Climate-driven changes in rainfall structure and urban landslide dynamics in subtropical Brazil

Climate change is expected to alter rainfall regimes worldwide, with growing evidence that changes in rainfall characteristics extend beyond precipitation totals to include shifts in intensity, intermittency, and temporal organisation. However, the extent to which such structural changes in rainfall patterns influence rainfall-triggered landslides remains insufficiently explored, particularly in tropical and subtropical regions where empirical evidence is scarce. Moreover, beyond aggregated seasonal or annual metrics, the role of rainfall structure in governing landslide triggering processes remains poorly constrained, motivating renewed methodological approaches to rainfall–landslide analysis.

In this contribution, we investigate multi-decadal changes in rainfall characteristics and their implications for urban landslide occurrence using a long-term daily rainfall record (1940–2024) from the Santos–Saboo rain gauge, together with a harmonised inventory of 2,252 urban landslides from Santos and Cubatão (420 km²), São Paulo state (Brazil) spanning 1988–2024. Rainfall structure is characterised through analysis of annual totals, annual mean event intensity, seasonal intensity patterns, and event-based intensity distributions. Long-term trends are assessed using Mann–Kendall tests with Sen's slope, together with the Standardised Precipitation Index (SPI) computed at multiple accumulation timescales. This approach explicitly evaluates shifts in rainfall concentration and temporal organisation beyond simple magnitude-based assessments.

Results show no significant long-term trend in total annual rainfall at Santos–Saboo. In contrast, clear modifications in rainfall structure are observed, characterised by increasing annual mean event intensity (+0.031 mm day^-1 year^-1, p<0.01) and amplification of extreme event intensity (95^th percentile: +0.081 mm day^-1 year^-1, p<0.05), with particularly strong winter intensification. These patterns suggest a tendency towards more concentrated rainfall delivery within events, which is consistent with expected climate-driven changes in precipitation regimes. Landslide frequency across the study period exhibits high inter-annual variability (±86 events/year) but no statistically significant long-term trend (p=0.09, Sen's slope=0.71 events/year), despite a weak positive tendency. Similarly, seasonal analyses show non-significant trends across all seasons. This points towards landslide activity remaining episodic and primarily controlled by individual extreme rainfall events, potentially obscuring long-term climatic signals in event frequency.

The findings highlight the importance of rainfall structure diagnostics for understanding climate-related changes in landslide hazard and for informing threshold-based early warning systems. Despite clear intensification of rainfall characteristics, the absence of increases in proportional landslide frequency suggest complex landscape responses, potentially influenced by countervailing factors such as improvements in urban drainage, slope stabilisation measures, changes in exposure, or early warning effectiveness. Overall, the study demonstrates the value of combining long-term station records with rainfall structure metrics. This approach provides a robust foundation for future methodological expansion, by using additional gauges, satellite rainfall products, and broader landslide inventories in underrepresented tropical and subtropical regions.