Scaling and trends of hourly precipitation extremes in two different climate zones: Hong Kong and the Netherlands Why hourly precipitation Long time series Extremes Trends in extremes Scaling Conclusions Geert Jan van Oldenborgh Lenderink, G., H.Y. Mok, T.C. Lee and G.J. van Oldenborgh, Scaling and trends of hourly precipitation extremes in two different climate zones Hong Kong and the Netherlands. Hydrol. Earth Syst. Sci. 15(2011)3033 3041 doi:10.5194/hess-15-3033-2011.
Why hourly precipitation Typically daily precipitation extremes are studied. Many observational records are daily (eg 8 8 volunteer stations in the Netherlands). Daily data is widely available (80369 series in GHCN-D v2). Standard output of climate models (eg CMIP3, CMIP5, RCMs). However...
Over land, daily extremes are very often in summer or tropics: convective events (thunderstorms) at the end of the afternoon. Kon ituut i n k l i j k N e d erlands Meteorologisch Inst Month of year of maximum daily precipitation
Spatial scale of extreme daily precipitation Two most extreme events in the ECA&D data in the Netherlands. max 148mm max 146mm The spatial extent of these extremes was < 10 km, too small to be resolved in GCMs, only in high-resolution RCMs.
Time scale of extreme precipitation The shower was almost exactly one hour long.
Hourly precipitation extremes Long hourly precipitation series exist, eg in De Bilt from 1906, Hong Kong from 1885 (with the exception of 1940 1945). Since then BMKG+KNMI have digitised data from Jakarta (1871 1970) and we found data in the U.S. (2 stations from 1901, 1875 stations from 1950). Note that clock hours introduce errors: a shower from 16:01 to 16:59 looks much more intense than the same shower from 16:31 to 17:29. Ten-minute data is better but only available for the last 15 years in the Netherlands
Extremes We look at the 90th, 99th and 99.9th percentiles of the amount of rain at wet hours. As it rains 7.5% of the time in the Netherlands, this translates to approximately 5, 50 and 500 days. These percentiles have been estimated directly and via a GPD fit to the top 4% of the data. Error bars have been derived from a non-parametric bootstrap procedure.
Observed hourly extremes in the Netherlands
Extreme precipitation and temperature trends 15-yr sliding window, regr( P h, Td ) = 11±3 %/K, regr( P h, Td ) = 11 %/K detrended. De Bilt May October Blue: extreme hourly precipitation (mean of 95th, 99th and 99.5th percentile) Red: dew-point temperature on day with extreme precipitation Black: mean dew-point temperature
CNT [Celsius] 16.50 16.00 15.50 15.00 14.50 14.00 13.50 13.00 Temperature and humidity trends Apr-Sep temperature CNT (cnt_v11) 12.50 1900 1920 1940 1960 1980 2000 2020 Temperature April September ug [1] Apr-Sep mean_daily_mean_relative_humidity De_Bilt (ug260_mean12 0.84 0.82 0.80 0.78 0.76 0.74 0.72 0.70 0.68 1900 1920 1940 1960 1980 2000 2020 Relative humidity Relative humidity is relatively constant, hence the trend in temperature is about equal to the trend in dew point temperature.
CNT [Celsius] 16.50 16.00 15.50 15.00 14.50 14.00 13.50 13.00 Temperature trends Apr-Sep temperature CNT (cnt_v11) 12.50 1900 1920 1940 1960 1980 2000 2020 Netherlands April September Ta [K] 0.8 0.6 0.4 0.2 0-0.2 Jan-Dec GISS_global_temperature (giss_al_gl_m) -0.4 1880 1900 1920 1940 1960 1980 2000 2020 World annual Temperature trends in the Netherlands and the world seem related: T NL = AT global +ε, with ε the weather noise and A = 1.9±0.3,r = 0.6, A = 2.1±0.7,r = 0.3 detrended (p < 0.01)
Extreme precipitation and temperature trends 15-yr sliding window, regr( P h, Td ) = 23±8 %/K, regr( P h, Td ) = 15 %/K detrended. Hong Kong May September Blue: extreme hourly precipitation (mean of 95th, 99th and 99.5th percentile) Red: dew-point temperature on day with extreme precipitation Black: mean dew-point temperature
Scaling with temperature Why is there a connection between the temperature trend and the extreme precipitation trend in the Netherlands, but no connection in Hong Kong? It is easier to see the connection on a day-by-day basis. Heavy thunderstorms tend to occur after hot days. Warm air can hold more moisture than cold air, the Clausius-Clapeyron Relation: approx. 7 %/K. Does heavy precipitation scale with temperature? Let s plot in overlapping 2 C bins.
Scaling with daily mean temperature Hong Kong Netherlands Approximately two times Clausius-Clapyron (14 %/K, red dots) up to 24 C, falls afterwards. Hot dry days do not produce heavy thunderstorms.
Scaling with dew-point temperature Extreme hourly precipitation depends on specific humidity: warm moist air. Consider dew-point temperature as measure of specific humidity. Best results for dew point temperature 4 hour before the extreme. Trends in dew-point temperature are similar to trends in temperature as the relative humidity is roughly constant.
Scaling with dew-point temperature Hong Kong Netherlands Same absolute behaviour between NL and HKO! 14%/K increase up to dt dew = 23 C, flat afterwards. In the Netherlands, all the extremes are located on the slope. In Hong Kong, they are all on the plateau.
2 CC scaling The intensity of showers increases as 1. The moisture content of the air increases (7 %/K) and 2. The intensity of the updraughts in the cloud increases (another 7 %/K). We do not know why the connection stops at 23 C.
Conclusions De Bilt Heavy hourly precipitation scales with 14%/K dew point temperature Temperature increased by 1.7 K since 1906, relative humidity stayed about the same, dew point temperature also increased by 1.7 K. The observed extreme precipitation trends agree with the ones expected from the scaling behaviour, and correlate well with dew-point temperature on days with heavy precipitation, and somewhat with dew point temperature. Hence the increase in heavy precipitation can be attributed to the increase in temperature. The increase in temperature is (at least partly) connected to global warming.
Conclusions Hong Kong rainy season Heavy hourly precipitation does not depend on dew point temperature (above 23 C) Temperature increased by 0.8 K since 1906, relative humidity stayed about the same, so dew-point temperature increased by 0.7 K. The observed extreme precipitation increase does not correlate with the temperature increase beyond the trend and also does not correlate well with dew point temperature on days with heavy precipitation, in accordance with no scaling on days with dew-point temperature above 23 C. The increase in heavy precipitation is due to other factors, for instance urbanisation. To be investigated further.
Questions To what extent is this scaling behaviour universal? We will investigate Jakarta, U.S.,..., series. What causes the transition from 14%/K to 0%/K at 23 C dew-point temperature? What causes the trend in extreme precipitation in Hong Kong? Urban effects? Can we make projections based on climate change and urbanisation projections?