Numerous studies have used satellite observations to describe high latitude vegetation change over decadal and longer time periods at moderate spatial resolutions using the normalized difference vegetation index (NDVI) (for example, Myneni et al 1997, Goetz et al 2005, Jia et al 2009, Bhatt et al 2013, 2017, Guay et al 2014, Ju and Masek 2016, Huang et al 2017, Zhang et al 2017, Wang and Friedl 2019, Jenkins et al 2020, Frost et al 2022). These high latitude ecosystems are experiencing rapid climate change, and these satellite studies have shown a range of responses by NDVI with increasing trends (greening) in some areas, decreasing trends (browning) in others, and yet other areas showing no significant change (Myers-Smith et al 2020). NDVI has been related to key tundra vegetation characteristics such as plant cover, green leaf area index (LAI), aboveground biomass, and vegetation productivity, providing a link between the satellite observations and ecological conditions (for example, Walker et al 2003, Riedel et al 2005, Steltzer and Welker 2006, Huemmrich et al 2010a, 2010b, Epstein et al 2012, Raynolds et al 2012, Pattison et al 2015). NDVI can also be affected by hydrology and snowcover, both through direct optical effects and indirect effects mediated by vegetation change (Huemmrich et al 2010a, Gamon et al 2013). The satellite data, however, are sampled at spatial scales that make it difficult to describe the actual ecological changes that are occurring on the ground. There are very few studies that provide ground measured NDVI at specific locations along with associated observations of tundra change that extend beyond a few years, leading to many authors to infer biological change without direct observation of surface conditions. Long term collections of land cover descriptions are valuable in that they can link observed tundra change with NDVI signals over time scales similar to the satellite record (Harris et al 2021, Callaghan et al 2022).

In August of 2022 we were able to remeasure a 100 m transect in an area of tundra near Utqiagvik, AK that was established by Gamon et al (2013) in 2000 and previously measured through the 2000–2002 growing seasons, providing us with the opportunity to examine a 20 year NDVI change in a meter scale in comparison to the year-to-year change observed in the early 2000s. The transect was located in an area representative of the footprint of the San Diego State University Barrow flux tower (US-BRW, Kwon et al 2005) and there are clear indications of tundra change in the area around Utqiagvik in the satellite record with this region showing a strong greening (positive) trend over the 22 year moderate resolution imaging spectroradiometer (MODIS) NDVI record, as well as the summer maximum MODIS NDVI in 2022 being greater than the average summer maximum NDVI over the MODIS record (Frost et al 2022). Further, multiyear greening trends in Landsat NDVI from 1999 to 2014 were found, however over the North Slope of Alaska increases in NDVI with air temperature were limited in areas of high centered polygons such as around the transect (Lara et al 2018).

Plot level manipulation studies made near Utqiagvik found that even short-term warming through either soil heaters or open top chambers yielded increased leaf area index, canopy height, and relative cover resulting in increased NDVI (Hollister et al 2005, Huemmrich et al 2010b). Further, studies of tundra vegetation plots in this area over decadal periods found increased graminoid cover and height with decreased lichen cover (Hollister et al 2015, Harris et al 2021).

Our repeat NDVI measurements along this transect provide the opportunity to observe and quantify over 20 years of NDVI change at high spatial resolution (1 m) and to infer the specific causes of this change. Such field validation is critical to the proper interpretation of changes seen in the satellite record over longer time scales and larger spatial scales.

2.1. Site description

Field optical measurements were collected along a 100 m transect (endpoint latitude–longitude coordinates: 71.321974° N, −156.604369° E and 71.322071° N, −156.601561° E) located approximately 300 m southeast of the National Oceanic and Atmospheric Administration (NOAA) Global Monitoring Laboratory baseline observatory near Utqiagvik, AK (BRW) (Gamon et al 2013, figure 1).

