and smaller lowlands of Chukotka represent the most extensive breeding area with the highest probability of occurrence (Figures 3 and 4). While there are suggestions that breeding ranges of West and East Asian sub-populations overlap between 103 and 118 E, our work did not confirm this. The flat and rolling subarctic tundra is among the most productive wetland system in north-eastern Russia (Gilget al . 2000). Vegetation characteristic in this area is typical tundra, southern tundra with shrubs and forest-tundra with sparse patches of larch (Larix spp.) Yurkovskaya (2011). A current IBA (Important Bird Area), including the four main deltas (i.e. the Kolyma, Indigirka, Yana and Lena), covers about 34% of the modeled breeding range (BirdLife International, 2017). However, the majority of the coastal plains, extending up to 450 km inland (Figures 3 and 4), and valleys of large rivers are not included in this IBA. Although there are several Wetlands of International Importance under the Ramsar Convention on the Kamchatka Peninsula, the closest to the study area (Parapolsky Dol) does not contain habitat the modelling suggests as suitable. Highly suitable habitats in the study areas have legal protection through declaration as Federal (State) Nature Reserves: Ust-Lenskiy, Olekminskiy and Magadanskiy, and also by Kytalyyk and Beringia National Parks.

Environmental characteristics of breeding habitat

The selection of environmental variables is a critical step in SDM (Araujo and Guisan 2006; Fourcade et al . 2018), and hundreds of environmental factors have been utilized in Maxent (Bradie and Leung 2017). These predictor variables can be loosely grouped into four main groups: limiting factors that control the ecophysiology of the species concerned (e.g. temperature, precipitation, pH); resource factors (e.g. vegetation, water areas), which are supplies needed by the organisms to survive; disturbance factors including anthropogenic and natural perturbations in the environment; and landscape factors, which can be related to the species dispersal limitations (Guisan and Thuiller 2005; Vuilleumier and Metzger 2006).
The geomorphological predictors (i.e. elevation, distance to streams and local deviation from global) collectively contributed to 61.4% of the model gain based on permutation test. This level of relative importance was considered very high for Maxent modeling (Bradie and Leung 2017). The decisive role of topography in controlling the distribution of summering grounds might be attributed to strong preference of river valleys and lowlands, especially considering reduced mobility of geese during breeding and molting periods (Akesson and Raveling 1982). Kosicki (2017) demonstrated the importance of topography for modeling the distribution of both lowland and upland bird species, and omitting topographic variables could lead to substantial overestimation of distribution range, especially for rare species. The response curves show that A. erythropus selects lowlands with a concave shape as preferred habitat, which is consistent with field observations (e.g. Artiukhov and Syroechkovski-Jr. 1999; Egorov and Okhlopkov 2007; Solovieva and Vartanyan 2011), which reported the bird bred and molt in river valleys.
The majority of Maxent models include climate variables as limiting factors, and most studies found temperature and precipitation were very important variables (Bradie and Leung 2017) as climate is believed to be the most important factor for species distributions (Gaston, 2003). It is therefore not surprising that climate variables including precipitation and temperature were also important for A. erythropus . A significant finding of the study is that there was an optimal window of mean summer temperature in 9-14oC (Fig. 5D) and dry continental or high Arctic precipitation of the wettest quarter in 55 mm (Fig. 5B), within which the habitat suitability is maximized.
Land cover is also important and contributes strongly to model performance (Table 2). The response curve indicates that two land cover types are favored by A. erythropus including shrubland and open-water areas. The land-cover preference can be linked to the requirement of nest shelters during breeding season (Hilton et al . 2004) and food resources. In terms of food resources, the A. erythropus is an herbivorous browser, i.e. it tends to increase the portion of the selective resources in their feeding range (Markkolaet al . 2003). The wet sedge meadows on the alluvial floodplains that are preferred by herbivorous geese (Sedinger and Raveling 1984), and are critical for brood rearing (Markkola et al . 2003) offer a range of highly nutritious species with an adequate protein–water ratio and low portions of cellulose and lignin, (e.g. grassesPuccinellia phryganodes , Phragmites australis and sedges