HESSHydrology and Earth System SciencesHESSHydrol. Earth Syst. Sci.1607-7938Copernicus GmbHGöttingen, Germany10.5194/hess-19-3903-2015Flood triggering in Switzerland: the role of daily to monthly preceding precipitationFroidevauxP.paul.froidevaux@giub.unibe.chSchwanbeckJ.WeingartnerR.ChevalierC.MartiusO.Oeschger Centre for Climate Change Research and Institute of Geography, University of Bern, Bern, SwitzerlandP. Froidevaux (paul.froidevaux@giub.unibe.ch)17September20151993903392412February201525March201530August201531August2015This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://hess.copernicus.org/articles/19/3903/2015/hess-19-3903-2015.htmlThe full text article is available as a PDF file from https://hess.copernicus.org/articles/19/3903/2015/hess-19-3903-2015.pdf
Determining the role of different precipitation periods for peak discharge
generation is crucial for both projecting future changes in flood probability
and for short- and medium-range flood forecasting. In this study,
catchment-averaged daily precipitation time series are analyzed prior to
annual peak discharge events (floods) in Switzerland. The high number of
floods considered – more than 4000 events from 101 catchments have been
analyzed – allows to derive significant information about the role of
antecedent precipitation for peak discharge generation. Based on the analysis
of precipitation times series, a new separation of flood-related
precipitation periods is proposed: (i) the period 0 to 1 day before flood
days, when the maximum flood-triggering precipitation rates are generally
observed, (ii) the period 2 to 3 days before flood days, when longer-lasting
synoptic situations generate “significantly higher than
normal” precipitation amounts, and (iii) the period from 4 days
to 1 month before flood days when previous wet episodes may have
already preconditioned the catchment. The novelty of this study lies in the
separation of antecedent precipitation into the precursor antecedent
precipitation (4 days before floods or earlier, called PRE-AP) and the short
range precipitation (0 to 3 days before floods, a period when precipitation
is often driven by one persistent weather situation like e.g., a stationary
low-pressure system). A precise separation of “antecedent” and
“peak-triggering” precipitation is not attempted. Instead, the strict
definition of antecedent precipitation periods permits a direct comparison of
all catchments.
The precipitation accumulating 0 to 3 days before an event is the most
relevant for floods in Switzerland. PRE-AP precipitation has only a weak and
region-specific influence on flood probability. Floods were significantly
more frequent after wet PRE-AP periods only in the Jura Mountains, in the
western and eastern Swiss plateau, and at the outlet of large lakes. As a
general rule, wet PRE-AP periods enhance the flood probability in catchments
with gentle topography, high infiltration rates, and large storage capacity
(karstic cavities, deep soils, large reservoirs). In contrast, floods were
significantly less frequent after wet PRE-AP periods in glacial catchments
because of reduced melt.
For the majority of catchments however, no significant correlation between
precipitation amounts and flood occurrences is found when the last 3 days
before floods are omitted in the precipitation amounts. Moreover, the PRE-AP
was not higher for extreme floods than for annual floods with a high
frequency and was very close to climatology for all floods. The fact that
floods are not significantly more frequent nor more intense after wet PRE-AP
is a clear indicator of a short discharge memory of Pre-Alpine, Alpine and
South Alpine Swiss catchments. Our study poses the question whether the impact
of long-term precursory precipitation for floods in such catchments is not
overestimated in the general perception. The results suggest that the
consideration of a 3–4 days precipitation period should be sufficient to
represent (understand, reconstruct, model, project) Swiss Alpine floods.
Introduction
River flooding is one of the most devastating and costly natural hazards in
Switzerland and worldwide .
Damaging flood events in the Alps are often caused by high precipitation
events that last for several days e.g.,. However,
river discharge during floods can also be influenced by both the spatial and
temporal characteristics of the precipitation event and by the state of the
catchment before the precipitation event, i.e., the antecedent conditions. One
of the most important antecedent factors is the total water storage in the
form of snow, soil water, ground water and surface water.
In particular, the importance of antecedent precipitation
for floods has long been emphasized (especially for large catchments). For
example, effort is invested in designing continuous hydrological simulations
which allow to account for year-long antecedent precipitation time series
when assessing discharge extremes see e.g.,for the Rhine and Meuse
basins.
For several recent catastrophic flood events antecedent water storage was
important. For example, point to the importance of a
positive water storage anomaly for the 2011 Missouri floods. The floods in
June 2013 in central Europe were preceded by above-average precipitation
during the second half of May that influenced the flood discharge by
presaturating the soils .
further show that this exceptional flood event resulted from the combination
of non-extraordinary precipitation with extremely high initial wetness. For
the floods of 2002 also in central Europe, describe
several intense rainfall episodes in the first half of August that finally
led to the extreme discharges. In southern Switzerland, severe flooding of
the Lago Maggiore in September 1993 was preceded by a series of high
precipitation events in the watershed .
Antecedent conditions might even be relevant for the development of flash
floods: found that the runoff coefficient,
i.e., the fraction of the total rainfall that is routed into runoff, of
58 flash floods in Europe was statistically higher for wetter antecedent
precipitation. They however also found that, although flash floods are more
frequent after wet antecedent conditions in central Europe, they primarily
occur following dry conditions in the Mediterranean region and show no
dependence on the antecedent conditions in the Alpine-Mediterranean region.
For large Swiss lakes and streams, underline the importance
of high soil saturation due to excessive water supply by enhanced melt and
precipitation over several months for the generation of historical floods.
However, damages in Switzerland often occur when small rivers overflow or when
surface runoff occurs outside of river beds .
The devastating event of 1993 is a memorable example of how a local river can
generate great damages . Local floods in Switzerland
result from a large variety of hydrological processes depending on the region,
floods may be driven by short but intense showers, continuous rainfall, rain
on snow, or snow and/or glacier melt; see.
Defining the influence of antecedent precipitation for this large variety of
flood types is a complex task. A modeling study by showed
that soil saturation can play a paramount role in mediating the discharge
response of a small Pre-Alpine catchment. The initial conditions
also significantly affect flash flood forecasting in the Southern Swiss Alps .
However, found that the impact of initial moisture
conditions on the runoff coefficient during floods is important only for
catchments with intermediate subsurface water storage capacity; i.e., the role
of initial moisture conditions is negligible for catchments with either very
large or very small storage capacity. Also, reports from
on observed floods in mesoscale Alpine
catchments with relatively shallow and permeable soil layers conclude that
“… values of antecedent precipitation do not dramatically
affect the resulting runoff coefficient, at least during major floods. This
indicates a smaller sensitivity to initial soil moisture conditions than
generally assumed …”.
A better understanding and quantification of the role played by antecedent
precipitation in the development of floods is crucial for flood hazard
management for two reasons:
Because future flood frequency changes might depend on the role of
antecedent precipitation. Future changes in precipitation for Switzerland are
still uncertain but general tendencies can be derived from the
projections. In summer, the most important season for Alpine floods, a clear
decrease in mean precipitation (due to drier soils) is expected to be
accompanied by a weak increase in extreme daily precipitation due to
warmer air, see. Thus, depending on whether
short-term or long-term precipitation is more important for floods, flood
frequency might increase or decrease in the future.
Due to the relatively long residence time of water in catchments with significant
moisture storage capacity, information regarding the current moisture state
can help to improve medium-range flood forecasting. Identifying catchments
where the amount of antecedent precipitation is particularly determinant for
floods may help to determine critical regions where an efficient use of that
information is primordial for flood forecasting systems. For example, it is
now possible to derive water storage information from satellite data, and
demonstrate a great potential for warning systems at
weekly to seasonal lead times.
Here, we do not aim to quantify the role of antecedent precipitation by
calculating runoff coefficients like e.g., in ,
, or
. Instead, following the idea of large
sample hydrology e.g.,, we make use of two extensive
networks of rain gauges and river discharge stations to derive robust
statistics from an important number of catchments and events. The underlying
hypothesis is that if a period of antecedent precipitation influences the
amplitude of peak discharges, floods should be significantly more frequent
after wet conditions during that period provided that a sufficient sample of
events is investigated. The following questions are addressed in particular
for different precipitation periods before floods (e.g., 0–1 days, 3–14 days before floods):
In the past 50 years, have floods in Switzerland been significantly
more (or less) frequent after wet conditions during that period?
If they were more frequent, can we define catchment properties that
determine whether and how strongly that period influences flood probability?