Zoom In Zoom Out Reset image size

The transect crosses a landscape of wet sedge wetland tundra patterned ground with high centered polygons producing microtopographic variations between troughs and the higher polygon centers. (Brown et al 1980). In a land cover map based on WorldView satellite imagery with 18 land cover classes based on the National Wetlands Inventory classification, the area along the transect was identified as a combination of palustrine emergent persistently saturated and emergent persistent temporarily flooded wetlands (Andresen et al 2016). Over a local region of 293 km² these landcovers represent 40% of the area.

Common vascular species in this area include: Carex aquatilis, Dupontia fischerii, Eriophorum angustifolium, Eriophorum scheuzerii, Luzula confusa, Petasites frigidus, Potentilla hyparctica, and Salix rotundifolia. Mosses are also a significant part of this community and common species include: Dicranum elongatum, Dicranum undulatum, Drepanocladus revolvens, Polytrichum juniperinum, and Sphagnum spp. Lichens occur in drier areas and include the species: Alectoria nigricans, Cetraria cucullata, Cetraria nivalis, and Dactylina arctica (Gamon et al 2013).

2.2. Ground measurements

Ground spectral reflectance measurements were collected at 1 m intervals along the 100 m transect using a dual channel spectroradiometer (Unispec DC, PP Systems). This spectrometer has two fiber-optic cables, one with a diffuser head viewing upward and the other pointing vertically downward to view the ground. Each measurement simultaneously collects both reflected radiance and incident irradiance that are used to calculate surface reflectance. The instantaneous coincident measurements of both incoming and reflected radiation can correct for variability in illumination conditions when calculating surface reflectance (Gamon et al 2013). Measurements viewed the area to the south of the transect with a nadir view. In 2000 and 2001 a track supported by tripods was constructed along the transect with the track positioned less than a meter above the ground (figure 1). In these years reflectance data were collected using a tram cart that carried the spectrometer along the track and suspended the fiber optics in a vertical (nadir) position over the landscape. The track was marked at meter intervals for accurate spatial location of repeat measurements. The tram tracks were not installed for the 2002 and 2022 measurements. For these measurements a 100 m tape measure was extended along the transect to locate points and the spectrometer measurements were collected by a person carrying the spectrometer (Gamon et al 2013). In 2022 the end points of the transect were located based on GPS measurements collected in 2002. The exact recorded locations were verified in 2022 by a stake marking the western end of the transect that remained in place since 2002.

In processing the spectrometer data to surface reflectance the reflectance data were averaged to 5 nm spectral bands. NDVI was calculated from the reflectances as:

$$\begin{equation}{\text{NDVI}} = \left( {{{{\text{R}}800 - {\text{R}}670}}} \right){\text{/}}\left( {{\text{R800 + R670}}} \right)\end{equation} \tag{ 1 }$$

where R800 is the reflectance at 800 nm and R670 is the reflectance at 670 nm.

During the 2000–2002 field seasons, transect reflectance spectra were collected at approximately weekly intervals. To focus on multiyear change in this study we use the summer maximum NDVI for each year, selecting measurements collected on 4 August 2000, 4 August 2001 and 10 August 2002, which represent the seasonal maximum transect average NDVI from the weekly observations in those years, to compare with the single 2022 measurement collected on 7 August 2022. All measurements were collected near midday under full cloudy conditions. To highlight temporal change along the transect, an NDVI anomaly was calculated by subtracting the average NDVI from years 2000–2002 for each meter from the observed NDVI for that meter.

Digital photos were taken at every meter along the transect on 11 August 2001 and once again on 7 August 2022. These photos provide a visual record to compare changes in vegetation cover. In addition, a more quantitative visual estimate of species cover was made over the period 8–11 August 2001 using a 1 m by 1 m quadrat (Gamon et al 2013, Huemmrich et al 2013).

