Climate change is enforcing physiological changes in Arctic Ecosystems

Remote sensing and site-level observations have shown unprecedented changes in Arctic ecosystems owing to climate warming. These observations include greening and browning trends in Arctic vegetation as well as changes in species composition both in the tundra and the boreal forests. Here, we leveraged solar induced chlorophyll fluorescence (SIF) to study changes in ecosystem phenology across the pan-Arctic domain from 2000 to 2020. Ecoregions, as a proxy for plants’ functional traits, were the single most important variable to explain the spatial and phenological heterogeneity in observed SIF trends. We observed unique regional trends in ecosystems responses to climate change affecting the timing of spring photosynthesis onset, magnitude of peak productivity during the growing season (GS) and fall senescence. While, Photosynthetic activity in the early GS showed increasing trends across the vast majority of the pan-Arctic, it tends to decline at the end of the season for nearly half of the land area, including parts of North America but more significantly in central Siberia. The observed changes in phenology highlight the role of biodiversity in regional climate sensitivity and emphasizes the need for better representations of sub-biomes to community level information in Arctic ecosystem process models and projections. These results also highlight the importance of phenology information in ecosystem models for better understanding of the timing and magnitude of carbon uptake in the Arctic domain with accelerated changes in climate.


Introduction
Productivity in the Arctic tundra and taiga ecosystem has historically been temperature-limited (Churkina and Running 1998, Nemani et al 2003, Madani et al 2017a, Liu et al 2020a. Climate warming in the last few decades has triggered increased vegetation growth in the Northern Arctic, leading to widespread increases in vegetation greenness (Zhu et al 2016, Berner et al 2020, Myers-Smith et al 2020, Berner and Goetz 2022. Underlying driving processes such as earlier and longer growing seasons (GSs) could in theory reinforce the productivity rate and create far-reaching natural negative feedbacks to mitigate the impact of increasing atmospheric CO 2 concentrations. However, the assumption that reduced temperature constraints and enhanced CO 2 fertilization could increase total ecosystem productivity in the northern high latitudes (Piao et al 2014, Forkel et al 2016 ignores other potentially growth limiting factors including water availability (Jiao et al 2021), disturbance (Rocha et al 2018, Wang and Friedl 2019, Madani et al 2021, photoperiod limitations (MacIas-Fauria et al 2012), phenological changes (Richardson et al 2010, Zheng et al 2020, and species-specific responses to changes in climate (Tilman et al 2012, Fei et al 2017. Each of these factors may reinforce or offset vegetation greening's impact on carbon uptake. Hence, the extent to which Arctic ecosystems respond to the changing climate remains highly uncertain. Recent studies have highlighted the spatial heterogeneity in Arctic ecosystem responses to climate change that not only includes greening trends, but notable vegetation-browning signals (Bhatt et al 2013, Appenzeller 2015, Myers-Smith et al 2020, Huemmrich et al 2021, Berner and Goetz 2022. The seemingly inconsistent responses to increasing temperatures are attributed to significant climate events and physical and biological forces that could result from the changes in water availability between the onset and ending of the photosynthetic period (Park et al 2019). An earlier warming-induced spring greenup could shift peak photosynthesis away from climatological values, resulting in earlier soil water depletion and reduced photosynthetic rates in the mid-to late-stages of the photosynthetic GS (Hu et al 2010, Liu et al 2019. This phase shift in the timing and length of the GS and seasonal amplitudes of carbon uptake can significantly influence exchanges of carbon, water, and energy with the atmosphere and have downstream impacts on ecosystem properties including species habitats, food availability, and disturbance regime (Polgar and Primack 2011).
