The Fire and Disaster Management Agency estimates that over 200 houses have been damaged fully or partially, and 5,000 more have been affected by the floods and landslides. Evacuation orders were issued to 1.9 million people and advisory notices to 2.3 million people as of 8 July 2018.
For some extreme weather events we cannot do a full attribution study, usually because of the lack of data (observations and/or models). Instead we provide a quick look which is a preliminary analysis of an extreme event with some elements of an attribution analysis but without an attribution statement.
Globally, intense rainfall is increasing, but local trends can deviate substantially from the global tendency (see, e.g., these global maps). We cannot do a full attribution analysis of the impact of anthropogenic climate change on the extreme rain in southwestern Japan at this time, due to lack of manpower and of enough suitable climate models to produce a robust quantitative assessment. Here we report a quick look at the event, analysing observations and the one model available, resulting in the qualitative conclusion that, based on these lines of evidence only, climate change increased the probability of extreme rains such as observed at the beginning of July in Japan.
A previous attribution analysis of extreme rainfall in this area, triggered by the 2012 extremes, focused on the occurrence of circulation patterns that caused higher rainfall in this region (Imada et al, 2013) as no model available at the time of the study could resolve the precipitation event itself. The study did not detect a shift due to climate change in the main pattern that is sensitive to SST variations. Observational analysis of quality-controlled station data from 1901 to 2004 found a 45% linear increase in the 1 in 10 year extremes summed over western Japan, mainly in autumn (Fujibe et al, 2005, Fujibe et al, 2006).
Here we focus instead on the rainfall itself, in observations and in a single model operating at high enough resolution to resolve the small-scale precipitation extremes that caused the flash floods and damage in Japan in early July 2018. This analysis therefore includes thermodynamic factors such as the increase in moisture with temperature according to the Clausius-Clapeyron relation. It therefore gives a more complete analysis than Imada et al, 2013 who only considered circulation changes.
We base the observational analysis on the GHCN-Daily v3 station data augmented with real-time data from the GTS system that meteorological institutes use to exchange data for weather forecasting. The results were checked against the 0.5º CPC analysis and GPM/IMERG satellite estimates of precipitation.
Figure 1 shows that on 6 and 7 July a few stations recorded very large amounts of precipitation; up to 338 mm/dy at Iizuka. The extremes were very local and at different locations on these two days, so we consider the highest daily precipitation of the year (RX1day) at station locations as our event definition. In Figure 2 this variable is shown for 2018 (up to 10 July) and as a comparison for 2012, which was analysed by Imada et al, 2013. In 2012 there were more stations that recorded more than 250 mm/dy, but mainly on the southern coast. In 2018 the rains fell more in the mountainous interior.
The GPM/IMERG satellite observations show extreme rainfall on the same two days, but much higher amounts, up to 500 mm/dy in the mountainous interior. This can be due to the fact that the highest rainfall intensities were on mountain slopes with no rain gauges. The intensity of the flash floods is qualitative evidence for this. However, there may also be calibration problems of the retrieval for such high precipitation amounts from the satellite instruments (cf our Houston analysis). The extreme rainfall was caused by a strong south-westerly flow bringing moisture from the subtropics and convergence over southwestern Japan, the mechanism also described by Imada et al, 2013.
Climatology and anomalies
The south of Japan has its highest rainfall and extremes in the June-July ‘Baiu’ rainy season (Imada et al, 2013). This is caused by a convergence zone that lies over the southern part of Japan and the adjoining oceans. There, a normal year would be called extreme in many other parts of the world, with many stations regularly receiving more than 100 mm/dy on the wettest day of the year, especially on the southern coast (Figure 3). Due to the sharp gradient in the amount of rain on the wettest day of the year between southern and northern Japan we restrict the further analyses to the area south of 36 ºN (indicated by a line of Figure 4), and consider stations in 25-36 ºN, 125-145 ºE (the area below the line, this includes a few stations in South Korea).
Relative to this climatology, we found that the 2018 event was exceptional, with some stations in the interior measuring more than two times the usual amount for the wettest day of the year (Figure 5). In 2012 the extremes on the coast were about 50% higher than normal with two stations in the interior having double the usual amount. So in contrast to the absolute RX1day amounts, the relative deviations from normal were larger in 2018 than in 2012. This may explain why the number of casualties and damages were also larger.
