Uncovering the role of solar radiation and water stress factors in constraining decadal intra-site spring phenology variability in diverse ecosystems across the Northern Hemisphere

Abstract

Introduction

Spring phenology in the Northern Hemisphere marks the onset of leaf development and plays a critical role in regulating various terrestrial surface biophysical and biochemical processes. These processes include changes in land surface albedo, land-atmosphere carbon and water exchanges, forest productivity, and nutrient cycling (Fang et al., 2020; Gerst et al., 2020; Huang et al., 2023). Additionally, spring phenology influences numerous biotic interactions, such as intra- and interspecies competition for resources and trophic interactions with other living organisms (Cohen & Satterfield, 2020). Furthermore, vegetation-mediated climate feedback is impacted by spring phenology, as it can lead to earlier soil water depletion (Lian et al., 2020) and an increased risk of summer droughts (Vitasse et al., 2021; Li et al., 2023). Despite the importance of spring phenology in ecological and Earth surface processes, our understanding of the drivers behind its variability across large vegetated landscapes and extended periods remains incomplete, creating considerable amounts of uncertainty when estimating how future climate change will affect spring phenology and other related biological processes (Geng et al., 2020; Xie & Wilson, 2020; Adams et al., 2021).

To better understand and represent the mechanisms underlying plant spring phenology in response to climate change, researchers have developed numerous prognostic models, such as growing degree day (GDD) models, sequential models, and parallel models (McMaster & Wilhelm, 1997; Melaas et al., 2016; Zhao et al., 2021). These models incorporate key environmental indicators, such as temperature and photoperiod, to predict the leaf unfolding data (LUD) and other phenological timing (Chuine et al., 2000, 2013). The majority of these models attribute phenological shifts to chilling, that is, the exposure of plants to cold temperatures to break dormancy, forcing, which involves exposure to warm temperatures, and the photoperiod effect to promote growth (Heide, 2003; Schwartz et al., 2006). In addition to these models that only consider temperature and photoperiod, researchers recently have developed the eco-evolutionary optimality (OPT) theory and associated OPT-based spring phenology model as a more comprehensive and innovative hypothesis for spring phenology modeling (Fu et al., 2020; Wang et al., 2020b; Meng et al., 2021). This theory posits that the LUD in plants results from trade-offs aimed at maximizing photosynthetic carbon gain and minimizing frost risk. This hypothesis was supported by Gu et al. (2023), who analyzed data from hundreds of temperate ecosystem sites in the north and east of the United States, highlighting the significant, yet overlooked role of solar radiation (SR) in having a greater impact on early-season plant photosynthesis potential than photoperiod. These prognostic models have significantly enhanced our ability to quantify phenology shifts in response to climate change, therefore improving the assessments of how these shifts impact various ecosystem processes and functions, such as carbon, water, and nutrient cycles (Nord & Lynch, 2009; Buermann et al., 2018; Lian et al., 2020; Zhou et al., 2022), as well as related climate change feedback (Richardson et al., 2013; Shen et al., 2015).

Despite these advancements in our understanding and modeling of spring phenology, key knowledge gaps remain. First, most prior large-scale assessments of prognostic phenology models have primarily concentrated on a narrow range of environmental variables, such as temperature and photoperiod (Singh et al., 2017; Richardson et al., 2018b; Meng et al., 2021), neglecting other influential environmental variables. For instance, variables such as soil moisture (SM) and vapor pressure deficit (VPD), which affect plant water availability and transpiration, have demonstrated an impact on the timing of leaf unfolding and other phenological events (Li et al., 2021; Zhang et al., 2021), yet receive inadequate attention in current models. Similarly, SR, which drives plant photosynthesis and growth, also impacts phenology (Z. Ren et al., 2022; Gu et al., 2023). This omission makes the mechanisms and modeling underlying large-scale spring phenology incomplete. Second, recent efforts focusing on more environmental variables were often confined to a select number of sites within the United States (e.g. Gu et al., 2023). It remains unknown whether these findings can be extended to large, spatially continuous landscapes over the entire Northern Hemisphere. Furthermore, several empirical studies have reported various environmental variables on water stress conditions, such as precipitation (P) (Ganjurjav et al., 2020), SM (Tao et al., 2020), and VPD (Grossiord et al., 2020). These variables can influence spring phenology in temperate ecosystems, and their relative importance often varies with different plant functional types (PFTs) (Nemani, 2003; Caldararu et al., 2016). This implies a more complex regulation mechanism of spring phenology in temperate ecosystems than that which is included in state-of-the-art prognostic models. In other words, it remains unclear whether the variables represented in current models are comprehensive and if the dominant environmental variables would vary with different PFTs (Hmimina et al., 2013; Ji et al., 2021). Should such variability across PFTs exist, identifying the overarching mechanisms that account for the diversity in the phenology–environmental cue relationship across broader landscapes becomes crucial and warrants immediate exploration.

Several recent technical advances offer a unique and timely opportunity to bridge the aforementioned knowledge gaps: the availability of long time-series phenology datasets covering the entire Northern Hemisphere (Friedl et al., 2019; Zhang et al., 2020) and the recently developed diagnostic approach for exploring the potential impact of comprehensive environmental variables on spring phenology modeling (Gu et al., 2023). One typical example of this long-term phenology dataset is the LUD extracted from the Global Inventory Modeling and Mapping Studies (GIMMS) NDVI3g dataset (C. Wu et al., 2022); this dataset is developed based on algorithms retrieving the phenology transition dates, providing us not only extensive spatial coverage (i.e. the entire Northern Hemisphere) but also covering a long time duration (i.e. 1982–2015). Concurrently, Gu et al. (2023) developed a holistic analytical framework that not only examined the performance of existing prognostic models for spring phenology modeling but also established a quantitative pipeline that assessed the partial correlation between the model residuals and other environmental variables not yet included in the current models, by which the effectiveness and the most sensitive variable would be identified. As a result, we expect that the integration of these two advances will provide relevant quantitative results across the entire Northern Hemisphere, helping to address the key knowledge gaps identified earlier.

This study thus aims to assess and improve the existing model of spring phenology (e.g. the OPT model introduced in Meng et al., 2021) for intrasite decadal spring phenology modeling across the entire Northern Hemisphere by exploring other variables beyond temperature and photoperiod. Specifically, we address three key questions: (1) How does the OPT model perform in simulating spring phenology on a hemispheric scale? (2) What are the key environmental variables responsible for intrasite model residuals, and whether these variables differ across PFTs? (3) How significantly can the new prognostic models enhance spring phenology modeling? To answer these questions, we used satellite land surface phenology products and environmental datasets over the Northern Hemisphere for the years 1982–2015. We trained and evaluated the existing OPT model developed by Meng et al. (2021), which has demonstrated superior model performance compared with other prognostic models of GDD models and sequential models (Gu et al., 2023). Given that the hemisphere-scale spring phenology dataset has a coarse spatial resolution of 1/12° and long-term data often come with significant climate anomalies, we further investigated whether landscape PFT heterogeneity and climate anomalies contribute to the model residuals. Subsequently, the relationships between the model residuals and climate variables were examined, including SR and three water stress factors: P, SM, and VPD. This analysis aimed to identify additional environmental factors, not accounted for in current models, that have a substantial impact on the variability of spring phenology, and to determine whether the dominant environmental variable varies across PFTs. Last, we revised the OPT model by incorporating the identified dominant environmental variable(s) and investigated whether these revisions could enhance the model's performance.