Continental Portuguese Territory Flood Susceptibility Index - Contribution for a Vulnerability Index

1 This work defines a national flood susceptibility index for the Portuguese continental territory, 2 by proposing the aggregation of different variables which represent natural conditions for 3 permeability, runoff and accumulation. This index is part of the national vulnerability index 4 developed in the scope of Flood Maps in Climate Change Scenarios (CIRAC) project, supported 5 by the Portuguese Association of Insurers (APS). 6 This approach expands on previous works by trying to bridge the gap between different floods 7 mechanisms (e.g. progressive and flash floods) occurring at different spatial scales in the 8 Portuguese territory through: a) selecting homogeneously processed datasets; b) aggregating 9 their values to better translate the spatially continuous and cumulative influence in floods at 10 multiple spatial scales. 11 Results show a good ability to capture, in the higher susceptibility classes, different flood types: 12 fluvial floods and flash floods. Lower values are usually related to: mountainous areas, low 13 water accumulation potential and more permeable soils. Validation with independent flood 14 datasets confirmed these index characteristics, although some overestimation can be seen in the 15 southern region of Alentejo where, due to a dense hydrographic network and an overall low 16 slope, floods are not as frequent as a result of lower precipitation mean values. 17 Future work will focus on: i) including extreme precipitation datasets to represent the triggering 18 factor; ii) improving representation of smaller and stepper basins; iii) optimizing variable 19 weight definition process; iii) developing more robust independent flood validation datasets. 20


Introduction
The focus of this work will be on susceptibility to floods, for the Portuguese continental 38 territory, which is defined as the propensity of an area to be affected by floods. This propensity 39 is given by the territory intrinsic characteristics such has slope, geology, river network, and land 40 use. The present work is part of a flood vulnerability study for the Portuguese continental 41 territory, developed in the Flood Risk Mapping in Climate Change Scenarios (CIRAC) project. 42 Section 2 presents a state of the art review of concepts and methods implemented to translate 43 flood susceptibility and its relation with flood vulnerability and provides insight on the current 44 work contribution to improve flood susceptibility mapping at the national scale. Section 3 is 45 divided into three subsections describing the study area hydromorphological characteristics, the 46 different used datasets and the methodology followed to design and implement the national 47 susceptibility index map. Section 4 presents the main results, including intermediate and final 48 index maps, provides a first overall interpretation of its advantages and limitations and validates 49 them through a comparison with historical flood events. Finally section 5 analyses the main 50 findings, the contributions for the state of the art and the impact of the results in the Portuguese 51 context. 52

53
The crucial factor on turning a flood on a potential damaging event for communities and 54 ecosystems is the proximity to prone areas such as floodplains which determines their 55 vulnerability to the phenomena (Cutter et al., 2008). The IPCC (2012) presented vulnerability as 56 being the "predisposition, susceptibilities, fragilities, weaknesses, deficiencies, or lack of 57 capacities that favor adverse effects on the exposed elements". This is a general concept that 58 introduces susceptibility as one of the different dimensions that contribute to and should be 59 contained in a vulnerability assessment (Figure 1). Adger (2006) also relates both concepts by 60 defining vulnerability as the susceptibility to harm from exposure to a change on the 61 environment or on the society and the incapacity to adapt to those changes. The juxtaposition 62 and interdependency between vulnerability and susceptibility is evident, leading sometimes to 63 inconsistencies in their definition, depending on the researching perspective. 64

