CARMA’s MERRA-based caribou range climate database

1 Yukon College, Box 10038, Whitehorse, YT, Y1A 7A1, Canada (Corresponding author: don.russell@ec.gc.ca). 2 Department of Earth Sciences, Simon Fraser University, Burnaby BC, V5A 1S6, Canada. 3 Department of Statistics, Simon Fraser University, Burnaby BC, V5A 1S6, Canada. 4 368 Roland Road, Salt Spring Island, BC. V8K 1V1, Canada. 5 Institute of Arctic Biology, University of Alaska Fairbanks, Fairbanks, AK, 99775, USA. 6 Aurora Wildlife Research, 1918 Shannon Point Road, Nelson B. C., V1L 6K1, Canada.


Introduction
e CircumArctic Rangifer Monitoring and Assessment (CARMA) network's primary goal is to monitor and assess impacts of global change on caribou (Rangifer tarandus).One core approach is conducting cross-herd comparisons and contrasts to gauge how herds are similar and how they di er in their responses to climate.By understanding regional climates in which seasonally migratory tundra caribou herds have evolved, we can better assess strategies and mechanisms that Rangifer employ to cope with environmental stress.Climate has a strong in uence on caribou ecology through its e ects on forage growth and availability, its in uences on snow conditions, and on insect abundance that can harass caribou and cause changes in caribou movements and redistribution (Gri th et al., 2002;Bergerud et al., 2008;Couturier et al., 2009).We therfore need regional climate datasets that allow direct comparison of environmental attributes across and between continents.Although climate data are available from meteorological stations, those stations are relatively few, unevenly distributed across herd ranges, and often measure climate using di erent protocols.
An alternative to assessing regional climate based on meteorological stations is NASA's Modern Era Retrospective Analysis for Research and Applications (MERRA) dataset (http:// gmao.gsfc.nasa.gov/research/merra/).MERRA was undertaken by NASA's Global Modeling and Assimilation O ce with the objectives of placing the observations from NASA's Earth Observing System satellites in a climate context, and improving upon the hydrologic cycle represented in earlier generations of reanalyses (Rienecker et al., 2011).e resolution of the MERRA grid is 1/2 degrees latitude by 2/3 degrees longitude and data are provided on a daily time step for most variables.MERRA was chosen over other datasets because it covers the modern era of remotely sensed data (from 1979 through the present), attempts to address problems with previous reanalysis products, and is , 2012 focused on the hydrological cycle.Other longterm reanalyses of the Earth's climate have high levels of uncertainty in precipitation and interannual variability.MERRA also has better coverage north of 60° than other datasets such as NCEP and data are normally publicly available within a few months.

Methods
Climate data were summarized for 22 herds into as many as 8 polygons for each herd.For most herds separate shape les were constructed for ve seasons (calving, summer, fall, winter, and spring), for tundra and taiga portions of the range, and for the annual range.For some herds, the two Greenlandic herds and the Iceland reindeer, only the annual range was used as their distribution is small and there is no taiga.For most North American herds, all Greenlandic herds and Iceland reindeer, polygons were determined from radio-collar data (Table 1).Fixed kernel polygons (90% utilization distributions) were produced, using standard settings (href, raster resolution set to 120) from the Rodgers and Carr (1998)  erefore for those herds older data (pre-1995) were obtained with permission from GNWT-ENR.e basis for the resultant polygons was from work done pulling together surveys, maps, and collar data.Leslie Wakelyn, on behalf of the Beverly-Qamanirjuaq Caribou Management Board (BQCMB), produced seven overall seasonal ranges that were a general amalgamation of the data for 10-25 years over time periods that generally began in the late 1950s and extended to the early 1980s to mid-1990s.Wakelyn's seven seasons were amalgamated to produce polygons for the ve seasons used by CARMA.Because there is no history of collaring reindeer in Russia, for the ve Russian herds, seasonal polygons were developed by combining seasonal maps produced from aerial surveys by Russian management agencies combined with personal contact to nalize seasonal distributions.To produce the taiga and tundra polygons, a global treeline shape le was taken from the Circumpolar Arctic Vegetation Map from the Alaska Geobotany Center (http:// www.geobotany.uaf.edu/), with the original coverage supplied from (http://www.arcticatlas.org/).
In order to extract information stored in these downloads, readings of each variable of each grid points were rst extracted using "ncdf " package (Pierce, 2011) in the open source statistical software R (R Development Core Team, 2012).e procedures of extraction are listed as the following: 1) coordinates of the quali ed MERRA grid points were obtained through overlaying herd ranges to MERRA grids, 2) median values for each variable among the quali ed grid points were further extracted by ranking and locating the 50th percentile of all gridded data, and 3) daily climate les for each year, each herd, and each range were constructed and written into comma-separated values (csv) formatted les.erefore, each MERRA reading in the database represents the median daily averaged value with its MERRA grid point falling inside or close to the herd range in study.Using median readings instead of mean values within each range avoids making normality assumptions and reduces bias.
All snow variables were considered in snowyears only.A snow-year is de ned as starting from the 184th day of the year prior to the 183rd day of the current year.us, fall and winter periods of a snow-year consist of 182 days (or 183 days if the year prior is a leap year), and spring and summer periods of a snowyear consist of 183 days.Daily minimum and maximum temperature variables were extracted from the hourly MERRA assimilated temperature data with the following procedures.Hourly maximum and minimum readings were rst extracted for each MERRA grid point within the study range.Median readings were then extracted by ranking to locate the 50th percentile.
Based on these 36 variables (Table 2) we then produced a "caribou-relevant" dataset that includes 25 variables for each herd range.Some MERRA variables could be directly used and for others we derived variables based on the MERRA variables for each herd and seasonal range.Equations used to derive the variables were written into R source codes.

Discussion
Currently, CARMA has compiled the herdspeci c data at the scale of seasonal herd ranges and has distributed the datasets to caribou management agencies.Additionally, CARMA has initiated summary analysis of the key variables.
e MERRA dataset, although only including post-1979 data, will allow us to select climate variables to assess large-scale global oscillations which switch between climate phases at ap-proximately decadal timescales.Over decades these large-scale climate patterns are indexed as switches between positive and negative phases and in uence caribou ecology (Couturier et al., 2009;Joly et al., 2011).
To make the dataset readily available for CARMA members, these data have been organized into a menu-driven Microsoft Access® database.CARMA's intention is to provide the original dataset and annual updates through its website (www.ca.is/carma).CARMA is undertaking a series of validations by comparing datasets, for example, growing degree-days derived from MERRA data were compared to the Normalized Di erence Vegetation Index (NDVI), a variable that has been used to measure "green-up" patterns on caribou range (Gri th et al., 2002).e climate database has also been used to provide driving variables for CARMA's energy-protein model (Russell et al., 2005) and will be central in CARMA's cumulative e ects modeling program.

Table 1 .
e source of range information and time periods they represent for CARMA's 22 migratory tundra herds.