Verifying the relationships among the variabilities of summer precipitation extremes over western Japan in the d4PDF climate ensemble, monsoon activity, and Pacific sea surface temperature

Abstract. Upper 99th percentile hourly and 90th percentile daily rainfall over western Japan was calculated for June–July every year, using two observation-based and one simulation-based datasets. These were 54 rain-gauges over the 1952–2022 period, 1 km resolution radar/rain-gauge merged precipitation data over the 2006–2022 sub-period, and the 5 km resolution d4PDF (database for Policy Decision-making for Future climate changes) climate ensemble over the 1952–2010 sub-period. Grid-points over western Japan were clustered by applying the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) algorithm. Spearman correlation was calculated between rainfall extremes of the clusters, and standardised scores of four modes from the rotated extended Principle Component Analysis of Pacific Sea Surface Temperature (SST) anomalies. These modes represent ENSO (El Niño-Southern Oscillation) growth, ENSO decay, warming trend, and PDV (Pacific Decadal Variability). Based on the clustering, 10 sub-regions were selected for analysis. The correlation coefficients between rainfall extremes and SST modes were at most moderate (|R| ⩽ 0.60) over most sub-regions, reflecting correlation with ENSO decay and warming trend, both directions with a spatial pattern for ENSO growth, and anti-correlation with PDV. These relationships could be partially explained through the strength and location of the monsoon jet in relation to different ENSO phases. d4PDF reflected similar relationships for the first three modes, although it showed both spatial and strength biases. Correlation between rainfall extremes and ENSO decay was likely excessively strong in d4PDF due to the regional climate model's over-response to the monsoon jet wind speed. For the PDV mode, the model could not reproduce the observed relationship of spatially widespread anti-correlation with rainfall extremes. Based on the mostly weak long-term correlations, we concluded that individual SST modes modulated rainfall extremes but were not controlling factors in their occurrence. However, we hypothesize that multiple modes may stack in ways that greatly strengthen their modulating effect, and recommend further investigation in this using sensitivity simulations on case studies.