Did extreme floods follow wetter antecedent conditions than smaller discharge peaks?
Which precipitation accumulation period is most closely related to flood occurrence?
How many days of antecedent precipitation are relevant for floods?
We aim to explicitly separate short-range and long-range antecedent
precipitation and thus discuss the temporal separation of different
precipitation accumulation periods. The analysis comprises thousands of
annual maximum discharge events in a large sample of catchments
representative of the various hydrological regions of Switzerland. This
analysis is unique for Switzerland with regard to the number of floods
considered and, to our knowledge, also unprecedented worldwide.
Data
The events analyzed in this study are 4257 annual maximum instantaneous
discharge measurements (called floods hereafter). They were recorded at
101 stations during the period 1961 to 2011. The data are provided by the Swiss
Federal Office of the Environment (FOEN)
http://www.bafu.admin.ch/index.html?lang=en
. The
stations measure water level from which a discharge value is obtained through
a rating curve that is based on regular discharge measurements. In the case
of extreme floods, the discharge values have been manually checked and, if
required, have been corrected by hydraulic modeling and expert judgment. All
annual maximum discharge events are denoted HQ hereafter. HQs exceeding the
5-year and the 20-year floods will be denoted HQ5 and HQ20, respectively.
Note that HQs of estimated return periods of more than 100 years have been
recorded in the last decades. Here those floods are simply included in the
HQ20 sample (return period larger than 20 years). The distinction of higher
return periods than 20 years is avoided in order to maintain a large sample
size. Empirical return periods have been used for simplicity. The empirical
return period of a HQ is given by the length of the time series divided by
the rank of the HQ (in decreasing order of discharge).
We use gridded daily precipitation accumulations constructed from
interpolation of a dense network of rain gauges
see. The daily sums (from 06:00 to 06:00 UTC) are
available on a 0.02 by 0.02 degrees grid covering the Swiss territory for the
period 1961–2011 hereafter RhiresD, see. The number of
gauges varies from approximately 400 to 500 throughout this time period. The
effective resolution of the data set, given by the typical inter-station
distance, is approximately 15–20 km. Some of the smallest catchments
investigated here may not contain any rain gauge but the results from
Sect. show that the flood-relevant precipitation is adequately captured
in each catchment.
MethodsSelection and classification of catchments
We selected 101 catchments based on the following criteria:
the discharge time series must cover at least 20 years during the
period 1961–2011;
the catchment must be larger than 10 km2 and its area must
be covered > 90 % by the precipitation data set;
the possible human influence on the HQs must be minimal;
a homogeneous representation of the Swiss territory is ensured and
multiple counting of basins, i.e., small catchments located in larger
catchments, is minimized.
The selected catchments were subdivided according to their size into
microscale catchments (Micro, 10–100 km2), mesoscale catchments (Meso,
100–1000 km2) and macroscale catchments (Macro, > 1000 km2).
Catchments within the same size category never overlap spatially, but Micro
catchments can be contained in Meso and Macro catchments and Meso catchments
in Macro catchments.
Assessment of human influence on peak discharges (e.g., hydropower dams and/or
discharge regulation) requires detailed knowledge about water management in
each catchment. Some of this information is available within the Hydrological
Atlas of Switzerland see table of plate 5.6 from.
Only Micro and Meso catchments with no or low human influence were selected.
Some human influence was tolerated for Macro catchments. Discharge is
regulated at the outlets of the majority of large Swiss lakes and the lake
outlet stations are analyzed separately (hereafter “Lake
Outlets”). Karstic catchments with very complex underground
flow were removed based on expert knowledge.
Swiss river discharge stations selected for this study. Colors refer
to the hydrological regimes in the legend. Stations at lake outlets are shown
by triangles to highlight the strong anthropogenic influence on the discharge
(lake outlets are thus analyzed separately). The numbers refer to
Table which provide brief descriptions of the catchments.
The Swiss landscape contains distinct geographical and hydrological regions:
The Alps (Prealps, High Alps, Southern Alps), the Swiss Plateau and the Jura
Mountains. Each region shows specific hydro-meteorological properties. In
order to account for this diversity, a typical hydrological regime has been
attributed to each Micro and Meso catchment (see Fig. ). This
classification of hydrological regimes follows ; see
also . A first set of separation criteria is the mean
elevation and the glacier coverage. These properties allow us to distinguish
between Glacial (mean altitude > 1900 m and glacial coverage > 6 % or
mean altitude > 2300 m and glacial coverage > 1 %), Nival ( mean
altitude > 1200 m) and Pluvial regimes. The mean annual cycle of the runoff
in Pluvial, Nival, and Glacial catchments is mainly dominated by rain water,
snow melt, and glacier melt, respectively. Then, all catchments from the
southern side of the Alps were joined in a separate group. The specific
precipitation regime and flood seasonality
of this group, as well as the specific geology
(crystalline, poor infiltration rates, steep slopes, and weak soils)
motivated this choice. called this group
“Meridional” to emphasize its southern
location. Similarly, the catchments in the Jura Mountains were joined in the
Jurassien regime type because of their shared specific morphology and geology
(high plateaus, gentle slopes, high infiltration rates and important network
of underground streams due to the calcareous and karstic bedrock).
From Glacial to Nival to Pluvial, the flood seasonality decreases but a
maximum flood frequency in summer is maintained. Meridional catchments are
characterized by a maximum flood frequency in fall and summer and Jurassien
catchments by winter floods with rain on snow as a major flood process
see e.g.,.
In summary, the different catchment subsamples are: Micro (52 catchments),
Meso (35 catchments), Macro (8 catchments), Glacial (19 catchments), Nival
(17 catchments), Pluvial (31 catchments), Meridional (8 catchments),
Jurassien (12 catchments) and Lake Outlets (7 catchments). See Table for a brief description of
each catchment.
Derivation of precipitation time series for each catchment
We identified catchment area boundaries for each discharge station by
applying a purely topography-based approach to a digital elevation model (DEM)
with a 10 m resolution. For most of the Swiss territory, the
effective drainage areas of the stations can be expected to be reasonably
close to the catchments derived from the DEM. Critical regions are the highly
karstic areas in the Jura Mountains and some areas of the Prealps, where the
hydrological and topographical catchments tend to be significantly different
because of the complex underground flow see e.g.,. The
most critical catchments were not considered for the analysis.
The different precipitation accumulation periods (PAPs) used in this
study.
D0–1climatological percentile of the2-daysprecipitation sum0 to 1 daysbefore the flood dayD2–3climatological percentile of the2-daysprecipitation sum2 to 3 daysbefore the flood dayD0–3climatological percentile of the4-daysprecipitation sum0 to 3 daysbefore the flood dayD4–6climatological percentile of the3-daysprecipitation sum4 to 6 daysbefore the flood dayD4–14climatological percentile of the11-daysprecipitation sum4 to 14 daysbefore the flood dayD4–30climatological percentile of the27-daysprecipitation sum4 to 30 daysbefore the flood dayD0–30climatological percentile of the31-daysprecipitation sum0 to 30 daysbefore the flood dayAPI2climatological percentile of theAPI2 daysbefore the flood dayAPI4climatological percentile of theAPI4 daysbefore the flood dayPRE-APall precipitation accumulation periods excluding the last 3 days before the flood day (here D4–6, D4–14, D4–30 and API4)
Area-averaged precipitation time series were obtained by combining the
gridded precipitation data with the topographical catchment areas.
Definition of precipitation periods
The first challenge is to distinguish between event and pre-event
precipitation. Flood triggering precipitation can be in the form of
synoptically driven precipitation (periods lasting between a few hours to
several days when the synoptic situation is particularly conducive to
repeated precipitation events) and/or localized and short lived high
precipitation events (typically convective). Ideally, a flood-by-flood
analysis using a hydrological model should be performed to determine the
exact time lag between the most intense precipitation rate and the discharge
peak and to merge all precipitation events that can be attributed to a
particular synoptic situation, such as the passage of a cyclone. However, a
case-by-case analysis is beyond the scope of this study first because the
daily resolution of the data does not allow for an evaluation of
precipitation rates on sub daily timescales and second because of the very
large number of events considered. Instead, we search for simple indices
(precipitation accumulation periods, PAPs), that will (on average) best
represent the precipitation associated with all floods in Swiss rivers.
