Delineating Forest Canopy Phenology: Insights from Long-Term Phenocam Observations in North America

Abstract

This study utilized the North American PhenoCam network to evaluate phenological characteristics and their relationships with geographic and climatic factors across deciduous broadleaf (n = 39) and evergreen needleleaf (n = 13) forests over the past decade. Using high temporal resolution near-surface imagery, key phenological indicators including the start, end, and length of growing season were derived and analyzed using linear regression and structural equation modeling. The results revealed substantial spatial variation; the evergreen needleleaf sites exhibited earlier starts to the growing season (112 vs. 130 Julian date), later ends to the growing season (286 vs. 264 Julian date), and longer lengths for the growing season (172 vs. 131 days) compared with the deciduous broadleaf sites. Latitude was significantly related to the start of the growing season and the length of the growing season at the deciduous broadleaf sites (R² = 0.28–0.41, p < 0.01), while these relationships were weaker at the evergreen needleleaf sites, and elevation had mixed effects. The mean annual temperature strongly influenced the phenology for both forest types (R² = 0.18–0.76, p < 0.01), whereas longitude, distance to the coast, and precipitation had negligible effects. Temporal trends in the phenological indicators were sporadic across both the deciduous broadleaf and evergreen needleleaf sites. Structural equation modeling revealed distinct causal pathways for each forest type, highlighting complex interactions among the geographical and climatic variables. At the deciduous broadleaf sites, geographical factors (latitude, elevation, and distance to the nearest coast) predominated the mean annual temperature, which in turn significantly affected phenological development (χ² = 2.171, p = 0.975). At the evergreen needleleaf sites, geographical variables had more complex effects on the climatic factors, start of the growing season, and end of the growing season, with the end of the growing season emerging as the primary determinant of growing season length (χ² = 0.486, p = 0.784). The PhenoCam network provides valuable fine-scale phenological dynamics, offering great insights for forest management, biodiversity conservation, and understanding carbon cycling under climate change.

1. Introduction

Bioclimatic law is often used to illustrate how environmental gradients such as latitude, longitude, or altitude influence the distribution or growth cycles of plants and animals, serving as an important empirical rule for understanding spatiotemporal patterns in ecology. Hopkins [1] observed phenological events, namely the periods of mass growth or decline of pests in North America, and discovered a consistent delay of approximately four days for every one-degree increase in latitude northward, five-degree movement in longitude westward, or 400-foot increase in altitude. This empirical rule was used to propose guidelines for planting wheat to avoid pest infestations and improve yields in the subsequent years [1]. Increases in latitude are generally associated with reduced solar radiation and lower average temperatures, resulting in delayed onset of spring and later snowmelt, which is more pronounced in high-latitude regions compared with mid- and low-latitude regions. Such delays significantly influence the availability of food sources of birds, including insects, plant seeds, and other resources, which in turn affect bird migration patterns and breeding success [2,3,4]. Longitude influences plant phenology through a combination of geographic factors. For example, warm ocean currents in southeastern areas and cold currents in northeastern areas contrast with the drier inland climates of the west, which are shaped by the distance from the ocean and mountain barriers, leading to variations in phenological events [5]. Altitude effects are complex, with high-altitude areas presenting significant adaptive challenges for plants and animals due to low temperatures, intense radiation, and shorter growing seasons. Vitasse et al. [6] observed that high-altitude plants exhibited delayed budburst and greater sensitivity to climate change compared with those at lower elevations in the European Alps.

However, application of the bioclimatic law framework faces challenges. Firstly, species show considerable variability in their responses to geographic gradients. For instance, evergreen conifer forest ecosystems exhibit high phenological variability driven by the complex interplay between temperature and precipitation, often deviating from the predictions of this law [7]. Additionally, extreme climatic events such as droughts and storms can further disrupt the relationship between vegetation phenology and geographic gradients. Polgar and Primack [8] found that while rising temperatures generally lead to earlier plant green-up, extreme temperature fluctuations can cause some species to deviate from expected latitudinal trends in growth patterns. Land use changes, including deforestation, urbanization, and agriculture, significantly alter native species distributions and phenological patterns [9]. Urban heat islands, for example, advance spring phenology in cities, contrasting sharply with the surrounding rural areas [3]. Although field-based phenological observations have been conducted for over a century, their spatiotemporal coverage remains limited due to inconsistencies among observers and restricted monitoring at larger scales [10,11]. Phenology is highly sensitive to both short-term weather fluctuations and long-term climate changes, affecting the seasonality of numerous ecosystem processes [12,13,14] and profoundly influencing ecosystem functioning and management strategies [15,16]. Understanding and adapting to climate change is one of the greatest challenges of the 21st century, with one of its most evident impacts on terrestrial ecosystems being the alteration of phenology, namely the timing of seasonal activities in plants and animals.

