14 T.A. Cinco et al. / Atmospheric Research 145–146 (2014) 12–26 combined to create significant negative impacts for a developing country with a growing population (Yumul et al., 2011, 2013). Indeed, the Philippines has been identified as one of the most vulnerable nations to the impacts of climate change; a function of the density of populations living near to the coast and in marginal upland settings and the frequency and intensity of extreme events including tropical cyclones, associated flooding and rain triggered landslides (Yumul et al., 2011; Yusuf and Francisco, 2009). Identifying, quantifying and understanding the threats and associated risks are therefore of the utmost importance for decision makers in a range of sectors to provide accurate and timely information to those communities at risk. Recent efforts to predict the impacts of a changing climate across a range of sectors including agriculture, have led to the integration of climate change oriented policies and the creation of new institutions and bodies to ensure that social and economic interests are protected (PAGASA, 2011; Republic of the Philippines, 2009, 2010, 2011). Such institutional tools designed to improve the Philippines' resilience and adaptive capacities are only effective if they are guided robust scientific data. 3. Methods 3.1. Data gathering The Philippines Atmospheric Geophysical Astronomical Services Administration (PAGASA) under the Department of Science and Technology (DOST) is responsible for managing a countrywide network of over 50 synoptic weather stations which have captured daily rainfall and temperature data since the 1950s and as early as 1911 in some cases, and from which the observed data discussed here is retrieved. The records gathered from the selected stations have been compiled to form a time series data set for daily near surface temperature and precipitation. Data observed and collected between 1951 and 2010 was used and the baseline or reference period against which anomalies or departures from this norm are measured is 1961 to 1990 for mean, minimum and maximum temperatures and extreme temperature and precipitation events. In order for the synoptic station to be included in the sample, certain criteria relating to the quality of the data produced had to be met including; records were for as long as possible including the reference period 1961–1990; less than 20% of the daily values were missing in each year (i.e. ≥293 days data available); the stations were of high quality and well-maintained; and the station had been located at a single site during the period of record. Fig. 1 shows the location of the stations from which data was retrieved, in their topographic context. Most of the stations are located in the coastal areas of the Philippines which is where many of the population centres are situated. However, two of the stations – those at Baguio and Malaybalay – are located in upland areas of Luzon and Mindanao, respectively, which is important to note when considering the results, particularly for rainfall extremes. Appendix A provides available metadata relating to the location of the stations used in this study. 3.2. Data quality and inhomogeneities Whilst every effort was taken to select the synoptic weather stations with the highest quality, most complete data, some discontinuities in the data are inevitable and were therefore statistically treated to remove inhomogeneities which can affect the mean climatic values and in turn the anomalies and extremes (Manton et al., 2001). Common causes of data discontinuities include the relocation of synoptic stations, any change in the measurement or data gathering techniques used as well as mechanical damage to the station (De Lima et al., 2013; Griffiths et al., 2005; Shahid et al., 2012). In the Philippines, there has been a rapid increase in population since the middle of the 20th century which has led to a commensurate increase in population density (from 68 capita km−2 in 1950 to 303 capita km−2 in 2010) and growing urbanisation. Unfortunately, this means that many of the stations included here are now situated in urban locations and subject to the heat island effect, a fact which should be considered when assessing the results. An initial visual inspection of the complete datasets was conducted in order to identify and eliminate any obvious outliers which were outside three standard deviations from the daily mean values, although this form of initial subjective treatment may mean that some genuine trends have been excluded. To identify and eliminate discontinuities, the data was subsequently arranged in series, creating mean, minimum and maximum daily temperatures and annual precipitation for each station. This data was then subjected to a form of nearest neighbour statistical analysis using the Multiple Analysis of Series for Homogenization (MASH) software following Manton et al. (2001) which compares the target station data with that of climatically similar stations to identify significant changes in the target station. MASH was originally developed by the Hungarian Meteorological Service and has become widely used as an acceptably accurate means of conducting homogeneity tests and for smoothing time series data (Lakatos et al., 2013; Szentimrey, 2011; Zhen and Zhongwei, 2010). Once discontinuities were identified, if they could be linked to changes in the weather station or data management technique, then they were excluded otherwise they were included in the final dataset. However, overall this technique may still allow for the inclusion of undetected inhomogeneities and this should be considered when reviewing the results. The overall mean annual temperature used data from all 34 synoptic stations but for the extreme rainfall events a number of stations were discounted on the basis that some of the daily data was not wholly reliable. 3.3. Data analysis and trend identification Remaining data from the identified highest quality synoptic stations was further analysed to identify climatic trends for both rainfall and temperature. 3.3.1. Mean temperature trends The first step was to arrange the observed annual mean of daily temperature for each of the selected synoptic stations to create a time series 1951–2010. National values were calculated by taking the arithmetic mean from all the stations included in the study. Using this data the mean annual