Quantile functions were used to derive new IDF curves for the 5–100 year RP events,
which were compared to the originally derived IDF curves. Gaps in short duration events, less than 24 h, were filled with IDF scaling relationships. Both the Chowdhury model (Rashid et al., 2012) and Nhat model (Nhat et al., 2006) were used (see Eqs. (1) and (2) below). The basis of the Chowdhury model Roxadustat order is the 24-h event rainfall depth, P24 (see Eq. (1)). Optimizing functions are used to derive the best fit values for its exponent (E) and constant (C). The Nhat model is based on the simple scaling of time and scale invariance of daily rainfall to derive intensities for shorter durations (Eq. (2)). Alpelisib The process relies on equating the probability of distribution of the parent duration (typically the 24-h) and any other duration (d). Parameter values for the exponent in the Nhat model are optimized to fit a training set, i.e. after the data sets for NMA and SIA are split into a training set and a verification set. The goodness of fit (CC i) for predicted (P′d,iP′d,i) versus observed (Pd,i) rainfall depth for both models for each duration for the ‘ith’ year was optimized until both had approximately the same root mean square of errors (RMSE). The performance
of both models was compared to the AMS data, for the period 1957–1991, for variations in performance for each duration, and the optimal relationships used to fill the gaps. Modified Chowdhury (Rashid et al., 2012) of the Indian Meteorological Department (IMD) empirical reduction formula for estimation of rainfall depths, P (mm), for various durations (d) from Annual Maxima values. Where E and C are constants to be determined equation(1) Pd=P24d24E+C Simple scaling factor for derivation of shorter duration events intensities (id) by equating the frequency distributions, after Nhat et al. (2006) equation(2) iddist¯¯λd−Hd⋅iλd Gaps in the long duration events (2 days and longer) were filled using an artificial neural network (ANN) (see Appendix A) driven by National Centers for Environmental
Predictions and National Center for Atmospheric Research (NCEP/NCAR) re-analysis data (Kalnay et al., 1996 and NOAA, 2012). ANN is a statistical downscaling method that develops non-linear relationships between input Pregnenolone global gridded data and output at-station precipitation predictions. ANNs are described by Rumelhart et al. (1986) and Gegout et al. (1995). The method represents a good downscaling option for this study since previous studies suggest that it performs credibly and comparably to other downscaling methods (Goodess, 2007 and Abebe et al., 2000). Once calibrated, it can be deployed to determine future climates using projections from Global Climate Models (GCMs). No previous studies were found in, which a feed-forward ANN was used in a Caribbean extreme precipitation study.