Modeling Joint Relationship and Design Scenarios Between Precipitation, Surface Temperature, and Atmospheric Precipitable Water Over Mainland China

• Main Authors: Olusola Olaitan Ayantobo, Jiahua Wei, Guangqian Wang
• Format: Article
• Language: English
• Published: American Geophysical Union (AGU) 2021-04-01
• Series: Earth and Space Science
• Subjects: Bayesian framework; copula function; Markov Chain Monte Carlo simulation; precipitation; precipitable water; temperature
• Online Access: https://doi.org/10.1029/2020EA001513

Abstract This study employs copula model to discuss joint relationship and design scenarios between precipitation (P), surface temperature (T), and precipitable water (PW) in seven regions and mainland China over 1980–2016 within a bivariate framework. Markov Chain Monte Carlo modeling in a Bayesian framework was applied to calculate copula parameters while best fitted copulas were assessed using Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), Root Mean Squared Error (RMSE), Nash‐Sutcliffe Efficiency (NSE), and maximum likelihood criteria. Results showed that the spatial variations of P decreased from South to Northwest, T decreased from South to North and from East to West. Distributions of P and T were similar, with higher values in regions c and d. PW was higher in the south, with more than 64.54 mm in the southwest, 25 mm in the north‐central. The correlation between PW and P as well as PW and T were higher than 0.74. Birnbaum Saunders distribution was considered appropriate to fit P in regions a and f while P and PW in other regions are good with Generalized Pareto. T performed better with generalized extreme in all regions. Moreover, Gumbel, Frank, and Joe copulas provided good fit for PW and P in regions b, f, and g, respectively while in regions a, c, d, and e, Nelson provided perfect fit. For PW and T, Nelson was a good fit in all regions. The bivariate probabilities and design scenarios in various regions suggest tremendous variations, with regions having high probabilities associated with low return periods. We showed that joint analysis of climate variables gives more robust design quantiles.