Predicting precipitable water vapor by using ANN from GPS ZTD data at Antarctic Zhongshan Station

Abstract

Precipitation plays an important role in human activities, and accurate prediction of precipitation is expected to make the arrangements accordingly, especially in Antarctic area with complicated weather conditions. Since directly forecasting precipitation usually requires a lot of meteorological data, which is difficult to be collected in Antarctic area, the precipitation is usually predicted indirectly by using precipitable water vapor (PWV). The PWV can be calculated by Hopfield model using GPS zenith tropospheric delay (ZTD) data if only the temperature and pressure data is available. In this paper, we adopt the artificial neural network (ANN) with genetic algorithm (GA) to predict the PWV of the Zhongshan Station in 6 and 12 h by four different input schemes, including ZTD, ZTD with real-time meteorological data, PWV, and intrinsic mode functions (IMFs) of PWV. The predicted results show that the worst prediction is got by using ZTD sequences with the correlation coefficient of about 0.50. The results of using ZTD with real-time meteorological data and PWV have approximate correlation coefficient of about 0.80. The best prediction is obtained by using IMFs of PWV sequences to predict 6 h PWV with the correlation coefficient of 0.95.