Abstract
Despite increasing evidence of intensification of extreme precipitation
events associated with a warming climate, the magnitude of extreme river
flows is decreasing in many parts of the world. To better understand the
range of relationships between precipitation extremes and floods, we
analyzed annual precipitation extremes and flood events over the
Contiguous United States from 1980 to 2014. A low spatial correlation
(less than 0.2) between changes in precipitation extremes and changes in
floods was found, attributable to a weak causal relationship. The
co-variation between precipitation extremes and floods is also
substantially low, with a majority of catchments having a coefficient of
determination of less than 0.5, even among the catchments with a
relatively strong causal relationship. The findings indicate a need for
more investigations into causal mechanisms driving a non-linear response
of floods to intensified precipitation extremes in a warming climate.
1 Introduction
Among the most important implications of global climate change is the
intensification of the hydrologic cycle [Huntington , 2006],
including the intensification of rainfall extremes [Westra et
al. , 2014]. As air temperature rises, the water vapor held in the
atmosphere also increases following the Clausius-Clapeyron relation
[Clausius , 1850]. This relationship has been documented
extensively in the climate literature [Donat et al. , 2013;Guerreiro et al. , 2018; Papalexiou and Montanari , 2019;Westra et al. , 2013], and has led to concerns of a future
characterized broadly by an increase in the magnitude of global flood
events.
Large scale investigations into changes in floods, however, indicate a
broader range of global flood response, with many studies documenting
sites with a decrease in flood magnitude [Do et al. , 2017;Do et al. , 2020b; Gudmundsson et al. , 2019; Hodgkins
et al. , 2017; Kundzewicz et al. , 2004; Lins and Slack ,
1999]. These somewhat unexpected relationships between trends in
extreme precipitation and trends in extreme discharge can be attributed
to the influence of other flood generation mechanisms such as soil
moisture [Ivancic and Shaw , 2015; Wasko et al. , 2020;Ye et al. , 2017] and snow dynamics [Berghuijs et al. ,
2016; Blöschl et al. , 2017; Do et al. , 2020a;Ledingham et al. , 2019; Stein et al. , 2020]. Even when
floods are triggered by precipitation extremes, the relationship between
precipitation magnitude and flood magnitude is likely non-linear
[Sharma et al. , 2018], owing to the complex interactions of
many variables which have undergone substantial changes such as land
cover [Keenan et al. , 2015; Lambin et al. , 2003],
river channels [Slater et al. , 2015; Yamazaki et al. ,
2014] and evapotranspiration [Bosilovich et al. , 2005;Gronewold and Stow , 2014].
However, there is still a lack of quantitative understanding of the
relationship between precipitation extremes and floods, partially due to
unavailable discharge observations in many parts of the world
[Do et al. , 2018; Do et al. , 2020b]. Even for regions
with relatively good streamflow records, empirical investigations have
primarily focused on the consistency between the timing of precipitation
extremes and that of floods [Berghuijs et al. , 2019;Blöschl et al. , 2017; Do et al. , 2020a; Ivancic and
Shaw , 2015; Stein et al. , 2020; Wasko et al. , 2020]
rather than co-variation between precipitation extreme intensity and
flood magnitude. As a result, it is difficult to promote generalized
statements about global relationships between changes in precipitation
extremes and changes in floods, which is essential to the design of
robust flood prevention and mitigation strategies in a warming climate
[Milly et al. , 2008].
We aim to fill in this gap through an empirical assessment of the
co-variation of precipitation extremes and flood magnitude using a large
sample (671) of catchments across the Contiguous United States (CONUS)
(Section 2.1). We used annual maxima streamflow from 1980 to 2014 from
these catchments as the flood population, and we used three metrics of
annual maxima precipitation to represent precipitation extremes (Section
2.2). Temporal changes in floods and precipitation extremes were then
estimated at each catchment and the correlation between the spatial
patterns of these trends was assessed (Section 2.3). The ordinal date of
precipitation extreme events was then compared to that of annual flood
events (Section 2.4) to assess potential causal relationships between
these hydro-climatic extremes. Finally, the co-variation between the
intensity of precipitation extremes and flood magnitude across
catchments was assessed (Section 2.5) to evaluate the appropriateness of
using changes in extreme precipitation as a proxy for changes in floods.
2 Data and Methods
2.1 Data
Data for our analysis was derived from the Catchment Attributes and
Meteorology for Large-sample Studies (CAMELS) dataset [Addor et
al. , 2017a; Newman et al. , 2015]. The CAMELS database
aggregates a variety of hydrometeorological variables (primarily derived
from other studies) for 671 catchments across the CONUS (the outlets of
CAMELS catchments are shown in Figure 1). The catchments in the CAMELS
database are intended to reflect relatively natural hydrologic
conditions (the impervious surface area of each catchment is less than
5% of the total catchment area; see Newman et al. [2015] for
more information). These catchments have a relatively small size (the
median catchment area is 340.7 km2) and cover a range
of climatic conditions (e.g., dry, temperate, and continental climates)
as well as geographic features (e.g., mountains and deserts). Other
variables in the CAMELS database include daily streamflow (originally
obtained from the United States Geological Survey), catchment-average
daily precipitation and temperature (derived from the Daymet dataset
[Thornton et al. , 1997]), and daily evapotranspiration,
simulated by the conceptual SAC-SMA model [Burnash et al. ,
1973].
In addition to the hydro-meteorological data available through CAMELS,
we also obtained soil moisture data from the NOAA Climate Prediction
Center [Van den Dool et al. , 2003]. This dataset provides
monthly soil moisture water height equivalent, simulated by a leaky
bucket model, with a spatial resolution of 0.5-degree latitude x
0.5-degree longitude (i.e., a cell-size of about 2,500
km2). We used monthly soil moisture from the cell
containing each catchment outlets as a proxy for catchment-wide to
obtain soil moisture conditions from 1980 to 2014 for each of the CAMELS
catchments. We judged this approach to be appropriate for the CAMELS
catchments, which have a relatively small size.