Land carbon uptake from plant photosynthesis is the major part of the global carbon cycle and yet there is no consensus on total estimation of productivity among global ecosystem models (Anav et al 2013(Anav et al , 2015. A major source of uncertainty in productivity estimation in global ecosystem models is related to simulation or observation of temporal patterns of interannual variability in plant phenology (Jolly et al 2005, Peano et al 2021, Song et al 2021. A number of studies have used the reflectance based Normalized Difference Vegetation Index (NDVI) (Huete and Justice 1999) derived from Advanced Very High-Resolution Radiometer observations or the Moderate Resolution Imaging Spectroradiometer (MODIS) for tracking plant phenological changes and trends in productivity. These sensors provide continuous detection of land surface biophysical properties including phenology since 1978 and have enabled breakthrough insights into the evolving picture of vegetation greening and browning trends during recent decades associated with structural changes in photosynthesis (Polgar and Primack 2011, Lu et al 2018, Wang et al 2018, Berra and Gaulton 2021. Strong relationship between vegetation indices and canopy leaf area index (Myneni et al 2002) provides a suitable means by which to monitor canopy interannual dynamics. However, greenness-based satellite products have limitation in detecting plant phenological phases (e.g. leaf flush, leaf growth, and dormant phase) (Berra and Gaulton 2021, Zolles et al 2021) in addition to other problems such as the impact of soil background reflectance on greenness signal (Rocha and Shaver 2009). Due to these limitations, satellitebased ecosystem productivity models, also known as light use efficiency models, are using additional proxies such as temperature and moisture constraints to better account for the impact of environmental stress on ecosystem productivity and to better characterize seasonal plant phenology dynamics (Zhao et al 2005, Jones et al 2017. Unlike greenness indices, satellite observations of solar induced chlorophyll fluorescence (SIF) provide key complementary information on functional changes in photosynthesis related to physiological responses to temperature (Liu et al 2020a) and water stress , enhancing our ability to detect short and long term phenology of vegetation function (Jeong et al 2017, Lu et al 2018, Magney et al 2019, Pierrat et al 2021. SIF satellite retrievals are available from multiple sensors over the last few decades. However, the limitations such as short observation records from individual sensor, sensor degradation, and inconsistency in retrieved signal across different instruments make the multi-decadal analysis on vegetation dynamics challenging (Zhang et al 2018b, Wen et al 2020. Recently, efforts have been made to extend SIF observations, specifically from the Orbiting Carbon Observatory-2 (OCO-2) (Zhang et al 2018a) by utilizing MODIS reflectance retrievals in a machine learning environment, to provide a contiguous SIF (CSIF) records spanning multiple decades (Zhang et al 2018a, Zheng et al 2020. In particular, hybrid SIF datasets such as the CSIF data records use the advantage of OCO-2 based SIF phenological information in addition to MODIS long-term high spatiotemporal records and are expected to improve our understanding of dynamics of gross primary productivity (GPP) and plant interannual dynamics relative to changes in climate.
The relatively long temporal coverage of the hybrid SIF product (2000-2020), leveraging high accuracy but limited temporal frequency and duration of OCO-2 spaceborne SIF retrievals (2014present) (Zhang et al 2018a), provides an opportunity to track changes in plant functional phenology over the last two decades.
Here, we leverage the CSIF data to assess changes in vegetation phenology across the pan-Arctic in response to climate variability for 2000-2020. We perform phenology analysis (see methods) to estimate annual start of the season (SOS), GS, and endof-season (EOS) metrics in northern high-latitude tundra and taiga vegetation (>50 • N). We then estimate linear trends to track changes in phenology metrics from 2000 to 2020 at grid-point, regional, and continental scales. We provide new a perspective on greening/browning trends by integrating and characterizing multiple aspects of phenological changes over the full seasonal cycle for a more accurate assessment of climate sensitivity as a function of regional differences in climate, hydrology, geology, and community structure.

Datasets
We used global CSIF data from 2000 to 2020 to estimate trends and derive phenology metrics. Additionally, we acquired several ancillary datasets representing climate, plants' physical characteristics, and topography to assess the factors that explain changes in plant phenology in the pan-Arctic domain. Additional information regarding the datasets used in the study are available in the supplementary material.