Advances in satellite remote sensing technology, such as the Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat imagery, have greatly enhanced the monitoring of vegetation growth dynamics and land surface phenology at regional and global scales over the past decades [17,18]. Schwartz and Reiter [3] reported that the long-term normalized difference vegetation index (NDVI) data revealed a notable advancement in spring green-up across the mid-to-high-latitude regions of the Northern Hemisphere strongly linked to rising global temperatures. Regional climate drivers such as precipitation and radiation also play crucial roles in shaping vegetation dynamics in arid and tropical regions [5,19]. Phenological indicators derived from vegetation index time series, such as the start (SOS), end (EOS), and length (LOS) of the growing season, are closely tied to seasonal climatic conditions and have cascading effects on ecosystem hydrological and biogeochemical cycles, which are emerging as critical issues [16,20,21,22,23,24]. Atmospheric interference from frequent cloud cover and difficulty in monitoring specific tree species pose limitations for large-scale remote sensing, which often relies on mixed pixels that represent entire landscapes. In regions with complex terrain, such as mountainous areas and tropical or subtropical forests, subtle phenological differences among tree species are frequently overlooked, despite their importance in assessing climate change impacts [25]. Ground-based observations play a vital role in addressing these limitations, improving the accuracy and ecological relevance of remote sensing data. The integration of multi-scale observation methods, particularly near-surface remote sensing techniques like digital cameras or PhenoCam, which are less affected by cloud interference, can effectively reveal the relationship between geographic environmental gradients and phenological changes [26,27].

PhenoCam, which uses high-frequency image assessments and calculates the green chromatic coordinate (GCC) from red, green, and blue intensity values, has proven highly effective for monitoring phenological changes across various plant functional types [28,29,30]. A study on budburst phenology in boreal coniferous forests found that the GCC predicted budburst with greater accuracy than other vegetation indices (R² > 0.90) [31]. Similarly, the GCC provided more precise predictions of photosynthetic phenological phases in temperate evergreen and deciduous forest canopies compared with the green-red vegetation index [32]. Toomey et al. [33] demonstrated that the SOS and EOS of evergreen needleleaf forest are closely related to temperature and precipitation changes, achieving higher precision than MODIS imagery. Furthermore, PhenoCam time series effectively capture fine-scale canopy changes in high-latitude regions, such as an earlier SOS driven by spring warming, which are often missed in satellite observation [34]. The continuous accumulation of phenological data provides a critical foundation for analyzing seasonal dynamics across diverse vegetation types, enhancing our understanding of the relationship between vegetation phenology and climate change. Global observation networks are progressively being established, including the PhenoCam Network in North America (https://phenocam.nau.edu/webcam/ (accessed on 20 December 2024), Phenological Eyes Network in Asia (Japan, http://www.pheno-eye.org/ (accessed on 20 December 2024) [25]), the European Phenology Camera Network (https://fmiprot.fmi.fi/index.php?page=EUROPHEN/ (accessed on 20 August 2024)), and the Australian Phenocam Network (https://ecoimages.tern.org.au/ (accessed on 20 December 2024)). Long-term PhenoCam observations at fixed locations are invaluable for investigating phenological trends and variability at both the organism and canopy scales, providing insights that support land use and ecosystem management [31,35,36].

Using data from the North American PhenoCam monitoring network, observational records from 2000 to 2023 have been compiled into PhenoCam Dataset V2.0 and updates in V3.0 (https://phenocam.nau.edu/phenocam_explorer/ (accessed on 20 December 2024)). Among these, 52 stations representing evergreen needleleaf (EN; n = 13) and deciduous broadleaf (DB; n = 39) forest sites have been identified (Figure 1), with over a decade of observations, making them the most extensively and consistently monitored vegetation types to date [37]. First, we hypothesize that the timing or duration of phenological stages (SOS, EOS, and LOS) for EN and DB forests will show distinct latitudinal and climatic gradients, with greater sensitivity in deciduous broadleaf forests. Second, we expect long-term trends to reveal earlier SOSs and delayed EOSs, particularly at higher latitudes, reflecting regional responses to recent climate change. Third, we hypothesize that geographic variables (e.g., latitude, elevation, longitude, and distance to the coast) and climatic factors (e.g., temperature and precipitation) exert differential influences on the phenological metrics between EN and DB forests, mediated by forest type and regional context. Therefore, this study utilizes the continuous data of EN and DB sites to (1) examine phenological patterns (SOS, EOS, and LOS) across latitudes and climatic conditions, (2) investigate the phenological trends over the past decade (10–18 years) across the regions, and (3) evaluate how geographic conditions (e.g., elevation, longitude, and distance to the coast) and climate factors (e.g., temperature and precipitation) affect phenological changes in and trends for EN and DB forests, using phenological metrics derived from both the GCC and structural equation modeling (SEM). To the best of our knowledge, this study is the first to analyze long-term trends using data from the North American PhenoCam monitoring network to disentangle complex interactions among geographical, climatic, and vegetation factors influencing forest phenology. These findings advance our understanding of forest ecosystem responses to climate and inform effective management strategies.

