Successful evacuation planning is critical for emergency managers and local decision-makers to save lives from potential future tsunamis. As in all evacuation models, there is no way to create ideal evacuation planning; hence some assumptions have to be made according to the applied model. In this section, the assumptions that are made through this study is discussed.
Tsunami numerical simulations are performed according to the present mean sea level; however, it continuously changes due to tides or storms, hence for different seasons, lunar cycles, sea level fluctuations and climatic changes a sensitivity analysis should be made followed by a multitemporal evacuation analysis.
The inundation is assumed as composed of only water, not the debris material, during the numerical simulations. However, the damage is mostly due to the debris materials carried by tsunami waves, such as boats, cars, parts of trees, and concrete particles detached from the ground or houses.
In an actual tsunami event, the arrival of the first wave to the land differs, but it is assumed for LCD models that all locations are inundated simultaneously. Therefore, in some places, there may be more time for individuals to initiate an evacuation.
According to the resultant evacuation walk time maps (Fig. 5), it seems like there is enough time for individuals to reach safety from the shortest route with slow walking speed since the arrival of the first tsunami wave is higher than the calculated evacuation time required. However, the tsunami warning system’s reaction time, including the institutional notification and the decision time to notify the at-risk community (Post et al. 2009), is also neglected since there is no effective tsunami warning system in the study area currently, despite a pilot local tsunami early warning system has been deployed as part of the Last Mile (Necmioglu et al. 2019) project and being tested. In addition to the delay in the tsunami warning process, the different evacuation behavior depending on individual awareness is not included in the LCD model. Furthermore, the effect of possible crowd behavior, such as panic and chaos, has not been taken into consideration (Liu et al. 2018).
The earthquake’s impact on buildings, roads, bridges, or any structure is neglected, which may slow the evacuation dramatically, or even preclude at certain evacuation routes. Additionally, all the buildings are assumed as reinforced concrete and not damaged by the tsunami waves. However, in coastal resorts like Bodrum, there are lots of unstructured temporal buildings, shelters, etc., that are vulnerable to earthquakes and tsunami waves. In the case of a strong earthquake, reinforced concrete buildings may collapse along with unstructured ones towards the important roads, which leads to block the evacuation suddenly and cause congestion. In PEAT, roads are assumed as the safest places to evacuate for an individual. However, in addition to the collapse of buildings, the road network may be destroyed by an earthquake due to liquefaction.
The routes in docks and breakwaters are where evacuations are difficult since there is only one narrow path to reach the safe zone (Wood and Peters 2015). Although it does not reflect the reality, it is assumed that all the roads have identical SCVs providing the easiest way for a pedestrian.
In this study, we implemented the LCD model instead of ABMs to understand the spatial distribution of the required evacuation time for a pedestrian based on landscape properties. The LCD models ignore the individual behavior and interactions during an evacuation. In contrast, ABMs assess hundreds of individuals’ decisions and their interactions based on a set of rules (Wang et al. 2016). Understanding and modeling individual decision-making behavior during a panic situation like evacuation is a very complex phenomenon, which depends on the different awareness of all individuals affected by near experiences and ancestral knowledge of that region (Tufekci et al. 2021). In addition to that, LCD models ignore congestion of roads leading to panic among evacuees that causes prolonged the required evacuation time. Depending on the roads’ congestion level, evacuees’ travel speed may vary; however, in LCD models, it is assumed that all individuals in a community have the same and constant speed of evacuation (Wood and Schmidtlein 2012). There is no information on the human capacity in the available road dataset, which is essential to prevent congestions during an evacuation. It is vital knowledge for emergency managers and local decision-makers to offer the routes in evacuation planning.
PEAT requires lots of complex data representing landscape properties that affect pedestrians’ walking pace during an evacuation, such as slope, barriers, roads, or water bodies located within the hazard zone (Wood and Schmidtlein 2013). In order to create successful evacuation planning, the data must be accurate, up-to-date, and with high resolution. In this study, the available data is provided by Bodrum Municipality and open sources like Google Earth Images and Open Street Map. Since the shoreline of Bodrum is continually changing and developing due to the construction of new coastal facilities such as marinas, renewed buildings, constructions of new buildings like hotels and roads, the data used for the LCD models should be updated regularly to avoid misleading an evacuee. In addition to that, resultant evacuation time maps (Fig. 5) are sensitive to the assumptions made in the land-cover and slope SCVs (Schmidtlein and Wood 2015). Due to the lack of available data in this study, land-cover data are produced from Google Earth Images, which is user-dependent. The created land-cover data are classified, and SCVs are given according to Wood and Schmidtlein (2012).
Bodrum is one of Turkey’s largest holiday towns and also attracts many tourists from abroad; hence the population rises to ~ 2 million people in the summer seasons (Erdogan 2016). The spatial distribution of the population is time-depended and changes dramatically based on seasons in holiday towns. Since the occurrence time of a tsunami cannot be estimated, the detailed dataset that should represent the spatial distribution of population both in summer and winter seasons should be included in the evacuation plans. In addition to seasonal changes, even both day and night population continuously changes. No such data represents all population variability; hence, in the study, the population is ignored.
Furthermore, evacuation is assumed to performed only horizontally; vertical evacuation to the buildings is ignored. Likewise, multi-evacuation is not allowed in the model, only self-initiated pedestrian evacuation is permitted as it is the most useful way for sudden hazards like tsunami, debris flow, etc.
4.1. Validation of resultant evacuation time surfaces
The resolution of input data for PEAT totally affects the interpretation of resultant tsunami evacuation maps (Fig. 5) by an individual during an evacuation. According to the model sensitivity analysis conducted by Wood and Schmidtlein (2012), PEAT tended to underestimate the required evacuation time with coarser-resolution elevation data. Therefore, DEM resolution plays a critical role in the resultant maps and validation is required to make sure accuracy of resultant evacuation time maps (Fig. 5). For this purpose, four different locations having maximum evacuation time to reach the safe zone and their shortest path to reach the safe zone are selected as validation routes (Fig. 6–9). While choosing the validation road, it was prioritized that the road was wide and the fastest for pedestrians. The validation is performed by a nearly constant speed not exceeding 1.1 m/s. During the walk, actual time and distances are recorded (Table 2). The time calculated by PEAT in the final resultant evacuation time maps at the locations where the validation route starts (Fig. 6–9) is compared with the time passed to reach the safe zone during the walk (Table 2).
According to validation results (Table 2), the measured time to reach the safe zone from selected validation routes is lower than it is calculated by PEAT. Therefore, for those three routes (Fig. 6, 7, and 9), evacuation is possible with 1.1 m/s speed according to the proposed evacuation time of PEAT. However, in the case of the Bitez-2 validation route (Fig. 8), field experts thought that the selected route was within the hazard zone and walked 40 m more than it is calculated by PEAT for that route even there is no panic situation during validation. As shown in Fig. 8, there is a valley where the hazard zone continues 40 m more right next to the validation route. Increasing the accuracy and decreasing the misinterpretation and confusion of evacuees depends on using detailed and high-resolution data.
Table 2
Measurements during the walk on validation routes
Validation Route | Evacuation Time from PEAT Results (min) | Measured Distance (m) | Measured Time (min) | Average Speed (m/s) |
Gumbet | 3 | 124 | 2.32 | 0.82 |
Bitez-1 | 4 | 171 | 2.52 | 0.99 |
Bitez-2 | 5 | 356 | 5.35 | 1.06 |
Yahsi | 6 | 361 | 5.35 | 1.08 |