Phenology estimation from CSIF
Our phenology analysis is focused on extraction of pixel-level CSIF trends across three primary phenological stages over a 2-decade period from 2000 to 2020: (1) start of GS in the spring, denoted SOS, (2) end of GS in fall, denoted EOS, and (3) the GS period between SOS and EOS, denoted GS. We define SOS as the date when photosynthesis is reactivated in spring, following leaf-out in deciduous species and pigment transitions in evergreen species (increased chlorophyll:carotenoid ratio) needed for more permanent physiological shifts for the GS (e.g. (Pierrat et al 2021)). In evergreen species in particular, the transition from dormancy to the SOS is a gradual process associated with environmental triggers (warming, thawing, trunk dehardening) and biological change (deactivation of xanthophyll cycle, stem rehydration), then accelerates as optimal growing conditions and maximum photosynthetic capacity are reached. Likewise, we define EOS as the date when photosynthesis is deactivated in the fall, following seasonal leaf loss in deciduous species and reversal of pigment ratios during the transition back to winter dormancy.
These rapid transition periods are well confirmed by SIF (Zheng et al 2020, Pierrat et al 2021. We therefore focus our CSIF based phenological analysis on these periods, with the goal to (1) detect when SOS and EOS transitions occur in the climatological record of CSIF , and (2) study long term variability in photosynthesis as the timing of optimal growth conditions shifts away from climatological dates. Based on these definitions, we interpret 'greening' and 'browning' trends for each phenological growth stage. Increase in SIF signal at the SOS over time will drive up the mean SIF signal during the climatological SOS and subsequent increase in the integrated value of SIF during the GS (between climatological start and end dates), resulting from longer GS and/or larger peak in SIF values. Likewise, browning refers to decline in SIF values over time during different growth stages.
To detect these three stages, some preprocessing is needed to gap-fill and downscale to daily time scale. We used linear interpolation to convert four day CSIF data into daily time steps. We then used the daily interpolated SIF data to estimate SOS and EOS by using the R software package 'greenbrown' (Forkel and Wutzler 2015) based on the derivative of the seasonal curve (Tateishi and Ebata 2004). This approach defines the SOS and EOS based on the mid-points of spring green-up and autumn senescence, respectively (Forkel et al 2015). To assess the changes in the SIF values at the SOS and EOS, we extracted the SIF value of the corresponding day of the year derived from phenology analysis and retrieved the average SIF value within ±3 d. In order to isolate the impact of timing and magnitude of annual SIF in trend analysis, we used the SOS and EOS dates from the beginning of the time series to extract the corresponding annual SIF values each year from 2000 to 2020. We also analyzed the average SIF values between the SOS and EOS dates to focus on trends in SIF during the GS. Additionally, we calculated the SIF value at peak GS (PEK). This procedure allowed us to monitor any changes in SIF as a metric for productivity throughout the studied period. We also classified specific regions that shared common patterns of increasing or decreasing phenological trends. These patterns are identified based on analyzing SOS and EOS trends and previous reports of the role of phenology in regulating ecosystem processes or physiological activity (Richardson et al 2010). We first classified the regions based into two categories of earlier and later spring onsets by referring to SOS trends. We further referred to PEK and EOS of those regions and categorized the regions that showed increase or decline in SIF values at PEK and EOS. The retrieved information is helpful in defining the ecosystems' unique responses to changes in climate. Along with phenological analysis of SIF, we used MODIS NDVI and calculated phenological metrics in the same way as SIF data. We also compared the seasonality of SIF with that of MODIS NDVI at the regional level and among PFTs.

Data analysis
We created a RandomForest (RF) (Liaw and Wiener 2002) machine-learning model to analyze factors that may explain the heterogeneity in greening and browning trends across three phenological stages. In this regard, we aimed to attribute information on SIF trends concerning the potential underlying biotic and abiotic factors listed above. We trained the models at 0.5 • spatial resolution on 70% (randomly selected) of the pixel points from the study domain. Three RF models were created to assess the importance of the explanatory variables in explaining the heterogeneity in SIF trends across the season. We validated and tested the models' performance over the remaining 30% of the independent pixels, in which SIF heterogeneity at the SOS, GS, and at the EOS was predicted with 80%, 71%, and 76% accuracy, respectively. The models report an increase in mean squared error (IncMSE) after permuting each variable, indicating the importance of each potential factor in explaining the heterogeneity of the pixels relative to SIF trends. We performed additional analyses at regional scales and on specific ecoregions to monitor the changes in phenology in the last 2-decades. We assessed the regional changes in SIF climatology for each five year period for regions with distinctive phenological patterns.