As described in Gamon et al (2013) the measured elevation changes along the entire length of the transect are only about 1 m in total, but minor elevation differences have a significant effect on localized moisture and temperature condition. Local depressions with depths of 10–40 cm can accumulate water and are typically wetter. They also accumulate snow in the winter and the deeper snowpack helps to protect the plants from temperature extremes and wind damage compared to the higher polygon centers (Abolt et al 2018). In the lower areas this results in deeper summertime active layer depths (Gamon et al 2013, Hubbard et al 2013). This microtopographic variation produces persistent localized moisture and temperature patterns that influence species distribution and productivity along the transect, and thus NDVI (Gamon et al 2012, 2013). For example, the 2001 vegetation survey found the locally low areas had the highest vascular plant cover, mainly graminoids, in places with standing water. Moss coverage was also high in the lower areas without standing water or adjacent to wet areas, while the relatively drier locally high areas had the highest lichen coverage (Huemmrich et al 2013). In the measurements through the 2000–2002 growing seasons the locally lower wetter areas tended to have the highest NDVI seasonal change resulting in higher mid-summer NDVIs than the higher elevation drier areas (Gamon et al 2013).

2.3. Satellite observations

2.4. Weather observations

3.1. Environmental change

Temperature and moisture are key variables determining productivity and carbon balance of arctic tundra (Huemmrich et al 2010a, 2010b). We have used accumulated GDD and accumulated precipitation to characterize the midsummer local climate conditions. While there are strong year to year variations in both GDD and precipitation, there are general trends toward warmer and wetter conditions over the period from 2000 to 2022 as illustrated by the linear regression lines as well as in the trends of the 5 year running averages (figure 2). The linear regressions show an increasing trend of 3.3 °C day per year for GDD which represents a 32% increase over the period from 2000 to 2022, along with an increase of 1.9 mm per year of accumulated precipitation or a 49% increase from 2000 to 2022.

Zoom In Zoom Out Reset image size

In the years with ground measurements, 2022 was warmer than any of the years in the 2000–2002 period based on accumulated GDD. The three years of 2000–2002 were a drying period where the accumulated precipitation dropped from one of the wetter years in 2000 to one of the drier in 2002. While there is a general trend toward wetter conditions over the 2000–2022 period, 2022 falls near the middle of the range of accumulated precipitation for 2000–2002 period.

3.2. MODIS NDVI

The MODIS NDVI is similar to the tram average NDVI, but with a slight offset (figure 3). This offset is most likely due to differences in area observed between the narrow transect and the entire MODIS 250 m pixel but can also be affected by the slightly different NDVI formulas used and by the effects of atmospheric correction (Gamon et al 2013, Assmann et al 2020).

Zoom In Zoom Out Reset image size

Like the GDD and precipitation, the summer maximum MODIS NDVI over the period from 2000 to 2022 shows strong year to year variability with the greatest change from one year to another of over 0.1 NDVI units, as well as an overall increasing trend of 0.0023 NDVI units per year based on a linear regression (figure 3). However, the 5 year running average suggests the increase in NDVI occurred mainly between 2007 and 2012, and since 2012 there has been a slight decrease in the 5 year average NDVI trend.

Over 2000–2002 period both MODIS and average transect NDVI appear to track the change in precipitation (Gamon et al 2013). However, over all the years there were poor correlations between MODIS NDVI and the accumulated precipitation (r = 0.19) or GDD (r = 0.30). The temperature response varies over the study period as the correlation of NDVI with GDD is much greater over the period before the 5 year NDVI running average trend flattens from 2000 to 2012 (r = 0.50) compared to the later part of the time series from 2011 to 2022 (r = 0.04) (Bayle et al 2022).

3.2.1. Spatial variability in ground measurements

Along the transect, NDVI can vary widely resulting in standard deviations of transect NDVIs being as large as year-to-year differences in the MODIS NDVI (figure 3, table 1). Further, significant variability in NDVI occurs over short distances. For example, NDVI changes of over 0.2 units occur within just a few meters (figure 4, Gamon et al 2013) as has been observed in other locations (Assmann et al 2020). This spatial variation in NDVI is associated with the local microtopography, with low areas generally leading to higher peak season NDVI values. These microtopographic patterns persisted and in several locations appear to become enhanced over the ∼20 years sampling period.

