Introduction

European beech (Fagus sylvatica L.; in the following called beech) is one of the most widespread and dominant broadleaved tree species throughout Europe, with high economic and ecological importance (Leuschner and Ellenberg 2017). Consequently, beech has established itself as one of the most popular study species within the field of dendroscience, especially in light of climate change impact (Leuschner 2020). Despite high research efforts that focus on climate-growth relationships or climate adaptation mechanisms, a comparatively small number of studies investigated one of the main characteristics of this tree species: its pronounced masting behavior.

Generally, the term masting describes the spatially synchronized seed production across wide regions. Here, the magnitude of seed production fluctuates throughout a tree’s lifespan, resulting in years with lower or higher amounts of seeds (Kelly 1994; Pesendorfer et al. 2021). The so-called mast years are years in which exceptional high amounts of seeds are produced (LaMontagne and Boutin 2009), ultimately leading to a shift of leaf-fruit ratio within the canopy.

Most studies show that beech masting is driven by distinct temperature/drought conditions in summer over multiple years (Hacket-Pain et al. 2015; Vacchiano et al. 2017; Mund et al. 2020). More specifically, masting is commonly triggered by a sequence of cold/moist summers 2 years before masting (t−2) and warm/dry summers 1 year before (t−1) and during the mast year (t0). Beside these primary climate controls, other factors may impact the occurrence of masting as well, such as soil moisture (Müller-Haubold et al. 2015) or nutrient supply (Mund et al. 2020). Furthermore, Nussbaumer et al. (2020) highlighted that masting might be also interrupted by extreme summer drought and heat, eventually leading to the abortion of fruits. As a consequence, even if climatic conditions for masting were previously met, trees may not show a distinct masting behavior. While these additional factors (e.g., soil moisture, nutrient supply, and extreme summer drought) were shown to impact masting to certain extents, all three studies reported that climatic drivers, such as summer temperature and/or precipitation, had an overall stronger impact on masting. Consequently, these additional factors may be seen as secondary controls, which have the potential to alter masting occurrence or intensity. Because studies investigating the secondary controls remain scarce, we cannot say for sure to which extent masting is impacted by these parameters. Therefore, it is necessary to investigate masting within the context of multiple environmental drivers and not solely based on climate conditions.

Hacket-Pain et al. (2015) highlighted a strong negative relationship between masting and tree-ring width (in the following called TRW). While masting is primarily and predominantly controlled by summer conditions of the two previous years, t−2 and t−1 (Hacket-Pain et al. 2015; Müller-Haubold et al. 2015; Vacchiano et al. 2017; Mund et al. 2020; Bogdziewicz et al. 2021b), TRW is often reported to be sensitive to summer climate conditions of the current growing year, t0 (Scharnweber et al. 2011; Vacchiano et al. 2017; Harvey et al. 2020). Unfortunately, tree growth and consequently TRW are not impacted by climate alone but are also driven by site-specific soil water availability (Mund et al. 2010; Scharnweber et al. 2013), inter-specific competition/facilitation effects (Houpert et al. 2018; Vanhellemont et al. 2019), and even by masting itself (Hacket-Pain et al. 2017).

Previous masting research mainly focused on the interaction between climate, TRW, and/or masting (Mund et al. 2010, 2020; Hacket-Pain et al. 2015, 2018), spatial patterns in masting (Vacchiano et al. 2017; Bogdziewicz et al. 2021a), or mast-limiting climatic factors (Nussbaumer et al. 2020). Mund et al. (2020) emphasized that the relationship between masting and TRW cannot be considered a simple trade-off; instead it is a complex interaction network of different environmental drivers. Following, the relationship between TRW and masting plays a key role in the understanding of beech growth under climate change and new approaches are needed to investigate this relationship.

Here, the availability of RS data across large areas offers great potential for studying forest ecosystem dynamics under climate change, for example, by assessing the general forest condition (Buras et al. 2021), detecting and quantifying the impact of late frost (Menzel et al. 2015; Bascietto et al. 2018; Olano et al. 2021), and drought events (Shekhar et al. 2020), or by identifying and mapping insect-induced defoliation and consequent tree mortality (Townsend et al. 2012; Senf et al. 2017; Spruce et al. 2019). Rogers et al. (2018) showed that early warning signals of tree mortality are even evident up to 2 decades before tree death. In this study, we aim to disentangle the relationship of TRW and masting in regard to their diverse environmental drivers using RS data.