SIF representation of ecosystem phenology
The hybrid CSIF data product indicates significantly high correlation (R 2 = 0.98, P < 0.05) with selected tower GPP data at seasonal cycle (figure S5). In the pan-Arctic domain, CSIF indicates shorter GS compared to MODIS-based NDVI (72 ± 2.5 d pan-Arctic GS in CSIF, while NDVI detects the length of GS for 130 ± 6.3 d) at the beginning and end of the GS (figures 1, S6 and table S2).
CSIF phenology representation of ecosystems was tested over 32 tower sites with more than two years of CO 2 flux measurements over boreal and tundra biomes (figure 2). CSIF values are highly correlated with tower GPP estimates at monthly scales over the beginning of GS (January-May), peak of the season (June-August) and end of the growing season (Septembe-December). It should be noted that here we used the native spatial resolution of CSIF data (0.05 • ) and temporally matched the measured GPP years and months with their corresponding timestamps of SIF. Results indicate that SIF can provide valuable phenological information for studying ecosystem response, growth, development, and productivity over different growth stages. Moreover, SIF phenology information can be used to study the timing of vegetation growth and how it varies over time, which has important implications for understanding ecosystem function relative to changes in climate.

Greening and browning trends in the pan-Arctic
The pan-Arctic region shows contrasting patterns of SIF greening and browning trends that vary on seasonal time and regional spatial scales. Most regions show an increase in SIF signal at the SOS, with particularly strong positive trends in Alaska and northwestern Canada as well as northern portions of western and central Siberia ( figure 3(a)). The Hudson Bay Lowlands showed a declining SIF trend at the SOS after the year 2013, a notable exception. In contrast, a swath showing decreasing SIF trends during the GS between 55 N and 65 N is observed in North America and from Scandinavia across European Russia and into central Siberia ( figure 3(b)). EOS SIF trends exhibit extensive declines across most of Siberia, North Slope of Alaska, the Canadian Arctic, and the southern edge of the North American boreal forest ( figure 3(c)). Eastern Siberia and the Ob/Yenisei basins display increasing SIF trends for the SOS, GS, and EOS, but the increases are smaller for the GS and EOS than the SOS. The greening trend at the SOS in interior Alaska, continues during the GS, but tends to decline at the EOS, consistent with seasonal compensation patterns in Alaska and western Canada (Liu et al 2020b). The functional phenological change as observed via SIF is different from that shown by the NDVI, which shows similar greening pattern at the SOS and GS, but less EOS browning (figures S7-S8; median trend is positive [greening] for NDVI and close to zero for SIF). Notable differences between the signs of the SIF and NDVI trends occur in central Siberia, Northern Europe, and the North Slope of Alaska that can be due to higher sensitivity of SIF to LUE and closer relationship with plant phenology especially in the Arctic and Boreal forests (Wang et al 2020).
Analyses of selected phenological metrics for North America and Eurasia indicate that photosynthetic activity increased across the pan-Arctic from 2000 to 2020, but the vast majority of that increase was concentrated in the early GS. In fact, EOS photosynthetic activity actually declined significantly for ∼50% of the pan-Arctic during that same period, especially in North America (figure S9). Multiple factors can influence these phenological changes. We used a RF algorithm to quantify the ability of vegetation characteristics, climatic factors, topo-geographic variables and disturbance to explain the spatial and phenological heterogeneity in SIF trends shown in figure 3. Figure 4 shows that ecoregion was the single most important factor, removing ecoregion from the models would increase the mean square error (%IncMSE) between 65% and 75% in all three growth periods, while plant functional type (PFT) generally represented the least important factor (<20% IncMSE). Ecoregions characterize regionally distinct assemblages of species and communities within various biomes where ecosystems (and the type, quality, and quantity of environmental resources) are generally similar (Olson et al 2001).   . While studies have focused on heterogeneity in traits and trait adaptation to local environments as an explanation for ecosystem processes such as productivity, species-level responses to changes in climate from a phenological perspective have received less attention. Among the climatic variables, as expected, temperature played a larger role in explaining the pixel-scale heterogeneity of the trends at the start and middle of the GS, while factors linked to water limitations, including soil moisture (SM) and vegetation optical depth and their trends played a more important role at the EOS. At the EOS, tree height and percent-tree-cover were the most important factors after ecoregions and factors representing water and topographical information in explaining the SIF pixel-level greening/browning signals. The importance of the tree height variable may be related to higher tolerance of taller trees to drought (Westoby 1998).
Among the topographic-based variables, elevation and topographic wetness index remained important variables at all three stages of growth.
While other variables were significant in explaining the greening/browning signals, their importance was relatively lower than the named static landscape characteristics. These results indicate that phenological trends depend more on regional adaptations to climate change and landscape patterns due to species composition, growth forms, and traits of plant communities.

