Simulation of daily rainfall scenarios with interannual and multidecadal climate cycles for South Florida

Concerns about the potential effects of anthropogenic climate change have led to a closer examination of how climate varies in the long run, and how such variations may impact rainfall variations at daily to seasonal time scales. For South Florida in particular, the influences of the low-frequency climate phenomena, such as the El Nino Southern Oscillation (ENSO) and the Atlantic Multi-decadal Oscillation (AMO), have been identified with aggregate annual or seasonal rainfall variations. Since the combined effect of these variations is manifest as persistent multi-year variations in rainfall, the question of modeling these variations at the time and space scales relevant for use with the daily time step-driven hydrologic models in use by the South Florida Water Management District (SFWMD) has arisen. To address this problem, a general methodology for the hierarchical modeling of low- and high-frequency phenomenon at multiple rain gauge locations is developed and illustrated. The essential strategy is to use long-term proxies for regional climate to first develop stochastic scenarios for regional climate that include the low-frequency variations driving the regional rainfall process, and then to use these indicators to condition the concurrent simulation of daily rainfall at all rain gauges under consideration. A newly developed methodology, called Wavelet Autoregressive Modeling (WARM), is used in the first step after suitable climate proxies for regional rainfall are identified. These proxies typically have data available for a century to four centuries so that long-term quasi-periodic climate modes of interest can be identified more reliably. Correlation analyses with seasonal rainfall in the region are used to identify the specific proxies considered as candidates for subsequent conditioning of daily rainfall attributes using a Non-homogeneous hidden Markov model (NHMM). The combined strategy is illustrated for the May-June-July (MJJ) season. The details of the modeling methods and results for the MJJ season are presented in this study.