3.1.1. Population Increase and Decrease in Flood-Affected Areas

A comparison of population change before and after the flood shows that the inundated area had a decline of about 5% relative to the non-inundated area (Fig. 1). Previous studies have measured household-level depopulation in large-scale floods. Namikawa et al. ²⁵ reported a 6%–16% decline by age cohort within the inundated zone of Kurashiki City, Okayama Prefecture, after the July 2018 heavy rain, which caused 52 fatalities (excluding disaster-related deaths) and destroyed 4,645 houses, with an additional 1,331 houses partially destroyed, partially damaged, or flooded above floor level ²⁶. In Joso City, Ibaraki Prefecture, which was affected by the September 2015 heavy rain and levee breach along the Kinugawa River, population in the inundated zone fell by 1%–3% by age cohort. This event resulted in two deaths (excluding disaster-related deaths) and damage to 53 houses completely destroyed, 1,581 houses severely destroyed, 3,491 houses partially destroyed, and 3,216 houses flooded above or below floor level ²⁷.

Although only 11 houses were completely destroyed in Akita Prefecture during the July 2023 event, far fewer than in large-scale floods, the 5% population decline within the inundated area was greater than that observed in Joso City and only slightly less than in Kurashiki City. Kurashiki’s recovery plan included extensive levee reinforcement, new evacuation routes, and significant budgets for temporary housing, permanent rebuilding, and community restoration ²⁸. Joso City carried out basin-wide levee upgrades, channel dredging, and community and childcare support programs ^29,30. In contrast, Akita City’s actions have so far been limited to installing drainage pumps and conducting about 2 km of minor channel dredging ³¹. These limited structural measures may have contributed to the observed depopulation following this small- to medium-scale flood.

Within the inundated zone, the difference in average daily population for each mesh was calculated by comparing the month before and the month after the flood (Fig. 2). On average, population per mesh declined by 1.1%. The mesh with the largest decrease lost about 30% of its population, while the mesh with the largest increase gained about 40%. Although overall depopulation was generally smaller than that reported for large-scale floods, localized decreases sometimes exceeded those of large events. This finding indicates that small- to medium-scale floods can still cause severe population losses at fine spatial scales.

3.1.2. Regional Typology Based on Population Change

To examine spatial heterogeneity within the inundated zone, we clustered the population-change trajectories of all affected meshes (Fig. 3). We used the mean resident population during the month before the flood as the baseline. Based on this baseline, we classified each mesh into one of three groups: stable, decreasing, or increasing.

Based on previous studies ²², the initial number of clusters was set at approximately three to four. When three clusters were specified, the resulting groups most clearly represented the key characteristics of the data. In contrast, increasing the number of clusters to four or more mainly resulted in additional clusters with only a single element. Therefore, the number of clusters is set to three.

In the decreasing cluster, the population does not decline immediately after the FED but begins to decrease several days later. Although no specific event occurred during this period, one possible explanation is evacuation to secondary destinations. Because the data set in this study has a spatial resolution of 500-m meshes, population movements that occur entirely within the same mesh are not captured. Each mesh contains multiple primary evacuation shelters, but many of these facilities, such as elementary schools, are not suitable for long-term stays. Residents may have remained at primary shelters within the same mesh for several days after the FED, then relocated to more comfortable secondary evacuation destinations, such as hotels or relatives’ homes, in other areas.

The resulting clusters are color-coded on the maps in Fig. 4 (Akita City) and Fig. 5 (Gojome Town). The maps indicate significant depopulation in the heavily damaged town center of Gojome, while population growth was concentrated mainly in the southern part of Akita City. In contrast, the Higashidori, Hiroomote, and Narayama Districts, where Akita Prefecture ¹⁷ reported the highest numbers of damaged houses, did not show clear population decrease. Marked depopulation occurred in several districts with comparatively fewer damaged houses. These patterns support the view of Namikawa et al. ²⁵ that post-flood out-migration is influenced by factors beyond physical housing damage alone, even in small- to medium-scale floods.

Some meshes within the inundated area showed population increases. This result indicates that, despite the clear future flood risk, residential inflow may occur, which can increase the number of people exposed to future flooding.