A set of PAPs is defined (summarized in Table ). Most PAPs
represent a precipitation sum over a particular period before the flood day
and two more PAPs are based on the concept of antecedent precipitation
indices (API). A detailed description of the PAPs and the motivation for
choosing them is given in Sect. . For example, PAP D4–14 is the
precipitation sum that occurred within the period from 14 to 4 days prior to
the flood day. PAPs are calculated for each day of the catchment-averaged
precipitation time series (not only for flood days). The precipitation sums
corresponding to flood days are then compared to the climatological
distribution of all precipitation sums. The climatological sample is defined
by a 3-month moving window centered on each day of the calendar year. For
example, let us assume that a flood occurred on 1 June 2000. The
D4–14 of that day is compared to all 11-day precipitation accumulations
between 17 April and 16 July from 1979 to 2011 and the respective percentile
of D4–14 is calculated. For each flood event we can thus determine the
percentile value for each PAP. A 3-month moving window is an optimal
compromise between minimizing the effects of precipitation seasonality and
maximizing the climatological sample size (91 days per year times 20–50 years
means that each value is compared to 1820–4550 other values).
Beside the simple precipitation sums, more complex indices for antecedent
precipitation, i.e., APIs are used. APIs have been commonly used in hydrology
for decades see e.g.,. We follow the method
of :
APIi=Pi+KPi-1+K2Pi-2+…+KnPi-n,
where P is the daily precipitation sum, i is the day for which API is
calculated, K is the decay factor, and n+ 1 is the number of days since
measurements began. Here, a constant K value of 0.8 is used for all
catchments. The decay factor K is a proxy for diverse water fluxes that lead
to a reduction of the water stored in a catchment. In this study, a decay
rate of 20 % per day, i.e., K= 0.8, is chosen and reflects roughly typical
conditions in Switzerland . Results are
insensitive to a tested range of K between 0.7 and 0.9. We use the indices
API2 and API4 that include all days of the time series up to 2 and 4 days
before the flood day (hereafter also called PAPs).
Logistic regression
The underlying hypothesis of this study is that, if a PAP is important for
flood generation, a significant signal can be detected using the logistic
regression. A lack of significance on the other hand, implies either that the
PAP has no influence on flood probability or that this influence is too weak
to be significant during the investigated period.
In Sect. we assess the importance of the different PAPs for peak
discharge generation at each catchment. A test is performed for each
catchment and each PAP separately using a logistic regression model.
Binary daily time series of floods y(t) and precipitation PAPT(t)
are calculated. The time series contain approximately 7000 to 18 000 days t.
For days when floods were recorded y(t)= 1 and y(t)= 0 for all other days.
For days when the PAP exceeded a given percentile threshold T
PAPT(t)= 1 and PAPT(t)= 0 for all other days. The model is
then fitted as follows:
logit(p(t))=β0+β1PAPT(t),
where logit(x) =log(x/(1 -x)), and p(t) is the probability of
observing a flood at day t given the predictor, i.e., p(t):=P(y(t)= 1|PAPT(t)).
We are particularly interested in the value of β1. The odds ratio
(O=exp(β1)) is a measure for the increase (or decrease if O is
below 1) of the odds, p/(1 -p), of a flood occurring when the PAP
exceeds percentile T. Here, p is by definition small (we look at yearly
discharge maxima and even rarer events) and we can therefore set p/(1 -p) ≈p
and the odds ratio can thus be understood as a multiplicative
factor for the flood probability p. Statistical testing can assess the
significance of the predictor PAPT.
A significant p value implies that “the exceedance of a
given precipitation threshold significantly changes the flood probability”.
Note that working with binary predictors is not mandatory in logistic
regression. Here this choice offers the advantage of avoiding the assumption
that logit(p) is proportional to the percentile of the precipitation
period; an assumption for which no particular argument could be found. A
drawback is however that the regression can only be performed with predefined
thresholds. Here, the logistic regressions are tested for five different
thresholds (P50, P75, P90, P95, P99) and the p value of the most significant
test is selected (the corresponding thresholds and odd ratios are not discussed).
Results
Hereafter, we will use percentiles to describe precipitation quantities. To
simplify the language, we define a set of expressions (see Table ).
Defining different precipitation periods preceding Swiss floods
In order to determine the optimal separation of precipitation periods for the
sample of events considered, the precipitation distribution is first
investigated day by day. Figure a shows the distributions of
daily precipitation sums for every day prior to and after all floods. For
example, the boxplot at x=-10 represents the distribution of precipitation
sums recorded 10 days before all floods (4257 values of daily precipitation
recorded 10 days prior to the 4257 flood days). Moderate to high
precipitation is most often recorded 1 day before floods when the
80th local seasonal percentile is exceeded in 75 % of the cases and the median
precipitation sum corresponds to the 98th climatological percentile. During
flood days, the median precipitation only amounts to percentile 93. The
days -2 and -3 also show high precipitation sums with medians amounting to
climatological percentiles 75 and 60, respectively. From day -4 backwards,
the precipitation distribution is very close to climatology, although it
tends to be slightly enhanced up to 10 to 15 days before floods. Similar
results are observed when subsamples of catchments are analyzed (Fig. b–d).
The maximum median daily precipitation is recorded 0–1 days
before HQ days at Micro catchments and 1–2 days before HQ days at Lake
Outlets. A clearly enhanced median precipitation prior to 4 days before HQ
days is only found at Lake Outlets.
Expressions used to define different quantities of precipitation.
The distribution of daily precipitation before and after all flood
events is shown in (a). For example, the boxplot at x=-10 represents the
distribution of daily precipitation percentiles 10 days prior to the
4257 annual flood events analyzed in this study (all HQs from all catchments). The
middle line of the boxplots shows the median, the boxes comprise the 25–75 percentile
range, and the whiskers end at a deviation from the mean of 1.5
the interquartile range. (b)–(d) Same as (a) but for floods in Micro
catchments, Macro catchments and Lake Outlets. (e) The same procedure
as in (a), but applied to annual maximum precipitation days instead
of annual flood days.
Daily precipitation sums correspond to the 06:00 to 06:00 UTC accumulations and
are therefore shifted by 5 h compared to discharge peaks recorded on
calendar days. This partly explains the 1-day shift between maximum
precipitation and HQ occurrence, especially for the floods in Micro
catchments. The response time of catchments, i.e., the time between
precipitation and registration of the related runoff at the gauge, plays a
role as well. We therefore group the flood days and the preceding days
together (hereafter the PAP called D0–1; see also Table ).
This is the time range when high precipitation quantities are most likely. As
shown in Fig. b–c, this assumption is valid for Micro and Macro
catchments whereas for Lake Outlets the highest precipitation occurs 2 days
before floods (because of longer response times due to lake retention).
Intense precipitation events responsible for flood peaks might be very short
(hours or minutes in the case of flash floods) but the daily resolution of
the data and the shift between precipitation and floods does not allow for a
further separation of the time windows.
Summary of catchment properties for the selected stations. Catchments are sorted based on hydrological regime
and increasing size from top to bottom. Locations are given in Swiss coordinates (CH1903).
Absolute values of the climatological percentiles for the different
PAPs. Statistics from Macro (a–c) and Micro (d–f) catchments are shown on the
top and bottom row, respectively. Accumulations over 2 days which correspond
to the PAPs D0–1 or D2–3 are shown in (a, d). Accumulations over 11 days
corresponding to D4–14 are shown in (b, e). APIs are shown in (c, f). Variation
between catchments is visualized in boxplots.
Overview of all flood events. All river discharge stations (numbers
on the y axis, see Table ) cover at least 20 years in
the 1961–2011 period. For each annual discharge peak, the return period of
the 2-day precipitation sum (D0–1) is indicated by colors. HQ5s and HQ20s
are marked with squares and triangles, respectively. The catchments are
sorted by regime type and by increasing size from top to bottom. Hydrological
regimes are indicated by colors: blue = Glacial, cyan = Nival, green = Pluvial,
orange = Jurassien, red = Meridional, magenta = Macro, brown = Lake Outlets.