Regional patterns of ecosystem physiological changes
We identified four unique patterns of physiological change in the pan-Arctic based on shifts in timing of SOS, EOS, and amplitude of PEK (figure 5). The first pattern, denoted Universal Growth, is characterized by a longer, more productive GS, including earlier SOS, later EOS, higher PEK, and a net increase in annual SIF over each five year period from 2000 to 2020 ( figure 5(a)). The Universal Growth pattern covers the majority of the pan-Arctic land area (∼54%). These are relatively moist areas (humid category on aridity index), covering the western to northern regions of North America, most of Europe, and Eastern Siberia. MODIS land cover maps indicate these areas are covered by shrubs (∼50%) and boreal evergreen forests (∼25%) (figure 6(a)). Warming trends favor increased growth in spring (earlier SOS), and ample moisture availability supports enhanced assimilation and growth through the entire GS.
The Later Amplified Growth pattern, characterized by delayed spring onset (∼9% area in our study domain), is mostly located in eastern Canada and includes the Southern Hudson Bay Taiga and Central Canadian Shield Forests ecoregions. The reduced temperature in early spring inhibits growth at the early stage of plant activities, but physiological activities tend to increase at the EOS ( figure 5(b)). The delay in spring onset in these regions is consistent with the increased snow cover in early spring (figure S10). Additionally, SIF values at the PEK in this pattern tend to increase through time, but are offset by reduced growth at the SOS, such that the total annual SIF did not change significantly over our 20 yr study period.
The Earlier Amplified and Earlier Dampened patterns (figures 5(c) and (d), respectively) both show reduced growth at the EOS. In the former case, representing ∼18% of the pan-Arctic area, increased spring activity leads to higher carbon assimilation during the GS. In the latter case, representing ∼11% of the pan-Arctic area, photosynthetic activity after early spring onset declines through the rest of the season. The aridity index classifies these regions as drier. They cover some ecoregions in western North America, including the Interior Yukon-Alaska alpine tundra, Yukon Interior dry forests, Arctic coastal tundra, Northern Canadian shield taiga, Muskwa-Slave Lake forests as well as Taimyr-Central Siberian tundra and Eastern Siberian taiga in the Eurasian domain (figures 6(c) and (d)).
The decline in growth at the EOS after an earlier spring onset (figure 6(c)) could also be due to leaf aging effects, while the reduction in growth during the GS that extends to the EOS (figure 6(d)) could be related to moisture constraints (Richardson et al 2010). In the Eastern Siberian taiga, we found high correlations between interannual variability in SIF values at the EOS with SM interannual variability (figure S11), showing that increased water constraints in the region could inhibit growth at the EOS. This suggests that EOS water limitation in central Siberia is Variable importance is derived from a Random Forest analysis of the relationship between SIF trend values for each period with selected explanatory variables representing plant community structure (yellow), climate (blue) and other landscapes features (red). Variables with higher importance increase the percent mean squared errors (%IncMSE) when removed from the models. occurring earlier than expected from the RCP4.5 and RCP8.5 scenarios . Our results further indicate that in general, increasing LST trends at the SOS have provided suitable conditions that resulted in the Universal Growth pattern (figure S12). On the other hand, negative trends in SM have caused the 'Damped Growth pattern' , while 21% of the variability in ecosystem responses within the growth forms to trends in temperature and moisture can solely be explained by ecoregions.
Overall, our SIF analysis indicates significant greening patterns across each phenological stage, including Universal Growth across more than half of the pan-Arctic land area. Nearly one-third of Figure 6. Regional pattern of net annual SIF within detected growth form in the pan-Arctic domain including: Universal Growth (a), Later Amplified Growth (b), Earlier Amplified Growth (c) and Earlier Dampened Growth (d). Plots show regional SIF patterns, PFT composition, and aridity classification of four growth forms as well as trends in regional mean start of season (SOS), peak of the growing season (PEK), and mean end of season (EOS) values in SIF from 2000 to 2020.
the land area, however, shows significant late-season browning, including parts of North America, but more significantly in central Siberia. The Universal Growth based on the NDVI analysis covers 61% of the pan-Artic domain, indicating that NDVI tends to overestimate the area of greening (larger fraction of greening land area, figure S7) and underestimate the rate of greening in the early to middle GS, (faster growth rates in greening pixels, figures S7, S13 and table S3). These results indicate that reliance on reflectance-based greenness indices, limited to phenology structural information (Jeong et al 2017), may overestimate the beneficial influence of warming temperatures on vegetation growth.

