Introduction
The intensified urbanisation caused by rapid population growth and
increasing rural migration is in many studies found to be a major cause of
amplified peak flows and increased flood risk in cities worldwide
(Chang and Franczyk, 2008). Changes in the urban land cover
are important factors for the increased susceptibility to flooding. Thus,
the increasing presence and location of impervious surfaces (IS) like roads,
buildings, parking lots and other paved areas strongly affect the amount and
velocity of run-off during high-intensity precipitation events
(Arnold and Gibbons, 1996). Climate change is expected to
exacerbate the flood risk as heavy precipitation is projected to increase in
intensity and frequency over many areas of the globe, including central,
northwest and northern Europe (Field et al., 2012; EU and EEA, 2012).
Most major cities in Europe are characterized by IS fractions ranging from
50–75 % or even higher (EU and EEA, 2012). If present trends
towards increasing IS fractions continues this is expected to further
aggravate the risk of pluvial flooding. On the other hand the introduction
of more pervious surfaces could serve as adaptation to climate change.
Hence, detailed knowledge of the importance of both urban land cover changes
and climate change for the risk of urban areas towards flooding provides
substantial insight into how to plan for future climate proof cities.
Simulations of urban flooding based on more or less detailed hydrodynamic
models are commonplace most places in Europe, where they are used for e.g.
urban drainage design and/or emergency management purposes. Some of these
studies take climate change into account, whereas others consider past,
present or future urban development. Integrated maps at the European level
(EU and EEA, 2012) or even at the country level are, however, not readily
available due to these significant differences in modelling approaches and
tools.
Relationship between NDVI and actual imperviousness for sample
area (a), and relationship between NDVI based imperviousness and
NDVI based imperviousness for sample area (b). RMSE = Root Mean
Squared Error, MAE = Mean Absolute Error.
![]()
Design intensities and climate factors for extreme precipitation in Odense
under current and future climate, RP5 = return period of 5 years. CF's for
RCP4.5 from Gregersen et al. (2014) and RCP 8.5 from
Arnbjerg-Nielsen et al. (2015).
RP100
RP50
RP20
RP10
RP5
Current climate
Total precipitation (mm)
59
51
42
35
30
Maximum intensity (mm h-1)
43
37
29
25
21
RCP 4.5
Climate factor
1.6
1.4
1.29
1.22
1.2
Total precipitation (mm)
83
66
51
42
33
Maximum intensity (mm h-1)
60
47
36
30
23
RCP 8.5
Climate factor
2.2
1.98
1.84
1.8
1.6
Total precipitation (mm)
130
101
77
64
48
Maximum intensity (mm h-1)
94
72
54
44
33
Landsat TM image Odense 1984 (a), Landsat OLI image
Odense 2014 (b), Absolute change in imperviousness in Odense
during the 1984–2014 period (c).
![]()
In this paper we present a methodology suitable for investigating the
relative and combined influence of historical urban land cover changes and
climate change for the exposure of cities to pluvial flooding. The combined
remote sensing and flood modelling approach enables the quantification of
the flood risk relative to changes in imperviousness and climate change. The
methodology is validated using data from the Danish city of Odense,
comparing the influence of the past 30 years (1984–2014) of urban land cover
changes on the exposure to flooding under present day conditions and future
climate conditions. The method is, however, easily applicable for the
majority of cities in Europe.
Methods and materials
Urban land cover change
Estimated changes in IS during the period 1984–2014 may generally be
derived from analyses of Landsat moderate resolution (30 m) satellite
imagery, assuming a relationship between the vegetation cover and the
fraction of IS in the urban environment (Bauer et al., 2008;
Skougaard Kaspersen et al., 2015). A regression model
was developed for the city of Odense using the Normalized Difference
Vegetation Index (NDVI) as a measure of green vegetation while IS were
manually digitized from a high-resolution aerial photography
(Fig. 1). To exclude major areas characterized by
bare soil, which are spectrally similar to man-made impervious surfaces, the
city's boundary was established prior to the analysis. Inside the city
boundary urban areas are implicitly assumed to consist of a combination of
impervious surfaces and green vegetation and only marginally by areas of
bare soil. Random training and validation areas of approximately 500 m × 500 m
(314 Landsat pixels) covering the full range of IS values were chosen for
the development and validation of the regression model. A non-linear
function was found to best describe the relationship between NDVI and IS
fractions (Root Mean Squared Error = 9.1 %) and to provide the basis for
the most accurate estimate of IS fractions based on NDVI (Mean Average
Error = 6.1 %) (Fig. 1). For the current analysis
the city boundary was digitized manually from the 2014 Landsat imagery and
used as delineator of the urban area for both 1984 and 2014.