We assume that masting events can be detected with multi-spectral RS observations and consequently derived products because masting is associated with a reduction in leaf area (Müller-Haubold et al. 2015) and hence should cause a reduction in the absorption of blue and green light as well as a reduction of reflectance in near-infrared. However, as observations within a pixel of a RS image are usually a mixture of the reflectance from tree canopies, adjacent herbaceous vegetation, and soil/background, disentangling the effect of masting in single remote sensing images might be challenging. Nonetheless, small inter-annual differences in leaf development and cover can be commonly retrieved from time series of spectral vegetation indices or of biophysical parameters such as leaf area index (LAI) in studies of land surface phenology (Rodriguez-Galiano et al. 2015; Forkel et al. 2015). For example, Jolly et al. (2022) showed that heavy flowering, as proxy for subsequent masting, was detectable within the canopy of southern beech forests (genus Fuscospora and Lophozonia) using RS data. Many studies around the world set a strong focus on early masting detection and monitoring to enable preventive pest control of seed predators (Elliott and Kemp 2016; Jolly et al. 2022). European masting research, on the other hand, focuses more often on growth-productive traits, such as tree-ring width. Here, Decuyper et al. (2020) reported that TRW of European beech is assessable with remotely sensed enhanced vegetation index data (EVI derived from MODIS satellites) if physical damage occurs to the tree canopy. While the aforementioned study did not incorporate the masting aspect, it nonetheless shows that canopy-impacting characteristics should be detectable with RS data.

Thus, we are convinced that a study incorporating masting, TRW, and RS data will improve our understanding of the complex interaction network of different environmental drivers of masting and TRW. At the same time, this study enables us to assess the potential of remote-sensing approaches to estimate masting. This is especially important because the possibility of acquiring masting data via remote sensing approaches harbors significant merit for masting research by enabling constant monitoring of seed production across large scales. Up to this moment, we are not aware of existing studies investigating masting and tree-ring width using RS data in a combined approach.

With the help of already available time series data of TRW, in-field measured masting, and remotely sensed LAI, we aim to disentangle the relationship between these target parameters. We hypothesize that (I) the target parameters are strongly interrelated, with TRW and LAI negatively correlating to masting, while LAI and TRW positively correlate with each other. Furthermore, secondary controls may impact masting, and thus we expect that (II) masting abundance and intensity increase with higher site-specific water availability. Lastly, we assume that the relationship between the target parameters is predominantly climate-impacted; more specifically, we hypothesize that (III) the three target parameters are driven by summer climate.

Materials and methods

The BDF-network

In this study, we employ a network of 19 beech sites situated along a precipitation gradient in the north-eastern German lowlands (Fig. 1a). This region is characterized by a transition from a more maritime climate in the west to a continental one in the east (659–458 mm annual precipitation). The regularly managed beech sites are part of a permanent observation network of the state forestry service, Landesforst Mecklenburg-Western Pomerania. Soil status (soil type, water availability, and nutrient levels) and tree vitality (crown condition and fructification) are recorded at distinct time intervals (for more details, see Dieckmann 2004).

Fig. 1
figure 1

Map of the BDF-Network and raw data time series. a The remotely sensed leaf area index (LAI) of forest areas in Mecklenburg-Western Pomerania. Shown is the maximum LAI (SPOT-VGT/PROBA-V data) averaged for the months of June–August for the time period 1999–2000. Forest areas are classified with ESA CCI land cover map (ESA 2014), while non-forest areas are masked. The white circles and labels represent the 19 beech study sites. b Annual precipitation sums (columns) and annual mean temperature (line) averaged across all 19 study sites for the time period 1960–2020. The black bars/ red shadings represent standard deviations for precipitation and temperature, respectively. c Abundance of mast events for the time period 1992–2020. Different categories of masting intensity are displayed by gray shading: 0 = no; 1 = low; 2 = medium; 3 = heavy masting. d Mean tree-ring width (TRW) chronology of all 19 study sites (black line) for the time period 1961–2017. Gray shading represents standard deviation. e Remotely sensed LAI series for the time period 2000–2020. Red line/shading represents PROBA-V data across all study sites with respective standard deviations. Gray line/shading shows MODIS data

Full size image

Climate data

Average monthly temperatures and total monthly precipitation sums were derived as 1 × 1 km2 grid data for the time period 1900–2020 from the German Weather Service (DWD Climate Data Center 2020a, b). We extracted the climate data for each study site (Fig. 1b) and calculated the standardized precipitation evapotranspiration index (SPEI6-month) using the R-package SPEI aggregated across a 6-months scale (version 1.7, Beguería and Vicente-Serrano 2017).

