Contrasting Trends in Flood Occurrence Rates in Three Polish Catchments During 1951–2024: A Seasonal Point-Process Analysis

Flood non-stationarity is usually assessed from annual maxima, which merge seasons and magnitudes and can conceal changes confined to one part of the extreme-value distribution. Here, the Mudelsee point-process framework—comprising peaks-over-threshold sampling, declustering, kernel estimation of the time-varying occurrence rate λ(t), bootstrap confidence bands, and the Cox–Lewis test—is combined with a fully seasonal design and three disjoint magnitude classes and applied to daily discharges during 1951–2024 from three contrasting Polish catchments: the lowland Mogilnica, the lake-district Rega, and the Carpathian Poprad. Seasonal thresholds Q80, Q90, and Q95, derived separately for the cold (November–April) and warm (May–October) half-years, yield 18 occurrence-rate panels. Two of these depart significantly from a constant rate in opposite directions: the occurrence rate of extreme (>Q95) warm-season floods declined in the Rega River (u = −3.171, p = 0.001; λ^ ≈ 1.20 → 0.45 yr−1) and increased in the Poprad River (u = 1.893, p = 0.029; ≈1.35 → 2.00 yr−1), whereas in the Mogilnica River, no panel reached significance. Only the decline in the Rega River survives within-river Benjamini–Hochberg correction (padj = 0.006); this is corroborated by classical trend tests, by the concordant sign of all six of its occurrence-rate panels (p = 0.031), and by a 33% loss of the warm-season quickflow volume. Both signals are robust to the declustering interval and to the bandwidth, but neither remains significant when the thresholds are allowed to track the shifting flow distribution, so they describe the frequency exceeding a fixed discharge rather than a change in the point-process intensity itself. A Monte Carlo power analysis shows that none of the sixteen non-significant panels contain an effect large enough to be detectable at its event count; these null results are uninformative rather than evidence of stationarity. Flood-regime change in Poland is thus season- and magnitude-specific, and largely invisible to annual-maximum analysis.