This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 93.153.182.114

This content was downloaded on 08/11/2016 at 10:10

Please note that terms and conditions apply.

You may also be interested in:

Land cover and land use changes in the oil and gas regions of Northwestern Siberia under changing

climatic conditions

Qin Yu, Howard E Epstein, Ryan Engstrom et al.

Regional and landscape-scale variability of Landsat-observed vegetation dynamics in northwest

Siberian tundra

Gerald V Frost, Howard E Epstein and Donald A Walker

Satellite observations of high northern latitude vegetation productivity changes between1982 and

2008: ecological variability and regional differences

Pieter S A Beck and Scott J Goetz

Potential change in forest types and stand heights in central Siberia in a warming climate

N M Tchebakova, E I Parfenova, M A Korets et al.

3D simulation of boreal forests: structure and dynamics in complex terrain and in a changing

climate

Ksenia Brazhnik and Herman H Shugart

Climate-induced larch growth response within the central Siberian permafrost zone

Viacheslav I Kharuk, Kenneth J Ranson, Sergei T Im et al.

Spatial heterogeneity of greening and browning between and within bioclimatic zones in

northern West Siberia

View the table of contents for this issue, or go to the journal homepage for more

2016 Environ. Res. Lett. 11 115002

(http://iopscience.iop.org/1748-9326/11/11/115002)

Home Search Collections Journals About Contact us My IOPscience

Environ. Res. Lett. 11 (2016) 115002 doi:10.1088/1748-9326/11/11/115002

LETTER

Spatial heterogeneity of greening and browning between and withinbioclimatic zones in northernWest Siberia

VictoriaVMiles and Igor EsauNansen Environmental andRemote SensingCenter/Bjerknes Center, Bergen, Norway

E-mail: victoria.miles@nersc.no

Keywords: greening, boreal forest, tundra, Siberia, satellite remote sensing, vegetation productivity

AbstractStudies of the normalized difference vegetation index (NDVI) have found broad changes in vegetationproductivity in high northern latitudes in the past decades, including increases inNDVI (‘greening’) intundra regions and decreases (‘browning’) in forest regions. The causes of these changes are not wellunderstood but have been attributed to a variety of factors.We useModerate Resolution ImagingSpectrometer (MODIS) satellite data for 2000–2014 and focus on northernWest Siberia—a hot spotof extensive landcover change due to rapid resource development, geomorphic change, climatechange and reindeer grazing. The region is relatively little-studied in terms of vegetation productivitypatterns and trends. This study examines changes between andwithin bioclimatic sub-zones andreveals differences between forest and treeless areas and differences in productivity even down to thetree species level. Our results show that only 18%of the total northernWest Siberia area hadstatistically significant changes in productivity, with 8.4% increasing (greening) and 9.6%decreasing(browning).We find spatial heterogeneity in the trends, and contrasting trends both between andwithin bioclimatic zones. A keyfinding is the identification of contrasting trends for different specieswithin the same bioclimatic zone. Browning ismost prominent in areas of denser tree coverage, andparticularly in evergreen coniferous forest with dark (Picea abie,Picea obovata) or light (Pinussylvestris) evergreen and evergreen-majoritymixed forests. In contrast, low density deciduous needle-leaf forest dominated by larch (Larix sibirica), shows a significant increase in productivity, evenwhileneighboring different species showproductivity decrease. These results underscore the complexity ofthe patterns of variability and trends in vegetation productivity, and suggest the need for spatially andthematically detailed studies to better understand the response of different northern forest types andspecies to climate and environmental change.

1. Introduction

The climate of theArctic is changing, and there is stronginterest in understanding how vegetation will respondto and contribute to this change. The ongoing changesin plant productivity will affect many aspects of north-ern systems including changes in active-layer depth,permafrost, biodiversity, wildlife and human use of theregion (Bhatt et al 2010). However, the direct impact ofthe climate warming factors on boreal and arcticvegetation cannot be establishedunambiguously.

Satellite observations of vegetation dynamics haveshown that this warming leads to increased vegetationgrowth and a marked ‘greening’ trend in non-forested

tundra (Walker et al 2009). Reflecting temperaturechange, a widespread ‘greening’ indicated by theincreasing growing-seasonmaxima of normalized dif-ference vegetation index (NDVI) has been reportednorth of 64°N (e.g., Walker et al 2009, Epsteinet al 2012). However, since 2003, the greening in tun-dra has slowed down (Bhatt et al 2013) and the surfacetemperature trends have become negative or at leastmore spatially fragmented in the tundra zone (ComisoandHall 2014).

The region south of 64°N in the boreal forests(taiga) experienced much less greening. In contrast, awidespread ‘browning’ indicated by the decreasingmaxima of NDVI (hereafter NDVImax) was found, in

OPEN ACCESS

RECEIVED

14May 2016

REVISED

30 September 2016

ACCEPTED FOR PUBLICATION

7October 2016

PUBLISHED

26October 2016

Original content from thisworkmay be used underthe terms of the CreativeCommonsAttribution 3.0licence.

Any further distribution ofthis workmustmaintainattribution to theauthor(s) and the title ofthework, journal citationandDOI.

© 2016 IOPPublishing Ltd

particular, in the most biologically productive forestecosystems (Beck and Goetz 2011). The browningseems to be a relatively recent and northward advan-cing tendency, as earlier studies found the NDVImaxincreased in 1981–1999 in the boreal forest zone(Zhou et al 2001). The reports of extensive tree growthdecline or mortality in northern mid- and high- lati-tudes (Bunn et al 2007) highlight the large spatial–temporal variability in tree growth responses toclimate change in northern areas. It indicates that theresponse to warming varies substantially between dif-ferent tree species (Goetz et al 2011). Elevated tem-peratures enhance growth in deciduous species morethan in evergreen, and the main boreal forest mayrespond with a loss of evergreen trees and a shifttoward deciduous trees (Way and Oren 2010). At thesame time climate warming induces species- and for-est-type-shifts and northwardmigration. Siberian for-estsmay collapse in south areas and become greener inthe north (Bonan 2008). The projection is that ecosys-tems unique to larch forests occurring on permafrostwillmove farther north, and Siberian pine, spruce, andfir will expand northward into traditional larch habitat(Tchebakova et al 2009). It is very probable that thisprocess will be driven substantially by changes in per-mafrost regimes. Changing species distribution andbiodiversity will have important biological, ecologicaland social consequences. Tree species play a key role,providing habitat, food or mutualisms with many ani-mals, fungi, microorganisms, and other plants in addi-tion to other ecosystem services and resources forhuman use. Forests also contain around three quartersof the Earth’s terrestrial biomass and thus are tightlylinked with atmospheric carbon budgets (Aitkenet al 2008).

