Extreme precipitation trend estimation in Conterminous United States (CONUS)

There have been studies showing upward trends in the occurrence of extreme precipitation in the United States (e.g., Kunkel et al., 2013). However, the quantification of trend remains a challenge due to data representativeness. We approach the problem by using an explicit statistical model with temporal covariate, namely, a linear trend generalized extreme value (LTGEV) distribution model (Coles, 2001). This model simplifies the non-stationary component by making only the location parameter time-dependent while keeping other parameters, scale and shape, time independent. We define extreme precipitation as a daily precipitation amount which exceeds a threshold set as the top one percentile. Percentile is commonly used in the climate community (e.g, Groisman et al., 2012) to identify extreme events, rather than a fixed threshold, because all stations are treated equally. So, the threshold value set by the top one percentile varies from station to station. Data are COOP daily precipitations available during 1949-2008 (60 years), but stations with consecutive missing observations greater than one year are deleted. The consecutive missing observation (CMO) criterion was an effective approach to de-select stations of questionable quality. We also show 50-year return level fields of LTGEV, and that of no-trend GEV model (NTGEV) to demonstrate LTGEV is a viable option to operationalize climate information products. Preliminary results show that: i) a strongest but localized up-trend in Midwest and Gulf Coast region, ii) a modest up-trend in the New England region, iii) no-trend or down-trend in the most of Western Rockies. Several features are under further investigation to isolate from data representativeness issues.