Enhancement of river flooding due to global warming | Scientific Reports – Nature.com
Extreme weather events (e.g., heavy precipitation and flooding) can have devastating effects on human society and the environment. Heavy precipitation events, a major contributor to flooding, have increased in recent years, especially in many parts of the Northern Hemisphere, due to human-induced climate change (increases in greenhouse gases, GHGs)1,2,3,4. Despite the fact that floods are projected to occur frequently in many regions as the climate warms further5, the effects of past human-induced climate change on river flooding remain unclear due to the complexity of runoff generation and streamflow processes.
The latest 6th Intergovernmental Panel on Climate Change (IPCC) Assessment Report stated that there is low confidence regarding the changes in past floods on a global scale due to human-induced climate change because of limited evidence5,6. One reason for this is that there are spatial and temporal limitations in the available observed data (especially from data-poor regions such as Asia and Africa), as well as in the tools used to gather such data7. Fortunately, in recent years it has become possible to investigate historical flooding using newly developed models, such as the global river inundation model, which can replicate changes in long-term river flow, and large ensemble climate experiments. A study based on these tools found that the spatial patterns of observed trends in river flooding can be reproduced only when human-induced climate change is considered8. Another recent advancement in attributing past flooding changes is event attribution (EA) using the large ensemble climate experiments derived from Atmospheric General Circulation Models (AGCMs), which is a method of probabilistically evaluating the occurrence of extreme phenomena over a specific period in which extreme phenomena are observed9. These climate models have the advantage of being able to represent most important atmosphere and land processes using specific input conditions (e.g., sea-surface temperature [SST], sea ice cover, and human-induced climate changes including concentrations of atmospheric CO2 and aerosols). Moreover, the models are able to create a factual world with climate warming (referred to as HPB, see Methods) and a hypothetical counterfactual world without climate warming (referred to as NAT). The experiments aim to reproduce the atmospheric conditions under which extreme phenomena occur by providing the same oceanic natural variability of an extreme phenomenon10. The probability of event occurrence can then be estimated by repeating the simulation many times (ensemble). It is possible to quantitatively show the effects of human-induced climate change on past extreme phenomena by comparing the probability of the occurrence between factual (with climate change) and hypothetical counterfactual climate simulations. Thus far, the EA has been used to measure quantitatively the contribution of human-induced climate change to past specific extreme phenomena1,10,11,12,13. Applying this EA method to globally distributed river flooding indicated that the probability of occurrence of 14 of the 22 flood events analyzed between 2010 and 2013 was influence by human-induced climate change13. Spring floods due to snowmelt are more susceptible to historical warming due to increased precipitation in the Northern Hemisphere and decreased snowfall due to increased temperatures13. However, we cannot clarify whether the impact of flooding has changed over time, since climate experiments are available only for a short target period. To address this limitation, we extended this study to include a larger number of flood events that occurred over a longer time period (1951–2010) using data from a large ensemble climate experiment, the “Database for Policy Decision-Making for Future Climate Change” (d4PDF)14, derived from the Meteorological Research Institute (MRI) AGCM 3.2 and CaMa-Flood (Catchment-based Macro-Scale Floodplain) model15.
Flood events were determined from the Emergency Events Database (EM-DAT), news media, streamflow records, global daily S14FD discharge reanalysis (Supplementary Materials S1), and existing publications. Data were collected that included the year of occurrence, location, reported damage, number of victims, and main causes of flooding. For selected flood events, we then confirmed whether the flood magnitude in the year of occurrence statistically exceeded the magnitude of a 10-year flood based on available observations, S14FD discharge reanalysis, and HPB simulation. In total, 52 globally distributed flood events (Fig. 1 and Supplementary Table 1) were identified. Forty-one of the 52 flood events in 28 river basins occurred after 1980. Nearly 38% of flood events occurred in Asia, followed by South America (12 events), North America (9 events), Europe (6 events), Oceania (5 events), and Africa (3 events). The fraction of attribution risk (FAR)16 was used to quantify the influence of human-induced climate change on these historical events by comparing the probability of the flood events in the historical climate simulation (hereafter called as HPB) and NAT (counterfactual experiments without human-induced climate change) (see Methods). The bootstrap resampling technique was used to quantify uncertainty in the FAR by calculating the FAR 1000 times using randomly selected ensembles with replacement of all available ensemble simulations.