We fitted the observed precipitation series to a GEV distribution in which the position and scale parameters vary with the smoothed global mean temperature as a proxy for global warming (see van der Wiel et al, 2017 for a full description). This gives return times in the current climate, which are more than 200 years at five stations for the 2018 extremes (Urmi, Hiroshima, Okayama, Fukuyama and Matsuyama). Note that due to to the small spatial scale of the extremes, the probability of finding such extremes anywhere in the region is much higher. We estimate there are about 10 degrees of freedom, so that the probability of having a station with a 200-yr return time anywhere in southwestern Japan is about once every 20 years.
It makes little sense to analyse single precipitation series for trends, which in Japan usually have 50 to 60 years of data. The signal/noise ratio is so poor that the trends are dominated by the random weather. We therefore considered all stations in the area together, each fitted to the non-stationary GEV described above. Figure 6 shows that there are more stations with positive trends (a higher probability of extreme rainfall now) than with negative trends. The number of stations with negative trends is compatible with what one expects from pure chance: 6/94 stations (6%) have p<0.1, 3/94 stations (3%) have p<0.05. The number of stations with positive trends is higher: 21/94 (22%) have p<0.1, 10/94 (11%) have p<0.05. This implies that about half the positive trends in Figure 6 are due to natural variability. The other half are due to systematic trends, the most plausible being climate change due to greenhouse gases and aerosols. It is impossible to state which half, but the area of the 2018 floods is almost uniformly blue, i.e., almost all stations there have positive trends up to 2017 (the event under study, 2018, has not been included in the fits).
Some model results
In order to ascertain the cause of the trends in the station data we have to use climate models. State-of-the-art global climate models with a resolution of about 200 km show an increase in precipitation extremes in the region. However, due to the small scale of the extreme events and the fact that some are due to typhoons that are not resolved by these models, these results are not trustworthy. On physical grounds we require climate models with about 25 km resolution or higher to undertake a full attribution study. Only at that resolution models start to represent tropical cyclones well enough and should have a better representation of the mountainous terrain. However, model output for this region at this resolution is not available to us at the moment. The best we could find were the EC-Earth T799 (~30 km) Future Weather experiments. These are four time slices of five years around the years 1850, 2000, 2030 and 2100 with 6 ensemble members run with RCP4.5 greenhouse gas and aerosol concentrations and observed SSTs in 2002-2006, the SSTs were adjusted to the other periods using the ESSENCE ensemble of ECHAM5/MPI-OM runs under the forcing scenario SRES A1B.
The climatology of the wettest day of the year looks realistic in this model, with the east-west band of high extremes ending at about 36 ºN and the highest extremes on the south coast, just like the observations (compare figures 4 and 7). The seasonal cycle also has the right structure, although the rainy season peaks two weeks earlier than in the observations and the model seems to overestimate the extremes relative to the CMORPH satellite retrieval.
Figure 8 shows the trends in this model. The left panel compares the 2000 runs with the 1850 runs, and the right panel uses all four time slices up to 2098. They both show on average increases in these precipitation extremes,. The differences are due to the obviously higher natural variability and the effects of aerosols, which have large uncertainties. However, southwestern Japan exhibits positive trends in RX1day over both intervals. Because of the uncertainties in the effects of greenhouse gases and aerosols and the large contribution of natural variability to local trends we cannot draw a quantitative conclusion from this single model beyond the statement that the trend is positive.
On 6 and 7 July 2018 extreme amounts of precipitation fell in single days in the interior of southwestern Japan, causing extensive flooding. This is the main rainy season in this region, but the daily extremes were about twice the usual amount on the wettest day of the year in the five stations that experienced the largest rain amounts. This translates into local return times of more than 200 years in the current climate.
The observed precipitation series show significantly more stations with positive trends than negative trends. The latter are compatible with chance behaviour from natural variability, while there are two times more stations showing an increasing trends in the probability of these kinds of high precipitation extremes than the number expected due to natural variability. This agrees with earlier studies that found positive trends in the 1 in 10 year extremes in this region of Japan.
A single model, EC-Earth T799, appears to reproduce the main spatial and seasonal features of the rainfall extremes qualitatively. It also shows positive trends in the wettest day of the year in southwestern Japan both over the historic time period 1850-2006 and the time period spanning past, present and future 1850-2098.
The observational and model evidence agree that climate change has made extremes like the ones observed in 2018 more likely. However, the data on which our observational analysis is based is too noisy to compute by how much. Based on past experience on how much different models diverge in their representation of extreme rainfall events, we choose not to quote quantitative model results from a single model either. A full attribution study will have to be based on multiple models, after a formal model evaluation step.