Figure 1 65
For instance, according to Balica et al. (2012), "a system is susceptible to floods due to 66 exposure in conjunction with its capacity/ incapacity to be resilient, to cope, recover or adapt 67 to". The authors connected susceptibility with exposure, considered as the hydro-geological 68 component, and also with the institutional and socio-economic systems. 69 Collier and Fox (2003), despite not discussing directly the susceptibility concept, identified 70 some components to describe a baseline susceptibility to flash floods that were mostly derived 71 from inherent characteristics of a specific basin. Those characteristics are: the likelihood of 72 unimpeded flow and the existence of channel constrictions, catchment slope, ratio of catchment 73 area to mean drainage path length, ratio of land use to vegetation type as a proxy of urban 74 extension. This approach to susceptibility leads to the definition adopted in this work and also 75 indentified in other studies (Verde and Zêzere, 2007;Zêzere et al., 2005), where flooding 76 susceptibility is a characteristic of an area, given by its natural terrain configuration and 77 occupation and that determines its propensity to flooding. 78 The several steps included in the methodological approach to susceptibility estimation, from 79 variable and source data selection, to the composition of indicators, depend not only on the 80 chosen definition of susceptibility but also on the spatial scale of analysis. The work presented here contributes to the improvement of the current state of the art in the 108 susceptibility evaluation field by designing and implementing, for the first time, a flood 109 susceptibility index for the Portuguese territory. Some innovative methodological features are 110 also introduced to overcome the limitations stated above, regarding the determination of flood 111 susceptibility at a national scale. Variable selection tries to reflect the different flood dynamics 112 that occur in the Portuguese territory. Selected parameters include flow accumulation potential, 113 topographical and land use/soil permeability characteristics, representative of processes at 114 different scales and influent in both progressive and flash floods. The selection process also 115 reflects the need to reduce index complexity by choosing fewer input variables and select 116 datasets that are uniformly processed across the Portuguese territory, to minimize index 117 misinterpretation due to possible spatial inconsistencies at a country scale. The exclusion of 118 precipitation reflects a focus on the territory characteristics, but also a difficulty of having a 119 dataset that could efficiently represent the reality and not hide the susceptibility in the Alentejo 120 and Algarve regions, both located in the south of Tagus River where the mean annual 121 precipitation is much less then northern Tagus River and which is less affected by frontal 122 systems than the north. The inclusion of precipitation would require a different scale of analysis, 123 namely a regional index. Also, a double evaluation for types of episodes and events, extreme 124 rainfall and annual mean rainfall. Finally, the presented methodology applies an aggregation 125 methodology to some of the chosen variables, described in more detail in section 3.3, to better 126 represent the spatially continuous and cumulative nature of their influence in flood generating 127 mechanisms, across increasingly higher spatial scales. 128

Study area 130
The study area is the continental Portuguese territory (Figure 2 (i)), part of the Iberian 131 Peninsula, located in the southwest of Europe. 132 Historically, and due to climatic characteristics, this territory has frequently registered flood 133 occurrences. According to Quaresma (2008) ( Figure 3). The first two describe the potential water accumulation in the riverbed and adjacent 155 areas, while the last assesses soil permeability based on land use and geology. 156 The Hydrosheds (Hydrological data and maps based on Shuttle Elevation Derivatives at 157 multiple Scales) Digital Elevation Model (DEM) was used to obtain two of the three final 158 variables and several other auxiliary variables. Hydrosheds data is derived from the Shuttle 159 Radar Topography Mission (SRTM) at 3 arc-second resolution (90 meters) and is freely 160 available online (http://hydrosheds.cr.usgs.gov). The original data has been hydrologically 161 conditioned in order to be used in regional and global watershed analysis. Furthermore it has an 162 adequate scale for country scale flood susceptibility analysis, allowing for a homogeneous and 163 spatially continuous processing of the different datasets. The Hydrosheds DEM was used to 164 derive slope, flow accumulation and direction and the hydrographic network. All original and 165 subsequently processed datasets were converted to the WGS1984 coordinate system and 166 resampled to a 90 m resolution grid. 167   factors. This also allows the possibility of including, on a later stage, a precipitation theme or a 225 combination of precipitations themes (e.g., mean annual precipitation or a set of maps with the 226 interpolated ground station precipitations for different return periods and durations (Brandão et 227 al., 2001) to better reflect flood susceptibility for any specific climatological time period. 228

Methods 229
The main objective of the methodology presented in this section is to produce, using the above 230 described datasets, a spatially continuous flood susceptibility index for the Portuguese territory, 231 varying from 0 to 1, where the highest values correspond to a higher flood propensity. To 232 achieve this, a four stage approach was followed, including: a) an aggregation process for the 233 flow number dataset to better represent, for each cell, the cumulative influence of its upstream 234 to downstream spatial distribution; b) a normalization process for all variables to rescale them to 235 common 0 to 1 range, where higher values represent areas more susceptible to floods (  In order to evaluate the quality of FSI model a further validation was carried out, based on the 264 DISASTER hydro-geomorphologic database. A classification of FSI values in 6 classes shows that nearly 62% of the occurrences lie in the 281 0.45 to 0.5 susceptibility class (see Figure 6A). Values below 0.3 are not coincident with 282 occurrences and these ones are present residually in class 0.6 to 0.95 (about 0.6%). The non-283 increasing occurrence frequency, from the lowest to the higher susceptibility class, is also 284 associated with differences in class frequency in mainland Portuguese territory (see Figure 6B). 285 The calculation of occurrence densities eliminates the influence of the frequency of each class in 286 the results; thus calculating this density (number of occurrences per km 2 ) allows to accept the 287 results obtained for FSI as representatives for the entire mainland Portugal (see Figure 6C). In 288 fact, the FSI value of 0.5 appears to provide a critical threshold above which the relatively high 289 hazard map for the main Portuguese Rivers allowed an accurate assessment of the two higher classes. 320 In fact, most of the values included in those classes are within the limits of those flooded areas. 321 As can be seen in Figure 7, the adjacent areas to all major and medium sized rivers in the 322 Portuguese territory are also included in these higher classes. This demonstrates the FSI ability 323 to better identify regions susceptible to fluvial floods in the highest class (see section 3.4) due, 324 as stated above, to the higher importance given to the flow accumulation and cost distance 325 variables. 326 Hydrosheds DEM, slope and land use. All information related to the final set of classes is given 330 in Table 2. 331

