.png){kind=icon}
.png){kind=icon}
Return
Data-Driven Sparse Sensor Selection for Observing-Network Optimization and Its Impact on Data Assimilation
.png){kind=icon}
PRE
AI
DOI：10.1175/mwr-d-25-0267.1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study primarily focuses on the optimized planning of observational networks, using data assimilation and numerical weather prediction (NWP) as evaluation frameworks. The target observations are ground-based Global Navigation Satellite System (GNSS) precipitable water vapor (PWV) retrievals currently operated in Japan. Using a greedy sparse sensor placement algorithm, 500 operational GNSS observation sites were ranked according to their contribution to minimizing the domain-averaged PWV reconstruction error over a 10-yr period, with respect to the NWP analysis fields used as initial condition in Japan's operational NWP system. The ranked observation sites were divided into three groups top, middle, and bottom 100 and each group was assimilated separately into the data assimilation system to evaluate its impact. Assimilating the top 100 sites yielded better results than assimilating the middle or bottom 100 sites. Signifi-cant improvements were found in the PWV analysis field, mid-lower-tropospheric humidity, and temperature throughout the troposphere. Although the magnitude of improvement decreased with forecast lead time, assimilation of the top-ranked sites led to statistically significant gains in PWV forecasts and maintained significant improvements in humidity and temperature at multiple vertical levels. These results demonstrate that assimilation of objectively selected, high-ranked observations consistently enhances forecast skill compared with lower-ranked observations. Previous studies have rarely examined whether objectively ranked sparse subsets of observations can produce meaningful impacts within operational systems. The novelty of this study lies in linking climatology-based sparse sensor ranking with an operational data assimilation framework. SIGNIFICANCE STATEMENT: This work addresses how to strategically thin a dense network of ground-based Global Navigation Satellite System precipitable water vapor observations while preserving forecast accuracy. By ranking sensors using long-term climatological data, we show that assimilating only the most informative stations yields better forecasts than using the least informative ones even with fewer sensors. The results highlight that uniform spatial coverage and objective sensor ranking can benefit real-world atmospheric models without continuously relocating instruments. This approach can guide the design of efficient, low-cost observation networks for weather forecasting across various variables, not only for climate monitoring networks.
Keywords:
Bayesian methods
Optimization
Data assimilation
Numerical weather prediction/forecasting
Journal
IF:
3
Papers:
103
Citations:
2.9W