Comparing CLIGEN-generated storm patterns with 1-min and hourly precipitation data from China

Authors: WANG, WENTING - Beijing Normal University; YIN, SHUIQING - Beijing Normal University; Flanagan, Dennis; YU, BOFU - Griffiths University

Submitted to: Journal of Applied Meteorology and Climatology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/23/2018
Publication Date: 8/20/2018
Citation: Wang, W., Yin, S., Flanagan, D.C., Yu, B. 2018. Comparing CLIGEN-generated storm patterns with 1-min and hourly precipitation data from China. Journal of Applied Meteorology and Climatology. 57(9):2005-2017. DOI: 10.1175/JAMC-D-18-0079.1.
DOI: https://doi.org/10.1175/JAMC-D-18-0079.1

Interpretive Summary: Assessment of runoff, sediment loss, and chemical losses from landscapes is important in determining flooding risks, erosion levels, and threats to off-site water quality. However, measurement of these factors is difficult, expensive, and impractical for the majority of lands, thus computer simulation models are typically applied to estimate them. The most important input driving runoff and related processes is rainfall depth and intensity, which can be obtained from weather stations, if available. Rainfall measurements for 30-minute periods or finer are needed to define rainfall intensity parameters for storms, however, many stations only record hourly rainfall (or coarser). This is especially true for locations outside of the United States. In this study we used detailed 1-minute measurements of rainfall depths at 18 stations located in China and aggregated them into hourly values. We found that the maximum 30-minute rainfall intensities computed with hourly rainfall data were lower than those computed using the 1-minute intensities, but there was a clear bias that could be corrected. After correcting for the bias, we found that the hourly data could be used to determine the time to peak intensity and the maximum 30-minute intensity values, and subsequently some related soil erosion model inputs. These results impact research scientists, conservation agency personnel, and others applying natural resource models that require generated climate, and who have observed rainfall station data at coarse scales (1-hour measurements). Application and testing of this approach at other locations in the world may yield similar results, possibly with different bias adjustment factors.

Technical Abstract: CLIGEN (CLImate GENerator) is a stochastic weather generator which has been widely used in the United States and some other regions around the world to generate daily precipitation amount and storm pattern. To compute the two parameters, namely TimePk (cumulative distribution of the time to peak intensity) and MX.5P (the mean daily maximum 30-min intensity for each month) for generating the storm pattern properly, rainfall data at interval = 30-min are required. High resolution rainfall data, however, are not widely available around the world. One-min precipitation data from 18 stations in eastern and central China were aggregated into hourly intervals to evaluate various methods to optimally prepare TimePk and MX.5P for generating storm pattern with CLIGEN. Five sets of the two parameters were used to run CLIGEN to generate weather sequences for comparison: C0 - using the original observed 1-min data; C1a - replacing the TimePk used in C0 with the average TimePk values for the 18 stations; C1b – replacing the TimePk values used in C0 with that computed with hourly data without any adjustment; C2 – replacing MX.5P values for C0 with that calculated with hourly data and modified by an adjustment factor; C3 – replacing both TimePk and MX.5P parameter values for C0 with those calculated with hourly data with adjustment for the peak 30-min intensity. Results showed that: 1) The difference in generated storm patterns between C0 and both C1a and C1b was insignificant; 2) MX.5P computed with hourly data were systematically lower than that computed with 1-min data, and the bias could be corrected by multiplying MX.5P values with an adjustment factor of 1.40 for all months and all 18 stations; 3) Storm pattern generated from C2 and C3 agreed well with that using the original parameter set using the original 1-min data. The mean absolute percentage error for storm duration and peak intensity was less than 8% for the mean, standard deviation and skewness; (4) The R-factor for the Revised Universal Soil Loss Equation and the 10-year storm erosivity (EI) were in good agreement using CLIGEN output C0 and C3. Hourly precipitation data can be used to prepare CLIGEN parameter values for generating storm patterns for sites where data at less than or equal to 30-min intervals are not available.