Possibility of GPS precipitable water vapour for reservoir inflow forecasting

We investigated the possibility of using GPS precipitable water vapour (GPS-PWV) for forecasting reservoir inflow. The correlations between monthly GPS-PWV and the inflow of two reservoirs were examined and the relationship tested, using a group method of data handling (GMDH) type neural network algorithm. The daily and monthly reservoir inflows were directly proportional to daily and monthly GPS-PWV trends. Peak reservoir inflow, however, shifted from the peak averages for GPS-PWV. A strong relationship between GPS-PWV and inflow was confirmed by high R2 values, high coefficients of correlation, and acceptable mean absolute errors (MAE) of both the daily and monthly models. The Ubon Ratana reservoir model had a monthly MAE of 54.19·106 m3 and a daily MAE of 5.40·106 m3. By comparison, the Lumpow reservoir model had a monthly MAE of 25.65·106 m3 and a daily MAE of 2.62·106 m3. The models using GPS-PWV as input data responded to extreme inflow better than traditional variables such that reservoir inflow could be predicted using GPS-PWV without using actual inflow and rainfall data. GPS-PWV, thus, represents a helpful tool for regional and national water management. Further research including more reservoirs is needed to confirm this preliminary finding.