Precipitation 2 to 3 days before floods is also greater than climatology in
all catchments and, interestingly, precipitation remains also greater than
climatology 2 days after floods in Fig. a. An explanation for
this phenomenon can be found in Fig. e, which shows the results
of an analysis similar to the one of Fig. a but applied to
maximum precipitation days instead of flood days. In Fig. e, the
precipitation distribution is similarly enhanced ±2 days around high
precipitation events like it is enhanced around flood events. The typical
timescale of precipitating weather systems over Europe leads to some
persistence of the daily weather situations so that daily precipitation time
series are autocorrelated. Figure a thus highlights a time
window centered between day -1 and day 0 and ranging from day -3 to day +2
when precipitation is clearly higher than usual. We identify it as the time
range when the flood-producing weather situations generate high
precipitation. Two more PAPs are thus defined which range back to 3 days
before floods in order to capture precipitation associated with
longer-lasting weather events (periods D0–3 and D2–3). The
“precursor antecedent precipitation” (PRE-AP) is subsequently
defined as the period finishing 4 days before floods. PAPs representing
PRE-AP are D4–6, D4–14 and D4–30. To complete the set of PAPs, a similar
separation is also applied to APIs (see API2 and API4, stopped 2 and 4 days
before floods, respectively). Hereafter, the analysis is based on seasonal
percentiles of the PAPs. For comparison, precipitation sums [mm]
corresponding to percentiles of different PAPs are shown in Fig. .
For example, the P99.9 of D0–1 in summer is summarized for all
Macro catchments by the rightmost orange boxplot in Fig. a. The
P99.9 exceeds 94 mm for 50 % of the Macro catchments and reaches 156 mm at
one catchment. The P99.9 of D0–1 at Macro catchments is in general lower in
winter than in summer (compare the orange and the blue boxplot). Note that
API2 and API4 result from the same calculation (see Eq. 1) applied at
different days i. Their climatology is therefore the same and Fig. c
and f are valid for both API2 and API4.
In hydrology, “antecedent precipitation”
typically implies all the precipitation preceding the very last
flood-triggering event. Here we separate flood-preceding precipitation into
the short-range antecedent precipitation and what we define as the precursor
antecedent precipitation PRE-AP. Although this sharp separation (between
days -3 and -4) is only based on averaged statistics and although flood-triggering
events can be defined over a wide range of timescales; we choose this simple
formulation to distinguish explicitly long-range antecedent precipitation
from a period when unusual precipitation is obvious in rainfall time series.
We strongly emphasize that hereafter PRE-AP excludes the last 3 days before
floods (see Table ).
Overview of the precipitation associated with Swiss floods
We start the analysis with an overview of the variability of the
precipitation associated with Swiss floods (event and pre-event precipitation).
The 2-day precipitation
Figure shows the 2-day PAP (D0–1) associated with each annual
maximum discharge (HQ) of each catchment. The return periods of D0–1 vary by
several orders of magnitude between different events. Very high precipitation
(with a return period longer than 100 days) is frequently associated with
floods, but a majority of catchments also experience HQs during low or
moderate precipitation. A return period of D0–1 shorter than 10 days
corresponds to a percentile lower than 90 and thus to less than 20–30 mm in
2 days (see Fig. a and d). There are more floods without high D0–1 in
Nival and Glacial regimes as compared to the Pluvial regime. The D0–1 in
Jurassien and Meridional groups is comparable to the Pluvial group. D0–1 is
slightly lower in Macro catchments and clearly the weakest for Lake Outlets.
HQ5s and HQ20s tend to be associated with longer return periods of D0–1 than
HQs, although they can also be triggered by weak or moderate precipitation
(return periods shorter than 10 days), especially at Lake Outlets, as well as
in Glacial and Nival catchments. Interestingly, extreme D0–1s often occur
simultaneously in several catchments, indicating widespread events. Most of
them correspond to extraordinary flood events in 1978, 1987, 1990, 1999,
2002, 2005, and 2007 and involve several HQ20s.
Precursor antecedent precipitation
Figure is similar to Fig. but shows the PAP
D4–14, i.e., the accumulated precipitation between day -4 and -14
(PRE-AP). The large majority of floods are associated with return periods of
PRE-AP shorter than 10 days, i.e., not unusually wet. In general HQ5s and
HQ20s are not associated with higher PRE-AP than HQs and the rare cases of
unusually wet PRE-AP typically occur simultaneously at many catchments (like
in 1972, 1993, 1999 and 2006).
The logarithmic scale of return periods in Figs. and
underlines the fact that return periods of D4–14 are several
orders of magnitude shorter than those of D0–1. However, one cannot expect
D4–14 to be systematically extreme as this 11-day period often excludes the
heavy precipitation (which happens just before the flood).
Quantification of the precipitation during different periods preceding Swiss floods
The overview of flood-precipitation in the last 50 years
revealed that precipitation during PAP D0–1 was high or extreme for a
majority of floods but PRE-AP (during PAP D4–14) was not. This raises the
question of whether D4–14, although not extreme before floods, still tends to
be wetter than climatology.
Figure shows the distribution of PAPs for different
flood samples (deviations from climatology significant at the 99 % level are
outside of the gray zones). The gray zones are based on binomial
distributions and represent the 99 % level of significance of the variations
of relative frequency in case of independent events. In the case
investigated, the independence of events cannot be assessed in a purely
quantitative way but the flood events are likely dependent, i.e., there are
more simultaneous flood occurrences than expected from a random process,
because floods in neighboring catchments can be triggered by the same weather
event. The significance shown is hence likely too high (the zones too small)
but the gray zones are still drawn as indicators of the minimum amount of
random noise that can be expected. Note that it is strongly dependent on the
sample size, i.e., on the number of flood events.
For HQ5s in Micro catchments (Fig. a), precipitation
during D0–1 was very high (higher than P99) for 61 % of the floods and high
(higher than P90) for 90 % of the events. Only 10 % of the floods were
preceded by no or moderate precipitation (lower than P90). For D2–3, high and
very high precipitation was also significantly more frequent than usual
although the deviation from climatology is very weak compared to D0–1. Drier
percentiles of D2–3 were also significantly less frequent than usual (only
35 % of the cases are below P50). On the other hand, no significant departure
from climatology is found for the PRE-AP PAPs (D4–6, D4–14, D4–30). This
means that, as a general rule, the conditions were not significantly wetter
than usual earlier than 3 days before floods in Micro catchments.
The statistics of Meso and Macro catchments (Fig. b–c)
resemble the ones of Micro catchments.
In contrast, HQ5s at Lake Outlets (Fig. d) were triggered
by significantly higher than usual precipitation during all PAPs (and not
only during D0–1 and D2–3). For example, a percentile of D4–14 higher than 99
is as frequently observed as a percentile lower than 50.
Figure e–f show the results for HQs and HQ20s in all
catchments. During D0–1, very high precipitation is twice as frequent prior
to HQ20s (80 % of all floods) as it is prior to all annual HQs (45 % of all
floods). However, the precipitation prior to HQs and HQ20s is surprisingly
similar during the other periods (D2–3 is only slightly higher for HQ20s than
for HQs and PRE-AP is basically the same).
In summary, the flood events considered in this study, with the exception of
Lake Outlets floods, frequently co-occur with high precipitation during the
flood day and/or the day before (D0–1). Longer-lasting multi-day events also
generate high precipitation during D2–3. The slightly larger departure from
climatology during D2–3 at Macro compared to Micro catchments indicates a
higher importance of longer-lasting events.
already showed that larger catchments are more sensitive to longer-lasting
precipitation at the sub-daily scale; here we can extend those findings to
multi-day events. Regarding precipitation 4 or more days before HQ days, a
significantly enhanced frequency of wet weeks is only found for Lake Outlets.
For other catchments, floods did not happen after significantly wetter nor
drier PRE-AP in general.
Although no significant signal is found, PRE-AP was nevertheless slightly
wetter than climatology before floods in Switzerland. Consequently, more
detailed analyses are presented in the next sections to explore the
correlation between PRE-AP and floods for particular catchments, particular
flood types, and particular flood seasons.
Same as Fig. but for PRE-AP (D4–14).
Relative frequency of precipitation percentiles for several PAPs
before floods. Each colored line represents a PAP. (a)–(d) HQ5s in (a) Micro
catchments, (b) Meso catchments, (c) Macro catchments and (d) Lake Outlets
catchments. (e) All HQs and (f) HQ20s in all catchments. Gray shadings
represent the 99 % level of significance of the frequency of each percentile bin.
The relevance of the different precipitation periods for the
occurrence of annual floods is tested using logistic regression for each
precipitation period and each catchment (a) D0–1, (b) D2–3, (c) D4–6,
(d) D4–14, (e) D4–30, (f) D0–30, (g) API2, and (h) API4. Several thresholds are
tested (P50, P75, P90, P95, P99) and the most significant P value is
displayed symbolically (squares, dots and triangles indicate a non-, weakly-,
and strongly significant influence, respectively). The colors of the symbols
refer to the hydrological regimes of the catchments. Circles denote a
negatively significant correlation, i.e., the exceedance of a given
precipitation threshold significantly reduces flood probability.
