Sensor networks are used for applications in monitoring harsh environments including reconnaissance and surveillance of areas that may be inaccessible to humans. Such applications depend on reliable collection, distribution and delivery of information to processing centres which may involve multi-hop wireless networks which experience disruptions in communication and exhibit packet drops, connectivity loss and congestion. Some of these faults are periodic, attributed to external, recurring factors. In this paper, we study an effective way to forecast such repetitive conditions using time-series analysis. We, further, present an application-level, autonomic routing service that adapts sensor readings routes to avoid areas in which failures or congestion are expected. A prototype system of the approach is developed based on an existing middleware solution for sensor network management. Simulation results on the performance of this approach are also presented. © 2011 IFIP.