36 and 0.39 mm for NMIA and SIA versus 0.86 and 2.56 mm determined with the Weibull PDF. These differences indicate that biases for the Weibull are 2.4–6.4 times higher AZD4547 research buy than the Gumbel and indicate the ability of both PDFs to fit the AMS. Finally, frequency analysis performance was sensitive to PDF and confirmed the advantages of the Weibull PDF in
Experiment 4, in comparison to the Gumbel and Logistic PDF. Biases were the distinguishing GOF as CC was all close to 1. Biases determined from the Weibull experiment were lower than both Gumbel and Logistic with values of 0.32 and 0.35 mm for NMIA and SIA stations versus 0.58 and 0.55 mm for the Logistic (Fig. 2 bottom row). Gumbel performed similar to Weibull but with higher biases. The experiments suggest that bias and correlation vary for the same configuration
from station to station, which makes distinguishing the optimal model challenging. Some configurations perform better than others (including the control) regardless of the metric used. For ERK inhibitor example, the Logistic PDF has the lowest correlation coefficient of all three models for both stations and does not appear suitable for this data set. However, the Hosking PPF, and both Weibull and Gumbel PDF, perform credibly. There were no differences in the performance of the PEM. It was decided to continue the frequency analysis investigations detailed in the following sections with Weibull PDF, L-Moments PEM and Hosking PPF. The extension and infilling process will likely include
more outliers and it is believed that this configuration will prove robust based upon previous literature and the performance noted above (Overeem et al., 2008). Re-analysis CYTH4 of the existing data with Weibull PDF, L-Moment PEM and Hosking PPF yielded more intense IDF curves in comparison to those determined previously by UWA (see Fig. 3). Firstly, there were spatial differences where NMIA IDF curves were very similar and differed by only 2% on average. However, SIA’s new IDF curves were higher than those determined previously by 2% to 238%, with a difference of 41% (see Fig. 4 top panels). This is particularly interesting as the small differences in the GOF measures did not suggest considerable increases in quantile predictions. The differences also increase in intensities with increasing RP between the existing UWA analysis results and the new Weibull PDF results from the experiments. For instance, the Weibull and Gumbel correspond almost identically for the 5 and 10 year RP for both stations. However, the differences increased for the 50 and 100 year RP. For instance, the differences for the NMIA and SIA stations 5 year RP predictions were zero and 2% respectively. However, they increased steadily to 5% and 85% for the 100 year RP. Re-analysis of the existing data with the Weibull L-Moments frequency analysis configuration also points to extreme events being more frequent than suggested by the former UWA analysis.