Climate change has increased the frequency and intensity of climate hazards, increasing the importance of monitoring systems for disaster response. Existing climate-monitoring platforms rely on manual database query approaches, which make real-time failure detection difficult. This study proposes a system architecture in which data-collection adapters expose status metrics in a standardized format via real-time endpoints, enabling automated collection via a centralized monitoring system. The system was implemented using open-source technologies (Spring Boot 3.5.6, Java 17, PostgreSQL 17, Docker) and validated through a 48-hour continuous operation experiment and fault-injection testing across five failure scenarios, collecting weather data from the Seoul and Bundang regions via the Korea Meteorological Administration API. The experiment achieved a 100% collection success rate across 192 collection attempts, collecting 6,528 meteorological records with an average collection time of 372 ms. System resource usage remained minimal, with an average memory usage of 481 MB and 0.01% CPU utilization, demonstrating a lightweight operation. The proposed metric-based monitoring approach reduced the failure detection time by approximately 95% compared with manual database query methods (from hours to days to minutes), with over 99% reduction for infrastructure-level failures. Fault injection experiments verified that infrastructure-level failures (network disconnect and container crash) were detected within 5 min through Prometheus target status changes. In contrast, application-level failures (API failure, database failure, and API timeout) were detected within approximately 40 min through counter-based monitoring with automatic recovery in all cases. This study presents a generalized design applicable to various climate hazards, including heat waves, cold waves, droughts, and floods, thereby enhancing the reliability of climate hazard monitoring infrastructure.
Kim et al. (Tue,) studied this question.