This study utilizes NDVI data products from theModerate Resolution Imaging Spectroradiometer(MODIS) onboard of the Earth Observing System-Terra platform satellite. Several previously publishedstudies by Frey and Smith (2007), Epstein et al (2012),Macias-Fauria et al (2012) and others demonstratedthat MODIS NDVI can be successfully used in vegeta-tion productivity assessment for Siberian biomes. Weused MODIS NDVI 16-day composites with 250 mspatial resolution (MOD13Q1). This product is widelyused in phenological and vegetation dynamics studiesin the Arctic and boreal zone (e.g., Blok et al 2011). Inthis paper, we update, detail and expand upon pre-vious satellite studies of high northern latitude vegeta-tion productivity. Relevant previous studies have hadeither a broader scope (e.g., Zhou et al 2001, Bunn andGoetz 2006, Beck and Goetz 2011, Barichivichet al 2014) or else covered different regions of northernEurasia (e.g., Elsakov and Teljatnikov 2013) includingarctic tundra sub-regions ofWest Siberia (e.g., Walkeret al 2009, Frost et al 2014). Here we focus on the entireNWS spanning four bioclimatic zones: tundra, forest-tundra, northern taiga and middle taiga. Two novelaspects should be mentioned: (1) whereas previous

studies analyzed coarse-resolution data, which is likelyto exaggerate the extent and magnitude of NDVItrends, we use moderate-resolution (250 m) data; and(2) moderate-resolution data give an opportunity totrace the changes within the same bioclimatic zone,and help to reveal differences between forest types andeven down to the species level. The objectives are to:(1) identify the recent (2000–2014) spatial–temporalpatterns of variability and trends of NDVImax acrossNWS in detail, (2) examine the variability of trends inNDVImax for different bioclimatic zones, and (3)examine within-group variability of trends in NDVI-max. We assess the distribution of different foresttypes and species within the same bioclimatic zone inrelation to the observed increases or decreases of pro-ductivity. In particular, we compare and contrasttrends between main forest forming species in NWS:evergreen coniferous forest, with the dark (Picea abie,Picea obovata) or light (Pinus sylvestris) evergreen andevergreen-majority mixed forests, and deciduous nee-dle-leaf forest dominated by larch (Larix sibirica)species.

2. Study area, data andmethods

2.1. Study area and its vegetationWest Siberia is the territory east of the Ural Mountainsbut west of the Yenisei River. The West Siberianlandscape is nearly flat, with large winding rivers thatflow thousands of kilometers to the Arctic marginalseas. The Western Siberian plain is the most boggedregion of the world (in total 106 km2); in some parts,up to 70%–80% is covered by mires (Kirpotinet al 2009). Forest covers only 36% of northern WestSiberia (hereafter NWS) (Sedykh 1997). The length ofgrowing season in the northern latitude is short: 50–60days in the tundra zone and 60–90 days in taiga zone.June, July and August are the months with highestbiomass level, with the peak of growing activity in July.Vegetation zones across NWS are delimited in termsof phytogeographic zone or bioclimatic (latitudinal)zone, according to Sedelnikov et al (2011). The fourmajor bioclimatic zones within NWS, from north tosouth are: tundra, forest-tundra, northern taiga andmiddle taiga (figure 1).

In the tundra zones of northern NWS, vegetationdevelops under limiting conditions of a short growingperiod, low air and soil temperatures, and highhumidity. Tundra expands on the continuous perma-frost region and keeps it frozen most times of the year.The permafrost active layer here is approximately25–100 cm (Ukrainceva et al (2011)). The main vege-tation types are lichens and bryophyte and also grami-noids, forbs, and shrubs. Larix sibirica together withSiberian spruce (Picea obovata) forms the polar limit oftree-stand distribution (see figure 1) over the territoryacross NWS between the Ob and Yenisei Rivers. Trees

2

Environ. Res. Lett. 11 (2016) 115002

cover only 2% of the tundra zone and comprise only0.5%of the entire area ofNWS.

The forest-tundra is a transition zone or ecotone

that expands on sporadic permafrost. It has low and

open vegetation canopy and forest covers slightly

more than 30%, comprising 2.6% of the total NWS

area. Evergreen forests here are often swampy, with

abundant dwarf shrubs and green mosses dominated

by spruce (Picea obovata) and pine (Pinus sylvestris).Better-drained localities are occupied by larch forest

(Larix sibirica).In the northern taiga, the deciduous needle-leaf

forest dominated by Larix sibirica and light evergreen

coniferous dominated by Pinus sylvestris are the major

forest type. Forests cover 52%of northern taiga, which

is 16% of the total NWS area. In themiddle taiga, ever-

green light forest (Pinus sylvestris) is common on the

north while in the south deciduous broadleaf birch

(Betula) forests, which mainly represent secondarysuccession, are widely distributed. Forests cover 49%of themiddle taiga and it is 17%of the total NWS.

In NWS, forests develop generally in proximity torivers where drainage is better and, because of theexceptional swamping, they do not occupy large areasbut stretch for great distances along rivers. Forests alsooccur on elevated drier sites in wetlands. However,these forests are sparse and tree growth is impeded(Shahgedanova 2003).

2.2.Data andmethods2.2.1. NDVI dataIn this study we examine changes in vegetationproductivity in NWS (2000–14) using variations inNDVI, which is a well-established proxy for grossphotosynthesis at different spatial scales (Goetzet al 2005) and an index of vegetation greening, density

Figure 1.NorthernWest Siberia (NWS) land-surface cover and bioclimatic zones. The land-surface cover categories are comprised offive (5) forest types, unforested area andwater bodies. Thematicmap produced using digital data fromBartalev et al (2014).

3

Environ. Res. Lett. 11 (2016) 115002

and development. NDVI is defined as a normalizedratio of reflectance factors in the near infrared (NIR)and red spectral radiation bands

( )r rr r

=-+

NDVI 1NIR RED

NIR RED

where rNIR and rRED are the surface bidirectionalreflectance factors for their respective bands. NDVI isbased on the contrast between red andNIR reflectanceof vegetation, as chlorophyll is a strong absorber of redlight, whereas the internal structure of leaves reflectshighly in the NIR (Gamon et al 1995). The greater thedifference between the reflectance in the red and NIRportions of the spectrum, the more chlorophyll in thevegetation canopy. Vegetation yields positive NDVIvalues and approaches +1 with increasing plantchlorophyll content or green biomass. Being a synthe-sizing dimensionless indicator, NDVI has strongcorrelations with the productivity and biomass ofvegetation for various ecosystem types (Gamonet al 1995, Raynolds et al 2008, Epstein et al 2012).

