Impacts of climate change on reproductive phenology in tropical rainforests of Southeast Asia

The study investigates how climate change affects community-wide mass flowering and fruiting (general flowering) in less-seasonal Southeast Asian tropical forests. General flowering involves irregular, multi-year synchronous events across many species, prominently Dipterocarpaceae. Proposed environmental drivers include drought associated with ENSO, high solar radiation under cloud-free conditions, and drops in minimum night-time temperature; endogenous nutrient status (notably phosphorus) may modulate readiness but is unlikely to drive long (>2-year) intervals. With warming temperatures and more variable rainfall projected for Southeast Asia, and a paucity of long-term phenology records and mechanistic models for the tropics, the study aims to quantify past trends and predict future reproductive phenology across species using a long-term arboretum dataset from Peninsular Malaysia and a process-based phenology model linking low-temperature and drought cues to flowering.

General flowering in Southeast Asian dipterocarp forests is a well-documented but incompletely understood phenomenon, with events spanning large spatial scales and irregular intervals. Prior work implicates drought (often ENSO-linked), increased solar radiation, and transient cool temperature drops as proximate cues, while internal nutrient dynamics (especially P) can constrain or enable flowering after mast events. Mechanistic modeling at Pasoh, Malaysia, and other sites has shown that synergistic effects of cool temperatures and drought can trigger florogenesis in Dipterocarpaceae, and a model using cumulative cool units (CU) and drought units (DU) successfully predicted species-specific flowering. Broader climate change literature shows phenological advances in temperate/boreal biomes with warming, and hypotheses suggest tropical species, evolved under low seasonality, may be highly sensitive to climatic shifts. However, quantitative assessments of tropical tree reproductive phenology responses to future climate have been scarce due to limited long-term datasets and predictive frameworks.

Data: Monthly presence/absence of flowering and fruiting were compiled for trees in the FRIM arboretums (near Kuala Lumpur; 3°24′N, 101°63′E; elevation ~80–97 m) from April 1976 to September 2010 (417 months). Observations were made monthly by trained staff using binoculars, recording flowering (bud to anthesis) and fruiting (immature to ripe) status. Initial dataset: 112 dipterocarps and 240 non-dipterocarps. Species filtering applied five criteria: (1) ≤50% missing values; (2) stable flowering period (CV of flowering duration < 1.0); (3) flowering period ≤ 12 months; (4) no drastic shifts in flowering/fruiting frequency between halves of the record (to remove unreliable series); (5) removal of overlapping species, herbs, and unknown species. Final dataset: 95 Dipterocarpaceae and 115 non-Dipterocarpaceae (210 total). Phenology seasonality was assessed via monthly histograms for flowering, fruiting, and inferred seed dispersal (month when fruiting switched from 1 to 0). Climate: Daily minimum temperature and precipitation from the FRIM Kepong station were used (1 Mar 1973–31 Mar 1996 and 23 Jul 1997–20 Apr 2005). Missing runs >5 days were removed; short gaps (<3 days) were infilled using adjacent 3-day means. Solar radiation was unavailable but precipitation served as a proxy (negatively correlated regionally). Trend analyses used Mann–Kendall tests and linear regressions on anomalies smoothed with 30-day and 12-month moving averages. Modeling flowering drivers: A process-based model quantified cumulative environmental cues during a pre-induction window of n1 days and a development lag of n2 days. Cool unit (CU) at time t: CU(t) = max{C − x(t), 0} (threshold C, daily minimum temperature x(t)); Drought unit (DU) at t: DU(t|θ) = max{Σ_{t−n1}^{t} y(t) − n1·D, 0} with rainfall y(t) and threshold D. Two logistic models were fit with presence/absence of first flowering per month as dependent variable: (i) DU-only (drought-induced flowering); (ii) CU × DU (multiplicative cool- and drought-induced flowering). Model fitting focused on Dipterocarpaceae due to fewer missing values. Time-series clustering (hierarchical, DTW distance; R package TSclust) grouped the 95 dipterocarps by flowering phenology. The optimal number of clusters was chosen by minimizing AIC via forward selection over assignments of DU vs CU×DU per cluster. Training period: May 1976–Mar 1996; validation: Jul 1997–Apr 2005 (gap in climate data in between). Clusters with fewer than five species and three independent species were excluded from forecasting; six clusters remained. Model performance was assessed by AUC for training and validation. Future projections: Bias-corrected daily minimum temperature and precipitation at 0.5° resolution from ISIMIP (GFDL-ESM2M, IPSL-CM5A-LR, MIROC5) were used for historical (1 May 1976–31 Mar 1996) and future (1 Jan 2050–31 Dec 2099) periods under RCP2.6 and RCP8.5. Additional bias correction adjusted precipitation variance seasonality using a gamma-distribution-based quantile mapping preserving variance ratios between GCM and observed monthly deviations. For FRIM, when site location was not at grid center, a distance-weighted average of four surrounding grid cells was used. Flowering probabilities were predicted monthly per cluster for historical and future periods; future values were normalized to historical baselines. Regional generalization: The calibrated cluster models were applied to three additional Southeast Asian sites (Trang, Thailand; Lambir Hills, Malaysia; central Kalimantan, Indonesia) using corresponding GCM grid data to compare seasonal flowering distributions historically and under future scenarios. Statistical environment: R 3.6.3.

