Here we test the predictive power of the methodology based on two real tsunamis in Tonga: the January 15, 2022 and September 29, 2009 events. We also show the relevance of this methodology for other types of natural hazards such as storms, volcanic eruptions and earthquakes.
4.1 Case study #1: 2022 tsunami
Satellite imagery has revealed considerable damage at locations on Tongatapu, ‘Eua and Ha’apai islands, suggesting that run-ups reached the 15 m contour (Government of Tonga 2022; UNOSAT 2022). Figure 4 indicated that 62% of the population (n = 62,677) lives below the 15 m contour in all of Tonga (61% when restricting to Tongatapu, ‘Eua and Ha’apai islands). The CARE (2022) report recorded 84,776 affected people from the tsunami and volcanic fallout combined. Our map-derived estimations indicate that 74% of buildings in Tonga are located in a coastal tsunami hazard area below 15 m, meaning all those buildings may have suffered from tsunami run-ups in 2022. The additional impact of thick ash fall causing roofs to collapse tallies with reports claiming that this double disaster affected 80% of Tonga (OCHA 2022a) and caused almost 100% damage to buildings (GFDRR 2022). Here we gather information restricted to tsunami damage and discuss links with the results obtained through the methodology.
UNOSAT (2022) provides an overview of the building damage for several districts. An example of this information is given in Table 4, with direct links to the number of buildings, populated areas and inhabitants for each district analysed. As shown, the information supplied is imprecise but can be advantageously refined by using the dasymetric maps produced in this study for obtaining a rapid assessment of the situation and a first-order, reasonably accurate estimation of vulnerable buildings and number of residents in population clusters. As explained in Fig. 5, ‘Eua and Ha’apai shows sharp contrasts in topography, but the tsunami had similar impacts on structures standing below the 15 m contour on both islands.
Table 4
Comparison of damages to buildings between UNOSAT (2022) data and results from this study.
Information provided by UNOSAT (2022) | Information gained from this study |
Island | District | Buildings damaged | No. of populated area units | Existing buildings | Estimated no. of inhabitants |
‘Eua | Fo’ou | 1 | 13 | 776 | 2150 |
Prope | 139 | 21 | 1058 | 2795 |
Ha’apai | Lulunga | 11 | 7 | 325 | 923 |
Mu’omu’a | 145 | 4 | 305 | 432 |
‘Uiha | 4 | 5 | 407 | 695 |
UNOSAT Maps (2022) provides an atlas of disaster-related damages on Tonga based on satellite imagery. The assessment focuses on a selection of villages highly affected by the tsunami, with a confirmed number of damaged buildings. The methodology implemented here shows an improvement in detecting the number of potential buildings impacted, while also linking the geographic information directly to an estimated maximum number of vulnerable inhabitants (Table 5).
Table 5
Comparison of damages to buildings in specific populated areas.
Information provided by UNOSAT Maps (2022) | Information gained from this study |
Location | No. of buildings damaged | Populated area | Elevation (m) | No. of buildings | No. of inhabitants |
Tungua | 11 | Tungua | < 10 | 82 | 178 |
Fonoifua (Mu’omu’a) | 30 | Fonoifua | < 10 | 30 | 25 |
‘Atata | 72 | ‘Atata | < 10 | 89 | 122 |
‘Eua | 48 | ‘Ohonua (Prope) | < 15 | 118 | 254 |
6 | Ta’anga (Prope) | < 15 | 6 | 42 |
1 | Futu (Fo’ou) | < 20 | 7 | 23 |
Mango (Mu’omu’a) | 26 | Mango | < 10 | 26 | 27 |
Nomuka (Mu’omu’a) | 61 | Nomuka1 | < 10 | 184 | 227 |
4 | Nomuka2 | < 15 | 4 | 28 |
Our estimation of buildings damaged is often close to, although occasionally much higher, than the imagery-based number reported by the UNOSAT survey. However, this overestimation, compared to the existing dataset of buildings per village, clearly highlights an improvement in the results (Table 6). A comparison coefficient c, defined below, is calculated to quantify the gain in accuracy when estimating vulnerable buildings using the three datasets: (i) the number of buildings in the village (bv) based on the TSD (2019) data, (ii) the number of buildings in the populated area identified in this study (bpa), and (iii) the number of effectively damaged buildings (bd) using UNOSAT Maps (2022).
$$c=\frac{(bv-bpa)}{(bv-bd)}$$
Coefficient c ranges between 0 and 1. The closer c gets to 1, the closer the estimation from this study approximates the actual damage estimated by UNOSAT. As c approaches 0, the estimation obtained stays close to the TSD data, with little improvement. Among the examples compiled in Table 6, improvement in the precise knowledge of vulnerable buildings is gained for 4 locations, where c > 0.85. Tungua, ‘Atata and Nomuka1 perform less well, perhaps because our approach only considers the run-up distance of the tsunami, i.e., the elevation range of the impact, and not the form of tsunami propagation or disparities in the intrinsic mechanical strength of the buildings exposed to it. Fonoifua and Mango were completely destroyed by the tsunami. This is confirmed by numbers from the TSD dataset as well as the study dataset, and in itself is a validation of the methodology. Furthermore, any overprediction of building damage in this study may advantageously compensate for likely errors in the definition of the coastal belts when based on SRTM elevation data, which have generated underprediction in other studies (Kulp and Strauss 2016).
