At the occurrence of the Tokyo Bay northern earthquake (M 7.3), about 3.39 million evacuees are presumed to be occurred in the entire Tokyo area, and it is assumed that about 2.2 million people will be forced to live in evacuation shelters. Based on this assumption, the Tokyo Metropolitan Government has designated evacuation shelters that can accommodate a maximum of approximately 3.28 million people, including the primary evacuation shelters and secondary evacuation shelters. In order to suppress various confusion and troubles at evacuation shelters, it is necessary to predict the degree of congestion at each evacuation shelter, and to consider about setting up evacuation shelters and measures for evacuees. However, the extent of physical damage, the number of evacuees, the scale of evacuation shelters, and their spatial distributions are not necessarily balanced, so it is not easy to estimate the degree of congestion at evacuation shelters. In this paper, we propose a model describing evacuation behaviors caused by physical damage at a large earthquake, and estimate the degree of congestion at each evacuation shelter in the Tokyo ward areas assuming the Tokyo Bay northern earthquake (M 7.3). Also, we examine countermeasures for reducing the congestion degree.

 First, we proposed a shelter selection model that estimates physical damage (building damage, fire damage, water-supply failure, power failure, elevator inoperability) on a building unit basis, a model for estimating whether to evacuate or not, which evacuation shelter will be selected. The parameters of the shelter selection model were estimated using the survey data of the Kobe earthquake (1995) and showed that the size of evacuation shelters, the distance from home to evacuation shelters, and the school districts are strongly influencing. Furthermore, when these estimated parameters were applied to the survey data of the Kumamoto earthquake (2016), it was confirmed that the results show the same compatibility as the survey data of the Kobe earthquake.

 Next, simulation analysis based on the evacuation behavior model was carried out assuming the Tokyo Bay northern earthquake (M 7.3). As a result, it was found that about 2 million people will evacuate to shelters throughout the area, and about 700,000 people will evacuate due to building damage and water-supply failure, respectively. Also, after 48 hours, more than half of the evacuation shelters showed the possibility of exceeding the number of persons that can be accommodated. In particular, in Arakawa, Sumida and Ota wards, the possibility that severe congestion exceeding the congestion degree of 3.0 (the number of evacuees is 3 times or more than the capacity of shelter) was shown.

 Finally, paying attention to water-supply failure which is one of the major evacuation factors, we calculated the expected value of the number of evacuees caused by water-supply failure for each water-supply pipe (the value of sheltering importance degree coefficient). We executed a simulation to verify the effect of congestion reduction by preferentially earthquake-proofing from the water pipeline where this value is large. As a result, the number of evacuees was reduced by about 240,000 even by treating 0.25% (about 47 km) of pipelines of non-earthquake-resistance pipeline to earthquake-resistant. Also if the 2.0% (about 376 km) of non-earthquake-resistance pipeline was improved to earthquake-resistant, the number of evacuees decreased to about 170,000. We demonstrated that promoting earthquake resistance of water pipeline by referencing the value of sheltering importance degree coefficient is very effective and efficient means to reduce the congestion of evacuation shelters.