Masting data

The information about masting occurrence and intensity was obtained from field observations done by the state forestry service. For the period 1992–2012, masting data was acquired annually. Since 2015, the observations took place every 5 years (Fig. 1c). The masting intensity was visually estimated for each tree individually by 4 categories: 0 = no masting, 1 = low masting, 2 = medium masting, and 3 = heavy masting. Afterwards, the masting data was averaged for each site, resulting in 19 masting time series. In the last step, these averages were assigned to the masting classes 0, 1, 2, and 3 using the thresholds 0–0.5, > 0.5–1.5, > 1.5–2.5, and > 2.5, respectively.

Tree-ring width data

The data on TRW was taken from a previous study focusing on the BDF-Network (Stolz et al. 2021). For each study site, raw tree-ring series were detrended using a fixed cubic smoothing spline with a 50% frequency cut-off at 30 years. Site chronologies were calculated using a bi-weight robust mean (Cook and Kairiukstis 1990), resulting in 19 TRW chronologies with a common overlap period ranging from 1961 to 2017 (Fig. 1d). Calculations were done with the R-package dplR (version 1.7.2, Bunn 2008, 2010) in R (R Core Team 2021).

Leaf area index data

We used two different datasets of remotely sensed LAI with medium spatial resolution to account for differences in temporal coverage, sensor properties, and the applied corrections for atmospheric distortions and clouds. The first dataset employs merged observations from the SPOT-VGT and PROBA-V sensors (in the following called SPOT-VGT/PROBA-V). The SPOT-VGT/PROBA-V LAI (Copernicus Service Information 2017) has a spatial resolution of 1 km with observations at a 10-day interval. The dataset has been already intensively smoothed, corrected for distortions, and gap-filled (Fuster et al. 2020). The Moderate Resolution Imaging Spectroradiometer (MODIS)-LAI (Knyazikhin et al. 1999) has a spatial resolution of 500 m with an observation frequency of 8 days. The original MOD15A2H LAI product comes with quality flags, which we used to exclude all observations that were not flagged as good quality (i.e., observations with dead detectors, un-produced pixels, water, cloud, cloud shadows, or snow cover).

After the exclusion of observations with poor quality as described above, we applied a 3-time-step rolling maximum filter (calculation across three consecutive observations) to each LAI time series to reduce further noise and applied then a smoothing spline to interpolate missing values. Temporal means and maximum values were computed from this smoothed time series. This resulted in the following LAI parameters: mean LAIannual, mean LAIMay-Jul, max. LAIannual, max. LAIApr-May, max. LAIMay–Jul, max. LAIJun–Aug, max. LAISep–Nov. In the end, the maximum LAIJun–Aug processed with a rolling maximum and smoothing spline showed the most abundant and highest significances throughout all study sites. As the literature reports that the summer months are of utmost importance for TRW and masting of European beech, we worked with LAIJun–Aug (in the following called LAI) for the rest of our analyses. This evaluation process resulted in 19 LAI time series for the SPOT-VGT/PROBA-V and MODIS dataset each, ranging from 1999 to 2020 and from 2000 to 2020, respectively (Fig. 1e).

Water gradients

Site-specific water availability may affect the abundance and intensity of masting events. Consequently, we investigated if the mast frequency and intensity are related to the water availability along different water gradients. To account for possible variations in the water availability proxies, we choose three different datasets: (i) a climatic gradient employing the annual precipitation sums for each study site, (ii) in-field measurements of the effective field capacity (= amount of water available for tree growth), and (iii) remotely sensed soil water index (= SWI) at different T-values (10–100) using the ASCAT SWI dataset (Copernicus Service Information 2018). A higher T-value represents stronger time-lagging between surface soil moisture and soil moisture at certain depths and thus describes deeper soil layers. A pre-evaluation showed that the chosen T-value had no drastic effect on the SWI observations (Fig. S1, Supplements), thus we decided to go with the SWI_020 representing soil layers at T-value = 20, which was proven to be a constant value with rather low error rates (Wagner et al. 1999). To keep consistency in data availability across our network, we chose the time period 2001–2011 in which masting data for all 19 study sites was available. The remotely sensed SWI was limited to the time period 2007–2011. We counted the number of occurring mast years of each masting category and present these data in sorted order along the three different water gradients.

Data analyses

To investigate the relationship between our target parameters, we conducted correlation analyses for each parameter against each other. We used Spearman rank correlations for the relationship between TRW against mast and LAI against mast, while Pearson correlation was used for TRW against LAI. Correlation analyses were done for the BDF-network as a whole, as well as for each single study site to account for site-specific relationships.