The results demonstrated that ongoing climate change affected the occurrence of flood events. Among the 52 selected flood events for the period 1951–2010, human-induced climate change changed the occurrence of 20 flood events with 90% likelihood; most of which occurred in Asia and South America (Fig. 1, Table 1, and Supplementary Fig. 4). The occurrence of 14 flood events in Asia (Yangtze River, Indus River, Brahmaputra River, Tapi River, and Chao Phraya River), South America (Magdalena River, Amazon River, and Sao Francisco River), and Africa (Niger River) increased. These flood events were caused mainly by heavy rainfall, including monsoon-induced heavy rainfall in Asia and rainfall associated with El Nino (Yangtze River) or La Nina (Magdalena River). Assessment of rainfall-related indices also indicated that heavy precipitation in the above-mentioned flood events occurred more frequently in HPB experiments than in NAT experiments. In particular, the quantity of short-duration heavy precipitation (annual maximum 1-day or 5-day precipitation) was greater in HPB than NAT for these 14 intensified flood events, except for one event in the Sao Francisco River in 1979. Similarly, more precipitation occurred in almost all events with insignificant but enhanced flooding (Table 1). These results were consistent with those of previous studies, suggesting that short-duration heavy precipitation has intensified, and that this trend has been significantly driven by human-induced thermodynamic changes, including increases in atmospheric moisture1,5,6,8,17,18.
In contrast, human-induced climate change suppressed the occurrence of floods in Asia (Songhua River in 1975 and Mekong River in 1996), South America (Rio Paraguay River in 1980), North America (Mississippi River in 1975), and Europe (Rhine River in 1983 and Elbe River in 1995). The attenuation of these events may be explained by the decreased maximum precipitation in the HPB experiment compared with the NAT experiment (Table 1), as well as increased temperatures in the HPB experiment that reduced the likelihood of snowfall17 and decreased the snow melting peak in some snow-affected rivers (e.g., the Mississippi, Rhine, and Elbe Rivers). The reason for the significant suppression in flooding over the two Asian rivers is unclear, as both rivers have similar annual and seasonal (May to August) mean precipitation (Supplementary Fig. 5, Supplementary Fig. 6). Higher evapotranspiration or less snowmelt (Songhua and the upper Mekong) in HPB than NAT could explain this flood suppression.
More than 50% of flood events shown in Fig. 2 occurred in the most recent 20 years of the study period, partly because more events have been registered in the disaster database in recent years and partly because human-induced climate change altered the probabilities of 20 of 52 analyzed flood events, and that half of these events occurred after 1990. Even ignoring sample size variation among decades, the probability of finding a significant enhancement has been higher in more recent years (Fig. 2b). This illustrates that the increase in flooding has become more prominent in recent years.
Note that this study examined more flood events from the same period (2010–2013) than our previous study using MIROC5 AGCM13. The difference in the spatial resolutions of the previous study (1.4 degrees) and this study (0.5625 degrees), as well as in the models’ abilities to replicate past climates, may explain why the present study was able to identify a larger number of flood events, as we selected flood events only when the frequency in HBP of flooding (e.g., larger in magnitude than the 1/10 return period) in the specific flood year was larger than that of the average of whole period. Comparing the previous study13 and the present study for the same period, both studies indicated that human-induced climate change increased the probability of the Indus River flood in 2010, although this increase was non-significant. By contrast, non-significant enhanced flooding occurred in the Fitzroy River in 201013. These two findings may have resulted from differences in skill and predictability, or the use of different AGCMs to generate large ensemble climate experiments, each of which has developed with its own algorithm to simulate physical atmosphere, ocean, cryosphere, and land surface processes. Conversely, they may have been cause by the application of different methods to generate NAT simulations for these datasets. In a previous study13, the 2010 Fitzroy River flood was attributed to climate warming based on 50th percentiles of FAR scores between two different NAT experiments (ALL and NAT_dtr); this result showed the opposite direction of change between the 50th and 90th percentiles, which signaled that the effect of anthropogenic climate change was not strong in this flood event.
The large ensemble discharge simulation used in this study was derived from a single general circulation model, MRI-AGCM3.2, integrated with the CaMa-Flood model. Therefore, the results of this study depend on uncertainties in the hydrologic processes of these models. For example, one study indicated that intense tropical cyclones are not well represented due to the lack of horizontal resolution14. The consideration of neglected human water management and biases in AGCMs, and the implementation of a single method for removing past climate change signals in NAT, simulation could add further uncertainty, whereas the simulated discharge in HPB was comparable to the range of variability in the observed discharge (Supplementary Fig. 3, Supplementary Fig. 4).
In summary, it is clear that human-induced climate change has increased the occurrence of extreme river flooding events for the past several decades, mostly in South Asia and South America. Furthermore, a decline in the occurrence of river floods was also identified in some regions (e.g., North America and Europe), possibly due to decreased snowfall with warmer surface temperature. Interestingly, human-induced climate change had a stronger effect on floods that occurred in recent years. This study attempted to quantify the contribution of human-induced climate change on a per-event basis. It provides important insights into adaptation strategies for river flooding for which long-term observations are lacking.