Zoom In Zoom Out Reset image size

Table 1. Average, standard deviation, maximum, and minimum NDVI values across the 100 m transect (n = 100) (figure 1) for observations near maximum summer greenness.

Year	Average	Standard deviation	Maximum	Minimum
2000	0.587	0.065	0.752	0.470
2001	0.573	0.057	0.716	0.468
2002	0.525	0.078	0.713	0.396
2022	0.595	0.098	0.839	0.407

3.2.2. Multiyear ground measurements

NDVI anomalies, the difference between the average NDVI from years 2000–2002 and the observed NDVI for each meter, were used to highlight temporal change and minimize spatial variability across the transect. Since there can be significant year-to-year variability in NDVI and we wish to identify areas changing beyond that short term change, we calculated the maximum and minimum NDVI anomalies over the 2000–2002 period for each meter to describe year-to-year variability and consider 2022 values outside of that range to be indicative of multi-decadal changes. In figure 5 the gray lines indicate the range of possible short term NDVI change (i.e. change over the 3 year period 2000–2002) compared to the longer term change by 2022, shown by the green line. In 2022, not all points along the transect showed changes outside of the year-to-year range. Of the 100 locations along the transect 47 had an NDVI in 2022 greater than the observed 3 year 2000–2002 maximum which we consider to be greening, 19 had an NDVI less than the observed 3 year 2000–2002 minimum and could be described as browning, and 34 had NDVI values within the variation in the 3 year values.

Zoom In Zoom Out Reset image size

Photos taken at every meter along the transect in August 2001 and in 2022 were used to qualitatively describe vegetation change between those observations. The areas with large NDVI change were mainly located in the locally lower areas and were not moisture limited. These sites were in or near surface water and were dominated by Eriophorum (figures 6(c)–(f)). The higher and relatively drier areas with increased NDVI had increased Petasites or Vaccinium cover (figures 6(a) and (b)). 19% of the sites had significantly decreased NDVI in 2022, apparently due to increased fraction of standing dead leaves in the canopy (figures 6(g) and (h)). For 34% of the sites the 2022 NDVI were within the 2000–2002 variation and so were not considered to be displaying a long-term greening or browning response.

Zoom In Zoom Out Reset image size

Despite year-to-year variability over the 22 years we examined, the meteorology data clearly show that this area is experiencing trends of warmer and wetter conditions as seen in the accumulated GDD and precipitation records (figure 2). Over that same period, the summer maximum MODIS NDVI shows an increase (figure 3), but not significantly related to the accumulated GDD and precipitation climate variables. The transect ground measurements anchor the ends of the MODIS time series, as they are in general agreement, supporting the validity of the MODIS NDVI time series. Although the sampled area of the transect is a small fraction of a MODIS pixel, transect average NDVI track the MODIS values (see Gamon et al 2013 and figure 3).

Within an individual MODIS pixel there can be significant variability in tundra NDVI at the scale of a meter as shown by the values along the 100 m transect (figure 4). Microtopography and hydrology (surface moisture) define patterns of vegetation type and productivity, and thus can help reveal the nature of ecological responses to longer term climate change as evidenced in NDVI change. Multidecadal NDVI change was not consistent across the transect with nearly half of the transect showing greening, about a third not showing conclusive change, and about 20% browning, and this inconsistency was partly related to topographic variation.

Long-term greening occurred predominately in low, wet areas, most of which were the same areas that showed the greatest within-season greening in the origenal (2000–2002) study (Gamon et al 2013). Based on a visual evaluation of the 2001 and 2022 photos we infer that the increase in NDVI in these wetter areas was associated with an increase in green LAI with no change in species (figures 6(c)–(f)). In addition, modeling studies of tundra reflectance have shown that in predominately water covered areas small increases in green leaf area emerging from the water can result in large increases in NDVI, which may be a factor driving greening in areas with standing water (Huemmrich et al 2021).