To explain the resulting parameter correlations and to identify possible common climatic drivers, we conducted climate correlations for each target parameter. Climate correlations were done for a 3-year period, including the current year and the two previous years, because masting is commonly known to be influenced by climate conditions in this time period. The analysis period ranged from June to December 2 years before (t−2), across the entire year before (t−1), and up to September of the current growing year (t0). For consistency, we used this period for all three target parameters. We used the R package Hmisc (version 4.6–0, Harrell Jr. 2021) to correlate matrices of the target parameters (TRW, masting, and LAI) with climate data (temperature, precipitation, and SPEI6-month). Again, we used Pearson’s correlations for the relationship between climate × TRW and climate × LAI, while Spearman’s rank correlation was applied for climate × mast, resulting in climate correlation coefficients for each study site and each month within our analysis period. Both parameter correlations and climate correlations were done for the common data interval 2000–2012, as well as for three prolonged calculation periods distinctive to the 3 parameter combinations: TRW × mast (1992–2012), LAI × mast (2000–2012), TRW × LAI (2000–2017).

To discuss site-specific LAI correlations in more detail and to assess the applicability of the available LAI data, we checked the coverage and homogeneity of beech forests within a 1 km perimeter surrounding our study sites. We used species distribution maps (provided by the state forestry service Landesforst Mecklenburg-Western Pomerania) to extract species composition and coverage (%) for each study site. We chose a 1 km perimeter to ensure that both datasets, the lower resolved SPOT-VGT/PROBA-V LAI as well as the MODIS LAI data, are encompassed in this assessment. Afterward, we correlated beech cover with the correlation strengths of LAI × mast and TRW × LAI to assess the impact of beech coverage.

The visualization and statistics for all figures and analyses were done in R (R Core Team 2021) and RStudio (RStudio Team 2020).

Results

Parameter correlations

To assess the relationship between the three target parameters (TRW, masting, and remotely sensed LAI), we performed correlation analyses across the whole network as well as for individual study sites separately. At first, we analyzed the parameter correlations at the network scale for a data interval common for all three target parameters (2000–2012). The relationships between the target parameters correspond to our expectations, but only when including the LAI derived from SPOT-VGT/PROBA-V sensors (Fig. 2): TRW × mast (Fig. 2a) and LAI × mast (Fig. 2b) were related negatively, while TRW × LAI (Fig. 2c) showed a weak positive correlation. Contrary, MODIS LAI was not significantly correlated to either target parameter (Fig. S2, Supplements). Because the common data interval is rather short, parameter correlations were calculated also across the individual overlapping periods for each set of target parameters to tap the full data range. Thus, we used separate calculation periods for the three-parameter combinations: TRW × mast, LAI × mast, and TRW × LAI. While the parameter correlations with prolonged intervals showed slight significance (Fig. S3, Supplements), the overall response direction did not change compared to the results of the parameter correlations across the common interval 2000–2012 (Fig. 2).

Fig. 2
figure 2

Parameter correlations at network scale. Shown are the correlations for a tree-ring width × masting, b remotely sensed LAI × masting, c TRW × remotely sensed LAI. The LAI correlations originate from SPOT-VGT/PROBA-V data. The boxplots display: median, 25th and 75th percentiles together with the 1.5 interquartile range (whiskers) and outliers. The label above the boxplots and scatterplots show Spearman-correlation and Pearson-correlation coefficients and their p-values, respectively

Full size image

To untangle the relationship between TRW and LAI, we performed the target parameter correlation also for subsets of the data. When the data of the mast years, e.g., masting class = 3, was excluded, the relationship between TRW × LAI turned out to be insignificant (Fig. S4, Supplements). Additional group-wise correlations between TRW and LAI for each masting category supported this finding. Here, the relationship between TRW and LAI was insignificant for masting categories 0, 1, and 2, representing no to moderate masting (data not shown).

The site-scale analyses produced similar results for SPOT-VGT/PROBA-V data (Fig. S5, Supplements) and MODIS data (Fig. S6, Supplements), respectively. Here, most study sites show a trend of decreasing TRW with increasing mast intensity. While SPOT-VGT/ PROBA-V correlations showed a visual trend that is in line with the network-scale analysis, the direction of MODIS correlations differed drastically from site to site. The site-specific results are not as pronounced as the network-scale analysis because the general data range is restricted to the overlap period of TRW and mast data (2000–2017), with nmast ≤ 15 limiting the statistical power for site-specific analyses.

Water gradient

We found a higher number of non-mast events in comparison to low, medium, and high mast for all three water gradients (Fig. 3). Apparently, there was no visible trend, neither positive nor negative, in the number of mast events along the water gradients. Furthermore, mast events of the same intensity were equally distributed across all study sites, irrespective of water availability.

