Trust mechanisms are a promising solution to the security and reliability dilemma of city scale (smart city) IoT sensor deployments in non-critical scenarios. These deployments often employ low-cost sensor networks, where strict security policies are difficult to enforce, and whose IoT devices may exhibit incoherent sensing behavior. Because of the diverse disturbances that may affect the sensing data path, the integrity of the measurements has to be assessed by the data-consuming application, so that possibly misbehaving sensors and/or biased data can be identified. This paper proposes a trust model which assesses the trustworthiness of data collected from IoT sensors deployed at a city scale. The model is based on a set of context-aware trust metrics that leverage proximity information and context signatures. This intrinsic data-quality approach has the potentials to overcome the limitations of consensus-based metrics, which cannot easily detect anomalies affecting several neighbor sensors. Our new metrics are being validated over a real data-set coming from a wide LoRaWAN-based noise sensor deployment in a urban area.