Projections of extreme precipitation
High resolution climate projections for precipitation extremes may be
derived from global coupled atmosphere-ocean circulation models (GCMs) and
downscaled using regional climate models (RCMs) and/or empirical-statistical
approaches (Willems, 2012). To assert the considerable uncertainties
associated with future projections of changes in the frequency and intensity
of extreme events, such projections are generally based on multi-model
ensembles (van der Linden and Mitchell, 2009) rather than single
model projections, combined with different statistical downscaling methods
(Gregersen et al., 2013; Sunyer et al., 2014). Design
intensities and “climate factors” (i.e. the ratio between the expected
future and present design rainfall) for extreme precipitation in Denmark
under present and future conditions have previously been estimated from
regional climate projections (Gregersen et al., 2013). Here we
use recent results from the “RiskChange” project (Maule et
al., 2014; Arnbjerg-Nielsen et al., 2015) to generate
15 different time series for design precipitation, each with a duration of 4 h and representing precipitation events from 5 to 100 year return
values, using a Chicago Design Storm (CDS) rain calculator (IDA,
2014) (Table 1) for corresponding present-day and
future climate conditions (2071–2100; RCP4. and RCP8.5;
Meinshausen et al., 2011). Initially,
synthetic precipitation events based on statistical analyses of observed
precipitation from approximately 300 precipitation gauges in Denmark were
created for the specific geographical area of interest (Odense)
(Arnbjerg-Nielsen et al., 2006). Precipitation
intensities from the CDS rain time series under current climate conditions
were then multiplied by a CF to obtain corresponding intensities under
RCP4.5 and RCP8.5 climate change scenarios (Table 1).
Flood modelling
When modelling overland flooding several modelling concepts are available.
Some methods integrate 1-D modelling of flow in the sewage system with a 1-D
description of overland flow (Maksimović et al., 2009)
while others use modified software originally intended for estuarine and
coastal modelling to interact with the drainage system, e.g. MIKE FLOOD
(MIKE by DHI, 2014) or TUFLOW (TUFLOW, 2014). Pros
and cons of these methods have been widely discussed and choice of method
depend on the aim of the study, e.g. 1-D solutions are fast but offers a poor
approximation of complex (non-unidirectional) flows (Mark et
al., 2004; Leandro et al., 2009). When limited
information of the drainage system is available it can be necessary to
neglect the interaction with subsurface flows and model the overland flow by
use of 2-D software and a digital elevation map (DEM) only. The current
research uses such an approach as the occurrence of extreme precipitation is
simulated using the 2-D hydrodynamic overland flow model in MIKE 21 (MIKE by
DHI, 2014). The drainage capacity of the pipes is modelled by modifying the
precipitation, e.g. by assuming a max pipe capacity and then subtracting the
volume from the rain input. In order to include the effect of the drainage
system it is assumed that it is dimensioned perfectly according to the
current Danish recommendations. This means that flooding on terrain does not
occur for precipitation with a return period of 5 years (RP5) or less. As a
result we reduced all design precipitation time series with the amount of a
5-year event (RP5) prior to running the flood model simulations. Variation
in run-off from areas with different levels of imperviousness is modelled by
use of the evapotranspiration module in MIKE 21. Evapotranspiration is set
to equal precipitation for pervious surfaces (0 % IS = 0 % run-off)
while no evapotranspiration is assumed for completely sealed surfaces
(100 % IS = 100 % run-off). The EU-DEM (EU-DEM, 2014), which is
publicly available over most of the European continent and has a spatial
resolution of 25 m is used as input for the MIKE 21 model and provides the
basis for the overland flow calculations. The Landsat based impervious
surface maps are resampled from its original resolution (30 m) to match that
of the EU-DEM (25 m) as this is a requirement for the model to run.