Table 2 332
It should be noted that this susceptibility class definition methodology led to unequal interval 333 ranges for the different classes, as can be seen in the third column of Table 2. This was 334 somewhat expected since it was improbable that an index composed of three linearly 335 normalized and combined variables could translate flood susceptibility in a regular scale. In 336 fact, the variation of influence of each of those variables in flooding processes is, in most cases, 337 non-linear and therefore is associated with very different interval ranges. Therefore their 338 combination would most probably lead, as it was confirmed by this work, to susceptibility 339 classes defined by heterogeneous intervals. Moreover, some of the input variables also have 340 very unbalanced normalized values distributions, namely flow accumulation (high frequency of 341 low values and a few very high values) and cost distance (mostly high values), further distorting 342 the distribution of the final susceptibility values and consequently the definition of the 343 correspondent classes. 344 In addition to the above mentioned main rivers, FSI (Figure 8) for the Portuguese territory also 345 identifies some major cities like Lisbon, Coimbra, Aveiro, Setúbal, Faro and Oporto and some 346 small basin areas in the south part of Portugal (Algarve) as highly susceptible to floods (classes 347 3 and 4). This showcases the index sensitivity to identify also flash flood prone areas, 348 characterized by highly impermeabilized artificial surfaces situated in plain regions in the 349 vicinity of relevant water courses (see Figure 9, panels (ii) and (iii)). The Alentejo region, east 350 of Lisbon (Figure 9 (i)), is also classified as highly susceptible (class 3) due to its topographical 351 and geological characteristics, since most of the most the territory is plain, with a high 352 hydrographic network density and impermeable rocky (shale and marble) or clay soils. 353 Institute showed a general good direct correspondence between the frequency of flood points 361 and the magnitude of susceptibility values in the vicinity of those points (Figure 7 (i)). Looking 362 in greater detail the index confirmed its ability to capture: a) a higher flood susceptibility 363 associated with the main Portuguese rivers and their adjacent areas (example given for the 364 Tagus basin in Figure 9 (ii)); and b) flash flood prone urban areas like Lisbon and Setúbal 365 (Figure 9 (iii)). 366

383
The development of a national flood susceptibility index entails several challenges related to 384 difficulties in capturing the different flood dynamics usually occurring in a wide territory across 385 different spatial scales. The work presented here presents a first attempt to implement this type 386 of index for the Portuguese continental territory. 387 The first results are very promising with a consistent representation of the overall spatial 388 distribution of flood susceptibility. The presented methodological approach addresses some of 389 those scale issues by applying a spatial aggregation methodology that better characterizes the 390 cumulative influence of the different variables across spatial scales (from cell to basin and 391 higher). Furthermore the selection of only three variables that represent water accumulation 392 potential, topography and soil permeability allowed for a clear interpretation of the index and an 393 apprehension of different flooding phenomena, ranging from fluvial floods in large rivers to 394 urban flash floods. 395 Nevertheless some possible overestimation of flood susceptibility in regions of low precipitation 396 was observed and should be addressed in future work by including appropriate precipitation 397 datasets such as interpolated ground station precipitations for different return periods and 398 durations (Brandão et al., 2001). Other developments to be implemented in the future will be 399 focused on improving the representation of the higher susceptibility associated with smaller 400 basins or with stepper slopes due to a higher superficial flow generation potential and smaller 401 concentration times. In the future, this could be overcome by the inclusion of two themes 402 containing spatially aggregated values of slope (accumulated mean) and concentration time 403 (accumulated sum), following the methodology used in this work. 404  Tables   Table 1 -Information summary for all used