- Community trends: Proportions of flowering and fruiting species decreased from the mid-1970s to early 2000s (flowering: P=0.0021, MK test, two-sided, n=400; fruiting: P<0.0001, MK test, two-sided, n=400). Over the same period, minimum temperature increased at 0.39 ± 0.02 °C per decade (P<0.0001, n=10,330) and precipitation increased by 0.51 ± 0.26 mm/day per decade (P=0.027, n=12,668). - Synchrony and mass events: Fruiting fraction lagged flowering by 2 months (time-lagged cross-correlation = 0.77). Largest flowering/fruiting in 1985 (>35% of species). Six mass flowering events (>20% species) over 35 years, matched general flowering in natural forests. Dipterocarpaceae showed higher between-species synchrony than non-Dipterocarpaceae (P<0.0001, ANOVA; n=4465–6555). Coefficients of variation for flowering/fruiting proportions were roughly twice as large in Dipterocarpaceae (flowering CV 1.787; fruiting CV 1.583) versus non-Dipterocarpaceae (0.803; 0.753). - Seasonality and frequencies: Despite interspecific variation (17% of species flowered at least annually; 25% only once per 10 years), community-scale bimodal flowering peaks occurred in April and October, followed by fruiting peaks in June and December, and seed dispersal peaks in February and August. Moraceae (e.g., Ficus) flowered/fruited nearly year-round; many other families showed spring or bimodal peaks. - Modeling drivers and cluster structure: Time-series clustering of 95 dipterocarps yielded 10 phenological clusters (7 clusters + 3 independent species), with six clusters (≥5 spp.) used for forecasting. The CU×DU model best explained clusters 3 and 4 (27 and 28 species, respectively), indicating synergistic cool temperature and drought cues; other clusters were DU-only (drought-only sensitivity). Model discrimination was acceptable for most clusters (AUC training 0.64–0.78; validation 0.62–0.79), with lower reliability for cluster 1 (AUC 0.64 training; 0.62 validation). Overall, 57% of dipterocarp species responded to both drought and low-temperature cues. - Future projections (FRIM): Relative to 1976–1996, minimum temperature increases by 2050–2099 were 1.2 ± 1.1 °C (RCP2.6) and 3.1 ± 1.7 °C (RCP8.5). Flowering probabilities for clusters 3 and 4 decreased markedly due to reduced occurrences of low-temperature cues: under RCP2.6, to 57% and 49% of baseline; under RCP8.5, to 37% and 28% of baseline. Under RCP8.5, low-temperature triggers rarely occurred or disappeared for these clusters. Clusters sensitive only to drought showed little change because drought-related cues remained relatively stable. - Regional extrapolation: Applying models to Trang (Thailand), Lambir (Borneo), and central Kalimantan showed consistent declines only in clusters 3 and 4, with other clusters robust. Historical seasonal flowering patterns reflected a latitudinal gradient (spring peak in Trang; bimodal/weak bimodal in FRIM and Lambir; spring [Sep] peak in southern hemisphere central Kalimantan), and these seasonal distributions were predicted to be robust under future scenarios, indicating quantitative (probability) rather than qualitative (seasonal timing) changes.