Influence of urban land cover changes and climate change on the
exposure of cities to pluvial flooding
The combined impact of changes in the urban land cover and climate change
towards the exposure of Odense to pluvial flooding is investigated by
simulating the occurrence of 5 distinctive high-intensity precipitation
events (RP5, RP10, RP20, RP50, RP100) under both current and future climate
(RCP4.5, RCP 8.5) and for historical (1984) and current (2014) urban land
cover. The RCP4.5 scenario represent a fairly modest climate future with
increases in the near-surface air temperature towards 2100 in the range of
1.8 ∘C (1.1–2.6 ∘C) compared to the pre-industrial
(IPCC, 2013). Correspondingly, the RCP8.5 scenario represent a world where
the increased radiative forcing corresponds to an increase in near-surface
air temperature of 3.7 ∘C (2.6–4.8 ∘C) in the year 2100.
The primary outputs of the flood model are maps showing the maximum flood
depth and extent for each individual simulation. A pairwise cross-comparison
of the multiple flood maps enables a quantification of the relative
importance of land cover changes as compared to climate change for the
overall flood exposure from different high-intensity precipitation events.
Total area flooded by > 10 cm of surface water during
high-intensity precipitation events for present (2014) and historical (1984)
land cover conditions and for current climate and RCP85,
1984 = imperviousness of 32 %, 2014 = imperviousness of 51 %.
![]()
Effect of changes in imperviousness during 1984–2014
(32 → 51 %) for current climate and in a RCP4.5 and RCP8.5 world,
and impacts of climate change (RCP4.5, RCP8.5) for urban land cover as in
1984 (32 % IS) and 2014 (51 % IS) for total area flooded with
>10 cm of surface water during high-intensity precipitation.
![]()
Results
Land cover changes during 1984–2014
An urban land cover change analysis for Odense based on remote sensing
estimates shows an overall increase in absolute imperviousness of 19 %,
from 32 % in 1984 to 51 % in 2014 (Fig. 2c).
There is a somewhat clear pattern that the majority of the sealing occurred
at the fringes of the city during this period while the central parts are
characterized by little or no change (Fig. 2c).
Some of the increased sealing has occurred within the existing urban area
(in 1984), indicating that Odense has been experiencing both growth and
densification during the period of 1984–2014. Results are shown only for
those parts of the city which are currently connected to the urban drainage
system, which comprises the majority of the urban area.
Exposure to flooding due to land cover changes and climate
Figures 3 and 4 compiles the results of 30 simulations of urban flooding in
Odense, varying the land cover (1984 vs. 2014), the climatic conditions
(present day conditions, RCP 4.5 and RCP8.5 climate change scenarios) and
the severity of the event (from RP5 to RP100) as outlined above.
In general we see that results for the RCP8.5 climate change scenario for
2071–2100 using the estimate of urban land cover from 2014 clearly stands
out (Fig. 3). Whereas the total area flooded in most cases scale
approximately linearly with the severity of the extreme precipitation event,
in this case the scaling exhibits more of an exponential behaviour. We also
find that the effect of increases to IS in terms of the total area flooded
increases with the intensity of extreme precipitation and therefore also
with the degree of climate change. In the case of Odense the impact of the
observed change in urban land cover from 1984–2014 roughly compares to the
impacts of climate change under the RCP4.5 scenario.
