A hybrid LS-HE and LS-SVM model to predict time series of precipitable water vapor derived from GPS measurements

摘要: Precipitable water vapor (PWV) is one of the key parameters for analysis global climate systems, formation clouds, and short-term forecasts precipitation. Since highly variable both spatially temporally, it can be problematic, requiring modeling efficient techniques in order to measure its quantity. Global Positioning System (GPS) stations provide total column at high quality under all weather conditions with temporal resolution. The technique used this work Precise Point (PPP) single dual-frequency GPS receiver aided by precise orbit clock product. Zenith wet delay (ZWD) estimated through a stochastic process called random walk 5-min intervals. seasonal diurnal variations PWV four located Europe are investigated. Latitude station height most influential factors amount when we compare their amplitudes offsets NASA Water Vapor Dataset-M (NVAP-M) heights. area-averaged accumulated rain time series retrieved from Tropical Rainfall Measuring Mission (TRMM) product also shows that levels do not necessarily lead rainfall. semi-diurnal found much lower than those variations. In validate capability GPS-sensed measurements, two episodes wintertime summertime simulated using Weather Research Forecast (WRF) model. Both observational consistent each other. main objective paper model forecast which would useful information on climatology meteorology. Due distinct harmonic characteristics series, least squares estimation (LS-HE) applied derived 4-year observations. Subsequently, support vector machine (LS-SVM) optimized cross-validation strategy non-harmonic components signal. underlying motivation LS-SVM proficiency methodology precisely data usually non-stationary, defined priori, non-linear major data. modeled hybrid approach (LS-HE LS-SVM) efficiently filter white noise observations then perform forecasting. bias (∼0.37 mm) standard deviation (∼3 mm) observed predicted values presents sound proposed