2. Materials and Methods

2.1. Study Sites in PhenoCam Network and Datasets Acquired

The PhenoCam network, established in 2008, focuses primarily on monitoring terrestrial ecosystems across North America and has grown to include hundreds of camera sites contributed by participating researchers (referring to Seyednasrollah et al. [37]). Data are captured as daily images and processed using standardized methods, including the calculation of red, green, and blue digital numbers (DNs) and the GCC. These datasets are accessible via a user-friendly visualization interface (https://phenocam.nau.edu/phenocam_explorer/ (accessed on 20 December 2024)), enabling real-time viewing of greenness trends and phenological metrics across various vegetation types. The forthcoming PhenoCam V3.0 dataset is expected to include additional auxiliary data, such as meteorological variables (e.g., temperature and precipitation) and soil moisture measurements, while also expanding the spatiotemporal coverage of station data. These improvements will further facilitate analysis of long-term phenological dynamics across latitudes [26,38]. Observations are centered on sites with at least 10 years of data, with an initial selection of 52 stations meeting these criteria: 39 representing DB forests and 13 representing EN forests (Figure 1 and Table S1). Basic site information, including geographical variables (altitude [m], latitude [^o], longitude [^o], and distance to the nearest coast [km]) and climatic parameters (mean annual temperature [MAT; °C yr⁻¹] and precipitation [MAP; mm yr⁻¹]), was obtained from the network or, in the case of the coastal distance, analyzed using the Proximity tool in ArcGIS Pro.

2.2. Calculation of PhenoCam Green Chromatic Coordinate (GCC)

The PhenoCam time series image data for 52 sites were obtained from PhenoCam Dataset V2.0 and V3.0. For each image, a suitable “region of interest” (ROI) was identified based on the dominant vegetation type (Figure 2). Within the ROI, the DNs of red, green, and blue pixels were analyzed. The green chromatic coordinate (GCC), which measures the relative intensity of the green DN, has shown strong correlations with commonly used vegetation indices such as the NDVI and GPP [39,40,41]. It is calculated as follows:

$$GCC={\displaystyle \frac{G_{DN}}{R_{DN} +{ G}_{DN } + { B}_{DN}}}$$

(1)

where R_DN, R_DN, and G_DN, represent the DN values of the red, green, and blue reflectance intensities, respectively. To minimize the effects of lighting conditions, atmospheric interference, and weather fluctuations, this study selected the 90th percentile of GCC values for analysis [7,34,42,43].

2.3. Phenological Indicators Derived from PhenoCam GCC

After obtaining the GCC time series for each site, the data were smoothed using a Savitzky–Golay filter, which effectively preserves temporal vegetation growth dynamics while minimizing atmospheric contamination, particularly by filtering out low-value outliers and has been widely utilized across various regions [44,45,46]. This approach is also integrated into the processing of standard MODIS phenology datasets [47]. The smoothing function is expressed as follows:

$$y\left(t\right) = {\displaystyle \frac{c}{1 + e^{a + bt}}} + d$$

(2)

where t represents the day of the year (DOY), y(t) is the daily GCC at time t, a and b are fitting parameters related to the rate of change in the time series, c is the maximum value of y(t), and d denotes the initial background value of the GCC. We also have

$$C_{CR}=b^{3}cz\left\{{\displaystyle \frac{{3z(1-z)(1+z)}^3\left[{2(1+z)}^3+\right.\left.b^2{c}^{2}z\right]}{{\left[{(1+z)}^4+\right.\left.{\left(bcz\right)}^2\right]}^{\raisebox{1ex}{$5$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}}}\right.-\left. {\displaystyle \frac{{\left(1+z\right)}^2(1+2z-{5z}^2)}{{\left[{(1+z)}^4+\right.\left.{\left(bcz\right)}^2\right]}^{\raisebox{1ex}{$3$}\!\left/ \!\raisebox{-1ex}{$2$}\right.}}}\right\}$$

(3)

Once the data curve was fitted, phenological time points could be extracted using the curvature change rate (CCR), which identifies the timing of phenological events. The SOS (start) and EOS (end) of the growing season, expressed as the Julian date, were defined as the point at which vegetation indices rose or declined to 10% of their annual amplitude, as determined by a fitted function between the seasonal minimum and maximum values (Figure 2; [48,49,50]). We used the CCR (the second derivative of the fitted time series) to identify these phenological indicators. Specifically, z = e^a+bt, and the SOS and EOS correspond to the points in the time series where the CCR reached the local maxima and minima, respectively [49]. The length of the growing season (LOS, expressed in days) was then calculated as the difference between the EOS and SOS. The smoothing and calculation of GCC time series phenological indicators were performed using the TIMESAT analysis 3.3 tool ([16,48]; Lund University, Lund, Sweden; https://web.nateko.lu.se/timesat/timesat.asp, accessed on 20 December 2024).

2.4. Statistical Analysis

We used linear and nonlinear regression models to explore the spatiotemporal relationships between phenological metrics (SOS, EOS, and LOS) and climatic (temperature and precipitation) as well as geographic variables (longitude, latitude, and altitude). Phenological indicators were treated as response variables, while predictors included annual averages of the temperature and precipitation, latitude, longitude, and altitude, all widely recognized as significant factors influencing plant phenology [7,12]. The optimal model results would be identified based on the highest coefficient of determination (R²) [50].