Catchment by catchment analysis
Here, we use logistic regression to address the following question for each
PAP and each catchment: is the occurrence of HQs influenced by the amount of
precipitation? Or in other words: are floods more (or less) frequent after
wet periods? We thereby aim to investigate whether the large variety of Swiss
basins is associated with different flood responses to PAPs. Previous studies
showed that typical flood-triggering precipitation depends not only on
catchment size (investigated in the previous section), but also on various
catchment properties e.g.,.
Potentially important properties include mean elevation, slope, land cover,
soil type, geology and reservoirs (lakes, underground cavities). The
hydrological regimes encompass some of this variability and serve as a
framework for interpreting the following analysis.
Figure shows the results of the logistic regression for
the different PAPs (see details in Sect. ). For example,
triangles (P value < 0.001) in Fig. a indicate that, in
every catchment investigated, floods were significantly more frequent when a
particular threshold of D0–1 was exceeded. In other words, the amount of
precipitation that falls during D0–1 has a significant impact on flood
frequency. The amount of precipitation that falls during D2–3
(Fig. b) also significantly impacts the flood frequency in most
catchments, with the exception of most Glacial and few Nival and Pluvial
catchments. With regard to PRE-AP in D4–6, D4–14 and D4–30
(Fig. c–e), clear regional patterns can be distinguished. Wet
antecedent periods significantly enhance the flood frequency mainly in the
northwest and northeast Switzerland, as well as at the outlet of all lakes
except Lake Thun (no. 111). In contrast, floods were significantly less
frequent after wet periods in some Glacial catchments. Indeed, six catchments
show a significant P value with an odd ratio smaller than 1 for D4–14. These
are the exact six catchments with more than 25% glacial coverage. For the rest
of Switzerland, the amount of PRE-AP does not significantly affect the flood
probability. By comparing the results of D0–30 with D4–30, it emerges that
floods are significantly associated with wet months (D0–30) in a large
majority of catchments only because heavy precipitation 3–4 days before
floods leads to high monthly accumulations. Indeed, D4–30 indicates that
precipitation during the rest of the month has no significant impact on the
flood probability for most catchments.
A reduced flood frequency following wet periods (like found for the glacial
catchments) seems counterintuitive. The most significant negative correlation
is found for the most glaciated catchment (the Aletsch glacier catchment, no. 865).
The highest significance is obtained in this case with the threshold
P75 because none of the 51 HQs recorded correspond to the 25 % wettest D4–14.
The expected value is 51/4; i.e., approximately 12–13 HQs. It is almost
impossible to get 0 HQs just by chance and an explanation must therefore be
found. Glacial catchments are typically small and located at high elevations,
exhibit steep slopes and lack deep soils. They are characterized by very
short response times and a large runoff contribution from melt during the
flood season summer, see e.g.,.
The negative correlation is probably due to the fact that prolonged periods
of wet weather (lower temperature, reduced sunshine and hence reduced melt)
can lead to a lower baseflow in those catchments so that contributions from
short and intense precipitation events would be less likely to generate
annual discharge peaks. Indeed, discharge time series of glacial catchments
are typically characterized by a pronounced diurnal cycle in summer,
revealing the importance of high temperature and sunshine for melt and
discharge generation. The baseflow continuously rises from day to day in case
of extended periods of nice weather which are therefore particularly
conducive to floods. Hence, floods are less frequent after precipitation at
Glacial catchments, probably because of the reduced glacier melt.
Enhanced flood frequency after wet periods is less surprising. The Swiss
Plateau, especially the western part, is a relatively flat area characterized
by deep soils that need to be saturated before large runoff in the main
streams is recorded. Soils in the Jura are typically thinner but very
permeable and this region is well known for its underground karstic cavities.
A karstic underground network can contain important reservoirs, the water
level of which influences the flow response in surface streams see e.g.,.
In summary, the role of long-term antecedent precipitation for flood
generation depends strongly on the region and/or on the hydrological regime
considered. Wet PRE-AP periods enhance HQ probability where soil saturation
and reservoir filling are important processes and decrease HQ probability
where melt water is an important contributor to the floods discharges.
Antecedent precipitation indices (APIs)
We also tested the power of APIs (see Table ) for
statistically predicting floods as compared to simple precipitation sums.
API2, like D2–3, omits information about the flood day and the day preceding
the flood but accounts for the whole antecedent precipitation instead of for
only 2 days. The results for both periods are similar in most catchments.
D2–3 is a better (more significant) flood predictor than API2 for
12 catchments, and a weaker predictor for 11 catchments. API2 allows us to
distinguish the relevance of dry periods for flooding in Glacial catchments
but D2–3 is too short and too close to the flood to capture this signal.
However, combining D2–3 and D4–6 indicates that dry conditions followed by
wet conditions are important for flood formation in the Lütschine in
Gsteig (no. 387), for example. Both periods cancel out in API2 and no
significant signal is found. Searching for the best period also appears to be
complex with regard to PRE-AP. Each of the four periods (D4–6, D4–14, D4–30,
API4) is the most significant flood predictor at several catchments. D4–30 is
rarely the best predictor, indicating that the precipitation sum over a
monthly period is not a powerful measure for flood probability. API4 is
slightly more often a better measure than D4–6 and D4–14, although this is
not systematic. APIs are widely used in hydrology see
e.g., but our
integrative study cannot confirm that they explain flood frequency better
than simple precipitation sums.
Impacts of short-range precipitation and PRE-AP on flood magnitude
In the previous sections, the impact of PAPs on HQ probability was discussed
(i.e., whether floods are more frequent after wet periods). Here, the impact
on the flood magnitude is investigated as well (i.e., whether larger floods
follow wetter periods than smaller floods).
In Fig. , the flood-associated precipitation is simply summarized by the
median return period of the PAPs for a flood sample. This allows us to
compare various flood samples (different flood magnitudes, different
catchment groups, different flood seasons). Assuming that the precipitation
distribution is equal to climatology before floods, the median return period
should be equal to 2 days (delimited by solid lines in the graphs).
For the Micro, Meso and Macro catchments in Fig. a, larger
floods correspond to higher D0–1 than smaller floods (HQ20s are associated
with a median return period of D0–1 of 400–1000 days = 1–3 years while HQ1s
correspond to a median D0–1 of only 60 days). In contrast, HQ20s are related
to clearly higher D2–3 only at Macro catchments. At those catchments, as much
precipitation falls 2 to 3 days before the HQ20s as falls 0 to 1 days before
all HQs. At Lake Outlets, D2–3 is more extreme than D0–1 because of the long
time delay between precipitation and gauged discharge (see Sect. ).
Figure a can be directly compared to Fig. b.
For Micro, Meso and Macro catchments, the return periods of D0–3 in
Fig. b are similar to the ones of D0–1. On the other hand, the
median PRE-AP is remarkably close to normal for each catchment size (close to
the climatological median). Moreover, the PRE-AP was not higher before HQ20s
than before HQ1s. A change in PRE-AP with flood magnitude is only found at
Lake Outlets.
Figure c–f investigates different hydrological regimes and
different flood seasons. For no regime and no season is the amount of PRE-AP
precipitation linked to the flood amplitude. Even at Jurassien catchments,
where we found that floods are significantly more frequent after wet periods,
HQ20s are not associated with wetter periods than HQ1s.
Median return periods of flood-associated precipitation for different flood samples. The rows show
different catchment sizes (a, b), different hydrological regimes (c, d) and
different flood seasons (e, f). The left column shows D0–1 in x and D2–3 in
y and the right column D0–3 in x and D4–14 in y. The numbers 1, 5
and 20 indicate median return periods associated with all HQs, with all HQ5s, and
with all HQ20s, respectively. They are joined together by a line.
Flood-associated precipitation for different catchment samples: (a) Glacial, (b) Nival,
(c) Pluvial, (d) Jurassien, (e) Meridional, (f) Macro and (g) Lake Outlets. For each
discharge peak, D0–3 is shown in x and D4–14 in y. Annual floods are
shown by gray dots (shadings indicate the density of dots), HQ5s by green
dots and HQ20s by red triangles. Green lines show the linear regression of
the HQ5s.
Can weaker precipitation trigger floods if PRE-AP is higher?
