Adaptive Sampling for Spatiotemporal Anomaly Monitoring in Wireless Sensor Networks

A sentinel-assisted cooperative sensing-control pipeline combining Kalman filter sparse sampling, GLR-based detection, and local alert propagation

Guoqing Lu; Yixuan Sun; Yiwen Jiang; Bernard Butler· 2026· DOI 10.48550/arXiv.2607.15235

The core problem

Long-term environmental monitoring in wireless sensor networks (WSNs) frequently relies on sparse sampling to extend network lifetime. However, sparse sensing can miss short-lived, localized, and potentially diffusive anomalies. The paper addresses this trade-off by proposing a sentinel-assisted adaptive sampling framework that operates as a cooperative sensing-control pipeline for WSN anomaly monitoring. The core problem is that during normal periods, energy must be conserved, but during anomalous periods, high-resolution sampling is required to capture the spatiotemporal evolution of the anomaly. The authors aim to improve anomaly-window visibility without incurring excessive sampling cost. The proposed approach integrates three key mechanisms: (1) Kalman filter (KF) predictive uncertainty-driven sparse sensing during normal periods, (2) hybrid generalized likelihood ratio (GLR)-based detection with node-relative thresholds at continuously sampled sentinel nodes during anomalous periods, and (3) one-hop neighborhood wake-up with recovery-aware alert control triggered by local detections. The framework is evaluated on the Intel Berkeley Research Lab temperature dataset with abrupt

Innovation

Experiments were conducted on the Intel Berkeley Research Lab temperature dataset with abrupt random spatiotemporal anomalies. The primary metric is the anomaly-window sampling ratio (AWSR), defined as the fraction of samples taken during the anomaly window relative to the total possible samples in that window. The proposed method achieves an AWSR of 0.933 in the main experiment, compared to 0.439 for the baseline sparse sampling approach. This represents a significant improvement in capturing anomaly dynamics.Compared to AAS and Adapted e-Sampling, the proposed method improves AWSR while reducing total cost. Specifically, it reduces total cost by 15.4% compared to AAS and by 2.1% compared to Adapted e-Sampling. The total cost includes energy consumption for sensing, communication, and computation. The results demonstrate that the sentinel-assisted adaptive sampling framework effectively balances the trade-off between anomaly visibility and sampling cost.Key quantitative results are summarized in the table below:| Method | AWSR | Total Cost Reduction ||--------|------|----------------------|| Baseline Sparse | 0.439 | - || AAS | - | - || Adapted e-Sampling | - | - || Pro

Long-term environmental monitoring in wireless sensor networks (WSNs) frequently relies on sparse sampling to extend network lifetime. However, sparse sensing can miss short-lived, localized, and potentially diffusive anomalies. The paper addresses this trade-off by proposing a sentinel-assisted adaptive sampling framework that operates as a cooperative sensing-control pipeline for WSN anomaly monitoring. The core problem is that during normal periods, energy must be conserved, but during anomalous periods, high-resolution sampling is required to capture the spatiotemporal evolution of the anomaly. The authors aim to improve anomaly-window visibility without incurring excessive sampling cost. The proposed approach integrates three key mechanisms: (1) Kalman filter (KF) predictive uncertainty-driven sparse sensing during normal periods, (2) hybrid generalized likelihood ratio (GLR)-based detection with node-relative thresholds at continuously sampled sentinel nodes during anomalous periods, and (3) one-hop neighborhood wake-up with recovery-aware alert control triggered by local detections. The framework is evaluated on the Intel Berkeley Research Lab temperature dataset with abrupt random spatiotemporal anomalies.

The proposed framework operates in two distinct modes: normal and anomalous. During normal periods, all nodes perform sparse sensing driven by the predictive uncertainty of a Kalman filter. The KF models the spatiotemporal temperature field, and the predictive variance at each node determines the next sampling time. Specifically, a node samples when its predictive uncertainty exceeds a threshold, which is computed as:

Why it matters

The results indicate that integrating KF-based sparse sampling, sentinel GLR detection, and local alert propagation substantially improves anomaly-window visibility while maintaining a lower sampling-cost trade-off. The use of node-relative thresholds in GLR detection allows sentinels to adapt to local environmental variability, reducing false alarms and enabling reliable detection of subtle anomalies. The one-hop neighborhood wake-up mechanism ensures that the anomaly is monitored collaboratively without requiring global network-wide wake-up, which would be energy-inefficient. The recovery-aware alert control prevents unnecessary sampling after the anomaly has passed, further conserving energy. The comparison with AAS and Adapted e-Sampling shows that the proposed method achieves a better balance between detection performance and energy consumption. The 15.4% cost reduction relative to AAS is particularly notable, as AAS is designed for energy efficiency but may not prioritize anomaly visibility. One limitation is that the evaluation is based on a single dataset (Intel Berkeley Research Lab temperature) with synthetic anomalies. Future work could validate the framework on other environmental modalities and real anomaly events. Additionally, the sentinel selection strategy is not detailed in the abstract; further investigation could optimize sentinel placement for broader coverage. Overall, the proposed framework offers a promising approach for long-term WSN monitoring where both energy efficiency and anomaly detection are critical.

Who should read this

CS practitioners and researchers

Opening member content…