Structural equation modeling (SEM) is a powerful statistical approach for disentangling complex relationships among multiple variables by assessing both direct and indirect effects within a system [51,52]. Unlike conventional regression analyses, which evaluate one relationship individually, SEM enables multiple pathways to be tested simultaneously, effectively capturing the intricate interdependencies among variables. This makes SEM particularly valuable in ecological and hydrological studies, where numerous interacting processes influences outcomes [53,54,55]. In this study, SEM was used to examine how geographical variables, including latitude, longitude, elevation, and distance to the nearest coast, affect the MAT and MAP and how these climate factors, in turn, influence vegetation phenology, including the SOS, EOS, and LOS. SEM enabled us to quantify the direct effects of geographical gradients on climatic variation and the indirect effects of climate changes on phenological variability. By structuring multiple regression equations within a cohesive analytical framework, SEM enables a single variable to function simultaneously as both a dependent and independent variable across different equations. This approach clearly partitions direct effects (e.g., geographic variable influences on climate variability) and indirect effects (e.g., climatic influences mediated by geographical variables) on phenological dynamics while also identifying multiple pathways and the relative strengths by which one variable affects another [51]. Model performance was evaluated using standard fit indices: the chi-squared (χ²) test, root mean square error (RMSE), and goodness-of-fit index (GFI). A well-fitting model is indicated by a no-significance chi-squared test (p value > 0.05), a low RMSE (<0.05), and a high GFI (>0.95) [56,57]. The model was iteratively refined using the “modindices” function in the lavaan package in R, removing weak or unsupported paths and optimizing model fit based on the Akaike information criterion (AIC), which balances goodness of fit with model simplicity [58,59,60]. Importantly, non-significant or missing paths should not be interpreted as evidence of no relationship but rather as indicators of data limitations or high variability within the system. Nevertheless, SEM is valuable for disentangling causal pathways in complex environmental systems. It allows researchers to test theoretical frameworks, uncover feedback mechanisms, and identify key drivers of change across ecological and hydrological processes. However, it is important to note that SEM identifies statistical associations consistent with hypothesized causal structures but does not establish definitive causation [53,61]. Therefore, interpretation should be guided by both model outcomes and sound ecological reasoning within the appropriate context.

3. Results

3.1. Forest Phenological Characteristics Along Geographical and Climatic Gradients

The mean SOS and EOS at the DB sites were 92 (±5.0) and 286 (±7.8), respectively, at the lowest latitude and 151 (±6.3) and 257 (±8.2), respectively, at the highest latitude. Over the last decade, the SOS at the EN sites ranged from 126 (±38.3) to 131 (±5.8), and the EOS ranged from 251 (±3.7) to 279 (±8.9) along the latitudinal pattern. Consequently, the range in the SOS and EOS was 60 days for the DB sites, and 30 and 55 days for the EN sites, respectively (Figure 3a,b). As a result, the LOS at the DB sites extended from 86 to 194 days, and at the EN sites, it ranged from 121 to 197 days across the latitudes during this period (Figure 3c). A positive and significant relationship between the latitude and SOS was observed for the DB sites (R² = 0.41, p < 0.001), with a linear regression slope of 2.25, but not for the EN sites (R² = 0.09, p = 0.319; Figure 3a). In contrast, a weak negative relationship between the latitude and EOS was found for both the DB and EN sites (R² = 0.09–0.23, p = 0.062–0.101; Figure 3b). The LOS, calculated as the difference between the EOS and SOS, showed a significant negative correlation with the latitude for the DB sites (R² = 0.28, p = 0.001), with a linear regression slope of −3.19. No significant correlation was observed for the EN sites (R² = 0.20, p = 0.123; Figure 3c). At the DB sites, each one-degree increase in latitude was associated with a 2.3-day delay in the SOS and a 3.2-day shortening of the LOS (Figure 3a). Over the past decade, the average SOS at the EN sites was earlier (day 112) than at the DB sites (day 130), while the EOS occurred later at the EN sites (day 286) compared with the DB sites (day 264), resulting in a longer LOS at the EN sites (172 days) than at the DB sites (131 days) (Figure 3j–l).

