Decline in seasonal predictability potentially destabilized Classic Maya societies

The study examines how changes in the predictability of seasonal rainfall affected Classic Maya societies in the southern lowlands. Seasonal hydroclimate variability is critical for tropical agrarian systems, and the Classic Maya, reliant on rainfed agriculture, experienced demographic and political transformations during 750–950 CE. Prior work links decadal–centennial drying and conflict to the Classic Period Collapse (CPC), but the role of year-to-year seasonal predictability remains unclear. The authors aim to reconstruct seasonal-scale hydroclimate variability using a high-resolution, precisely dated speleothem (YOK-G) from Yok Balum Cave, Belize, to assess whether declining predictability of seasonal rainfall contributed to sociopolitical destabilization.

Archaeological and palaeoclimatic research has increasingly focused on seasonality and its societal impacts, with evidence from bones, shells, lake sediments, and speleothems. For the Classic Maya, many studies implicate climatic disturbances—particularly droughts between 500 and 900 CE—increased warfare, and demographic decline in the southern lowlands. However, past reconstructions often lacked the temporal control to quantify seasonality. Hypotheses to explain Classic Period droughts include ITCZ migrations, North Atlantic SST changes, persistent El Niño, and tropical cyclone variability or their combination. Regional heterogeneity in drought timing and persistence is noted, and the northern lowlands demonstrate more resilience. Recent work suggests that ITCZ dynamics may involve expansion/contraction and changes in residence time, not only latitudinal shifts. Modern observations show anthropogenic climate change reducing seasonal predictability, challenging smallholder farmers. The study positions itself to quantify sub-annual rainfall variability and predictability beyond the instrumental period.

Data and proxy: The study uses the aragonitic stalagmite YOK-G from Yok Balum Cave, southern Belize, comprising 7151 δ13C and δ18O analyses covering ~1600 years, constrained by 52 U/Th dates. δ13C is interpreted as a proxy for local effective rainfall (influenced by PCP, CO2 degassing, soil/vegetation processes), where higher δ13C indicates drier conditions; δ18O reflects regional hydroclimate including moisture source/transport, rainfall amount, and tropical cyclone influence. Age uncertainties (±2–12 years; ~±5 years typical) are propagated via COPRA Monte Carlo age-model ensembles. Time series preparation: Each COPRA realization is detrended using Singular Spectrum Analysis (trend window ~10 years) to isolate intra- and interannual variability. A most central realization (MCR) is used for illustrative purposes, but all analyses consider ensembles for uncertainty propagation. Extreme events detection: Annual-scale hydroclimatic extremes (dry or wet years) are identified using a peak-over-threshold method on detrended δ13C: years exceeding the 95th (dry) or below the 5th (wet) percentile are flagged across 2000 realizations, and the fraction of realizations per year is reported as an extreme-event frequency indicator. Time-frequency analysis of seasonality: Continuous wavelet transforms are computed on linearly interpolated series (Δt ≈ 0.30 yr; interpolation optimized via Lomb-Scargle periodograms). Significance is tested against irregularly sampled AR(1) surrogates. A seasonal cycle indicator is computed as the fraction of realizations with significant power in the 0.5–1.5 yr band. Potential aliasing is evaluated with sinusoidal tests; years with <3 samples are flagged as low-resolution. Seasonal predictability metric (τ_pred): Annual δ13C segments are compared using recurrence plots computed from an edit-distance metric tailored for irregular sampling, capturing similarity of sub-annual rainfall distributions between years. Mean diagonal line length yields a mean prediction time. To correct for sampling-rate bias, sampling rate-constrained surrogate (SRC) series are generated per realization; relative predictability τ_pred is defined as the ratio of observed mean prediction time to the 95th percentile of SRC-based prediction times. τ_pred ≥ 1 indicates predictability beyond that expected from sampling rate alone. Analyses use 200-year sliding windows (25% overlap) and 20 distinct age-model realizations. Downsampling/toy model assessment: A sinusoidal toy model with imposed downsampling/integration is used to estimate how much low growth/sampling resolution could artificially reduce τ_pred. Local rainfall coherency: Multi-decadal correlations between YOK-G δ13C and δ18O trends (via SSA) are computed to define a local rainfall coherency index; significant positive correlation implies local hydroclimate coherency with regional controls (e.g., ITCZ/N Atlantic SST), whereas insignificant/negative correlations suggest nonlocal or differing controls (e.g., TC influence). The coherency index is compared with Cariaco Basin foraminiferal Mg/Ca-based summer SST reconstruction. Archaeological datasets: Summed probability distributions (SPDs) of radiocarbon dates are constructed using a Bayesian KDE approach (OxCal KDE_Model) from a Maya Lowlands dataset (1,035 assays post-filtering from 80 sites) and a local Uxbenká dataset (338 dates). Monument production and warfare event frequencies are derived from the Maya Hieroglyphic Database, considering only securely dated dedicatory monuments and temporally grounded warfare-related events. These are compared against hydroclimate indicators. Data and code availability: YOK-G records and age-model ensembles are archived on Zenodo; Python code to reproduce figures is provided via Zenodo.

