Assessing Stationarity of Drought Records in the Amazon Basin using SPI and Record Theory

Unlike many natural hazards whose impacts are largely localized (e.g., volcanic eruptions), droughts can generate far-reaching spillover effects that extend well beyond their region of occurrence, producing  socio-environmental consequences at continental and even global scales. Moreover, severe seasonal droughts may occur even in regions typically characterized by high levels of humidity, challenging conventional perceptions of hydroclimatic vulnerability. In particular, droughts affecting the Amazon Basin – the world’s largest watershed, characterized by high water availability and exceptional biodiversity – pose significant risks to the global climate system. Given the basin’s central role in regulating the global hydrological cycle, drought events may propagate beyond local riverine livelihoods, disrupting large-scale hydroclimatic processes and ecosystem functioning.

This study assesses whether drought records in the Amazon exhibit stationary behavior by combining the Standardized Precipitation Index (SPI), a widely used multi-timescale indicator of meteorological, with record-based stationarity tests designed to detect non-stationarity specifically in distribution tails. Monthly precipitation series from 272 rain gauge stations, each with at least 30 years of data, were transformed into SPI at a 6-month timescale. The analysis focuses on October SPI values, which integrate precipitation anomalies accumulated over the preceding dry season, allowing a consistent seasonal basis for comparison across the basin.

Stationarity is tested under the i.i.d. record hypothesis (record probability ) using non-parametric statistics proposed by Cebrián; Castillo-Mateo; Asín (2022), from the RecordTest package including the record-count -test and a weighted variant with linear weights, the likelihood-ratio test (LR), and the Foster–Stuart test, all applied to lower records representing drought extremes. Statistical significance is assessed using Monte Carlo resampling with 10,000 simulations.

The application of record-based stationarity tests indicates that drought records are predominantly stationary across the Amazon Basin. Out of the 272 analyzed stations, approximately 82% show no statistically significant departures from the i.i.d. record hypothesis in any of the applied tests. Strong and consistent evidence of non-stationarity is rare, with fewer than 3% of the stations showing simultaneous rejection across all tests. Spatially, the stations identified as non-stationary are broadly dispersed across the domain, indicating the absence of coherent regional clustering or directional gradients. These results support the hypothesis that, for the SPI-6 October series representing dry-season accumulation, the statistical behavior of drought extremes remains largely stationary at the basin scale, despite recent severe drought events reported in the literature. Overall, the proposed framework is distribution-free, tail-oriented, and computationally scalable, offering a robust methodological basis for monitoring changes in drought extremes and supporting early-warning systems and long-term water resources management in a changing Amazonian climate.