Elevational changes revealed a significant effect only on the SOS for the EN sites (R² = 0.48, p = 0.008), the EOS for the DB sites (R² = 0.11, p = 0.043), and the LOS for the DB sites (R² = 0.17, p = 0.009), with all having nonlinear relationships (Figure 3d–f). However, the MAT appeared to have a more pronounced effect on the phenological shifts, SOS (R² = 0.32–0.76, p < 0.001), EOS (R² = 0.18–0.33, p < 0.01), and LOS (R² = 0.37–0.54, p < 0.05) for both the DB and EN sites (Figure 3g–i). The negative correlation between the MAT and SOS was stronger for the DB sites, with a linear regression slope of −3.72, compared with −1.94 for the EN sites (Figure 3g). In contrast, the positive correlation between the MAT and EOS was weaker for the DB sites (slope = 1.95) than for the EN sites (slope = 2.75) (Figure 3h). Consequently, the positive linear regression slope for the LOS was similar for both the DB sites (5.33) and EN sites (5.02) (Figure 3i). However, no significant relationships were found between the longitude, distance to the nearest coast, MAP, and phenological indicators for either the DB or EN sites over the past decade (R² = 0.007–0.25, p > 0.05) (Figure 4a–i). Therefore, the overall phenological patterns partially supported the first hypothesis, as the phenological metrics (SOS, EOS, and LOS) of the evergreen coniferous forests characterized by an earlier SOS, later EOS, and longer LOS were distinct from those of the broad deciduous forests. However, the results did not support the hypothesis that these three phenological metrics showed consistent relationships with the MAT; instead, the associations with the latitude, elevation, longitude, distance to the coast, and MAP were mixed or not significant.

3.2. Phenological Metric Interrelationships and Temporal Trends Across Sites

Regarding the interrelationships among phenological metrics at both the DB and EN sites, significant negative correlations were observed between the SOS and EOS (R² = 0.11–0.56, p < 0.05) and between the SOS and LOS (R² = 0.54–0.62, p < 0.001; Figure 5). Additionally, strong positive correlations were found between the EOS and LOS for both forest types, with a stronger association at the EN sites (R² = 0.85, p < 0.001) than the DB sites (R² = 0.72, p < 0.001; Figure 5).

Over the past decade, significant trends in the SOS were detected at 9 of the 52 sites, all within DB forests; two showed earlier spring onset, while seven showed delayed springs (Figure 6a). Similarly, significant trends in the EOS were found only at the DB sites (7 of 52), with five sites exhibiting delayed autumns and two showing advanced autumns (Figure 6b). For the LOS, significant temporal trends were observed at six sites: five at DB sites (three with lengthened LOSs and two with shortened LOSs) and one for the EN forests (shortened LOS (Figure 6c)). Only 5 of the 52 sites presented significant trends in at least two metrics simultaneously, indicating no consistent patterns for the SOS, EOS, or LOS across all DB and EN sites (Figure 6). For example, the DB site at Monture (northwest America) experienced an earlier SOS trend (9.6 days per decade) with no significant change in the EOS, resulting in a longer LOS (19 days per decade) between 2002 and 2018 (Figure 6d). The dispersed and mostly non-significant outcomes of long-term phenological trends refuted the second hypothesis, indicating a lack of widespread and consistent phenological response to current climate conditions.

3.3. Interrelations Among Geographical and Climatic Factors and Forest Growth

Based on the SEM analysis, where blue single arrows indicate positive effects and red single arrows indicate negative effects, the geographical variables (latitude, elevation, and distance to the nearest coast) had negative effects on both the MAT and MAP. Furthermore, the MAT was weakly negatively correlated with the MAP at the DB sites (Figure 7a). The MAT, in turn, had a negative effect on the SOS and a positive effect on the EOS such that an earlier SOS resulted in a longer LOS, and a delayed EOS also extended the LOS (Figure 7a). However, the MAP did not significantly influence the phenological development, SOS, EOS, or LOS over time (Figure 7a).

In the EN forests, SEM analysis indicated that latitude and elevation directly decreased the MAT, while a greater distance to the coast increased the MAT. In turn, a higher MAT led to an earlier SOS. These geographical variables did not affect the MAP (Figure 7b). Unlike at the DB sites, latitude directly shortened the EOS at the EN sites, and elevation both delayed the SOS and advanced the EOS. Increased distance from the coast also delayed the SOS. In addition, the MAP had a negative effect on the EOS, with a higher latitude, elevation, and MAP all contributing to a more advanced EOS (Figure 7b). Finally, among the phenological metrics, the SOS did not affect the LOS, while only a delayed EOS resulted in an extended LOS at the EN sites (Figure 7b). The results supported the third hypothesis that geographical and climatic variables exerted different effects on the DB and EN sites, as indicated by the SEM analysis. In the DB forests, geographical variables primarily influenced the MAT, which in turn played a dominant role in determining vegetation development (SOS, EOS, and LOS). In contrast, the EN forests exhibited a more complex association between geographical and climatic factors, which diminished both the direct and indirect contributions of these variables on vegetation phenology.