- Background hydroclimate: YOK-G δ13C indicates a drying trend beginning ~500 CE, with pronounced dry conditions between 600–800 CE, and the wettest conditions during the Little Ice Age (1400–1800 CE). This trend is consistent with δ18O and trace elements and aligns with regional records of multi-annual droughts. - Hydroclimate volatility: Fewer extreme events are inferred for 550–700 CE, followed by increased frequency of annual extremes (dry/wet years) between 700–900 CE, overlapping with the period of severe droughts and demographic contraction in the Maya Lowlands. Additional high-volatility intervals occur partly during the Medieval Climate Anomaly (1100–1300 CE) and the latter LIA (1600–1900 CE). - Seasonality detection: A significant annual cycle becomes more prevalent after ~1400 CE in both δ13C and δ18O; before 1400 CE the seasonal signal is muted, consistent with previous work. - Seasonal predictability (τ_pred): τ_pred declines below the reference threshold (τ_pred = 1) between 800–1000 CE, indicating seasonal rainfall became as unpredictable as a random process with similar sampling characteristics. Predictability begins to deteriorate after ~500 CE, with the sharpest decline after 800 CE; the minimum occurs around 1000 CE. Toy-model tests suggest only part of the τ_pred decline can be attributed to low sampling resolution; genuine seasonal irregularities are required to explain the observed reduction. - Nonlinear relationship: The relationship between average hydroclimate state (δ13C) and seasonal predictability is non-monotonic; highest predictability occurs at moderately wet conditions (δ13C ≈ −8.8‰ VPDB). The Terminal Classic shows particularly low predictability despite wetter average conditions compared to the late Classic peak drying. - Sociocultural linkage: The decline in τ_pred begins decades before and intensifies during the CPC (750–950 CE), coinciding with increased warfare and declining monument erection. At Uxbenká (near Yok Balum), demographic decline begins ~680 CE, the last monument is dated to 780 CE, and site abandonment follows ~30 years later—aligning with reduced predictability and severe drought (~750 CE). Across the lowlands, >63% of urban polities with dated monuments had disintegrated by 835 CE. - Mechanism via Atlantic SST/ITCZ: Multi-decadal periods of insignificant or negative δ13C–δ18O correlation at Yok Balum coincide with low tropical North Atlantic summer SSTs (Cariaco Basin). The authors infer reduced coherency of ITCZ-driven rainfall over the region (more ‘patchy’ convection, variable residence time/strength), increasing interannual volatility and reducing seasonal predictability. Periods of higher SSTs align with significant positive proxy correlations and greater coherency. Enhanced TC influence during low-SST periods could invert δ18O–δ13C correlation (low δ18O from TC rain despite local dryness), further increasing volatility. - Modern relevance: The findings parallel modern observations where declining seasonal predictability undermines smallholder farming strategies, suggesting increased vulnerability under ongoing climate change.

The analysis demonstrates that beyond multi-decadal drying, a pronounced decline in year-to-year predictability of seasonal rainfall likely undermined Classic Maya agricultural planning and surplus production. Reduced ability to anticipate wet-season onset, duration, and intensity would compromise land preparation, burning, planting, and harvest schedules, diminishing yields and surpluses needed to sustain large non-farming urban populations. The coupling of decreased predictability with severe droughts between 600–800 CE plausibly amplified food insecurity, social stress, warfare, and the disintegration of political institutions. The non-monotonic link between background hydroclimate and predictability implies that even during moderately wet background states (e.g., Terminal Classic), low predictability could impede recovery of Classic governance. Multi-decadal low North Atlantic SSTs likely reduced ITCZ coherency, increasing interannual volatility and sensitivity to transient disturbances such as tropical cyclones, consistent with the proxy coherency index and SST comparisons. The findings thus address the research question by identifying seasonal rainfall unpredictability as a critical, previously under-quantified climatic factor contributing to the CPC.

This study introduces a recurrence-based, sampling-corrected metric of seasonal predictability from a sub-annually resolved speleothem, revealing that declining predictability of seasonal rainfall—beginning after ~500 CE and reaching a minimum around 1000 CE—coincided with and likely contributed to the Classic Period Collapse, alongside severe droughts. The mechanism is consistent with periods of low tropical North Atlantic SSTs and reduced coherency of ITCZ-driven rainfall over the Maya Lowlands. The work reframes the CPC as driven not only by protracted drought but also by increased interannual volatility and reduced predictability of seasonal rainfall affecting agricultural systems and sociopolitical stability. Future research should expand high-resolution, well-dated regional records to test spatial coherence, refine corrections for sampling/growth effects, better quantify the role of tropical cyclones, and integrate climate–society modeling to assess thresholds and adaptive capacities.

- Proxy interpretation: δ13C reflects multiple processes (PCP, degassing, soil/vegetation dynamics, residence time, temperature); the study assumes dominance of local infiltration effects, though other factors may contribute variably. - Sampling/growth effects: Low stalagmite growth can suppress seasonal variance and reduce sampling resolution, potentially biasing τ_pred downward. Toy-model tests suggest only a fraction of the observed decline is due to downsampling, but exact magnitudes remain uncertain. Possible micro-hiatuses during severe multi-season droughts may go undetected. - Age uncertainties and irregular sampling: Age errors (±2–12 years) can blur spectral peaks; interpolation for wavelet analysis and correction via SRC surrogates mitigate, but cannot eliminate, these issues. - Single-site focus: Conclusions rely on one primary stalagmite (YOK-G); broader spatial replication is needed to generalize regional dynamics and disentangle ITCZ modes (shift vs. expansion/contraction/strength changes). - Extreme-event analysis: No explicit correction for varying sampling resolution is applied to the extreme-event frequency indicator. - Attribution complexity: Negative or insignificant δ13C–δ18O correlations may arise from multiple factors (e.g., TC rainfall, moisture source changes, wildfire impacts), complicating mechanistic attribution. - Archaeological proxies: SPDs and monument/warfare datasets entail assumptions (association with anthropogenic activity, sample-size effects, reporting biases) and potential chronological uncertainties.