3.2.1. Results on Characteristics of Respondents

Attributes and damage-related characteristics of respondents were aggregated for the three population-change clusters and tested for significance using \(\chi^{2}\) tests or one-way ANOVA. Variables with significant inter-cluster differences are shown in Figs. 6–13. At the 5% level, residence damage (Fig. 6), changes in expenditure (Fig. 8), damage amount relative to household income (Fig. 10), and score of stress (Fig. 11) differed significantly. Flood damage type (Fig. 7), changes in savings (Fig. 9), occupation (Fig. 12), and years of residence (Fig. 13) differed at the 1% level.

Figure 6 shows residence damage severity by cluster. In the stable cluster, 1.9% of households were classified as completely destroyed, 16.9% as medium-scale partially destroyed, and 20.0% as partially destroyed, resulting in 38.8% of households with damage at least at the partially destroyed level. In the decreasing cluster, no household was completely destroyed, and 25.9% were in the partially destroyed or worse categories. In the increasing cluster, no household was completely destroyed or medium-scale partially destroyed, and only 7.1% were classified as at least partially destroyed. Therefore, the proportion of severely damaged households was highest in the stable cluster, lower in the decreasing cluster, and lowest in the increasing cluster.

All three clusters showed approximately 20% “inundation-only” cases. However, the proportion of sediment intrusion was highest in the stable cluster and lowest in the increasing cluster (Fig. 7).

Flood impacts on household finances varied across clusters. The stable cluster had the highest proportion of respondents reporting increased expenditure (Fig. 8), reduced savings (Fig. 9), and losses exceeding 300% of annual income (Fig. 10). In contrast, the increasing cluster consistently experienced the lowest financial burden. Average stress scores followed a similar pattern, with the highest levels in the stable cluster and the lowest in the increasing cluster (Fig. 11). These results suggest that greater physical damage is associated with higher psychological distress.

Overall, residents in the stable cluster experienced the most housing damage, financial strain, and stress. Those in the increasing cluster experienced the least, while the decreasing cluster fell in between. These differences suggest that economic and psychological effects of flood damage persist for households that remain in severely affected areas. The increasing cluster had a significantly higher proportion of company employees and a much shorter average residence period (Figs. 12 and 13). This indicates that meshes in this cluster are likely newer residential areas with working-age households, which helps explain why population can increase even after a documented inundation event.

3.2.2. Changes in Preparedness Measures

A histogram of the change in the number of preparedness measures per respondent before and after the disaster (Fig. 14) shows a mean increase of 1.5 measures, indicating that preparedness generally increased after the event, even in a small- to medium-scale flood setting. For each of the 20 items, the difference between pre- and post-flood implementation rates was calculated, ranked by magnitude, and presented in Fig. 15. Respondents who left more than half of the items blank were excluded from this analysis.

The ranking shows significant variability. The largest increases occurred in soft measures, such as “Recognized that one’s own house lies within the inundation area” (Item 20), “Discussed disaster countermeasures with family” (Item 15), and “Consulted the hazard map” (Item 19). In contrast, hard measures to prevent inundation, such as “Raised the house” (Item 7) and “Performed waterproofing work on interior or exterior walls” (Item 3), showed smaller increases. Costlier actions intended to speed post-flood recovery, such as “Kept valuables on the second floor” (Item 4) and “Replaced tatami with wooden floor in preparation for flooding” (Item 1), ranked among the top five. After the flood, households tended to prefer hard measures that support rapid recovery rather than those that prevent inundation, while also assigning equal importance to soft measures.

Figure 16 shows a scatter plot with the pre-disaster implementation rate of each preparedness measure on the x-axis and the post-disaster implementation rate on the y-axis. Examination of the post-disaster rates indicates that even for the housing-related measures with the highest uptake (“Kept valuables on the second floor” (Item 4) and “Replaced tatami with wooden floor in preparation for flooding” (Item 1)) actual implementation remained below 30%. Because similar flood damage may occur again in the affected area, it is necessary to promote household-level housing measures that individuals can take, whether through public assistance or self-help.

Figures 17 and 18 show preparedness change in relation to residence damage and flood damage type, respectively. One-way ANOVA shows significant differences at the 1% level in both cases. Preparedness increased most among households that experienced mud and sediment inflow, likely because post-disaster cleanup was prolonged and labor-intensive, which increased motivation to reduce future damage. Households with greater overall damage showed a similar pattern.

Respondents were divided into tertiles based on stress-scale scores and the “community attachment and disaster heritage” factor of the disaster coexistence culture attitude scale. Figs. 19 and 20 present the relationships between these variables and changes in preparedness. ANOVA showed significant differences at the 1% level for stress and at the 5% level for disaster-culture attitude, confirming positive correlations: higher stress or cultural-attitude scores were associated with greater increases in preparedness.