MODIS data for the 2000–2014 summers (June–July–August, JJA) across NWSwere collected and pro-cessed for this study. Five tiles to cover the entire areaof interest (total 35 tiles per summer) were down-loaded and imported into the ArcGIS geographicinformation system (GIS). Images weremosaic and re-projected from original Sinusoidal (SIN) to the uni-versal Transverse Mercator projection (UTM Zone42N, WGS84 ellipsoid). NDVI data were quality-fil-tered by the MODIS reliability data provided togetherwith theMOD13Q1 product, to retain only data of thehighest quality, excluding snow/ice- and cloud cov-ered pixels (Solano et al 2010).

The data gaps in the raster mosaic pixels were thenfilled with information using the nearest neighbor sta-tistical interpolation from the surrounding pixels withdata. The percentage of excluded pixels is variable,ranging from 10%–30%. (Images with a higherpercentage of excluded pixels were eliminated fromthe analysis, usually one or two images per summer.)Because the spatial patterns of missing pixel values arenot random but naturally clustered (e.g., a significantMoran’s I test statistic value >0.8 (Moran 1948)), weassessed the potential problems when interpolatingover large-connected patches of missing values. Weexamined the spatial autocorrelation structure of theobserved values of NDVI, finding that autocorrelationremains positive and significant to spatial lag 10–20,e.g., ∼0.5 at lag10 in the tundra zone where missingvalues are most commonly found. This is larger thanthe typical size of the areas of excluded pixels and the9×9 kernel operator that we used in the interpola-tion.Moreover, we are focusedmore on the ecosystemresponse rather than a single pixel and consider thateach pixel-value is a part of surrounding ecosystemvalues. In our case we believe that interpolation of datais necessary, because by interpolation, each pixel con-tributes to the general ecosystem response value for

each date rather than using no data. This is importantfor producing the finalNDVImax composite.

The NDVI represents a measure of canopy ‘green-ness’ where values below 0.2 are generally non-vege-tated surfaces and green vegetation canopies aregenerally greater than 0.3 (Gamon et al 1995). A 0.3–1NDVI threshold was used to exclude water, bare soiland other non-vegetated pixel from the analysis.

In accordance with most remote sensing studies ofarctic and sub-arctic vegetation (e.g., Raynoldset al 2008, Walker et al 2009, Beck and Goetz 2011,Blok et al 2011), we use annual maximum NDVI(NDVImax). NDVImax values characterize the max-imum development and represent the peak greennessachieved by vegetation during the growing season.Summarizing NDVI into NDVImax composites elim-inates any seasonal variation in NDVI and reduces theerrors in beginning of phenological phases betweendifferent vegetation zones (Reidel et al 2005). AnNDVImax map for each year (summer, JJA)was com-piled by selecting the maximum NDVI value fromeach 16-day composite for each pixel. This results in a250 m resolution, 15-year dataset of NDVImax andgenerated an up-to-date 15-yearmeanNDVImaxmapthroughoutNWS.

2.2.2. Trend analysisStatistical methods were applied to the time series ofNDVImax and derived parameters, in order to providemetrics and to identify and test for changes. Timeseries characteristics of interest are mean, variabilityand trend. Temporal linear trends in yearly summerNDVImax over NWS for the period 2000–2014 wereestimated using Ordinary Least Squares (OLS) regres-sion for each pixel stack, with year as the independentvariable and NDVI as the response variable. Thepurpose is to do a pixel-wise trend analysis and extractonly significant trends over a certain period of time fora selected region for rejecting the null hypotheses. Toidentify pixels with statistically significant trends, wemasked out all pixels with an estimate p-value>0.05.The slope of the trend line is the annual rate of‘greening’ or ‘browning’ for pixel stacks with signifi-cant trends (p<0.05).

OLS trend analysis remains the most frequentlyused in environmental studies in general, and in arcticclimate change studies in particular. Our OLS trendanalysis can be easily compared with previous vegeta-tion trend studies and with the trend analysis of otherclimate-relevant quantities. The limitation of OLSestimates is that linear trendmodel is sensitive to unu-sual data values or outliers, particularly if they are fit-ted to small data samples (n=15 in this study). Thereare a number of robust linear regressionmethods (e.g.,Yu et al 2014b), which are less sensitive to the outliers.The difference becomes particularly significantwhen the sampled data are suspected to containheteroscedasticity.

4

Environ. Res. Lett. 11 (2016) 115002

To reduce the above-mentioned uncertainties, weperformed a comparison of the OLS regression versusthe non-parametric Theil–Sen regression (TSR)method. We found only small systematic OLS-TSRdifferences for the largest negative and positive trends—the TSR values are smaller than the OLS trendvalues, as expected. Furthermore, we tested the sig-nificance of the difference between trends using thevariance statistics for the OLS trends (e.g., Santeret al 2000). We found that the expected and revealeddata heteroscedasticity was surprisingly insignificant.As a result, the widely used OLS trends and the robustTSR trends here differ only statistically insignificantlyat the 95% confidence level.

To define the similarity/dissimilarities of NDVI-max changes between each bioclimatic zone and eachforest type, we calculated correlation matrices. Pear-son’s correlation matrix was used to describe the cor-relations among variables. These variables were meanNDVImax extracted every year for the area of eachzone and each forest type. The correlation coefficients(r) are between −1 and 1 (0: no correlation and ±1:high correlation). A positive (or negative) value indi-cates that the two correlated parameters increase (ordecrease) simultaneously.

We performed amore detailed examination of dif-ferent land cover and forest type interannual trends inproductivity and variation of summer maximumgreenness. To distinguish how productivity responsesdiffer between forest dominant species we calculatedthe proportion of areas browning and greeningfor each species. The number of pixels that exhibitedpositive and negative trends was determined for land-cover (forest versus non-forest treeless areas) and for-est types (forest type refers to the dominant treespecies in the overstory of a given site) across biomesinNWS.

Forest dominant species information (figure 1)was generated from Proba-V-TerraNorte digital mapof forest cover for Russia with 345 m spatial resolutionthat utilises the same forest types as Bartalev et al(2014). To avoid discrepancies between data, the forest

map was resampled to 250 m resolution using nearestneighbor interpolation methods, which is typicallyused for discrete data, such as a land-use classification,since it will not change the values of the cells. Themax-imum spatial error is one-half the cell size.