In the previous sections, the PAPs were investigated separately. Here we show
the combinations of PRE-AP and short-range precipitation events for single
floods. If the runoff coefficient is enhanced by wetter PRE-AP (and thus more
saturated soils), floods might happen in association with weaker triggering events.
Figure shows D0–3 and D4–14 of all flood events for different
catchment samples. As already inferred from Fig. , precipitation
accumulations before floods vary remarkably between single events and the
portion of floods lacking high triggering precipitation is highest in Glacial
and Nival catchments. The green lines in Fig. show the linear
regression between D0–3 and D4–14 for HQ5 events (only HQ5s are shown for
clarity). The regression lines address the following question: did wet
periods of PRE-AP allow weaker weather events to generate HQ5s? Indeed, it
seems that for the Jurassien, Meridional and Lake Outlets catchments, HQ5s
that were triggered by weaker weather events tend to be associated with
higher values of PRE-AP. This is in contrast to Glacial catchments where
weaker events trigger HQ5s after drier periods. Regarding flood forecasting,
it would be interesting to define which minimum threshold of event
precipitation is required to trigger a HQ5 given that PRE-AP is known,
similarly to the flash flood guidance (FFG) approach see
e.g.,. The scatter in observations shows that defining such a
threshold is impossible for Switzerland because floods can occur in
association with all types of precipitation. The only flood sample for which
such a threshold would be realistic is the set of HQ20s at Lake Outlets.
There, a HQ20 occurred without precipitation in the last 3 days but after an
exceptionally wet period of PRE-AP. In contrast, all HQ20s occurring after
not unusually wet periods of PRE-AP required at least a D0–3 of return period
of 100 days. There might be a minimum threshold of D0–3 for HQ20s in Macro
and Meridional catchments as well but it does not seem to depend on PRE-AP.
The lack of a minimum threshold of D0–3 for floods is probably due to the
very simple definition of precipitation used here and to the fact that the
precipitation thresholds vary between catchments. Finer and
catchment-specific approaches see e.g., are
required to formulate an FFG system for the catchments considered.
Discussion
A synoptic and statistical approach is used to separate event precipitation
and antecedent precipitation for several thousands of floods. We define weekly
to monthly precipitation periods preceding floods by more than 3 days
“PRE-AP” (PREcursor Antecedent Precipitation) periods. Flood-triggering
events are distinguished by D0–1, D2–3 and D0–3.
The relation between flood occurrence and the precipitation amount during D0–1
is stronger for Pluvial catchments than for Nival and Glacial catchments. We
attribute this observation to the fact that rain-on-snow events are more
common in Nival and Glacial catchments. During such events, the
transformation of precipitation into runoff is strongly influenced by the
presence of a snow cover through snow melt and complex snowpack runoff
dynamics see e.g.,. The Nival and Glacial
catchments are also at higher altitudes and typically smaller than Pluvial
catchments. They consequently react to shorter and more intense precipitation
events which do not necessarily correspond to high 2-days sums.
We attribute the weak relationship between the precipitation amount during
D0–1 and the occurrence of floods at lake outlets to the relatively strong
influence of the PRE-AP. PRE-AP is indeed significantly related to flood
occurrences at these catchments. This is most probably due to the large
reservoir capacities of the lakes; i.e., the lakes must first be filled before
floods can be recorded at their outlets.
The majority of the lake outlets is regulated. Small HQs after wet PRE-AP may
be triggered by the lake regulation itself (if the gates are opened after
long periods of precipitation resulting in high lake levels). However, we
expect the extreme discharge peaks after wet PRE-AP to be damped due to the
lake regulation. Despite the lake regulation, HQ20s at lake outlets are the
floods that are proportionally the most frequent after wet PRE-AP. Lake
regulation is often a compromise between the need to protect settlements
adjacent to the lake but also the downstream areas; its effect on extreme
floods is thus complex.
While PRE-AP is important at lake outlets, it is only weakly linked to flood
probability at the other catchments and its influence is region-specific:
(i) annual floods are significantly more frequent after wet PRE-AP periods in
most Jurassien catchments, in some Pluvial catchments of northwest and
northeast Switzerland, and at lake outlets. (ii) Annual floods are
significantly less frequent after wet PRE-AP periods in glacial catchments.
(iii) The amount of PRE-AP is not significantly related to the occurrence of
annual floods in the rest (the majority) of Swiss catchments. The fact that
PRE-AP is only weakly related to floods compared to D0–1 or D0–3 is not
astonishing. Indeed, we expected the highest precipitation amounts to fall
during and just before the flood days, rather than 4 to 30 days before.
More unexpected is the fact that more precipitation during PRE-AP is, in the
majority of catchments, not related to a significantly higher flood
probability, nor to a higher flood amplitude. For most catchments, floods and
precipitation amounts are not significantly related if we ignore
precipitation during the last 4 days. This observation may be most
convincingly reflected by Fig. b which shows that the median
PRE-AP of HQ20s is very close to the climatological median (except at lake
outlets). The idea that the flood risk remains enhanced for several days
after long periods of precipitation is strongly anchored in the general
perception. The influence of soil saturation on runoff formation is indeed
well established. Models showed that for the same triggering precipitation
event, variations in antecedent moisture can lead to strong differences in
discharge see e.g.,.
Also, artificial rainfall experiments showed that the runoff coefficient
changes strongly with the amount of antecedent precipitation for various soil
types in Switzerland e.g.,.
Moreover, weekly to monthly precipitation anomalies have been described as
important factors for the development of extreme European floods see
e.g.,.
Contrastingly, our results show that, in the majority of Swiss catchments and
for the period investigated, flood days are not significantly different than
other days regarding the amount of precipitation that fell earlier than 3 days before.
Our findings are, however, not in contradiction with the studies cited above.
First, we find that the role of PRE-AP is very dependent on the hydrological
regime of the catchments so that the absence of significant relationship
between PRE-AP and flood frequency/magnitude is specific to the Swiss
Pre-Alpine, Alpine (except glaciers) and southern Alpine catchments. Second,
several limitations inherent to the statistical experiment must be considered
in order to correctly appreciate the results.
The statistical results do not mean that the runoff coefficient is
independent of the amount of PRE-AP. Our analysis simply shows that this
dependence is too weak to generate a significant signal when 20–50 floods per
catchment are investigated. We nevertheless expect to be on the safe side
when stating that PRE-AP has no significant influence on the flood occurrence
at a particular catchment. Indeed, we performed 5 tests for each catchment
and each PAP (we tested if the exceedance of the P50, P75, P90, P95 or P99
of the PAP significantly changes the flood probability). Significance was
established even if only one of these five tests lead to a flood probability
change with a P value of 5 %.
Antecedent precipitation is not antecedent moisture. Extending the results to
the role of antecedent moisture would require to use land surface models
and/or extensive observations of soil moisture and ground water. This is
beyond the scope of our study given the large number of events considered. We
thus must emphasize that our results are limited to the role of antecedent
precipitation amounts and that the moisture state may better represent the
disposition of a catchment to generate discharge peaks, especially at the
timescale covered by PRE-AP.
The small-scale temporal and spatial distribution of precipitation is an
important determinant of the runoff coefficients of some catchments
e.g.,. Precipitation events can be very local and
imply rapidly varying rainfall rates. Some short and/or localized
precipitation events can thus be smoothed out or missed in the daily- and
point measurement-based precipitation data set used here. The PAPs are with
this regard very coarse representations of real precipitation events. While
this limitation prevents us from describing the sub-daily flood-triggering
precipitation characteristics, it is unlikely to impact the main findings of
our study; namely the role of PRE-AP.
Finally, the PAPs have a constant formulation for all catchments, regardless
of their diverse sizes and hydrological regimes. This limitation is inherent
to the nature of the experiment. The consideration of more than 100
catchments and several thousands of discharge peaks limits obviously the
possibilities of refinement. A catchment-specific formulation of the PAPs and
the APIs (a calibration of the K factor in Eq. () for e.g.) would
allow for a finer distinction of the triggering events and the antecedent
precipitation. Such a refinement would however require to determine typical
response times for all catchments. Moreover, a dynamical formulation of PAPs
and APIs would reduce the possibilities of comparing different catchment
types. Instead, a strict and simple formulation of PAPs like the one used
here maintains the experiment to an affordable level of complexity. This is
in our opinion primordial when investigating very large samples.