By contrast, the three population-change clusters showed no significant differences in preparedness changes. No significant correlation was found between preparedness change and other respondent attributes. These results indicate that preparedness behavior depends less on regional demographic characteristics or general household attributes, and more on direct disaster experience and individual psychological or cultural factors. Therefore, even in areas with clear future flood risk, households without direct disaster experience may remain reluctant to improve their preparedness measures.

3.2.3. Results of the Recovery and Reconstruction Calendar

Figure 21 shows the results of the recovery and reconstruction calendar. Cumulative percentages were calculated after excluding non-respondents for each milestone. Table 1 presents the number of hours required for each item to reach the 50% attainment threshold. At the aggregate level, “① I understood the extent of damage,” “② I felt safe,” and “③ I was prepared to have an uncomfortable life for a while” recovered first, which aligns with previous studies ^8,9. Recovery curves for the other items varied. “④ Business offices resumed operations,” “⑫ Local schools resumed operations,” and “⑪ Local roads were restored” (the relatively “hard” aspects) also reached 50% in about 102 hours, similar to the first three items. In contrast, “⑥ The disaster no longer impacted my household,” “⑦ Everyday routines resumed,” “⑧ Local activities were restored,” “⑨ I no longer considered myself a disaster victim,” and “⑩ The local economy was no longer influenced by the disaster” (the “soft” dimensions) took longer, requiring more than 103 hours to exceed 50%. This pattern is similar to that observed for the Kii Peninsula torrential rain disaster ⁸.

Previous work ⁹ has argued that, in earthquakes and large-scale floods, life recovery generally follows this order: first, victims prepare for a period of discomfort and understand the extent of damage; second, they feel safe, and offices and schools reopen; third, daily life stabilizes, housing issues are resolved, and personal finances are no longer affected by the disaster; fourth, individuals stop identifying as victims; and fifth, the local economy is no longer affected by the disaster. Our results show that, in the Akita heavy rainfall case, stages one and two were perceived at about the same time, followed by stages three, four, and five in sequence.

Recovery after small- to medium-scale floods is characterized by a faster initial restoration of hard aspects compared to earthquakes or large-scale floods, while the recovery of soft aspects takes a similar amount of time as in large-scale floods. Across all three clusters, more than 50% of respondents reported milestones such as “① I understood the extent of damage,” “⑪ Local roads were restored,” and “④ Business offices resumed operations” within one week, indicating that these hard aspects were perceived quickly. However, the pace of recovery for other items varied. The increase cluster recovered the fastest, with more than 50% of respondents reporting recovery for all items except “⑩ The local economy was no longer influenced by the disaster” in about 102 hours. The decreasing cluster was slower than the increase cluster but still surpassed 50% for all items except “⑩ The local economy was no longer influenced by the disaster” by 103 hours. The stable cluster was the slowest, taking more than 103 hours to exceed 50%. Differences between clusters were most evident in the perceived recovery of soft aspects related to stages (3), (4), and (5).

Table 2 presents the attainment levels for each item at the survey time point, approximately one year and eight months after the disaster. Several items had not reached 90%, showing that the effects of the flood continued. The perceptions “⑨ I no longer considered myself a disaster victim” and “⑩ The local economy was no longer influenced by the disaster” remained lower than other items, with about 30% of respondents still not feeling recovered. In small- to medium-scale floods, because large-scale destruction of housing and infrastructure is limited, hard aspects recover quickly. In contrast, soft aspects—“⑥ The disaster no longer impacted my household,” “⑦ Everyday routines resumed,” “⑧ Local activities were restored,” “⑨ I no longer considered myself a disaster victim,” and “⑩ The local economy was no longer influenced by the disaster”—recover gradually in stages, similar to other hazards such as earthquakes and large-scale floods. Therefore, even after water recedes and roads and buildings are restored, affected areas need long-term support for the economy, psychological well-being of victims, and community activities at a scale similar to that required for large-scale floods.

Disaster case management ³², which supports victims in their independent recovery, has received increasing attention in recent years. As flood damage becomes more frequent, promoting self-reliant reconstruction in areas affected by small- and medium-scale floods is expected to encourage the use of sustainable and adaptive countermeasures when large-scale assistance is unavailable.