In areas of the world where malaria prevails under unstable conditions, attacking the adult vector population through insecticide-based Indoor Residual Spraying (IRS) is the most common method for controlling epidemics. Defined in policy guidance, the use of Annual Parasitic Incidence (API) is an important tool for assessing the effectiveness of control and for planning new interventions. To investigate the consequences that a policy based on API in previous seasons might have on the population dynamics of the disease and on control itself in regions of low and seasonal transmission, we formulate a mathematical malaria model that couples epidemiologic and vector dynamics with IRS intervention. This model is parameterized for a low transmission and semi-arid region in northwest India, where epidemics are driven by high rainfall variability. We show that this type of feedback mechanism in control strategies can generate transient cycles in malaria even in the absence of environmental variability, and that this tendency to cycle can in turn limit the effectiveness of control in the presence of such variability. Specifically, for realistic rainfall conditions and over a range of control intensities, the effectiveness of such 'reactive' intervention is compared to that of an alternative strategy based on rainfall and therefore vector variability. Results show that the efficacy of intervention is strongly influenced by rainfall variability and the type of policy implemented. In particular, under an API 'reactive' policy, high vector populations can coincide more frequently with low control coverage, and in so doing generate large unexpected epidemics and decrease the likelihood of elimination. These results highlight the importance of incorporating information on climate variability, rather than previous incidence, in planning IRS interventions in regions of unstable malaria. These findings are discussed in the more general context of elimination and other low transmission regions such as highlands.