4. Discussion

4.1. Phenology Patterns Between Evergreen Needleleaf and Deciduous Broadleaf Forests

Analysis of the phenological metrics (SOS, EOS, and LOS) derived from time series PhenoCam images revealed significant differences between the deciduous broadleaf (DB) forests and evergreen needleleaf (EN) forests. Specifically, the EN sites exhibited earlier SOSs (112 vs. 130 Julian date), later EOSs (264 vs. 286 Julian date), and longer LOSs (172 vs. 131 days) compared with the DB sites over the past few decades. Additionally, the three metrics for the DB forests displayed more pronounced site-to-site variability than those for the EN sites (Figure 3j–l). Guo et al. [32] compared the photosynthetic phenology between evergreen pines (Cedrus deodara) and deciduous tree species (Quercus variabilis, Pistacia chinensis, and Gleditsia sinensis) in the Qinling Mountains of China (33–34°N). Their findings also indicated that the LOS of evergreen needleleaf species was approximately 40 days longer than that of the three deciduous broadleaf species. At the Turkey Point Flux Station in southern Ontario, Canada (42°N), the LOS of EN trees was observed to be approximately 150 days longer than that of DB trees [62]. The phenological differences between DB and EN forests tend to be pronounced with increasing latitudes, consistent with common knowledge. However, the divergent responses of DB and EN forests to ongoing climate warming remain insufficiently understood, particularly due to the scarcity of paired long-term observations needed to validate this hypothesis.

The interrelations among the SOS, EOS, and LOS, as assessed using near-surface remote sensing, revealed stronger and more consistent associations between the EOS and LOS (R² = 0.72–0.85) than those observed for the SOS and EOS (R² = 0.11–0.56) or SOS and LOS (R² = 0.54–0.62; Figure 5). These results suggest that the EOS may play a dominant role in determining the length of the growing season for both the DB and EN sites over the past decade. Field observations indicated that the extension of the growing season in temperate regions was primarily driven by earlier leaf unfolding in spring as a response to climate warming [63,64,65]. However, more recent studies highlighted that a delayed end of the growing season now contributes more to extension of the growing season than the earlier onset of spring greening does [66,67]. This shift was primarily attributed to ongoing increases in late-season temperatures and the delayed onset of autumnal frost events, which facilitate prolonged photosynthetic activity and leaf retention in temperate ecosystems [68,69,70]. Notably, autumn phenology is particularly sensitive to warming-induced changes at temperature extremes, allowing plants to extend their growing periods by postponing leaf senescence and abscission, thereby maintaining ecosystem productivity later into the growing season [68,71]. In contrast, in tropical and subtropical regions, the timing of spring greening or the SOS primarily drives the annual variation in the length of the growing season [16,18]. Therefore, the impacts of climatic variability on plant phenology and ecosystems require continued attention to develop a more comprehensive understanding amid intensifying climate warming.

4.2. Interactions Among Geographical and Climatic Factors Regarding Forest Growth

Over the last decade, only a few DB and EN sites exhibited significant trends for an advanced SOS, delayed EOS, or extended LOS, resulting in scattered patterns in temporal phenological trends (Figure 6). Substantial evidence indicates that plant phenology in mid-to-high-latitude regions is more sensitive to climate warming than in low-latitude regions, primarily due to the faster rate of warming in temperate and polar regions compared with tropical and subtropical areas [72,73,74]. For example, northern high-latitude regions (>60°N) warmed by approximately 1.36 °C per century from 1875 to 2008, nearly double the Northern Hemisphere average of 0.79 °C per century [73]. The disparity has grown in recent decades, with Arctic warming accelerating to roughly 1.35 °C per decade, underscoring pronounced polar amplification driven by mechanisms beyond ice–albedo feedback [73]. A multi-site study across sub-Arctic alpine and Arctic tundra regions (60–79°N) from 1992 to 2014 found that the timing of plant leaf emergence and flowering was significantly more sensitive to rising summer temperatures at colder, high-latitude sites compared with warmer locations [74]. In contrast, tropical and subtropical (low-latitude) regions have experienced smaller warming trends (approximately 0.2–0.3 °C per decade) in recent decades, resulting in less pronounced shifts in vegetation phenology compared with high-latitude areas [75,76]. Global satellite observations indicate that only 11–12% of vegetative surfaces, mainly at higher latitudes, have shown significantly earlier starts or longer lengths for growing seasons in recent decades [77]. In many tropical ecosystems, phenological shifts have been minimal, and they are often influenced by changes in seasonal irradiance and rainfall patterns rather than by temperature alone, reflecting the weaker warming effects at low latitudes [78,79,80].

Longer monitoring is essential for accurately tracking these dynamics. In cross-site phenological analysis, strong relationships between the SOS and MAT, as well as the LOS and MAT, particularly at DB sites, demonstrate the patterns observed between the SOS and latitude (Figure 3a–c). Nevertheless, the steeper regression slopes observed for the DB sites (absolute value: 3.7–5.3) compared with the EN sites (absolute value: 1.9–5.0) suggest that DB forests are more sensitive to temperature changes across regions (Figure 3g–i), further reflecting the influence of latitudinal gradients. Leaf-out in deciduous broadleaf forests is closely linked to spring accumulated warmth following adequate winter chilling, making these forests highly sensitive to temperature changes and prone to earlier budburst as the climate warms [81]. In contrast, evergreen needleleaf species primarily respond to photoperiod cues and retain their foliage year round, enabling continuous photosynthesis even under snow-covered conditions, leading to a more conservative phenological pattern that is less responsive to temperature fluctuations [82]. Moreover, variations in phenological indicators with the latitude and temperature are primarily driven by differences in solar radiation, as both annual radiation and the mean annual temperature decrease with increasing latitudes [83,84]. The weaker associations between the EOS and MAT suggest that EOS responses to temperature may be nonlinear [83,85]. While some studies reported delayed EOSs under current warming conditions [86,87], others found evidence for earlier EOS trends [70,88]. At higher latitudes, reduced solar radiation accelerates the accumulation of abscisic acid, decreases the chlorophyll content, and hastens leaf senescence [89], ultimately leading to an earlier EOS and a consequent shorter LOS. Mechanistically, in deciduous broadleaf forests, shorter autumn days act as photoperiod cues that trigger hormonal changes, particularly abscisic acid-induced chlorophyll degradation, leading to earlier senescence and limiting the extension of the growing season under warming conditions [90]. In contrast, evergreen needleleaf forests retain their foliage and sustain photosynthesis during warmer autumn periods, allowing for a delayed end and extended length of the growing season as long as environmental conditions remain favorable [91].

