This paper analyzes the spatiotemporal pattern of gunfire reports collected by the collaborative mobile app Fogo Cruzado in the Rio de Janeiro metropolitan area (Brazil). We apply nonparametric first and second-order inference tools to characterize gunfire behavior, and test whether gunfire patterns meet the assumptions of crime prediction models, such as kernel hotspot maps or self-exciting point process. The kernel intensity estimator describes the spatial distribution of gunfire and identifies chronic hotspots. The nonparametric test for comparison of first-order intensities found differences between gunfires with and without fatalities or police presence. The recently developed log-ratio based first-order separability test found that the spatial distribution of gunfire, fatalities and police presence varied over time. Finally, spatiotemporal inhomogeneous K-tests detected clustering between gunfire events, fatalities and police presence. These results suggest that a self-exciting point process with nonseparable background component is an acceptable model for future development of a suitable approach to forecast gunfire hotspots in Rio de Janeiro.

本文分析了协作移动应用程序Fogo Cruzado收集的枪声报告的时空模式在里约热内卢都会区（巴西）。我们使用非参数的一阶和二阶推理工具来表征枪声行为，并测试枪声模式是否符合犯罪预测模型的假设，例如内核热点图或自激点过程。核强度估计器描述了炮火的空间分布并确定了慢性热点。用于比较一阶强度的非参数检验发现，在有或没有死亡或警察在场的情况下，炮火之间存在差异。最近开发的基于对数比率的一阶可分离性测试发现，枪声，死亡人数和警察在场的空间分布随时间而变化。最后，时空非均匀K检验检测到枪声事件，死亡人数和警察在场之间的聚类。