Thanks to its relative simplicity, the method developed here can easily be
used anywhere on the globe provided that extensive observations are
available. Minimum requirements are multidecadal observations of discharge
peaks and daily precipitation, as well as an accurate digital elevation
model. The precipitation information may be the most critical to retrieve and
potentially useful data sets must guarantee a sufficient homogeneity in space
and time as well as a sufficient space resolution and coverage. The recent
daily precipitation data set from offers an
interesting opportunity to extend the method developed here to the whole
Alpine range. The high station density of the data set should also allow the
analysis of Meso- to Micro-scale catchments. Over areas of sparse rain gauges
networks, satellite or satellite-gauge daily precipitation climatologies may
alternatively be used see e.g.,.
Conclusions
We quantify statistically the influence of different precipitation periods
for the generation of thousands of annual floods in Switzerland. In contrast
to previous studies that define antecedent precipitation as all the water
that fell before the very last flood-triggering precipitation event, we
explicitly separate antecedent precipitation into the short-range and
long-range antecedent precipitation based on the autocorrelation of daily
precipitation time series and reflecting the synoptic timescale. The
short-range encompasses the 0–3 days period before floods and the long-range
the earlier period (called PRE-AP). This novel distinction allows to
specifically address the role of several antecedent precipitation periods for
flood generation.
At the short range, we do not separate antecedent precipitation from the
precipitation event directly triggering the discharge peak. Instead, we
consider accumulations over several days and address the following question:
over which preceding period is the amount of precipitation related to flood
frequency and flood magnitude?
The 2-day sum (0–1 days before floods) is clearly the best correlated with
both the flood frequency and the flood magnitude. The precipitation 2 to
3 days before floods also significantly affects flood frequency everywhere
except in the high Alps. It is moreover related to flood magnitude at lake
outlets and in large catchments. Regarding earlier periods however, we find
that PRE-AP has had no significant impact on flood frequency for the majority
of Swiss catchments in the last 50 years. Moreover, the magnitude of floods
was also independent of the magnitude of PRE-AP in all catchment types except
at lake outlets. The influence of PRE-AP is thus weak overall. We thus
suggest that researchers focus on 2 to 4 days precipitation periods when
reconstructing antecedent precipitation of past Alpine floods or when
inferring future Alpine flood risk from climate projections. Long-range
antecedent precipitation periods preceding the last 3 days before floods
are in contrast only relevant in the Jura Mountains, in the western and
eastern Swiss Plateau, as well as at lake outlets. The results presented here
may thus also motivate particular efforts to take benefit from information
about the antecedent precipitation for flood warning in areas where
antecedent precipitation significantly influences flood probability, given
that these areas are not covered by more sophisticated deterministic flood
warning systems.
Our findings are derived from extensive observations and can be expected to
be robust and representative of the various flood types encountered in the
Swiss territory. Although our results are specific to Swiss catchments, the
method presented here could be applied to other regions given that sufficient
data are available.
The large differences in return periods of precipitation prior to floods of a
similar magnitude indicate that catchment-averaged daily precipitation sums
only explain a limited part of the flood variability. Future work is required
to better characterize the short flood-triggering precipitation events at an
hourly and a kilometer scale. The advent of a new gridded precipitation data set
at an hourly resolution (combining rain gauges and radar) will offer new
potential with this regard although the use of radar data to achieve this
goal limits the time coverage to the 21st century. This analysis may also be
further expanded by including information about snow line, snow cover and
soil moisture.
Acknowledgements
The authors gratefully acknowledge the Swiss Federal Office of the
Environment (FOEN) for monitoring Swiss rivers, post-processing extreme
discharge values and contributing to this study by providing data on annual
maximum discharge events. This study would not have been possible without the
high-end interpolation of rain gauges made available by the Swiss Federal
Office of Meteorology (MeteoSwiss). We are thankful to Claudia Brauer and
Massimiliano Zappa for their pertinent and constructive comments.
Edited by: A. Weerts
ReferencesAschwanden, H. and Spreafico, M.: Flood Discharge – Analysis of
Long-Standing Measurement Series, Hydrological Atlas of Switzerland,
http://hades.unibe.ch/en/products/druckausgabe/gewaesser/tafe_l5_06
(last access: 29 January 2015), 1995.
Aschwanden, H. and Weingartner, R.: Die Abflussregimes der Schweiz, Publikation Gewässerkunde
Nr. 65, Geographisches Institut der Universität Bern, Bern, 237 pp., 1985.Baillifard, F., Jaboyedoff, M., and Sartori, M.: Rockfall hazard mapping along
a mountainous road in Switzerland using a GIS-based parameter rating approach,
Nat. Hazards Earth Syst. Sci., 3, 435–442, 10.5194/nhess-3-435-2003, 2003.
Ball, C. and Martin, J.: Spring Focus on Sustainability and the Environment:
Effects of Antecedent Hydrogeologic Conditions on Flood Magnitude and
Recharge to the Floridan Aquifer in North-Central Florida, J.
Undergrad. Res., 13, 10 pp., 2012.
Barton, Y., Giannakaki, P., von Waldow, H., Chevalier, C., Pfahl, S., and
Martius, O.: Clustering of Extreme Precipitation Events in Switzerland,
Mon. Weather Rev., in review, 2014.Berthet, L., Andréassian, V., Perrin, C., and Javelle, P.: How crucial is it
to account for the antecedent moisture conditions in flood forecasting?
Comparison of event-based and continuous approaches on 178 catchments,
Hydrol. Earth Syst. Sci., 13, 819–831, 10.5194/hess-13-819-2009, 2009.Bezzola, G. and Hegg, C.: Ereignisanalyse Hochwasser 2005, Teil 1 – Prozesse,
Schäden und erste Einordnung. Bundesamt für Umwelt BAFU,
Eidgenössische Forschungsanstalt WSL, Umwelt-Wissen, 7, 215,
http://www.bafu.admin.ch/publikationen/publikation/00044/index.html?lang=de (last access: 29 January 2015), 2007.
CH2011: Swiss Climate Change Scenarios CH2011, published by C2SM, MeteoSwiss,
ETH, NCCR Climate, and OcCC, Zurich, Switzerland, 88 pp., 2011.
Diezig, R. and Weingartner, R.: Hochwasserprozesstypen: Schlüssel zur
Hochwasserabschätzung, Wasser und Abfall, 4, 18–26, 2007.Fedora, M. A. and Beschta, R. L.: Storm runoff simulation using an antecedent
precipitation index (API) model, J. Hydrol., 112, 121–133, 10.1016/0022-1694(89)90184-4, 1989.Frei, C. and Schär, C.: A precipitation climatology of the Alps from
high-resolution rain-gauge observations, Int. J. Climatol., 18, 873–900,
10.1002/(SICI)1097-0088(19980630)18:8<873::AID-JOC255>3.0.CO;2-9, 1998.Grams, C. M., Binder, H., Pfahl, S., Piaget, N., and Wernli, H.: Atmospheric
processes triggering the central European floods in June 2013, Nat. Hazards
Earth Syst. Sci., 14, 1691–1702, 10.5194/nhess-14-1691-2014, 2014.Gupta, H. V., Perrin, C., Blöschl, G., Montanari, A., Kumar, R., Clark, M., and
Andréassian, V.: Large-sample hydrology: a need to balance depth with breadth,
Hydrol. Earth Syst. Sci., 18, 463–477, 10.5194/hess-18-463-2014, 2014.Heggen, R. J.: Normalized Antecedent Precipitation Index, J. Hydrol.
Eng., 6, 377–381, 10.1061/(ASCE)1084-0699(2001)6:5(377), 2001.
Helbling, A., Kan, C., and Vogt, S.: Dauerregen, Schauer oder Schmelze-welche
Ereignisse lösen in der Schweiz die Jahreshochwasser aus?, Wasser Energie
Luft, 98, 249–254, 2006.Hilker, N., Badoux, A., and Hegg, C.: The Swiss flood and landslide damage
database 1972–2007, Nat. Hazards Earth Syst. Sci., 9, 913–925, 10.5194/nhess-9-913-2009, 2009.Hohenegger, C., Walser, A., Langhans, W., and Schär, C.: Cloud-resolving
ensemble simulations of the August 2005 Alpine flood, Q. J. Roy. Meteorol.
Soc., 134, 889–904, 10.1002/qj.252, 2008.
Huffman, G. J., Bolvin, D. T., Nelkin, E. J., Wolff, D. B., Adler, R. F., Gu,
G., Hong, Y., Bowman, K. P., and Stocker, E. F.: The TRMM multisatellite
precipitation analysis (TMPA): Quasi-global, multiyear, combined-sensor
precipitation estimates at fine scales, J. Hydrometeorol., 8, 38–55, 2007.