Table 6
Comparison on the estimation of damages to buildings in specific populated areas.
Location | No. of buildings | Coefficient c |
Damaged bd | In the populated area bpa | In the village bv |
Tungua | 11 | 82 | 83 | 0.01 |
Fonoifua | 30 | 30 | 30 | 1 |
‘Atata | 72 | 89 | 104 | 0.47 |
'Ohonua Ta'anga Futu | 48 | 118 | 500 | 0.85 |
6 | 6 | 95 | 1 |
1 | 7 | 99 | 0.94 |
Mango | 26 | 26 | 26 | 1 |
Nomuka1 Nomuka2 | 61 | 184 | 249 249 | 0.41 |
4 | 4 | 1 |
The good match between the UNOSAT data and our mapping results presented in Tables 5 and 6 highlights the value of the methodology presented here. Geographic precision is additionally provided by the GIS maps of populated areas, here shown for ‘Eua in Fig. 7, and for the three islands of the Ha’apai group in Fig. 8, which were severely affected by the 2022 tsunami in all of Tonga. Entire villages on Mango and Fonoifua were swept away, 13 houses were flooded between the coast and the lake in Nomuka, and the shoreline retreated by up to 10 m on Nomuka and 30 m on Mango (CEMS 2022; GFDRR 2022; Pleasance 2022; UNOSAT Maps 2022).
In the context of a future tsunami event, maps such as Figs. 7 and 8 can easily be used to precisely locate in each village how many inhabitants were potentially vulnerable to that particular hazard and could have been affected by its estimated magnitude. Such prior knowledge should serve as an important decision-making tool. For example, by coupling the population cluster maps with tsunami hazard scenarios, this would help to focus attention on villages most exposed to a given tsunami type and propagation pattern. It would also help to calibrate the logistics of rescue operations (e.g., quantities of freshwater, first-aid equipment and food to be delivered), including passenger capacity on vessels or aircraft for the temporary relocation of disaster victims. Despite more abundant first-response and healthcare staff and facilities on this main island, residents on its west coast were nonetheless highly impacted by the 2022 tsunami. The maps and tables showcased in this study also highlight the value of precise and regular census data in geographically isolated islands, particularly given that urban growth and population migration between islands occur continuously and are unlikely to decline in the near future (Lolohea 2016). For example, more than 40% of the population has moved away from the Ha’apai island group since 2011, mostly settling on Tongatapu (GFDRR 2022).
4.2 Case study #2: 2009 tsunami
The second test case for the dasymetric mapping methodology presented here is the tsunami triggered in Samoa by the September 2009 earthquakes, claiming 9 lives on Niuatoputapu and Tahafi islands, both situated in the northeast of Tonga (Lay et al. 2010; World Bank 2022).
On Tahafi, run-ups reached 15 to 22 m on the southwestern side, damaging fishing boats and one house (Clark et al. 2011; Fritz et al. 2011; Okal et al. 2010; Wilson et al. 2009). Despite these high values, Tahafi is a steep-sided island with currently 31 inhabitants and 38 buildings counted, all above 30 m elevation. On Niuatoputapu, the villages of Falehau, Vaipoa and Hihifo are situated on the northwest shore, where run-ups reached the 5 m contour (Clark et al. 2011; Wilson et al. 2009). Several reports indicate around 135 to 145 buildings damaged out of a reported total of 225 to 228, mostly in the main village Hihifo and all occurring below 5 m elevation (Fritz et al. 2011; Government of Tonga 2009a, 2009b; WHO 2009; World Bank 2022).
Using the 2016 census data, Hihifo village layout is overlaid in Fig. 9 on the post-tsunami survey of buildings location in 2009 (Clark et al. 2011). Only 7 buildings below 5 m elevation were spared out of the 44 existing before the 2009 tsunami. Whereas 38 inhabitants are currently recorded as residing below the 5 m contour, population estimates at the time were closer to 242. These figures highlight the magnitude (57%) of post-tsunami emigration, not just in temporary buildings in Falehau (Clark et al. 2011; WHO 2009), but by a long-term decision to definitively relocate housing away from areas exposed to coastal hazards, generally to higher elevations. Niuatoputapu island nonetheless appears to benefit from the coral reef surrounding the island, which protected the villages from excessive run-up magnitudes on its northwest coast (Fritz et al. 2011). Although not universally verified, tsunami amplitudes can be mitigated by the presence of healthy coral reefs (Fernando et al. 2005; Ferrario et al. 2014; Hardy and Young 1996; Harris et al. 2018; Karim and Nandasena 2022; Kunkel et al. 2006; Monismith et al. 2015; Roger et al. 2014). This unique ecological asset should be integrated in future risk management studies on Tonga.