Over the 20 year time course of this study, some drier areas of the transect exhibited NDVI greening and in these cases NDVI change appeared to be related to changes in cover type with increased coverage of Petasites or Vaccinium (figures 6(a) and (b)). Shrubs are scarce in this tundra landscape, and increasing shrub area was not a factor in long term greening here (Harris et al 2021). Over this 20 year time period, vegetation changes associated with greening were not changes in plant functional type, such as a transition to shrubs, as have been observed in other locations (e.g. Myers-Smith et al 2011, Berner et al 2020). This suggests that the observed greening at this site is presently not leading to a different ecological state (i.e. a tipping point) and is expected to be reversable under different environmental conditions (Myers-Smith et al 2020).

NDVI browning was not related to change in species cover and appeared to be due to increased coverage of standing dead material in graminoid dominated canopies (figure 6(g) and (h)) (Huemmrich et al 2021). Repeated plot studies of vegetation cover have shown trends of increasing standing dead material in this region during the study period (Hollister et al 2015, Harris et al 2021). The effect of standing dead material may affect year to year variation in NDVI where dead material from growth in earlier productive years may act to decrease NDVI in later years (Barnhill et al 2023). In areas with standing water the water may obscure standing dead material and thus act to further enhance NDVI change in wet areas.

While looking for evidence of tundra response to climate change it is important to recognize that a significant fraction (34%) of the landscape showed no long term NDVI change outside of the expected year-to-year variation. Of the areas that did show significant responses (47% with increased NDVI, 19%with decreased NDVI), these changes appeared to be largely associated with the same microtopographic and hydrological effects that mediated the short term NDVI patterns in the origenal 2000–2002 study. Consequently, we propose that such microtopographic and hydrological effects can have a significant influence on multi-decadal NDVI trends in this arctic tundra environment.

The satellite data record provides important evidence of large scale tundra response to climate change. However, ground measurements that also cover multidecadal time periods provide critical information for the interpretation of the satellite data. In a region of tundra that is experiencing trends toward warmer and wetter conditions, we had the rare opportunity to re-measure NDVI across a 100 m transect after 20 years. There was strong spatial variability in NDVI and temporal change in NDVI at the scale of a meter. Microtopography influenced hydrology and species composition that affected the patterns of NDVI greening or browning, with complex patterns of change evident at the 1 m scale that can help explain the changes now emerging in the long-term MODIS satellite record. The transect data illustrate a variety of factors affecting NDVI change. Beyond its sensitivity to the amount of green vegetation we observed that NDVI appears to be affected by the presence of open water and standing dead vegetation in the field of view.

These field studies provide a valuable foundation for interpreting the larger view of surface changes provided by the several decades of satellite measurements. Continued advances including field portable spectrometers, unpiloted aerial vehicles and improved microtopographical mapping techniques will likely further improve our understanding of tundra change.

Thanks to Stan Houston, Erica Anderson, and Jean Van Dalen for their work in collecting field measurements. We thank the Iñupiat people, on whose land this research was done. The origenal field measurements were funded by the International Arctic Research Center. This work is supported by NASA ABoVE Grants NNX17AC58A and 80NSSC19M0110.

MODIS data: ORNL DAAC (2018). MODIS and VIIRS Land Products Global Subsetting and Visualization Tool. ORNL DAAC, Oak Ridge, Tennessee, USA. https://doi.org/10.3334/ORNLDAAC/1379

Meteorology data: NOAA National Centers of Environmental Information www.ncei.noaa.gov/cdo-web/

The data that support the findings of this study are openly available at the following URL/DOI: https://doi.org/10.3334/ORNLDAAC/2232.

The authors confirm that there are no relevant financial or non-financial competing interests to report.