Impact of plants' physiological changes on Arctic ecosystems
The heterogeneous response of Arctic ecosystems to climate change as detected here by SIF trends as well as in previous studies (Zhu et al 2016, Park et al 2019, Myers-Smith et al 2020, Piao et al 2020 could have broader impacts on ecosystem function and structure. The pervasive Universal Growth pattern observed across half of the pan-Arctic over the past 2 decades (figure 5(a)) helps explain long term increases in gross and net carbon uptake in northern high latitudes (e.g. Elmendorf et al 2012, Liu et al 2020a). However, the herbivory pressure (Cahoon et al 2012) and anticipated higher disturbance influence due to loss in leaf area (Phoenix and Bjerke 2016), as well as mortality and erosion (Grosse et al 2016) could adversely impact higher-biomass ecosystems and inhibit the beneficial impact of greening on CO 2 uptake. On average, the Universal Growth pattern shows a ∼6.2%/decade increase in SIF at the SOS, ∼3.1%/decade increase at the PEK, and 5.7%/decade increase at the EOS (table S3) The emerging browning trends detected in western North America and eastern Siberia in the second pattern, denoted Later Amplified Growth, are potentially alarming within ecosystem health and productivity perspectives, indicating earlier-thananticipated vulnerability to increased water limitation. In these ecosystems, we observed a delay in spring onset and −11.8%/decade reduction in SIF at the SOS, but an increased SIF through the remainder of the GS, including a 5.5%/decade increase in the PEK and 5.9% increase at the EOS ( figure 5(b)). The annual SIF increases from 2000 to 2010 as PEK and EOS positive trends exceed SOS negative trends, then equilibrates over the last decade (2010-2020) as SOS and PEK/EOS trends balance out. As such, late spring snowfall can inhibit total GS carbon assimilation even as cold temperature limitations are relaxed in a warming climate. This decadal change in the balance of greening and browning trends across the GS underscores the importance of characterizing all growth stages in Arctic greening analyses, rather than focusing on a single commonly used metric such as annual maximum summer NDVI (e.g. Berner et al 2020, Berner and Goetz 2022), which can bias our interpretation of vegetation change.
The third pattern, denoted Earlier Amplified Growth, shows earlier spring onset and enhanced activity in the early to mid-GS (16%/decade increase at the SOS, 3.4%/decade increase at the PEK), but reduced activity at the EOS (−3%/decade; figure 4(c)). The fourth detected pattern, denoted Earlier Dampened Growth, shows a shift toward earlier spring onset and positive SOS trends (5.7%/decade), but negative trends in PEK (−2.7%/decade) and EOS (−5.0%/decade), producing a decrease in net annual SIF from 2000 to 2020 (figures 5(d) and S16). The negative growth trends in eastern Siberian taiga, the habitat of Siberian larch forests (the largest distributed coniferous forests in the world (Kobak et al 1996)), is alarming. Even though these ecosystems have been affected by recent wildfires (Kim et al 2020), the pervasive browning signal at the EOS in these regions is beyond the influence of wildfires, as we observed similar browning trends in unburned pixels in the region (figure S14). Our results indicate that with the current rates of change in SIF (figure S15), ecosystems in Muskwa-Slave Lake forests, Midwestern Canadian shield forests and Eastern Siberian taiga are the most vulnerable, and we expect a significant decline in the total productivity and carbon uptake in these regions.
Our study highlights the significance of SIF in determining ecosystem phenological patterns and productivity. SIF compared to NDVI is closely related to physiological functions of the plants (Wang et al 2020) and the differences in SIF and NDVI phenological representation of ecosystems suggest that future efforts need to focus more comprehensively to understand vegetation dynamics and their responses to climate change. Nevertheless, the heterogenic phenological response of the Arctic ecosystems to changes in climate is a complex pattern. In this regard, other than the direct response of ecosystems to climate change, plant adaptation lag to climate change due to their physical traits and evolutionary adaptation to climate (Parker et al 2017, Madani et al 2018 may have a leading role in recent observed changes in the Arctic landscapes.