We calculated the area fractions covered by differ-ent forest types from the total area ofNWS in each bio-climatic zone (table 1) from Bartalev et al (2014) forestmap. The treeline was digitized using same data bydelimiting the edge of tree vegetation growth (figure 1)

3. Results

3.1. Patterns and trends inNDVImaxwithin biomesFigure 2 provide the region-wide overview of theNDVI pattern and its statistically significant trends forthe past 15 years. The mean NDVImax (figure 2(a)) inthe NWS generally decays from the southwest to thenortheast of the territory. The tundra naturally has thelowest NDVImax values. The central, swamped part ofthe NWS is characterized by a much lower NDVImax,and the largest NDVImax found for the most produc-tive vegetation along the Ob River and between the ObRiver and Ural mountains. Typically, the NDVImax issignificantly higher on river terraces with better-drained and therefore warmer soils.

Only 18% of the total NWS areas display sig-nificant trends, with 8.4% being an increase and 9.6%being a decrease in productivity. The map of statisti-cally significant (p<0.05) trends in NDVImax from2001–2014 (figure 2(b)) confirms continuing greeningin tundra and forest-tundra biomes. Indeed, morethan half of all positive changes in NWS are localizedin the tundra zones (table 2). However, this greening ishighly fragmented and overlay with patches of brown-ing and to the large degree could be associated withdifferent local disturbances. Themost significant areasof greening are found in Taz and southern Gydanpeninsulas. In general, almost all greening lies between65° and 70°N latitude (figure 2(b)). The main negativetrends in the tundra ecotones are in the northern part

Table 1.Proportion of different forest types in different bioclimatic zones inNWS, calculated here using digital datasetfromBartalev et al (2014). Description of forest types by dominant (at least 80%) or subdominant species (at least 20%–

60%) of the forest canopy (Bartalev et al 2014):EG—evergreen dark needleleaf forest consisting of spruce (Picea),fir(Abies), and Siberian pine (Pinus sibirica);EL—evergreen light needleleaf forest consisting of pine (P. sylvestris);BrD—deciduous broadleaf forest consisting of birch (Betula), aspen (Populus tremula), oak (Quercus), linden (Tilia), ash ( Frax-inus), maple (Acer), and some other deciduous broadleaf tree species;MBM—mixed broadleafmajority forest consisting ofdeciduous broadleaf species and evergreen needleleaf species;M—mixed forest—proportions of the evergreen needleleafand the deciduous broadleaf tree species are approximately equal;MNM—mixed needleleafmajority forest consisting ofthe evergreen needleleaf tree species and the deciduous broadleaf tree species;DN—deciduous needleleaf forest consistingof larch (Larix);UF—unforested areas—area not coveredwith trees.

EG EL BrD MNM M MBM DN UF

Tundra 0.03% 0.00% 0.00% 0.00% 0.00% 0.00% 0.48% 26.32%

Forest-Tundra 0.46% 0.00% 0.00% 0.00% 0.00% 0.00% 2.20% 5.60%

NorthernTaiga 3.48% 4.11% 0.94% 0.75% 0.14% 0.06% 6.67% 14.91%

Middle Taiga 1.90% 8.23% 2.96% 2.09% 0.54% 0.14% 0.68% 17.32%

Total per class 5.86% 12.35% 3.90% 2.84% 0.67% 0.20% 10.02% 64.15%

Note: total sums to 100%.

5

Environ. Res. Lett. 11 (2016) 115002

of the Gydan peninsula and eastern part of Yamalpeninsula are found along the river, in the coastal zoneand in thefloodplain landscapes.

In contrast, the taiga zones of NWS demonstratewidespread negative trends of NDVImax. The nega-tive trends are prevailing here and concentrating in the

Figure 2. (a) Spatial distribution of 15-year (2000–2014)meanNDVImax in northernWest Siberia (NWS). (b) Statistically significanttrends of greening (positive trend) and browning (negative trend) (p<0.05).

Table 2.Proportion (percentage) of statistically significant trends ofNDVImax in terms of theNWS total area,zone area and total trend, all stratified by bioclimatic zone.

%of the total NWS area1 %of the zone area %of the trend

Greening Browning Greening Browning Greening Browning

Tundra 18.4% 2.3% 89.0% 11.0% 42.8% 4.0%

Forest-Tundra 7.9% 1.1% 87.6% 12.5% 18.4% 2.0%

Northern Taiga 10.0% 17.7% 36.0% 64.0% 23.2% 31.0%

Middle Taiga 6.7% 36.0% 15.7% 84.3% 15.7% 63.1%

1 NWS areas with statistically significant trend. The values in the right columns (% of the trend) are plotted in

figure 2(a). 5.

Figure 3. Interannual variability and trends inmeanNDVImax averagedwithin four bioclimatic zones inNWS, 2000–14.

6

Environ. Res. Lett. 11 (2016) 115002

southwest, south and southeast parts of the NWS(figure 2(b), table 2).

The highest NDVImax in Tundra on record was in2012–14 (figure 3). For the transition forest-tundrazone and two taiga sub-zones, a dip in NDVImaxoccured in 2011, and recovery occurred in 2012. In2014 the NDVImax declined in tundra zone, howeverin other zones continued to rise. The correlation ishigher between neighboring zones (table 3). Foresttundra and the two taiga sub-zones show better corre-lation with each other than with the tundra zone. Thetemporal variability in annual NDVImax for the tun-dra is different than in other zones, suggesting that thecauses of changes in vegetation are different here.

3.2. Trends inNDVImaxwithin forest typesBoth positive and negative trends occurred in forestedand non-forested land of NWS, but declines aredominating for forested areas. The majority of foresttypes exhibit more negative than positive trends. Weobserve twomain tendencies: increasing share of areaswith positive trends from south to the north for eachforest type and non-forested areas and vice versa:increasing percent of areas with negative trends fromnorth to the south for each forest type and non-forested areas.

A significant portion of the area with declines isobserved in evergreen and mixed forest types (table 4,figure 4). Negative trends indicate decreased pro-ductivity in dark coniferous dominated by Picea. Inline with this observation we observe expansion ofbrowning in light coniferous dominated by P. sylves-tris. Dominant pattern of negative trend was detectedin the southwest part of NWS associated with declinein light evergreen forest, while in the south and south-east part it is mostly related to dark evergreen, decid-uous broadleaf and mixed forests. Declining spruceand pine productivity appears to be themajor driver ofthe downward trend in the NWS taiga zone (table 4).Nevertheless, decreasing productivity is observed inareas dominated by deciduous species as well. More-over, the negative trend that prevailed in taiga includ-ing a drop in greenness for non-forested areas aswell.