Fig. 3
figure 3

Site-specific masting abundance in relation to three different water gradients: a climatic precipitation gradient, b field measurements of effective field capacity, and c remotely sensed soil water index (SWI) at T-value 20. Single study sites (n = 19) are sorted with increasing water availability gradient at the x-axis, while the abundance of masting years is shown on the y-axis. The data time period ranges between 2001 and 2011 for a and b and between 2007 and 2011 for c. Different masting categories are decoded by gray shading: 0 = no; 1 = low; 2 = medium; 3 = heavy masting. Water gradients are ordered from left to right with increasing water availability

Full size image

Climate correlations

For TRW, we found significant positive correlations with SPEI in current June and with temperature in May one year before (t−1). Additionally, there were significant negative correlations with precipitation in October in t−2 (Fig. 4a).

Fig. 4
figure 4

Climate correlations for all three target parameters: a tree-ring width, b masting, and c SPOT-VGT/PROBA-V LAI. The climate correlations were calculated over the common data interval of 2000–2012. The panel columns display the correlation values for June to December 2 years before (t−2), for January to December 1 year before (t−1), and for January to September of the current growing year (t0). Depending on the input variable, values represent Pearson’s r (a and c) or Spearman’s rho correlation coefficients. The climate parameter used for the analyses are sorted from top to bottom: mean monthly temperature, monthly precipitation sum, and 6-month SPEI. Single rows display values for individual study sites (n = 19). Positive/negative correlation values are represented by red/blue color, respectively. Asterisks highlight significant correlations

Full size image

The climate correlations for masting showed the most pronounced signals of all three target parameters (Fig. 4b). For the summer month of July 2 years before (t−2), we see significant negative/positive correlations for temperature/precipitation and SPEI, respectively. This pattern is inverted in t−1 with significant positive correlations for temperature and negative correlations for precipitation and SPEI. The climate correlations in t0 are not as distinct as in the previous years. Still, there exists some significant negative/positive correlation for temperature/precipitation and SPEI, respectively. Furthermore, we found significant positive correlations for temperature in April of the current year.

The climate correlations for remotely sensed LAI are patchy for both SPOT-VGT/PROBA-V (Fig. 4c) and MODIS LAI (Fig. S7, Supplements). For SPOT-VGT/PROBA-V, we found some significant correlations for summer temperature in July and negative correlations of SPEI in t−2. Nonetheless, the correlations seem site-specific. Here, some study sites react either stronger to SPEI in t−1 or t−2, while others showed no significance at all. In comparison, we see no pattern in the climate correlation of MODIS data at all.

Impact of beech cover

While we found no significant impact of beech cover on the correlation strength between LAI and the other target parameters, tendencies of decreasing and increasing correlation strength for LAI × mast and TRW × LAI can be observed, respectively (Fig. 5). Contrastingly, for MODIS data, we see this trend only for LAI × mast (Fig. S8, Supplements).

Fig. 5
figure 5

Impact of beech coverage on the parameter correlation strength of SPOT-VGT/PROBA-V data. Here, plots display the strength of parameter correlations for each study site (n = 19) in relation to beech coverage (%) for a LAI × mast and b TRW × LAI. The labels above the scatterplots show Pearson correlation coefficients and their p-values

Full size image

Discussion

Complex relationships between TRW and masting are driven by summer climate

Throughout the literature, TRW of beech is reported to be impacted by a multitude of different climate factors ranging from temperature in spring (Mund et al. 2010, 2020) and October (Drobyshev et al. 2010) up to SPEI during the summer period of the current year (Hacket-Pain et al. 2015, 2017). Additionally, Hacket-Pain et al. (2015) showed that the temperature in July during the two years preceding the current growth year may be of similar importance. Generally, the most pronounced climate correlations are reported for the summer period May–July, in which beech growth mainly occurs (van der Maaten et al. 2018). Similarly, our study identified SPEIJUN in the current year, followed by May temperature in t−1, as main climatic drivers for TRW.

