Evaluation of the Reanalyses Products in Detecting Extreme Precipitation Trends over United States
To investigate potential trends and variability in the GEV distribution, the location parameter is considered as a time-variant variable based on a linear trend model. The slope of the linear model is interpreted as the linear trend in probability distribution function of extreme precipitation events. The results show that GEV trend estimates are generally positive but close to zero for the three data products, indicating that overall extreme precipitations have not considerably changed over the whole CONUS. However, regional positive trends appear more apparent in Southern and Eastern part of the U.S., and Gulf Coast area, meaning more extreme precipitations in these regions. The regional trend patterns in Central and Eastern United State seem quite different among the three data products. Compared to CPC gauge analysis, both MERRA and NARR are biased in the Gulf Coast regions, MERRA being more biased. We discuss that this result could be related to the Gulf Coast cyclones that hit the land in that part of the U.S. Moreover, MERRA shows a negative trend in the mid-U.S., while such a trend is not found in CPC gauge data. Generally, NARR is performing better in capturing the trend patterns similar to CPC; knowing that gauge data is included in the NARR data set.
To further understand whether or not potential trends in extreme precipitation events are sensitive to particular seasons, we also conduct a seasonal analysis. For this, we consider seasonal maximum precipitations to be the maximums of all daily precipitations occurred at each season. By narrowing down the analysis to seasonal scale, the signal from the tropical cyclones in Gulf Coast states is found enhanced, particularly during hurricane season. In addition, MERRA shows an incorrect trend in winter season (DJF), which obscures the results in the south central US.