Conclusions
Our study highlights the role of ecoregions, representing distinct assemblages of species and their communities, in controlling regionally unique and spatially coherent vegetation-climate feedback patterns. Plants' distinctive strategies that manifest as functional traits reflect their local habitats and environmental conditions (Wright et al 2004, Díaz et al 2015. Apart from climate impact on plant phenology, plants' phenological changes through time could also be linked to changing species composition, traits, and diversity that are currently hard to access at a large scale due to a lack of available long-term and expansive datasets on plant traits. The availability of dynamic species-trait data could help in analyzing the impact of species-level patterns and processes in the region. Furthermore, refining global ecoregion maps via satellite observations could greatly improve productivity estimates and global models that are currently lacking in terms of species' community information (Reichstein et al 2014, 2017b, Madani et al 2018. Nonetheless, our SIF-derived changes in Arctic phenology may not thoroughly represent carbon assimilation trends. However, these results highlight the importance of accurate phenological representations in ecosystem models in order to better represent ecosystem processes, such as carbon and nutrient cycling as well as changes in plant community distribution. Nevertheless, the dynamic and complex nature of seasonal to interannual changes in plant physiology are strong evidence that increased vegetation growth and carbon uptake are not a universal response to climate warming but rather more complex functions of diverse environmental drivers and unique resiliencies and vulnerabilities across ecosystems.

Data availability statement
All data used in this research are publicly available from the cited literature and the links below: CSIF data are available for public to download from: https://osf.io/8xqy6/ MODIS data are available to download from LP DACC data pool: https://lpdaac.usgs.gov/tools/datapool/ All data that support the findings of this study are included within the article (and any supplementary files).

Acknowledgments
This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). We acknowledge support provided from the NASA Interdisciplinary Science (IDS) and Terrestrial Ecology programs and the Arctic Boreal Vulnerability Experiment (ABoVE).
Copyright 2023, California Institute of Technology. All rights reserved. Government funding acknowledged.

Author contributions
N M implemented the analysis and led in the design and writing of the manuscript. All authors contributed to writing and provided feedback on development of the research.

Code availability
All applied methods and sample codes are freely available in the R software package 'greenbrown' (http:// greenbrown.r-forge.r-project.org/). R codes to reproduce the results and to conduct similar research are available via contacting the corresponding author.

Conflict of interest
The authors declare no competing interests