While the climate signal of TRW may vary to certain extents, masting is reported to be driven by a distinct climate pattern. Here, Mund et al. (2010) reported that masting is triggered by a sequence of moist and cool summer conditions in t−2, followed by warm and dry conditions in t−1. In agreement with the literature, we found strong climate signals in June for temperature, precipitation, and SPEI throughout the whole analysis period. The high correlation coefficients, especially within the 2 years before the current growing year, mark the strong climate dependence of masting in European beech. The pronounced climatic pattern of cold summer conditions in t−2 and warm summer conditions in the following year, t−1, might be interpreted as evidence of the ∆T-hypothesis. Here, Kelly et al. (2013) state that mast behavior is promoted by the temperature difference of the 2 years prior to masting. Nonetheless, underlying mechanisms driving masting are still partially understood, and other non-climatic factors might control masting behavior, too. For example, masting could be driven by precipitation recharge and consequent higher nitrogen uptake (Geßler et al. 2005), which was previously proven to facilitate masting in Fagus crenata (Han et al. 2014). Furthermore, higher soil water availability positively impacts fine root abundance (Müller-Haubold et al. 2015), which eventually leads to increased nutrient uptake. While we found strong and uniform correlations between masting and summer precipitation in t−1 and t−2 which support our assumption that water availability is of high importance for masting, we found no tendencies of increasing masting abundance nor intensity along the water availability gradient. One possible explanation for these contradicting findings may be that the water availability gradient within our network is plainly not distinct enough. Here, our 19 study sites range between almost optimal conditions for all three water availability proxies: precipitation (575–745 mm), effective field capacity (130–381 mm), and RS soil water index (54–63%), thus reacting similarly across the complete spatial scale. A second possible explanation might be that the found pattern in July precipitation in t−2 and t−1 could be a cofounded impact of solar radiation. Müller-Haubold et al. (2015) emphasized that the solar radiation in June and July in t−1 was the main impact factor of masting. Thus, an underlying solar radiation signal below our precipitation signal might explain the significant values within the climate correlations while the findings across the water gradients remain insignificant. Following, we cannot agree to our second research hypothesis that masting abundance and intensity are increasing with higher site-specific water availability. Here, further investigations with stronger differences in site-specific water availability and quantitative measure of masting intensity may be needed to provide conclusive answers.

Many studies on beech masting report a strong negative interrelationship between TRW and masting (Drobyshev et al. 2010; Hacket-Pain et al. 2015, 2017; Mund et al. 2020). Still, interpretations differ between either a straight trade-off of wood production and fruit biomass (Drobyshev et al. 2010; Müller-Haubold et al. 2015) or complex system of sink and source limitations (Mund et al. 2020). For example, Hacket-Pain et al. (2015) highlighted that TRW may be either impacted directly by summer climate or indirectly by masting, which is driven by summer climate, suggesting an interacting control of climate and masting on TRW (Hacket-Pain et al. 2017). In our study, we report a strong negative correlation between TRW and masting. The analyses with remotely sensed LAI data support the assumption of a complex climate-driven interplay of TRW and masting.

Remotely sensed LAI is not suitable for tree-ring width estimation

In this study, we showed that TRW and remotely sensed LAI are only slightly correlated to each other when considering the individual data interval (Fig. S3, Supplements). To disentangle the relationship between TRW and LAI in more detail, a subsequent investigation for each individual masting category highlighted that only TRW during heavy masting (category 3) exhibited this weak positive relationship (Fig. S4, Supplements). This means that TRW is not assessable with RS LAI during years of no, low, or medium masting intensity. These findings are in line with Decuyper et al. (2020), who highlighted that TRW is only assessable with RS data (EVI from MODIS satellites) if physical damage occurs to the tree canopy directly. Additionally, the climate-parameter correlations (Fig. 4) suggest that TRW and LAI are driven in different manners. While TRW was found to be largely impacted by summer climate in the current growing season, LAI showed no distinct climate correlation pattern. Following and in line with Decuyper et al. (2020), we highlight that more discrete climatic effects that cause no physical damage to the tree canopy are hard to disentangle. Regarding our initial hypotheses, we can say that TRW and LAI are not related to each other (hypothesis I) and that this relationship is not commonly driven by summer climate (hypothesis III).

Mast detection via remote sensing—is LAI an appropriate proxy?

Most studies on masting detection employ spectral or vegetation indices, such as EVI and NDVI (Fernández-Martínez et al. 2015; Garcia et al. 2021; Jolly et al. 2022), as surrogates of photosynthetic active leaf area. But spectral indices may introduce a certain uncertainty when investigating masting. For example, Wang et al. (2005) reported a strong linear relationship between LAI and NDVI during the period of leaf unfolding and senescence but found no clear relationship during periods of closed canopy, in which masting would mainly occur. Consequently, vegetation indices such as NDVI may not be appropriate for masting identification. Under the assumption that “heavy flowering leads to eventual masting,” studies on RS masting detection also often concentrate on time periods prior to masting, thus enhancing possibilities for early seed predator pest control (Jolly et al. 2022). Still, Nussbaumer et al. (2020) showed that masting could be canceled by secondary controls, i.e., extreme summer heat and drought, even shortly before masting itself. Consequently, we highlight the necessity to directly detect masting during the time it occurs and, from this, derive methods to forecast masting to enable management adjustments as early as possible. In our study, we showed that masting is slightly correlated with SPOT-VGT/PROBA V LAI when regarding a sufficiently long data period (Fig. S3, Supplements). Following, we state that remotely sensed LAI might bear the potential to detect and assess masting.