Nemani et al. [92] analyzed global satellite-based estimates of annual net primary productivity (NPP) from 1982 to 1999 and found that tropical regions and high latitudes accounted for 80% of the observed increase, primarily driven by reduced cloud cover and increased solar radiation. However, a subsequent analysis covering the period from 2000 to 2009 revealed a contrasting trend, namely a global reduction in NPP by 0.55 Pg of carbon, attributed to drying effects from continued warming and increased evapotranspiration [93]. As warming persists, its effect on plant phenology is expected to diminish, with co-limiting factors becoming increasingly influential [14,15,94]. For example, in nutrient-poor ecosystems, shifts in plant phenology are driven by soil nutrient availability rather than by temperature, whereas in humid forest systems, light availability becomes the primary limiting factor [94,95,96,97]. These interactions may collectively explain the observed weakening of plant phenological responses over time. Further investigation is needed to quantify the relative importance of these factors and their integrated impacts on ecosystem dynamics.

Although no consistent significant relationships were found between the SOS, EOS, LOS, and latitude (Figure 3), this suggests that vegetation growth may be only partially explained by temperature variations across latitudes and altitudes [83,98], as well as forest adaptation mechanisms [99]. Elevation also shaped forest phenology, with several studies, including our findings, reporting earlier SOSs and later EOSs at lower elevations and shorter LOSs at higher elevations [7,100,101]. This pattern is likely due to colder air and soil temperatures at high altitudes, which constrain tree growth rates [102]. Precipitation, which typically decreases with increasing latitudes or longitudes (from coastal to inland areas), can mitigate water deficit-induced leaf senescence and help sustain the temperature-driven responses in autumn phenology [12,83,103]. Additionally, seasonal rainfall played a key role in driving phenological divergence between the DB and EN forests. DB forests, which shed their leaves to avoid hydraulic stress during dry seasons, depend heavily on rainfall timing to initiate leaf-out and senescence [104,105]. Conversely, EN forests adopt more conservative water-use strategies, such as deep root systems and strict stomatal regulation, supporting sustained leaf function and reducing their phenological sensitivity to precipitation fluctuations [106]. In the seasonal dry tropical forests (SFTFs) of the Caatinga region in northeast Brazil, characterized by high temperatures, prolonged dry periods (>6 months), and an MAP < 720 mm, the LOS and NPP are positively correlated with precipitation [107]. However, in our study, most DB and EN sites were relatively humid (MAP > 1000 mm), and the MAP did not significantly influence the three main phenological stages. Globally, water stress has surpassed cold stress as the primary growth-limiting factor for forests, owing to changes in precipitation regimes, shifts in annual or seasonal rainfall redistribution, and changes in the proportion of rainfall versus snow, all under a changing climate [108].

In contrast to macroclimate influences such as temperature, precipitation, and solar radiation, which drive photosynthesis, the local terrain and canopy structure can filter radiation and buffer against temperature extremes, wind exposure, and reduced evapotranspiration [70,109,110,111]. For example, dense litter layers and extensive root networks beneath a mature forest canopy help retain soil moisture and stabilize microclimates, thereby affecting nutrient cycling, decomposition, and overall productivity [112]. Although these mechanisms were not included in the present analysis, they significantly mediate spatial variations in phenological development [12]. Incorporating the interactions among the canopy structure, microclimate, and phenological events is essential for accurately understanding forest ecosystem feedback and biogeochemical cycling in response to climate change [113,114].

