Physics-based logic puzzle games present diverse challenges for AI, primarily related to coping with uncertainty in complex continuous environments. As these challenges tie in with the goals of qualitative reasoning (QR), in particular to plan, predict and diagnose physical mechanisms, these games pose a natural benchmark for QR. In this paper we consider the projectile collision games Angry Birds by Rovio Entertainment on which the AI bench- mark AIBirds is based. We investigate whether a qualitative approach to action planning under uncertainty proposed by Ge et al. (2016) can be adapted to the domain of Angry Birds in order to identify targets hittable by multiple rebounds. It turns out that the search space for solving Angry Birds is complexly structured and a fine-grained decomposition is required, leading to high computational costs. This paper presents an analysis of the problem and discusses means to overcome the problems faced. We propose an improved qualitative prediction method that is able to solve hard levels, yet it also suggests the need of complementary methods.