2.1. Rain gauge data
We examined daily precipitation data at 27 stations available for North Korea from 1983 to 2007, and at 60 stations in South Korea from 1973 to 2007. For North Korea, we used data provided by the World Meteorological Organization (WMO) to the Korea Meteorological Administration (KMA). For South Korea, the KMA has maintained the daily precipitation data of observatory stations since the early 1970s. Among the 76 stations run by the KMA, we chose 60 stations to exclude those stations which are missing more than 3 days per year and the stations established later than 1973. With regard to the quality of data in South Korea, Chang and Kwon (2007) analysed the station data with Worsley likelihood ratio, cumulative deviations, and the Bayesian procedure, with the result that only a few stations exhibited homogeneity with significance slightly less than 95% confidence level by one or two criteria.
We calculated the total precipitation (Ptot), the number of days without precipitation (Ndry), and the number of days in which the daily precipitation is greater than 50 mm (number of days with the heavy precipitation: Nheavy) for each season, i.e. for the spring with March, April and May (labelled MAM); the monsoon summer with June, July and August (JJA); the fall with September, October and November (SON); the winter with December, January and February (DJF). These three indices are calculated in the station level as well as in the region level.
To detect positive or negative trends for the three indices (Ptot, Ndry, and Nheavy), the non-parametric Kendall's τ test and linear regression are used here. Kendall's τ coefficient is a measure of rank correlation, i.e. the similarity of the orderings of the data when ranked by each of the quantities (Gilbert, 1987). It ranges from − 1 (−100%) to + 1 (+100%) for negative association (i.e. decreasing trend) and positive association (i.e. increasing trend). Here we use τB, which makes adjustment for ties (i.e. tied pair). This test has been applied to other studies examining hydroclimatic trends (Chang and Kwon, 2007; George, 2007). In addition, the slope of linear regression represents the steepness of the linear line, estimated with the least square approach.
2.2. Gridded precipitation data
To confirm the validity of our findings with observations at stations, we used four gridded precipitation datasets. They include two monthly global datasets of the CPC Merged Analysis of Precipitation (CMAP) and the Global Precipitation Climatology Project (GPCP), which are the merged products of gauge observations and various satellite estimates, and two land-only datasets of the Precipitation REConstruction over the Land (PREC/L) and the Asian Precipitation-Highly Resolved Observation Data Integration Towards the Evaluation of Water Resources (APHRODITE), which are based on rain gauge observations. Below, we provide a brief description of each dataset.
The CMAP dataset (Xie and Arkin, 1997) was created globally on a 2.5° latitude–longitude grid from 1979 to near the present by merging several kinds of data sources. Data sources included gauge observations, estimates from satellite observations of infrared (IR), outgoing longwave radiation (OLR), Microwave Sounding Unit (MSU), and microwave (MW) scattering and emission from the Special Sensor Microwave Imager (SSM/I), and precipitation forecast by NCEP/NCAR Reanalysis.
The GPCP monthly precipitation analysis (Adler et al., 2003) was constructed at 2.5° latitude–longitude resolution from 1979 to near the present by combining the precipitation information from multiple sources. SSM/I data, IR estimates were merged, Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS), OLR measurements, and quality-controlled rain gauge data were merged based on hierarchical relations. Much of input data of GPCP and CMAP are in common, particularly over the land, but the merging algorithms of two datasets are different (Yin et al., 2004).
The PREC/L dataset (Chen et al., 2002) was constructed globally over land on 2.5°, 1.0°, and 0.5° latitude–longitude grids from 1948 to the present. The data were derived from gauge observations collected in the Global Historical Climatology Network (GHCN), and the Climate Anomaly Monitoring System (CAMS) dataset, using the Optimal Interpolation (OI) technique.
The APHRODITE dataset for 1958–2007 (Yatagai et al., 2009) was created on 0.50° and 0.25° latitude–longitude grids in a daily time step by collecting gauge observations over three different regions, including Monsoon Asia, Russia, and the Middle East. For this dataset, three categories of data were used: datasets from the Global Telecommunication System (GTS) network, precompiled datasets, and individual data collected by APHRODITE project. Here, the GTS over the entire globe was also used for the above-mentioned global datasets of CMAP, GPCP and PREC/L. The pre-compiled datasets include different sources such GHCN-Daily, GWEX Asian Monsoon Experiment.
In this study, we evaluated the seasonal precipitation (Ptot) for gridded datasets. While some of the datasets extended back to 1948, here we analysed the data from 1983 to 2007 except for PREC/L (1983 ∼ 2006), to be consistent with the rain gauge station data. We chose to use the data at the finest grids when multiple datasets were available, and for PREC/L and APHRODIE we used the data at a 0.5° and 0.25° grid, respectively. PREC/L data at the 0.5° grid resolution were available only up to July of 2007.