While RS LAI might be an appropriate proxy for masting, one question remains: Is the LAI representative for our study sites? We found clear but weak relationships between our target parameters when including the SPOT-VGT/PROBA-V LAI. The findings incorporating MODIS data revealed some uncertainties. This poses the question: what may be the difference between SPOT-VGT/PROBA-V and MODIS LAI? Initially, we assumed that the finer-resolved MODIS LAI would represent our data better at stand level because of less background noise from the surrounding area. To investigate this, we used forest coverage maps with a perimeter of 1 km surrounding our study sites, completely encompassing the extent of SPOT-VGT/PROBA-V (1 km) and MODIS (500 m) LAI data. Surprisingly, we found that beech coverage seems to have a greater impact on correlation strength for the coarser resolved SPOT-VGT/PROBA-V data than for the finer-resolved MODIS data. Following, differences between both RS datasets are not driven by spatial resolution alone. The better performance of the lower resolved SPOT-VGT/PROBA-V data could be explained by a higher quality of temporal smoothing and gap filling. Here, validation studies show that the SPOT-VGT/PROBA-V data performs better than MODIS LAI and fAPAR products (Camacho et al. 2013; Fuster et al. 2020; Nestola et al. 2017) or other LAI data at higher spatial resolutions (Brown et al. 2020). Thus, patterns between tree growth and climate might be stronger at larger scales (or at coarser spatial resolution) because site-specific factors are reduced. Similar findings were reported by Zhao et al. (2020), who found that PROBA-V LAI at 1 km resolution performed better than MODIS LAI at 500 m for study sites in tropical and subtropical evergreen forests in China. Nonetheless, they also highlighted that the generally reported uncertainties of RS products might be substantially underestimated (Zhao et al. 2020).

In that sense, we conclude that SPOT-VGT/PROBA-V LAI may be a suitable proxy for masting assessment and bears promising potential for masting assessment in the future, but the findings should be evaluated in light of possible measurement uncertainties accompanying RS data in general.

Limitations of masting detection using RS data

While current ready-to-use SPOT-VGT/PROBA-V and MODIS LAI products are convenient, they also hold some limitations. Here, the short temporal range of RS LAI time series often does not match the longer TRW and masting time series. Consequently, low observation numbers decrease the statistical power of our analyses, especially when sub-setting to individual site scales. Additionally, the coarse spatial resolution (500 m/1 km) of RS LAI presents a challenge for our study sites, which extend only up to 5000 m2. A possible solution would be to use combined high-resolution time series from Landsat and Sentinel-2 observations that are becoming more frequently used to study phenological events (Bolton et al. 2020; Kowalski et al. 2020). Garcia et al. (2021) showed that it is possible to detect masting in white spruce with Landsat data by applying moisture-oriented vegetation indices, which even perform better than photosynthesis and color-oriented indices. However, the short temporal overlap of combined Landsat/Sentinel-2 observations (since 2016) and the low temporal frequency of older Landsat observations still hamper investigations of long-term climate-tree growth relationships. Ongoing developments produce longer and continuous LAI time series by applying temporal high-resolution meteorological data (Zhu et al. 2020) or by combining MODIS LAI (Moreno-Martínez et al. 2020) with Landsat imagery. While many of these methods are new or still in development, our focus was on ready-to-use RS-LAI products that are easy to use for forest ecologists who want to tap into the vast research field of remote sensing.

Next to resolution-related issues, mast-induced changes in the tree canopy are often hard to distinguish from other impacts, such as defoliation by late frost, drought, or pests (McDowell et al. 2015; Buras et al. 2021). Thus, adding environmental information for RS approaches, i.e., climate station data, is highly recommended (Buras et al. 2021). McDowell et al. (2015) showed that these limitations could be partly offset by identifying distinct spectral signatures in canopy change. For example, bark beetle-caused mortality is characterized by a pattern of needle discoloration and consecutive needle loss (Wulder et al. 2008). Following, knowledge about canopy changes throughout time is highly valuable, and RS masting detection would certainly improve by such a spectral signature approach in the future.

