Development of bias-correction PERSIANN-CDR models for the simulation and completion of precipitation time series

Abstract

The availability of continuous long-term precipitation time series with uniform spatial distribution has a significant role in hydrological studies and applications from simple water budget calculation to infrastructure design. However, in numerous parts of the world precipitation observations are sparse, patchy and uneven and hence hydrological research and hydroclimate-related projects are conducted with uncertainty. Therefore, filling gaps in precipitation time series and estimating precipitation in ungauged regions are of prime importance. Satellite-based precipitation measurement methods, among rainfall measuring tools, seem to be the most promising method to provide continuous and high spatiotemporal precipitation data. Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR) is one of the most commonly used precipitation retrieval algorithms, with long term data and high spatiotemporal resolutions that is employed in this study. Biases in the satellite-based precipitation data, as some studies have reported, demand the development of bias-correction models to enhance the accuracy of the satellite-based estimates. First, the performance of the PERSIANN-CDR algorithm is assessed based on 35 years of monthly, seasonal and annual observed precipitation at 49 stations as the benchmark in Fars province, Iran. Results indicate that while PERSIANN-CDR mimics the precipitation time series with a high degree of correlation, it underestimates the precipitation in the province with the Root Mean Squared Error (RMSE), Relative Mean Error (RME) and BIAS of (15.23 mm, 0.23 and 0.77), (37.68 mm, 0.24 and 0.76) and (96.07 mm, 0.24 and 0.76) for the monthly, seasonal and annual precipitation, respectively. Then, a set of bias-correction models is developed to enhance the PERSIANN-CDR estimates with the reduced monthly, seasonal and annual RMSE, RME and BIAS of (9.09 mm, −0.01 and 1.01), (19.36 mm, −0.01 and 1.01) and (41.23 mm, 0.0 and 1.0), respectively. The Bias-correction models are validated with the precipitation time series that were not used in the development phase. The validated models proved as an effective and robust tool in improving the accuracy of the PERSIANN-CDR datasets. We also demonstrate how the proposed models are used to estimate precipitation at ungauged sites and to fill in the gaps in the ground-based precipitation time series. The proposed models provide the practitioners, planners, and decision-makers with a huge precipitation data bank for various hydro-climatological projects in the study area.