The only greening observed within the back-ground of browning is for forested areas dominated byLarix. Moreover, Larix stands are found to increase inNDVI in tundra and northern taiga zone (table 4).Forest-tundra and northern taiga are larch-dominatedzones and highest larch greening is seen there (table 1).

On the southern extent of its bioclimatic range (themiddle taiga zone) however, larch show prevailbrowning. The most interesting observation is thatamong variant type of forest stands growing close toeach other the Larix show positive changes when at thesame time the others are showing negative trend.

Changes in NDVImax were heterogeneous amongdifferent forest types (figure 5), although some com-mon features are evident. Similar interannual varia-tions were seen among the six forest types (ED, EL,Brd, MNM, M, MBM). Linear regression trends forthese types over the analysis period were significantlynegative (p<0.05). For these types, the peak vegeta-tion greenness was relatively low in 2001 and thenincreased with the highest peak in 2003 and a dropagain in 2004. In 2009, a sharp decline was observedfor each forest type and in non-forested areas. Morenotably, larch-dominated forest (DN) is only the for-est type with a significant positive linear trend and anegative correlation to other species (see table 5), e.g.,DN NDVImax increased in 2004 when other typesshows drop in greenness. The strongest positiveNDVImax anomalies are observed for DN in 2007,and after a slight decline in 2012 there was an ascend-ing trend. The linear trend for UF over the period ana-lyzed was slight but significantly positive (p<0.05),with an increase since 2012 after some years of decline.

4.Discussion

This study has shown contrasting trends betweenforest and treeless areas, and notably also within-biome differences down to the species level. Contrastsin NDVI tendency may depend on different factorscontrolling phytomass productivity. Despite therecent summer cooling (Bhatt et al 2013) the positiveNDVImax trends in tundra could be still linked towarmer temperature. The dip in ‘greenness’ in 2009was circum-Arctic and according to Walker et al(2011) was a response to generally cooler summertemperatures across the Arctic due to elevated atmo-spheric aerosols, including volcanic dust. Temporarydip in ‘greenness’ in 2009, followed by a circumpolarrecovery in 2010, when the mean NDVI for NorthAmerica and the Northern Hemisphere overall wasthe greatest on record. In the latest report tundragreenness, derived from remote sensing data, has beendeclining consistently for the past 2–4 years through-out the Arctic. Elsakov and Teljatnikov (2013) showeda dip in NDVImax for Gydan in 2006 andWest Yamalin 2007.Our record (figure 3) of interannual variabilityand trends for the tundra zone of NWS shows a slowdecline started in 2006, with recovery occurring in2011. The highest NDVImax on our record was in2012–14 contrary to the general declining trend ofgreening (Epstein et al 2015). For the transition forest-tundra zone and two taiga sub-zones, the dip inNDVImax happened in 2011, and recovery occurred

Table 3.Correlationmatrix (r) betweenmean annualNDVImax ineach bioclimatic zone inNWS from2000–14, n=15.

Forest-

Tundra

Northern

Taiga

Middle

Taiga

Tundra 0.43 0.09 −0.90

Forest-Tundra 0.68 0.19

NorthernTaiga 0.76

7

Environ. Res. Lett. 11 (2016) 115002

in 2012 when summertime surface air temperatures inthe NWS were 5 to 15 °C higher than those observedduring the previous three decades (Pokrovskyet al 2014).

The bioclimatic (latitudinal) NDVI variationsmostly related to changes in temperature, but at thesame time latitudinal variations in NDVI mostly rela-ted to change in precipitations and very much dependon site factor (elevation, hydrology, soils, plant den-sity, etc). For instance, the temperature stress on Larixsibirica is pronounced only in the southern part of itsareal range, whereas accelerated growth and its generalnorthward proliferation along river terraces wasreported in the northern part of the area (Tchebakovaet al 2009). Gridded climatic data associated with treering-width around the circumpolar boreal forest(Lloyd and Bunn 2007) show inverse growth responsesto temperature. Browning was mainly concentrated indark coniferous (Picea) species and occurredmore fre-quently in the warmer parts of species ranges, and theLarix species have a higher-than expected frequency ofgreening. Declining spruce productivity appears to bethe major driver of the decreasing trend in Alaskanboreal forest (Beck and Goetz 2011). Wilking andJuday (2005) found that longitudinal variation in treegrowth responses to recent warming depend on sitetype, tree density and precipitation pattern. Similarstudies from four sites along a 30 km north–south

transect in Alaska show diverse growth responsesbetween and within study sites (Driscoll et al 2006,D’Arrigo et al 2009). Wilking and Juday (2005) andBeck and Goetz (2011) found that low density standsshow higher sustainability either on latitudinal orlongitudinal variations; for example tundra and for-est-tundra characterized by low tree density showincrease in productivity whereas taiga with denser for-est showing decrease. Larch forests usually form low-density stands due to shade intolerance. Trends inproductivity are thus heavily dependent on covertype and the underlying vegetation density (Beck andGoetz 2011).

In general, most recent boreal forest declineappears connected with increased drought conditions(moisture stress) (Bunn et al 2007, Kharuk et al 2013).In northern West Siberia, however, climatic warmingactually results in increases in precipitation, rainfallfrequency, accumulated snow depth and a reductionin temperature range (Frey and Smith 2003, Bulyginaet al 2011). These factors cause bog developmentprocesses to become more active on the flat andpoorly-drained surface of the NW Siberian plain(Sedykh 1997, Crawford et al 2003,Moskalenko 2013).A reversal of succession from forest to bog resultsin a 26% decrease of all above-ground biomass(Moskalenko 2013).

Table 4.Proportion of positive and negative trends in each zonewithin each forest type.

EG EL BrD MNM M MBM DN UF

Negative Tundra 0.07% 0.46% 3.47%

Forest-Tundra 1.17% 2.10% 0.90%

NorthernTaiga 38.27% 23.53% 11.24% 24.20% 24.71% 17.51% 21.98% 13.25%

Middle Taiga 50.43% 59.59% 56.71% 73.42% 71.04% 70.59% 8.81% 26.22%

Total 89.94% 83.12% 67.96% 97.62% 95.75% 88.10% 33.35% 43.85%

Positive Tundra 0.04% 3.67% 31.91%

Forest-Tundra 0.62% 16.02% 10.05%

NorthernTaiga 8.04% 6.12% 4.43% 0.46% 0.18% 2.95% 44.15% 8.14%

Middle Taiga 1.37% 10.76% 27.61% 1.92% 4.07% 8.95% 2.81% 6.05%

Total 10.06% 16.88% 32.04% 2.38% 4.25% 11.90% 66.65% 56.15%

Figure 4.Trends inNDVImax between eight different forest/landcover types. The percentages correspond to the total area ofNWS.Forested area (F)-sumof areas. Note that only deciduous needleleaf (DN) and unforested (UF) areas havemore positive than negativetrends.

