Assessing Methods to Temporally Disaggregate Precipitation for Model Simulation and Calibration

While most hydrologic models require hourly inputs of precipitation for their simulations, observational and forecast data sets are typically not produced at such high temporal resolution. For example, outputs from weather and climate forecasting models are produced at temporal resolution of 3 or 6 hour accumulations.  Therefore, a statistical approach is needed to disaggregate the forecasts to hourly intervals. In this study, we evaluate three relatively simple and computationally inexpensive disaggregation methods to construct hourly precipitation time series from 3-hourly and 6-hourly precipitation using the NCEP Stage IV hourly precipitation analysis data. At each grid point, the observed hourly data were aggregated to 3-hourly and 6-hourly values to simulate a lower temporal resolution observation, and were then statistically disaggregated to yield hourly outputs. The first approach was to linearly interpolate between 3-hourly or 6-hourly accumulations to yield hourly values. The second approach was to disaggregate based on the tendency generated by a non-dimensional diurnal rainfall distribution computed based on historical high-frequency rainfall observations at each grid point, and for each season. The third method used the same approach as the second, but in this case a quantile-to-quantile mapping was applied in order to select only those historical diurnal cycles matching the total accumulation. Results show that the disaggregation of rain based on historical diurnal cycles is the most skillful in retaining the statistical characteristics of the hourly rainfall.