Higher run-ups occurred at the southern tip of Niuatoputapu along a near-shore fringing reef, with observed flow depths of up to 6 m (Wilson et al. 2009). This is consistent with the damage sustained by the airstrip, which is situated below 10 m and was partially inundated at the southern end of the runway (Clark et al. 2011). This is the only air connection to other islands in Tonga and it highlights the dangers of placing communications infrastructures at low elevations. The only undamaged public building on the entire island was the high school, which stands above the 5 m contour, emphasising here also the critical importance of choosing elevated ground for primary infrastructure whenever possible. Today, the spatial distribution of population densities and the spatial distribution of buildings appears to indicate that the hospital (completely destroyed in 2009), the primary schools, and the churches have been rebuilt above the 10 m contour. This is confirmed by several reports also indicating that water supplies, the police station, and 73 houses were built out of cyclone-resistant material and on safer and higher ground, testimony to a growing awareness of natural hazards in land-use planning on Tonga since 2009 (Government of Tonga 2009a; World Bank 2022).
4.3 Application to other natural hazards contexts
In the aftermath of the 2009 and 2022 tsunamis, the foremost concern was to relocate inhabitants, to clean up, and to stay on alert during the cyclone season. An accumulation of hazards hitting Tonga would significantly damage the infrastructure of the islands and strongly increase population vulnerability. The impact of the 2022 tsunami, for example, was compounded by the ash fall from the eruption, with damaging consequences beyond the tsunami-exposed coastal belt. Adding climate-change-related impacts, Tonga is thus exposed to a cumulative litany of disasters, highlighting the urgent need for disaster management strategies (Ammann 2013; GFDRR 2022; UNISDR 2005; UNISDR 2009).
Although focused on coastal inundation hazard on the basis of an elevation / run-up criterion, with necessary adjustments of parameters and suitable information sources, the methodology presented in this study can address any type of natural hazard. As shown by the latest eruption of Hunga Tonga-Hunga Ha’apai, ash fall, for example, can destroy buildings even kilometers away from the eruption site. By overlapping tracking models of volcanic ash clouds with the mapping toolkit of populated areas (Carn et al. 2009; Filizzola et al 2007; Searay et al 1998; Webley and Mastin 2009), the identification of vulnerable inhabitants can be easily assessed. Moreover, many hazardous volcanic sites, like Italy around Mt. Vesuvius (Gugg 2022), Hawai’i on Kīlauea (Meredith et al. 2022) and Indonesia on Mt. Merapi (Garcia-Fry et al. 2022), are densely inhabited and require regular updates of populated areas. For example, in September 1946 the Niuafo’ou volcano erupted, lava flows and ash clouds destroyed infrastructures and vegetation all over the island (Rogers 1981; Taylor 1991). The entire population (1300 inhabitants) was relocated permanently on other islands. The population data from the last census shows that 517 inhabitants returned permanently to the island among 8 different villages despite the volcano still being highly active (more than 10 eruptions since 1814; Taylor 1991). With 247 buildings scattered along the east coast of the island, the population is still highly vulnerable to volcanic hazards, and getting to know their precise location is a necessary feature for risk management in case of an eruption.
By its geographic position, Tonga is also exposed to strong earthquakes like the previously mentioned Mw 7.2 event in June 1977 causing serious damage on ‘Eua and Tongatapu islands. A similar event today would directly impact 79,556 inhabitants and could damage up to 33,268 service facilities and houses. Thus, a complementary feature to the dasymetric mapping toolkit would be an updated GIS of Tonga with all essential infrastructures explicitly positioned: civil security centers, hospitals, schools, food and freshwater supply centers. The building data could be enriched by an indication of their function in order to differentiate between residential areas and workplaces, which do not share the same population balance during day and night. A simple search on Google Maps and Maps.me confirmed how currently difficult (often impossible) it is to find accurate data about healthcare facilities on islands in Tonga. Thus, an open source GIS stands out as an essential tool for risk management. Similar issues arise in the context of a cyclonic event such as the strongest and dramatic category-5 cyclone (Ian) of January 2014 (Havealeta et al. 2017; Johnston 2015). Damages on the Ha’apai island group were considerable, with 18 villages affected, 1094 buildings destroyed, and 2335 people relocated (Government of Tonga 2014). This study documents 6125 currently vulnerable residents on Ha’apai, and indicates that along the path taken by the cyclone Ian, 2202 inhabitants are now located among 1158 buildings. Finally, reports indicate that 17 schools were destroyed, impacting 1293 children (Government of Tonga 2014), while at the same time many infrastructures already meeting cyclone-resistant building design were saved (World Bank 2022). These examples further highlight the urgent need for a full GIS database capable of specifying precise building functions.