The analyses link observed declines in community flowering and fruiting proportions to warming temperatures and reduced availability of low-temperature cues, demonstrating that many dipterocarps require a synergistic combination of transient cool temperatures and drought for floral induction. As regional warming progresses, the cool-temperature component becomes rarer, reducing flowering opportunities for the substantial subset of species dependent on both cues, while drought-only species maintain stable flowering probabilities. This heterogeneity in cue dependence implies differential reproductive success across species under climate change, potentially altering regeneration dynamics and community composition. Compared to temperate systems where phenological timing often shifts earlier with warming, tropical Southeast Asian trees appear to exhibit primarily quantitative reductions in flowering probability without major seasonal pattern changes, consistent with rainfall-driven seasonality persisting in the region. The findings underscore the potential heightened vulnerability of tropical species—evolved under low seasonal variability—to modest warming, and stress the importance of mechanistic, cue-based models for forecasting biotic responses and guiding conservation and restoration strategies in dipterocarp-dominated forests.

This work integrates a rare, multi-decade, multi-species phenology record with a mechanistic cue-accumulation model to quantify past trends and forecast future reproductive phenology in Southeast Asian tropical forests. It shows significant declines in flowering/fruiting proportions since the 1970s and predicts substantial reductions in flowering opportunities for approximately 57% of dipterocarps that require both cool and drought cues, driven by diminished low-temperature events under future warming. Species relying solely on drought cues are projected to be comparatively resilient, implying future shifts in reproductive output and potential community composition changes. Seasonal flowering distributions across a regional latitudinal gradient are expected to remain broadly intact, pointing to quantitative rather than qualitative phenological changes. Future research should expand to species-level models with longer, higher-resolution monitoring; incorporate additional environmental drivers (e.g., solar radiation), genetic/molecular markers of floral induction, and nutrient dynamics; validate projections across more tropical regions; and continue long-term phenology monitoring to enhance predictive accuracy and inform management and restoration planning.

- Data source and sampling: Phenology was monitored largely on one or two individuals per species in an arboretum setting, which may not capture full population variability or microhabitat influences, though patterns matched natural forests during mass events. - Missing data and gaps: Climate records had gaps (Apr 1996–Jun 1997), and some phenology series had missing months; species with high missingness were excluded, potentially biasing representation. - Model scope: Models were calibrated at the cluster (grouped) level rather than at the species level; one cluster (cluster 1) showed relatively low predictive performance (AUC ~0.62–0.64). - Environmental covariates: Only minimum temperature and precipitation were used; solar radiation data were unavailable and treated implicitly via its negative correlation with precipitation. - Bias correction: Despite bias correction, GCM precipitation variance biases remained seasonally structured and required an additional, assumption-dependent correction; dry-day frequency and wet-day intensity were not explicitly corrected. - Generalizability: Projections assume stationarity of cue-response relationships and may not account for acclimation or evolutionary changes; species with long generation times may respond slowly evolutionarily. - Causality: While synergistic cool and drought cues explain observed phenology, endogenous factors (e.g., nutrient status) and biotic interactions may modulate flowering readiness and were not explicitly modeled.