Adaptive Sensor Sampling and Data Center Offloading for Scalable Smart City GIS Models

Abstract

Smart city geographic information system (GIS) platforms depend on dense networks of heterogeneous sensors, including traffic loops, CCTV, air-quality nodes, water-level gauges, and mobile crowd-sensed streams. These sensors keep a city's spatial digital twin continuously updated. But sending every raw reading to a central data center does not scale. It saturates uplink bandwidth. It inflates data center compute and storage costs. And it adds latency that undermines time-critical applications such as incident response and adaptive traffic control. This paper proposes an artificial intelligence (AI) driven framework for adaptive sensor sampling and edge-to-data-center offloading, guided explicitly by GIS context. The framework splits decision-making across three tiers: an edge sensing tier, a GIS context and policy tier, and a data center analytics tier. It formulates the sampling and offloading problem as a context-conditioned Markov decision process (MDP), in which spatial priors, such as zoning, points-of-interest density, historical incident maps, and proximity to critical infrastructure, shape the sampling rate, feature compression level, and offload priority assigned to each sensor stream. A reinforcement-learning (RL) policy is trained to maximize a reward that jointly captures event-detection accuracy, communication cost, energy consumption, and data-center processing load. We evaluate the framework through simulation across four canonical GIS zone types: residential, commercial, transit hub, and incident zone. The proposed GIS-context policy cuts uplink data volume by up to 82% relative to fixed-rate sampling. It keeps event-detection accuracy above 94% in the 55–75% bandwidth-reduction operating region, outperforming both static and threshold-based adaptive baselines. These results show that spatial context is a first-class signal for scalable smart-city sensing pipelines, not merely an annotation added after data collection.