8 and Table 3). Frequency analysis performance measured by bias improved relative to the original data set. Infilling was beneficial in reducing bias between the frequency analysis quantile results and data. Bias was reduced from 3.1 and 3.6 mm for the original data set to 1.5 and 1.8 mm for the infilled data set. Change factors were used to determine the effects of extension and infilling on the IDF predictions. see more These were shown in Fig. 4 (bottom row). Intensities increased as a result of the frequency analysis of the extended and infilled data set and were noticeably greater for the longer durations and higher RP, for both stations. For instance, 24 h duration

intensities increased by 50% (7.3 from 5.8 mm/h) for the 5 year RP for NMIA, whereas, a much larger increase of 250% (17.8 from 6.5 mm/h) was realized for the 100 year RP for SIA. Increases in PDF

prediction of intensities CT99021 is likely to be due to the wide range of climate extremes experienced both pre-1957 and post-1991 and highlights the importance of using the longest possible data set to cover a range of climate variability (Koutsoyiannis, 2004). Increased intensity for higher durations was pronounced, with greater increases of 38–115% for 2 h or longer versus only 14–36% increases for shorter durations. Increases in longer duration intensities can be more devastating with more volume of runoff. Higher intensities were determined from the frequency analysis of the infilled data. Frequency analysis with temporal trends in the parameters for the present climate AMS data revealed that the models that allowed for temporal trends in the means, variance and skewness performed

better than the stationary model for both stations (WMO, 2009b) and confirms that the statistics are not stationary. Frequency analysis with temporal trends in the location, scale and shape parameters had high goodness of fit, with correlation of 0.92–0.99 and low RMSE of 10.3–20.8 mm. Quantile–quantile plots (Fig. 9) showed agreement for the four models (stationary with time; mean varying; mean and stand deviation varying; and mean, standard deviation and skewness varying with time) for the 200 mm and FAD smaller rainfall depths. Disparity emerged at the higher precipitation depths with the skewness varying model fitting better at the extremes, as expected. Skewness parameter with temporal trends enables better fitting at the extreme tail of the distribution. Given the high quality of fit evident in the models, it was decided to average the three time varying models’ 2100 predictions, in Table 4. A trend of reduction in the intensities of frequent events with RP less than 10 years, to increases for the less frequent events with RP greater than 25 years, emerged.