8

Environ. Res. Lett. 11 (2016) 115002

Kirpotin et al 2009 describes contrasting processesoccur in the southern and northern parts of the WestSiberian Plain. In the taiga zone bogs are expandingand these progressive swamping leads to forest death.However, in the tundra zone the situation is totally

different. Bogs there are reducing their area andbecoming actively colonized by shrubs and trees(Crawford et al 2003, Kirpotin et al 2009). Similareffect has been discovered during dendroclimatic stu-dies in mid- and northern Sweden: trees growing on

Figure 5. Interannual variability and trends inmeanNDVImax for eight different forest/landcover types inNWS.

Table 5.Correlationmatrix (r) betweenmean annualNDVImax for each forest class2000–14,n=15.

EL BrD MNM M MBM DN UF

ED 0.94 0.81 0.90 0.84 0.74 0.07 0.47

EL 0.89 0.88 0.86 0.75 −0.00 0.46

BrD 0.75 0.76 0.81 −0.10 0.35

MNM 0.96 0.77 −0.15 0.20

M 0.78 −0.28 0.21

MBM −0.37 0.23

DN 0.15

9

Environ. Res. Lett. 11 (2016) 115002

bogs in the forest zone are more sensitive to humidityvariations than trees in the north (Linderholmet al 2002). The predominance of relatively large dead-wood and young pines at ridges testifies the beginningof pine generation change in the tree stand in forest-mire ecosystem of middle taiga in West Siberia(Makhatkov 2010). In Eastern Siberia forest decline iscaused by high soil water conditions in a permafrostregion due to high amount in precipitations (Iwasakiet al 2010).

The observed contrasting trends indicate thatalong with climatic changes, site factors (soils, perma-frost, hydrology, etc) and plant factors (structure,health, phenology, species composition, etc) deter-mine vegetation vitality, growth and productivity inNWS. Greening is highly fragmented and overlay withpatches of browning and to the large degree could beassociatedwith different local disturbances.

Themost significant areas of greening are found inthis study in Taz and southern Gydan peninsulas. Thisfinding is in good agreement with previously reportedplot-scale studies using Landsat images (Frostet al 2014). In general, almost all greening lies between65° and 70°N latitude (figure 2(b)). The main negativechanges observed in this study in the tundra ecotonesare in the northern part of the Gydan and eastern partof Yamal peninsulas, also reported by Frost et al(2014). Moreover, vegetation in the area experiencesconsiderable pressure from various disturbance fac-tors. The territory of NWS is an extensive area of oiland gas exploitation and production. ContrastingNDVI trends in the same zone are probably due to acombination and variety of these factors. For example,NDVI patterns on the Yamal, Taz and Gydan are cor-related with the temperature and also strongly relatedto difference in landscape factors, erosion, reindeeryearly grazing and anthropogenic landscape transfor-mation (Walker et al 2009, Forbes et al 2009, 2010,Frost et al 2014, Yu et al 2015, Esau et al 2016). We didnot have enough auxiliary data to distinguish the nat-ure of the trend, but this must be done to be able tocharacterize the vegetation changes due to climateor anthropogenic influence on positive or negativechanges.

Consistent trends in the NDVI in the arctic andsub-arctic have been reported regardless of the datasetbeing analyzed (Goetz et al 2005, Bunn et al 2007, BeckandGoetz 2011, Bhatt et al 2013, Elsakov and Teljatni-kov 2013, Kharuk et al 2013, Epstein et al 2015 andetc). However, in other regions, the use of differentdatasets has led to conflicting findings, potentially dueto differences in the processing and correctionsapplied to the satellite data (Alcaraz-Segura et al 2010).In our study, the trend is generally consistent with pre-vious studies (e.g., Raynolds et al 2008, Walkeret al 2009, Beck and Goetz 2011, Blok et al 2011). Therobustness of the results is supported by severalaspects: (1)we usedNDVI, an establishmetric to studyvegetation change (Walker et al 2009); MODIS NDVI

series are widely used and well agreed with trends gen-erated from AVHRR series (Goetz et al 2005); (2) weused higher resolution data versus lower resolution,which can cause misleading trends (Alcaraz-Seguraet al 2010, Ju and Masek 2016); (3) our results are inagreement with previous high-resolution studies usingLandsat (Frost et al 2014); (4)we used time-tested dataprocessing and correction methods (Verbyla 2008;Bjerke et al 2014); and (5)we used standard OLS trendanalysis, which is themost frequently used in environ-mental studies in general and in arctic climate changestudies in particular (Bhatt et al 2013).

Several factors influence the stability of the satel-lite-derived NDVI, e.g., cloud contamination, atmo-spheric variability, and bidirectional reflectance(Gutman 1991). The changes in NDVI caused by thesefactors are seen as undesirable noise in vegetation stu-dies. Thus, compositingmethods have been developedto eliminate these effects. Three major compositemethods: bidirectional reflectance distribution func-tion composite (BRDF-C), constrained-view angle-maximum value composite (CV-MVC) and max-imum value composite (MVC) are applied to generateMOD13Q1 product (Huete et al 1999, van Leeuwenet al 1999, Huete et al 2002). It ensures the high accur-acy forMODISNDVI dataset (Yu et al 2014a). GeneralAccuracy Statement of MOD13 products are well dis-cussed at https://landval.gsfc.nasa.gov/ProductStatus.php?ProductID=MOD13.

It should be noted that Wang et al (2012) reportedon sensor degradation having an impact on trenddetection in North American boreal and tundra zoneNDVI with Collection 5 data fromMODIS. The mainimpacts of gradual blue band (Band 3, 470 nm) degra-dation on simulated surface reflectance was most pro-nounced at near-nadir view angles, leading to a smalldecline (0.001–0.004 yr−1; 5% overall between 2002and 2010) in NDVI under a range of simulated aerosolconditions and high-latitude surface types. This degra-dation has been reduced in theMOD13Q1 product bythe above-mentioned compositingmethods.

