As Needham (2015) concluded based on researching 700 storm surge events: “tropical cyclone induced floods are among the world’s deadliest and destructive natural hazards”. Due to this growing threat, there is a growing need for evacuation policies in case of an impending cyclone, in order to reduce the negative effects of a cyclone. The literature review in this research points out that there are many pre-disaster evacuation models, which focus mostly on either small scale evacuation of a city, or large scale evacuation with the use of motorized vehicles. However, in stretched out rural and developing areas, where access to motorized vehicles is limited or non-existent, residents need ample time to leave the area that is in danger of cyclone induced floods. This means that there should be enough time between the issue of the evacuation order and the moment the cyclone is forecasted to make landfall. However, the sooner the evacuation order is issued, the more uncertainty there is regarding the area that may be subject to the devastating consequences of the cyclone. That means that many residents would be unnecessarily evacuated. This is why the moment of the evacuation order should be at a point where the timeliness of the evacuation order and the uncertainty of the cyclone are optimally balanced. Currently, there does not exist a pre-disaster evacuation model for cyclone-induced floods that accounts for this trade-off. This research uses concepts of existing evacuation models and extends them with the trade-off between timeliness and uncertainty reduction in order to research the added value and applicability of this trade-off. This leads to the following research question: How to balance the trade-off between timeliness and uncertainty reduction in making shelter location decisions in rural and developing areas, under impending cyclone induced floods, while accounting for behavioral aspects of the vulnerable residents? The computer model that is developed to answer the research question consists out of three parts, which are called the building blocks of the model. The first building block translates discrete forecast reports about a cyclone into an area that is vulnerable and should be evacuated, because it will possibly be affected by the cyclone. These forecast reports form the decision points for an evacuation moment. Based on this vulnerable area, possible shelter locations are found in the surroundings of (horizontal evacuation), or on higher grounds within (vertical evacuation), the vulnerable area using a shelter searching algorithm. This algorithm uses a certain safety margin, which is the distance between the possible shelter locations and the vulnerable area to look for possible locations that can shelter the evacuees. The second building block is the optimization part. This sub model optimizes over the complete set of possible shelter locations and selects a given number of shelters that minimizes the weighted distance between the evacuees and the shelters. This weighted distance minimization is also known as the Minisum optimization model (Boonmee, 2017). The third and final building block simulates an evacuation using the previous generated data and captures the results in key performance indicators, so that the effect of the different policy levers can be compared and the balance between timeliness and uncertainty reduction can be found. The models are connected to each other using a Python interface. The model is applied to two different case studies, on which the results and conclusions are based. Three levers that define the shelter location decisions are found relevant in balancing the trade-off between timeliness and uncertainty reduction. The first is the evacuation moment. It shows that a later evacuation moment reduces the number of total evacuees, but it also reduces the evacuees that are saved from the impact of the cyclone. More precisely, the model shows a clear break point, which means that there is a point in time after which it is no longer possible to evacuate all evacuees in danger. The two case studies have shown that this break point is around two days in advance of landfall. This means that the balance in this trade-off lies before this break point. Too early evacuations however, result in a high number of total evacuees, which is not desired as well. This reduction in evacuees over time is not always linear and depends on how the cyclone is forecasted and the characteristics of the geographical area that is under threat. Therefore, it can be concluded that evacuation should happen before the break point, but the exact moment also depends on the forecast reports and the geographical terrain, and is also dependent on the other two levers. However, it is shown that vertical shelter locations significantly reduce the evacuation time and enable later evacuations or evacuations with less shelters. The second lever is the safety margin. This research concludes that a relatively high safety margin is advised in early evacuation moments, but in later evacuation moments it is advised to make use of vertical shelter locations, which means that a low safety margin should be used. The low safety margin is the only way, in later evacuation moments, to save as many evacuees as possible, but it also reduces the accessibility and security of the shelter locations. The third and final lever is the number of shelters. In early evacuation moments, there are many evacuees, which increases the need for sufficient shelters. Therefore, in early evacuations, it is shown that additional shelters have a relatively high reduction in travel distance and high increase of rescued evacuees when compared to later evacuation moments. However, the marginal benefit of an extra shelter is reduced with each additional shelter, which means that the cost of each additional shelter should be balanced against the reduction in travel time and the increase in safely evacuated evacuees. Furthermore, when a distance minimization model is used, the largest shelters will be located closest to the areas with the highest population density. Regarding the sizes of the shelters, later evacuation decisions often means there is need for more shelters. This means that those shelters tend to be smaller, but there will always be larger shelters because of the larger cities. In summary, three policy levers have been identified that define the shelter location decision and that have an impact on the balance between timeliness and uncertainty reduction. None of these levers can single-handedly define how the right balance, and they should therefore be used all-together to define the best balance the trade-off. However, it has also been found that in both case studies the cyclone evolved differently and the geographical area is far from identical as well, which also influences the right balance. This means that every answer about how to balance the trade-off, will also be different in every case. Additionally, it is concluded that the trade-off between timeliness and uncertainty reduction is especially relevant for evacuees who are evacuating by foot. When their travel speed increases, the relevance of the trade-off decreases. This confirms the hypothesis that most evacuation models with motorized vehicles do not account for this trade-off because evacuees have a higher travel speed. Furthermore, this research concludes that when a cyclone is advancing and there is no time to deploy an evacuation model, a heatmap of the population density, together with an elevation map, can give rough estimates of where the largest shelters should be located. The elevation map gives insights into the possible shelter spots because it will point out the elevated locations, either within or outside of the estimated vulnerable area. Those spots that are located closest to the most dense populated areas will probably prove to be suitable shelter locations. Furthermore, in both case studies it is shown that the latest evacuation moment is around two days in advance of landfall of the cyclone and that after that moment it is highly advised to make use of vertical shelter locations. To conclude, this research recommends that the evaluation of the model results will be changed from retrospective to prospective, which means that the model can be deployed in real-time disaster management. Only then, the real value of the model can be shown.