Fig. 1 shows the results of the analysis performed on the activity types and damage components, highlighting that damage composition, for each activity type, shows some common peculiarities in the case studies. The graphs for Commercial show a very similar behaviour in all events: about half of damage is related to stock, 30% to equipment and 20% to structure. A similar behaviour is described in Gissing and Blong (2004) for the flood that occurred in the commercial district of Kempsey (Australia), where damage to building was accounted for 15% and to contents 85% of the total. Even though damage to stock often represents the main portion of damage to commercial businesses, assessing its value is one of the most demanding challenges. The appraisal of the exposed value and the potential damage to stock must consider the variability of goods constituting the stock, the variability of their costs, and the variability of the amount of stock in different periods of time (during the day, the week, or in specific seasons). However, based on the achieved results, if damage to structure and equipment is known, it is possible to estimate the order of magnitude of damage to stock as well.
With respect to the other categories, also the Manufacturing shows similarities in all case studies. Damage to equipment is the most important component: it covers 44% of the total, stock covers 40% and structure covers the smallest part (16%). The category Restaurant shows that costs mainly derive from damage to equipment while damage to stock is less important. For this category, Lodi and Enza case studies are however very unrepresentative because of the low number of data available for the analysis. Concerning Office, most of costs are caused by damage to equipment with a mean damage of about 50% of the total. Lodi and Enza show a different trend of damage, due to the low amount of data.
For the “Office” category, considering that most of offices are located in civil buildings, a suitable strategy to assess damage to structure could be adapting models developed to assess damage to residential buildings. Indeed, damage assessment to the residential sector is based on more developed and well characterised models if compared to those applied to the economic activities (Molinari et al. 2019, Hasanzadeh Nafari et al. 2016).
With respect to the relation of damage with water depth, Fig. 2 shows an increasing trend of damage, in particular when it is not classified per activity component (i.e., total damage). Weakest trends observed for damage components may be linked to the low quality of data, related to both damage and activity surface. The former was derived from citizens’ declarations that could use different criteria to assess the costs; the latter was partly derived by citizens’ declarations, and partly computed as the entire area of the building where the activity is located, which may not coincide with the real activity area. Moreover, many activities declared a null damage (or did not declared damage at all) for some damage components (i.e., structure, equipment and stock). These null values were taken into account in the calculation of the average damage, as we could not speculate on their justification (e.g., whether it was actually null or, for example, paid back by insurance companies). We also tried to investigate the relation between damage and water depth per activity category (not shown here), but no trend was visible in this case as this operation implies significantly reducing the number of data available in each class. Similarly, the analysis of Gissing and Blong (2004) on the relationships between water depth and damage to contents and buildings did not highlight any significant correlation between the variables, although in that case the total damage was considered. Other studies showing increasing trends, as Kreibich et al. 2010, are characterised by a high number of data or do not distinguish among activity types or damage components.
With respect to the relation with the activity size, graphs in Fig. 3 confirm the expectation that damage increases with the activity surface and that the activity size is a significant variable in computation of damage magnitude.
5.1 Usability of results
Results shown in the previous sections represent a valuable knowledge for a first estimation of damage to economic activities in the Italian context.
Table 4 summarises the main findings of the analysis as reference damage values that can be used to assess the order of magnitude of damage in case the modeller knows, or not, the surface of the activities. Indeed,
Table 4 provides reference damage values expressed in €/unit or €/m2. The reference values “1.a” were computed as the ratio between the damage (sum of equipment, stock and structure) and the number of activities for the categories Manufacturing, Commercial, Restaurant and Office, while the value “1.b” as the mean damage for all the economic activities in the dataset (NACE Code B, C, F, G, I, J, K, L, M, N, R, see section 3). Analogously, damage values “2.a” and “2.b” were computed as the ratio between the damage and the sum of the surface of the activities. Damage per component (i.e., structure, equipment and stock), in both cases “1.a” and “2.a”, was derived by multiplying the damage by the percentages in Fig. 1. The reference values “3” can be used in case both the surface of the activity and the water depth are known. For this scenario, the table supplies the expected damage for different ranges of water depth, without differentiating by component and activity type, because a clear relation between damage and water depth was visible only for the total damage (Fig. 2).
Table 4 represents, in practice, a solution for a rough estimation of flood damage to economic activities in the Italian context. We implemented a simplified approach for verifying the consistency of results and the estimation bias, applying the model to the original case studies: Lodi, Secchia, Sardegna and Enza. Only Manufacturing, Commercial, Restaurant and Office are considered for the simulation and only activities with information about water depth and surface were used in the analysis. Table 5 shows the results of the analysis, comparing simulated and observed damage in total terms.
Table 5 shows that knowing the surface of the activities improves the damage estimation for most of the case studies. In fact, damages simulated with values “2” are more similar to the observed damage than those with values “1”. Damage of Sardegna is underestimated using reference damage expressed in €/m2, while that of Enza using reference damage in €/unit. For Sardegna, the average damage values in €/m2, calculated with data used for the validation, are significantly higher than the reference values, probably due to the low value of surface that characterizes the activities in this case study. For this reason, damage of Sardegna activity is underestimated using the reference values “2”. Differently, the case study of Enza contains data with high values of damage and surface, but they are considerably fewer if compared to the other case studies. For this reason, they have a lower weight in the computation of refence values and the damage is strongly underestimated using the reference value “1”, that does not account the surface.
Table 5 shows that the information about the activity type (reference model “a”) does not improve damage simulation in all case studies (e.g., for Lodi event, the simulations “2.b” is better than simulations “2.a”). Still, Table 6, that compares observed and simulated damage per activity category, without distinguishing among the case studies, shows that the information about the activity type tends to improve the results. Damage in the scenarios “1.b” and “2.b” was computed using the same reference values per activity type. The reference values “3”, which consider the water depth, provide similar results to the values “2”, since both models consider the activity surface. However, reference values “3” could be useful when mitigation measures must be defined/designed as they enable to assess the damage for different ranges of water levels.
Fig. 4 compares instead point values of observed and simulated damage, for two reference simulations, to give a picture of estimation uncertainty. Scatter plots in Fig. 4 show that, for the single activity, the damage could be significantly over or under-estimated. It implies that reference values can be used for estimating the overall damage to all affected companies, but they are not recommended for the estimation of damage at the microscale.