Quantifying biases in IDF values caused by temporally aggregating data, not considering all storm events, and using weekly rain gaging charts

---

Claudio Meier, Jorge Sebastián Moraga, Peter Molnár

Tuesday 30 june 2015

15:05 - 15:20h at North America (level 0)

Themes: (T) Extreme events, natural variability and climate change, (ST) Hydrological extremes: floods and droughts

Parallel session: 6I. Extreme events - Flood Drought

---

We used 19 years of daily rain charts at a research station in Concepción to derive IDF values using both the annual maxima and partial duration approaches. When comparing with official values, used for engineering design, we found a severe underestimation in the latter, increasing for shorter durations. For example, according to our data, the official 1-h, 100-yr design rainfall actually recurs every 3.8 years, a 27 times higher frequency of failure than intended. Part of the problem lies in the data: the best weather stations in Chile use good quality, paper-recording gages, but with weekly charts, whose resolution is not high enough to estimate extreme rainfall for shorter durations. Still, our results indicated strong underestimation even for longer durations (e.g., the 10-yr, 12-h rain actually recurs every 2.2 y). We hypothesised that there could also be methodological causes behind this large bias, potentially implying a nation-wide problem, which could very well explain the perennial failure of urban storm water systems in Chile. We specifically tested two alternative hypotheses to explain the bias: (i) the use of temporally-aggregated instead of continuous rainfall data, and (ii) the use of only the 4 or 5 largest storms per year when searching maxima, instead of sampling all events. In order to obtain robust results, we not only analysed with 32 years of weekly charts in Concepción, but also with 32 years of 10-min data at 52 locations in Switzerland. Results were similar at all locations, and actually quite surprising: Two widely accepted practices, the aggregation of rainfall data over even “reasonably short” periods and not considering all storms when searching for the maxima, result in large underestimation of extreme rains, which can easily range between 30 and 40% for durations equal or less than one hour. In the case of Concepción, these two effects fully account for the 45% underestimation in the 100-yr, 1-h rainfall.