Local Changes in Rainfall and Streamflow via Quantile Regression Methods and their relationship with the PDO, MJO and ENSO: Application to French Polynesia

<p>Flooding in the tropical high islands of French Polynesia located in the South Pacific represents a major threat for tens of thousands of people, in particular in coastal urban areas. In this context,&#160; this study aims to assess rainfall and streamflow changes at local scale and daily interval (hourly when data are available) over the Islands of Tahiti and Moorea during the period 1977-2020, and to provide new indicators for integrated water resource management under high anthropogenic pressure, as well as flood risk anticipation, especially over the northwestern urban part of Tahiti.</p><p>Our analysis is based on quantile regression combining linear and annual cycle components applied on 18 rain gauge and 2 river discharge time series. Results show a moderate to high decrease of extreme precipitation (percentiles > 0.90) on the western leeward side. This finding is in line with recent studies on trends in precipitation in the Pacific. More surprisingly, we also observe a moderate increase in rainfall for the windward (North-East) and southern areas, as well as an overall decline in streamflow over two watersheds encompassing the urban zone of North-West Tahiti. This last finding is correlated with the precipitation decline in leeward slopes.</p><p>The differences in observed rainfall and streamflow trends may be related to differences in forcings involved. To better understand the influence of the regional climate on our hydrometeorological observations, we investigate the modulating effect of the El Ni&#241;o Southern Oscillation (ENSO), the Madden-Julian Oscillation (MJO) and Pacific Decadal Oscillation (PDO) on rainfall and runoff by testing a fast inference for time-varying quantiles via a flexible dynamic quantile linear approach. Increase and decline in precipitation are tied to respective positive and negative PDO phases. &#160;However, further validation is required to disentangle the short-term influence of different modes of variability contained in distinct time scales.</p>