Our results indicate that a change in absolute imperviousness of 19 %
(32 → 51 %) during 1984–2014 has increased the overall flood
extent during extreme precipitation by between 6 and 26 % under current
climatic conditions (Fig. 4). If a similar change
in imperviousness took place under climate change conditions as expected in
the RCP8.5 scenario this would result in an increase in overall flood extent
of 21–36 %. Conversely, in a RCP8.5 scenario the exposure to flooding
increases by 84–104 % and 70–85 %, respectively, under current (2014)
and historical (1984) levels of imperviousness. The combined effect of both
land cover changes and climate changes increases the urban flood
vulnerability by as much as 112–152 % (moving from lowest to highest line
in Fig. 3). From the perspective of climate change adaptation this clearly
demonstrates quantitatively the effect of pervious surfaces as measures for
reducing flood exposure. Conceptually, if measures where implemented
reducing overall imperviousness by 19 % (same as the increase in the
period 1984–2014) flood exposure would decrease by 6–21 % under current
climate and by 13–24 % (RCP4.5) and 17–27 % (RCP8.5) under future
climatic conditions.
Discussion
In this research we demonstrate the applicability of a combined methodology
for investigating the exposure of cities to pluvial flooding due to changes
in the urban land covers (IS fractions) and climate. Whereas similar
modelling work are carried out for most European cities and often using
flood models of much higher complexity for the purpose of e.g. urban
drainage design and cloudburst management, the analytical framework
presented here is particularly suited for large-scale mapping of cities'
vulnerability to climate change. Thus, the combination of remote sensing
estimates with an easily customizable flood modelling approach principally
makes it feasible to create improved maps, which takes into account past,
present and future urban development as well as climate change in a
consistent manner, which is transferable across cities and countries.
Arguably, this may prove an important tool for policy-makers in facilitating
urban adaptation to climate change at the European level (EU and EEA,
2012).
The methods described here for the urban land cover change analysis and
flood modelling along with state of the art results from recent downscaling
exercises and extreme value analysis of climate projections are
well-described in literature. For example previous studies have shown that
the amount in impervious surfaces strongly influences total and peak
discharge, daily stream flows, run-off volumes and the hydrological response
time during medium/high-intensity precipitation events (Jennings
and Jarnagin, 2002; Chormanski et al., 2008;
Verbeiren et al., 2013).
The method applied to quantify changes to IS is sensitive towards the
presence of areas of bare soil. If large areas of bare soil were present
within the urban area in either 1984 or 2014 it could result in
misclassifications for those areas. However, as bare soil is only marginally
present within most urban areas in Europe, including Denmark, we assume this
not to have a large influence on the output of the Odense analyses, although
this might not be true for all cities.
The uncertainty associated with climate change projections and with
different downscaling methods is well documented (Frei et al.,
2006; IPCC, 2014; Sunyer et al., 2014). As a result the derived
climate factors for precipitation extremes may exhibit large variations even
within the same climate scenario. Adopting a conservative approach we have
in the current study used a set of climate factors for our design
precipitation time series based on multiple model runs, which represents the
“middle of the road” (Arnbjerg-Nielsen et al., 2015).
Arguably, if used for adaptation planning one could instead decide to focus
on best- and/or worst-case scenarios for robustness and economic
considerations.
The flood model used in this study does not include a module for spatial
representation of the drainage system in Odense, and the geographical
location of surface flooding may therefore be inaccurate for some locations
due to inconsistent dimensioning, since the actual drainage system in
practice is not likely to be perfectly dimensioned according to a 5-year
rainfall event for all locations. Evidently, this choice makes the model
easily transferable to other cities at the cost of precision. When neglecting
the sewage system, errors are introduced to the flood extents though these
effects may be assumed to be localized and their significance to decrease
with more extreme rain events (Paludan et al., 2012). Likewise, the use of a
rather coarse DEM (25 m resolution) does not allow for identification of
highly localised water flows and related flooding within urban areas
(Villanueva et al., 2008). Finally, the total volume and extent of surface
flooding (Fig. 3) must be interpreted with some caution and should be
considered as a conservative estimate, since this assumes run-off from green
areas to be negligible at any time during extreme precipitation (no
saturation occurs). Changes in imperviousness have previously been found to
be more pronounced for more frequent precipitation events (e.g. RP5) while
only to a lesser degree affecting the hydrological response for less frequent
events (RP100) (Hollis, 1975). The reason for this difference is that
pervious surfaces saturates faster during very extreme events, and begin to
behave as impervious areas rather quickly after the onset of the event.
Conversely, the saturation time is commonly much longer during less extreme
events.