Lastly, when working with RS LAI, we should not limit our considerations to technological restrictions alone but keep emphasis on ecological and physiological aspects as well. Within our study, we assumed that masting is theoretically similarly detectable with ground-based measurements and RS approaches. For white spruce, Garcia et al. (2021) reported that cone production during non-mast years occurs within the top one-third of the crown, whereas cones are more widely distributed throughout the canopy during mast years. What would be, if beech trees produce fruits at lower canopy levels while maintaining higher leaf biomass at the canopy surface to optimize photosynthetic gain? Following this train of thought, masting would be better detectable from ground observations, while RS measurements may detect the green canopy from up above only, thus underestimating masting. Horticultural studies on the vertical distribution of fruits within the canopy showed that fruit abundance and quality are changing within canopy layers (Farina et al. 2005; Feng et al. 2014). But, similar studies on forestry tree species are mainly missing from the literature. Consequently, phenological studies investigating the fruit distribution within the tree canopy may improve our understanding of masting ecology in forestry tree species, and we highlight the need to conduct such investigations in the future.

Mast in context of climate change

In the context of global warming, the climate report of Mecklenburg-Western Pomerania forecasts an increase in average temperature up to 3.7 °C by 2100 (DWD 2018). This might lead to prolonged vegetation periods, but also to concomitant extreme events and increasing risks for late frost damage. And this may have especial implications for European beech because late frost events may lead to defoliation up to nearly 100%, causing strong reductions in growth (Menzel et al. 2015; Principe et al. 2017). While this impact on growth was evident in southern Germany for a late frost event in 2011, our data indicates no effects of late frost within our study region overall. On the contrary, our data shows that the year 2011 is one of the stronger mast years within our data set (Fig. 1c), i.e., with a medium masting intensity across all sites (Table S1, Supplements). Whereas our study sites under more maritime and thus warmer climate might be protected to a certain extent against late frost events, the trees in southern Germany are exposed to a more continental climate. Nonetheless, global warming may have especial implications on beech populations at low elevations because an earlier start of the vegetation period increases the risk of late-frost damage (Bascietto et al. 2018). This is in line with our findings that displayed significant positive temperature correlations of masting during April of the current year (Fig. 4b). Alternatively, this correlation may be interpreted as a pollination signal with warm conditions in spring that may promote seed production through effective pollen flow (cf. studies on other species such as oak, Caignard et al. 2017; Lebourgeois et al. 2018).

In this study, we employed a space-for-time approach to infer information about masting under potential future climate conditions. But against our hypothesis, we found no such trend of increasing masting abundance or intensity along the gradient. In the first paragraph of the discussion, we highlighted two possible explanations for the missing trend: (a) a too-narrow climate gradient and (b) an underlying solar radiation signal possibly underlying the found precipitation signal (Fig. 4b). In line with Müller-Haubold et al. (2015), our study could not give evidence that water availability might impact masting behavior. Thus, forecasted change in precipitation regime (DWD 2018) might not impact masting behavior as strongly as other drivers. Here, temperature could play a more pronounced role. Recent studies showed that climate warming affects mean seed productivity, interannual variability, and synchrony (Bogdziewicz et al. 2021a, b). Furthermore, Nussbaumer et al. (2020) showed that mast can be canceled by extreme summer climate conditions due to early fruit abortion. Albeit these mast-limiting factors can be considered secondary control mechanisms, already observed and forecasted increases in climatic extreme events (DWD 2018) emphasize the growing risk of climate change.

Following, climate change impacts masting in a complex way with interactions between mast triggering and post-mast limiting factors (Bogdziewicz et al. 2021b). We state that RS masting detection based on pre-masting periods may be not as reliable, thus new or improved RS approaches are needed to detect and assess masting accordingly.

Conclusion

This study assessed the potential to estimate masting events and TRW with remotely sensed LAI. We used time series of TRW, categorical masting data, SPOT-VGT/PROBA-V, and MODIS LAI and investigated their relationship. Analyses including MODIS LAI did not yield any significant relationships, while partially significant correlations between our target parameters were found when using SPOT-VGT/PROBA-V LAI. This observation highlights that relying on single RS data sources may have great implications for study results. To decipher the causes of differences between RS products, devoted studies would be needed. While masting and TRW are commonly driven by summer climate, remotely sensed LAI shows only a few site-specific climate signals but is mostly decoupled from climate. Furthermore, it seems that masting is not driven by site-specific water availability. All in all, we can say that remotely sensed SPOT-VGT/PROBA V LAI bears the potential to observe masting but not TRW. Nonetheless, these findings should always be discussed in regard to general uncertainties of RS measurements and the complex interaction network of different environmental drivers of masting and TRW.