5. Conclusions

In this study, the summer maximum NDVI (NDVI-max) values from MODIS data have been aggregatedand analyzed for the entire northern West Siberia(NWS). The study provides a detailed analysis of 15years (2001–14) of NDVI variability and trends in fourdifferent bioclimatic zones, for forest and non-forestvegetation and different forest types in NWS. Themain conclusions are:

1. At the regional level, our analysis shows a generalzonal trend in the relative proportion of vegetationresponding positively or negatively. The generaltrend for the NWS is an increase in tundra anddecrease in boreal forest zone, which appears to be

10

Environ. Res. Lett. 11 (2016) 115002

comparable to the previous vegetation responsestudy in the Arctic (e.g., Raynolds et al 2008,Walker et al 2009, Beck and Goetz 2011, Bloket al 2011). About 18% of the region displaysstatically significant trend, with just over halfnegative and the reminder positive trend. Brown-ing of forest and non-forest vegetation has alatitudinal gradient and occurred more frequentlyin the warmer part of the species ranges; inparticular the middle taiga sub-zone shows signifi-cant decline in both forest and non-forestvegetation.

2. There are contrasting trends between bioclimaticzones. However there is substantial spatial hetero-geneity in the NDVI trends, with patches ofnegative and positive anomalies amongst a back-ground of insignificant change.

3. There are also contrasting trends for differentspecies within the same bioclimatic zone. Inparticular, most negative trends in the taiga appearto be related to a decline in evergreen coniferousforest, with the dark (Picea abie, Picea obovata) orlight (Pinus sylvestris) evergreen and evergreen-majority mixed forests showing highest decline inproductivity (76%of all negative trends for forestedareas) in every zone. In contrast, deciduous needle-leaf forest dominated by larch (Larix sibirica) showsa significant increase in productivity (54% of allpositive trends for forested areas), even whileneighboring different species show decreasingproductivity.

These contrasting results underscore the complex-ity of the patterns of variability and trends in vegeta-tion productivity. The results show the importance ofunderstanding climatic effects in combination withother factors that are causing the vegetation change.More detailed studies of the effect–response of vegeta-tion change in the region are necessary because it havedirect impact on local ecosystem and human well-being here. This suggests the need for spatially and the-matically detailed studies to better understand theresponse of different northern forest types and speciesto climate warming and other disturbances; higherresolution data, such as Landsat (Ju and Masek 2016)could be used to landcover change detection. Further-more, the results might be useful to coordinate thehumanpractice in the region.

Acknowledgments

This study was supported by the Belmont Forumthrough the Norwegian Research Council grantHIARC: Anthropogenic Heat Islands in the Arctic:Windows to the Future of the Regional Climates,Ecosystems, and Societies (no. 247268). We thank the

reviewers for his/her thorough reviews and highlyappreciate the comments and suggestions, whichsignificantly contributed to improving the quality ofthe publication. The authors thank Martin Miles fortechnical discussions and editing.

References

Aitken SN et al 2008Adaptation,migration or extirpation: climatechange outcomes for tree populationsEvol Appl. 1 95–111

Alcaraz-SeguraD et al 2010Debating the greening versus browningof theNorth American boreal forest: differences betweensatellite datasetsGlob. Change Biol. 16 760–70

Barichivich J et al 2014Temperature and snow-mediated controls ofsummer photosynthetic activity in northern terrestrialecosystems between 1982 and 2011Remote Sensing 61390–431

Bartalev S S et al 2014Assessment of forest cover in Russia bycombining awall-to-wall coarse resolution land-covermapwith a sample of 30 m resolution forestmaps Intl. J. RemoteSensing 35 2671–92

Beck P andGoetz S 2011 Satellite observations of high northernlatitude vegetation productivity changes between 1982 and2008: ecological variability and regional differences Environ.Res. Lett. 6 045501

BhattU S et al 2010Circumpolar arctic tundra vegetation change islinked to sea ice declineEarth Interact. 14 1–20

BhattU et al2013Recent declines inwarming and vegetationgreeningtrendsoverPan-Arctic tundraRemote Sensing5 4229–54

Bjerke JW et al 2014Record-low primary productivity and highplant damage in theNordic Arctic Region in 2012 caused bymultiple weather events and pest outbreaksEnviron. Res. Lett.9 084006

BlokD et al 2011The response of Arctic vegetation to the summerclimate: relation between shrub cover, NDVI, surface albedoand temperature Environ. Res. Lett. 6 035502

BonanGP 2008 Forests and climate change: forcings, feedbacks,and the climate benefits of forests Science 320 1444

BunnAG andGoetz S J 2006Trends in satellite-observedcircumpolar photosynthetic activity from1982 to 2003: theinfluence of seasonality, cover type, and vegetation densityEarth Interactions 10 1–19

BunnAG et al 2007Northern high-latitude ecosystems respond toclimate change EOSTrans. Am.Geophys. Union 88 34

BulyginaON2011Changes in snow cover characteristics overNorthern Eurasia since 1966Environ. Res. Lett. 6 045204

Comiso J andHall D 2014Climate trends in theArctic as observedfrom spaceWIREs: Climate Change 5 389–409

CrawfordRMM et al 2003 Paludification and forest retreat innorthern oceanic environmentsAnn. Botany 91 213–26

DriscollWG et al 2006Divergent tree growth response to recentclimatic warming, LakeClarkNational Park and Preserve,AlaskaGeophys. Res. Lett. 32 L20703

D’Arrigo RD et al 2009Tree growth and inferred temperaturevariability at theNorth AmericanArctic treelineGlob. Planet.Change 65 71–82

ElsakovV andTeljatnikovM2013 Effects of interannual climaticfluctuations of the last decade onNDVI in north-easternEuropeanRussia andWestern SiberiaContemp. ProblemsEarth’s Remote Sens. 10 260–71

EpsteinH et al 2012Dynamics of aboveground phytomass of thecircumpolar Arctic tundra during the past three decadesEnviron. Res. Lett. 7 015506

EpsteinH et al 2015TundraGreennessArctic Report Card 2015(www.arctic.noaa.gov/reportcard)

Esau I et al 2016Trends in normalized difference vegetation index(NDVI) associatedwith urban development in northernWestSiberiaAtmos. Chem. Phys. 16 9563–77

Forbes B et al 2009High resilience in the Yamal-Nenets social–ecological system,West Siberian Arctic, Russia Proc. NatlAcad. Sci. 106 22041–8

11

Environ. Res. Lett. 11 (2016) 115002

Forbes B et al2010RussianArcticwarming and ‘greening’ are closelytrackedby tundra shrubwillowsGlob.Change Biol.161542–54