4.3. Insights of Application of PhenoCam Networks

This study primarily utilized near-surface PhenoCam data to characterize fine-scale forest phenological patterns, providing high-frequency observations that revealed the nuanced timing of vegetation change beyond the reach of traditional ground monitoring. Based upon continuous PhenoCam imagery, key seasonal transitions, such as spring greening and autumn browning, were precisely tracked, and their timing was compared across diverse sites and regions. This enabled detecting rapid phenophase shifts that surpass satellites, which are limited by less frequent revisiting intervals. This approach is especially valuable in regions with frequent cloud cover or vegetation lacking distinct seasonality, and it helps avoid the mixed pixel issue common in satellite data [42,115]. In addition, PhenoCams can capture subtle phenological cues, including canopy coloration changes, that are often missed by satellites sensors [42]. Our results revealed pronounced spatial variation in phenology among deciduous broadleaf forests, where lower-latitude or warmer sites experienced earlier spring greening compared with cooler, higher-latitude locations, demonstrating that PhenoCam data can effectively capture geographically divergent patterns in seasonal development (Figure 3, Figure 4, and Figure 7). In contrast, the evergreen needleleaf sites exhibited more variable and subtle phenological changes throughout the year, reflecting complex interactions between the geographical and climatic factors that shape their seasonal dynamics (Figure 3, Figure 4, and Figure 7) [26]. These PhenoCam-based observations yield valuable insights into the stages of canopy development, which are closely tied to CO₂ exchange, reflecting seasonal patterns of carbon uptake and release [116]. In tropical forests with complex structures, PhenoCam indices have been shown to serve as the most reliable proxy for capturing seasonal fluctuations in productivity, outperforming traditional climatic variables [115]. By filling this critical observational gap, PhenoCams enhance our ability to disentangle the influence of leaf phenology on tropical ecosystem productivity.

Relying solely on PhenoCam networks also has limitations, as each camera covers a limited area and requires maintenance, and thus unrepresentative site coverage can occur, and PhenoCam observations alone cannot easily be scaled up to landscape or global levels [117]. In order to build a more comprehensive understanding of phenology, integrating PhenoCam data with on-the-ground observations or surveys for species-specific phenophases and remotely sensed datasets to provide regional-to-global coverage will provide a fuller picture of vegetation dynamics across scales [12]. Broader implications of these phenological insights are crucial for understanding climate change impacts. For example, shifts in phenology for earlier SOSs and longer LOSs have been observed, which significantly affect a forest ecosystem’s function and biodiversity by causing mismatches between plant development and wildlife feeding or breeding cycles [118,119]. For instance, warming-driven shifts in the timing of leaf emergence and fall (phenology) extend the growing season and thereby influence carbon sequestration. Recent studies showed that longer growing seasons generally enhance carbon uptake in temperate forests [120,121], even as other research cautions that heightened autumn respiration can offset some of these gains [122,123]. Furthermore, tracking site-specific phenological phases associated with seasonal climatic extremes, such as seasonal drought, wind storms, or cold snaps, can enhance our understanding of forest ecosystem responses and resilience across broad regions. From a management perspective, monitoring changes in the SOS and EOS is essential for optimizing planting and harvesting schedules and anticipating periods of elevated wildfire risk due to drying leaf litter [124]. Consistent phenology observations from the PhenoCam network capture fine-scale variations in forest phenology and help refine model projections of ecosystem processes under environmental change [112].

5. Conclusions

This study utilized the North American PhenoCam network to evaluate forest phenological patterns and their associations with geographic and climatic variables across deciduous broadleaf (n = 39) and evergreen needleleaf (n = 13) forests over the past decade. High temporal resolution near-surface imagery was used to derive key phenological indicators, namely the start, end, and length of the growing season, which were then analyzed using linear regression and structural equation modeling. The results revealed significant spatial variation, with evergreen needleleaf sites exhibiting an earlier start to the growing season (112 vs. 130 Julian date), later end to the growing season (286 vs. 264 Julian date), and longer length of the growing season (172 vs. 131 days) compared with deciduous broadleaf sites. Latitude strongly influenced the start and length of the growing season at the deciduous broadleaf sites (R² = 0.28–0.41, p < 0.01), whereas its effect was weaker and less consistent for the evergreen needleleaf sites. Elevation demonstrated mixed effects across both forest types. The mean annual temperature emerged as the primary driver influencing phenology for both forest types (R² = 0.18–0.76, p < 0.01), whereas precipitation, longitude, and distance to the coast had negligible impacts. Temporal phenological trends were sporadic across the sites. Structural equation modeling analysis highlighted distinct causal pathways. For the deciduous broadleaf forests, geographical factors primarily influenced the mean annual temperature, which in turn significantly affected phenological development (χ² = 2.171, p = 0.975), whereas for the evergreen needleleaf forests, more complex interactions emerged, with the end of the growing season predominantly determining the length of the growing season (χ² = 0.486, p = 0.784).

Future research should address spatial and temporal data gaps by expanding the coverage of the PhenoCam network, particularly in understudied regions. Incorporating additional environmental factors, such as soil moisture dynamics, nutrient availability, and species-specific physiological traits, could further elucidate the mechanisms driving phenological variations among forest types. Additionally, long-term, multi-platform monitoring that integrates PhenoCam data with broader-scale remote sensing technologies and field surveys would offer deeper insights into ecosystem dynamics. Such integrated approaches will enhance predictive modeling of phenological responses under future climate scenarios, assisting policymakers and forest managers in developing adaptive management strategies. Enhanced monitoring can improve forecasts of forest productivity, guide biodiversity conservation, and support assessments of climate-induced wildfire risks, thereby mitigating ecological and socioeconomic impacts associated with climate change.