Isotta, F. A., Frei, C., Weilguni, V., Perčec, Tadić, M., Lassègues, P., Rudolf, B., Pavan, V.,
Cacciamani, C., Antolini, G., Ratto, S. M., Munari, M., Micheletti, S., Bonati, V., Lussana, C.,
Ronchi, C., Panettieri, E., Marigo, G., and Vertačnik, G.: The climate of daily precipitation in the Alps:
development and analysis of a high-resolution grid dataset from pan-Alpine
rain-gauge data, Int. J. Climatol., 34, 1657–1675, 2014.
Kohler, M. A. and Linsley Ray K.: Predicting The Runoff From Storm
Rainfall, US Weather Bureau Research Paper, US Weather Bureau, US Deptartment of Commerce, Washington, 9 pp., 1951.Köplin, N., Schädler, B., Viviroli, D., and Weingartner, R.:
Seasonality and magnitude of floods in Switzerland under future climate
change, Hydrol. Process., 28, 2567–2578, 10.1002/hyp.9757, 2014.Liechti, K., Panziera, L., Germann, U., and Zappa, M.: The potential of
radar-based ensemble forecasts for flash-flood early warning in the southern
Swiss Alps, Hydrol. Earth Syst. Sci., 17, 3853–3869, 10.5194/hess-17-3853-2013, 2013.
Malard, A. and Jeannin, P.-Y.: Characterisation of karst aquifers in
Switzerland: the KARSYS approach, Eur. Geologist, 35, 59–63, 2013.Marchi, L., Borga, M., Preciso, E., and Gaume, E.: Characterisation of selected
extreme flash floods in Europe and implications for flood risk management, J.
Hydrol., 394, 118–133, 10.1016/j.jhydrol.2010.07.017, 2010.Massacand, A., Wernli, H., and Davies, H.: Heavy precipitation on the alpine
southside: An upper-level precursor, Geophys. Res. Lett., 25, 1435–1438, 10.1029/98GL50869, 1998.Merz, R. and Blöschl, G.: A process typology of regional floods, Water
Resour. Res., 39, 1340, 10.1029/2002WR001952, 2003.Merz, R. and Blöschl, G.: A regional analysis of event runoff coefficients
with respect to climate and catchment characteristics in Austria, Water
Resour. Res., 45, W01405, 10.1029/2008WR007163, 2009.MeteoSwiss: Documentation of MeteoSwiss Grid-Data Products, Daily Precipitation
(final analysis): RhiresD, http://www.meteoschweiz.admin.ch/home/suche.subpage.html/de/data/products/2014/raeumliche-daten-niederschlag.html
(last access: 29 January 2015), 2011.
Mogil, H. M., Monro, J. C., and Groper, H. S.: NWS's flash flood warning and
disaster preparedness programs, B. Am. Meteorol. Soc., 59, 690–699, 1978.Munich Re: NatCatSERVICE Database, Münchener Rückversicherungs-Gesellschaft, Munich,
http://www.munichre.com/en/reinsurance/business/non-life/natcatservice/index.html
(last access: 29 January 2015), 2014.
Norbiato, D., Borga, M., Degli Esposti, S., Gaume, E., and Anquetin, S.: Flash
flood warning based on rainfall thresholds and soil moisture conditions: An
assessment for gauged and ungauged basins, J. Hydrol., 362, 274–290, 2008.Norbiato, D., Borga, M., Merz, R., Blöschl, G., and Carton, A.: Controls on
event runoff coefficients in the eastern Italian Alps, J. Hydrol., 375,
312–325, 10.1016/j.jhydrol.2009.06.044, 2009.Paschalis, A., Fatichi, S., Molnar, P., Rimkus, S., and Burlando, P.: On the
effects of small scale space-time variability of rainfall on basin flood
response, J. Hydrol., 514, 313–327, 10.1016/j.jhydrol.2014.04.014, 2014.Pathiraja, S., Westra, S., and Sharma, A.: Why continuous simulation? The role
of antecedent moisture in design flood estimation, Water Resour. Res., 48,
W06534, 10.1029/2011WR010997, 2012.
Piock-Ellena, U., Pfaundler, M., Blöschl, G., Burlando, P., and Merz, R.:
Saisonalitätsanalyse als Basis für die Regionalisierung von
Hochwässern, Wasser Energie Luft, 92, 13–21, 2000.Pui, A., Lal, A., and Sharma, A.: How does the Interdecadal Pacific Oscillation
affect design floods in Australia?, Water Resour. Res., 47, W05554,
10.1029/2010WR009420, 2011.Rajczak, J., Pall, P., and Schär, C.: Projections of extreme precipitation
events in regional climate simulations for Europe and the Alpine Region, J.
Geophys. Res.-Atmos., 118, 3610–3626, 10.1002/jgrd.50297, 2013.
Ranzi, R., Zappa, M., and Bacchi, B.: Hydrological aspects of the Mesoscale
Alpine Programme: findings from field experiments and simulations, Q. J. Roy.
Meteorol. Soc., 133, 867–880, 10.1002/qj.68, 2007.Reager, J. T., Thomas, B. F., and Famiglietti, J. S.: River basin flood
potential inferred using GRACE gravity observations at several months lead
time, Nat. Geosci., 7, 588–592, 10.1038/ngeo2203, 2014.Schmidli, J. and Frei, C.: Trends of heavy precipitation and wet and dry spells
in Switzerland during the 20th century, Int. J. Climatol., 25, 753–771,
10.1002/joc.1179, 2005.Schröter, K., Kunz, M., Elmer, F., Mühr, B., and Merz, B.: What made the
June 2013 flood in Germany an exceptional event? A hydro-meteorological
evaluation, Hydrol. Earth Syst. Sci., 19, 309–327, 10.5194/hess-19-309-2015, 2015.Spreafico, R., Weingartner, R., Barben, M., Ryser, A., Hingray, B., Musy, A.,
and Niggli, M.: Hochwasserabschätzung in schweizerischen Einzugsgebieten,
p. 117, http://www.bafu.admin.ch/publikationen/publikation/00565/index.html?lang=de
(last access: 29 January 2015), 2003.Stucki, P., Rickli, R., Brönnimann, S., Martius, O., Wanner, H., Grebner,
D., and Luterbacher, J.: Weather patterns and hydro-climatological
precursors of extreme floods in Switzerland since 1868, Meteorol. Z., 21,
531–550, 10.1127/0941-2948/2012/368, 2012.Tramblay, Y., Bouaicha, R., Brocca, L., Dorigo, W., Bouvier, C., Camici, S.,
and Servat, E.: Estimation of antecedent wetness conditions for flood
modelling in northern Morocco, Hydrol. Earth Syst. Sci., 16, 4375–4386,
10.5194/hess-16-4375-2012, 2012.Ulbrich, U., Brücher, T., Fink, A., Leckebusch, G., Krüger, A., and
Pinto, J.: The central European floods of August 2002: Part 2 – Synoptic
causes and considerations with respect to climatic change, Weather, 58,
434–442, 10.1256/wea.61.03B, 2003.Verbunt, M., Gurtz, J., Jasper, K., Lang, H., Warmerdam, P., and Zappa, M.: The
hydrological role of snow and glaciers in alpine river basins and their
distributed modeling, J. Hydrol., 282, 36–55, 10.1016/S0022-1694(03)00251-8, 2003.Weingartner, R. and Aschwanden, H.: Discharge regime – the Basis for the
Estimation of Average Flows, http://hades.unibe.ch/en/products/druckausgabe/gewaesser/tafel5_02
(last access: 29 January 2015), 1992.Weingartner, R., Barben, M., and Spreafico, M.: Floods in mountain areas – an
overview based on examples from Switzerland, J. Hydrol., 282, 10–24, 10.1016/S0022-1694(03)00249-X, 2003.Wever, N., Jonas, T., Fierz, C., and Lehning, M.: Model simulations of the
modulating effect of the snow cover in a rain-on-snow event, Hydrol. Earth Syst.
Sci., 18, 4657–4669, 10.5194/hess-18-4657-2014, 2014.
De Wit, M. and Buishand, A.: Generator of rainfall and discharge extremes (GRADE) for the Rhine and
Meuse basins, Rijkswaterstaat RIZA Report 2007, 027/KNMI publication 218, Lelystad,
2007.