Analysis of Precipitable Water Vapor (PWV) Data Derived from Multiple Techniques: GPS, WVR, Radiosonde and NHM in Hong Kong

Abstract

Water vapor is a highly variable component in the atmosphere and plays a key role in many atmospheric processes. Accurate measurement of water vapor is vital for improving the predictability of regional precipitation, weather and visibility, especially for highly moist metropolis like Hong Kong. In this paper, we analyze precipitable water vapor (PWV) data in Hong Kong derived from four techniques, i.e. radiosonde, water vapor radiometer (WVR), Global Positioning System (GPS) and numerical weather prediction (NWP)’s non-hydrostatic model (NHM) during a period of approximately 6-month. Radiosonde observations have a low temporal resolution and the observation interval is usually 12 h. The quality of WVR PWV measurements is affected by rainy weather. GPS can provide PWVs at a relatively high temporal resolution, but the vertical resolution is poor. NWP’s NHM is developed based on sophisticated physical and numerical models. The Hong Kong Polytechnic University (PolyU) and Hong Kong Observatory (HKO) collaborate to perform three-dimensional (3D) water vapor modeling using PWV data derived from multiple techniques including GPS, WVR and radiosonde, with an aim to assimilate 3D water vapor data into NWP being run at HKO. This will allow the improvement of weather forecasting capability in Hong Kong. However, it is important to understand the quality and characteristics of the water vapor data obtained from various techniques prior to data assimilation. We analyze the correlation and comparison results of the four water vapor observation techniques. It is found that the correlation coefficients between PWVs from GPS, WVR, NHM and radiosonde data are 0.979, 0.983 and 0.975, respectively. Moreover, the correlation coefficients between NHM and GPS, NHM and WVR, WVR and GPS are 0.975, 0.962 and 0.968, respectively. Intercomparison results indicate that water vapor data derived from these four techniques have very good agreements with each other, with root mean squares error (RMS) ranging from 2.474 to 4.259 mm. Among the four techniques, we find that WVR and radiosonde water vapor observations have the best agreement, with a bias and RMS error of 0.176 and 2.474 mm, respectively.