FreyK and Smith L 2003Recent temperature and precipitationincreases inWest Siberia and their associationwith theArcticOscillationPolar Res. 22 287–300

FreyK and Smith L 2007Howwell dowe knownorthern land cover:comparison of four global vegetation andwetland productswith a new ground-truth database forWest SiberiaGlob.Biogeochem. Cycles 21GB1016

FrostGV et al 2014Regional and landscape-scale variability ofLandsat-observed vegetation dynamics in northwest SiberiantundraEnviron. Res. Lett. 9 025004

Gamon JA et al 1995Relationships betweenNDVI, canopystructure, and photosynthesis in three Californian vegetationtypesEcol. Appl. 5 28–41

Goetz S et al 2005 Satellite-observed photosynthetic trends acrossboreal NorthAmerica associatedwith climate and firedisturbanceProc. Natl Acad. Sci. 108 1240–5

Goetz S et al 2011Recent changes in arctic vegetation: Satelliteobservations and simulationmodel predictions EurasianArctic LandCover and LandUse in aChanging Climate edGGutman et al (Dordrecht: Springer)pp 9–36

GutmanGG1991Vegetation indices fromAVHRR: an update andfuture prospectsRem. Sens. Environ. 35 121–36

Huete A et al 1999A light aircraft radiometric package forMODLand quick airborne looks (MQUALS)EarthObserv. 1122–6, 39

Huete A et al 2002Overview of the radiometric and biophysicalperformance of theMODIS vegetation indicesRem. Sens.Environ. 83 195–213

IwasakiH et al 2010 Forest decline caused by high soil waterconditions in a permafrost regionHydrol. Earth Syst. Sci. 14301–7

Ju J andMasek JG 2016The vegetation greenness trend inCanadaandUSAlaska from1984–2012 Landsat dataRem. Sens.Environ. 176 1–16

KharukV I et al 2013 Siberian pine decline andmortality insouthern Siberianmountains Forest Ecol.Manage. 310312–20

Kirpotin SN et al 2009Western Siberia wetlands as indicator andregulator of climate change on the global scale Intl. J. Environ.Studies 66 409–21

LinderholmH et al 2002 Peatland pines as climate indicators? Aregional comparison of the climatic influence on Scots pinegrowth in SwedenCan. J. For. Res. 32 1400–10

LloydA andBunnA 2007Responses of the circumpolar borealforest to 20th century climate variability Environ. Res. Lett. 2045013

Macias-FauriaM et al 2012 EurasianArctic greening revealsteleconnectionsNature Clim. Change 2 613–8

Makhatkov ID 2010Relation of populations of forest-formingspecies in thefir-aspen-birch forests ofWest SalairContemp.Problems Ecol. 3 687–92

Moran PAP 1948The interpretation of statisticalmaps J. Royal Stat.Soc., Ser.B 10 245–51

MoskalenkoN2013 Impact of climatewarming onvegetation coverandpermafrost inWest Siberia northern taigaNat. Sci.5 144–8

PokrovskyOS et al2013 Impact ofwestern Siberia heatwave2012ongreenhouse gases and tracemetal concentration in thaw lakesofdiscontinuouspermafrost zoneBiogeosciences 105349–65

RaynoldsMC J et al 2008Relationship between satellite-driven landsurface temperatures, arctic vegetation types, andNDVIRem.Sens. Environ. 112 1884–94

Reidel SM et al 2005 Biotic controls over spectral reflectance ofArctic tundra vegetation Int. J. Remote Sens. 26 2391–405

Santer BD et al 2000 Statistical significance of trends and trenddifferences in layer-average atmospheric temperature timeseries J. Geoph. Res. 105 7337–56

SedelnikovVP et al 2011 Spatial-typological differentiation ofecosystems of theWest Siberian Plain. Communication: I.Plant CoverContemp. Probl. Ecol. 4 229

SedykhVN1997The Forests ofWestern Siberia and theOil andGasComplex (Moscow: Ekologica) 36 pp (in Russian)

ShahgedanovaM2003The Physical Geography of Northern Eurasia(Oxford:OxfordUniversity Press)pp 571

SolanoR et al 2010MODISVegetation IndexUser’s Guide (MOD13Series)Version 2.00,May 2010 (Collection 5)Vegetation Indexand Phenology Lab, TheUniversity of Arizona

TchebakovaNM et al 2009The effects of climate, permafrost andfire on vegetation change in Siberia in a changing climateEnviron. Res. Lett. 4 045013

UkraincevaNG et al 2011 Landscape indication of local permafrostvariability (UrengoyTerritory,West Siberia)EarthCryosphere 15 32–5

van LeeuwenW JD et al 1999MODIS vegetation index compositingapproach: a prototypewithAVHRRdataRem. Sens. Environ.69 264–80

VerbylaD 2008The greening and browning of Alaska based on1982–2003 satellite dataGlob. Ecol. Biogeog. 17 547–55

WalkerD et al 2009 Spatial and temporal patterns of greenness onthe Yamal Peninsula, Russia: interactions of ecological andsocial factors affecting the Arctic normalized differencevegetation index Environ. Res. Lett. 4 045004

WalkerD et al 2011Cumulative effects of rapid land-cover andland-use changes on the Yamal Peninsula, Russia EurasianArctic LandCover and LandUse in aChanging Climate edGGutman andAReissel (Berlin: Springer) pp 207–36

WangD et al 2012 Impact of sensor degradation on theMODISNDVI time seriesRem. Sens. Environ. 119 55–61

WayDA andOrenR 2010Differential responses to changes ingrowth temperature between trees fromdifferent functionalgroups and biomes: a review and synthesis of dataTreePhysiol. 30 669–88

WilkingMand JudayGP 2005 Longitudinal variation of radialgrowth at Alaska’s northern treeline—recent changes andpossible scenarios for the 21st centuryGlob. Planet. Change47 282–300

YuX et al 2014a Investigating the potential of long time seriesremote sensingNDVI datasets for forest gross primaryproductivity estimation over continental US Int. J. AppliedSci. and Tech. 4 55–71

YuC et al 2014bRobust linear regression: a review and comparison(arXiv:1404.6274)

YuQ, EpsteinHE, EngstromR, ShiklomanovN and StrelestskiyD2015 Land cover and land use changes in the oil and gasregions ofNorthwestern Siberia under changing climaticconditionsEnviron. Res. Lett. 10 124020

Zhou L et al 2001Variations in northern vegetation activity inferredfrom satellite data of vegetation index during 1981 to 1999J. Geophys. Res. 106 20069–83

12

Environ. Res. Lett. 11 (2016) 115002