On the use of Statistical Downscaling for the Study of Coastal Hazards in the Pacific

Abstract

In recent years, research efforts trying to understand and predict coastal hazards have increased as a result of the large number of studies predicting future intensification of storms and increasing rates of sea level rise. This has urged the need to develop and/or improve early warning systems in order to reduce the risk faced by coastal communities. This is particularly the case of the Pacific region, formed by a large number of low-lying islands, that even without accounting for large increases in sea level rise, are already facing numerous and devastating hazard events. These events are generally caused by the co-occurrence of a number of oceanographic processes that fluctuate at different spatiotemporal scales (e.g., water levels, waves), caused by different triggers such as tropical cyclones or distant source swells. Quantifying very low probability events from observations, results in analysis highly dependent of the length of the record, and even when available records may be enough for quantifying univariate extremes, compound events are impossible to characterize. Statistical downscaling is a computationally efficient and reliable technique for obtaining hydrodynamic components based on the relationship between large scale predictors and local predictands. Firstly, based on the relationship between sea level pressure fields and local storm surge in New Zealand, a database composed by reconstructed data for a hindcast period and future projections has been developed. The analysis of the results suggest that extreme storm surge will increase in the southern areas of New Zealand while a decrease will be experienced in the north. With the motivation of developing synthetic time series of not just univariate variables (i.e., storm surge), but multivariate wave and storm surge parameters for estimating shoreline evolution, a new climate-based emulator based on weather type techniques is presented. The emulator is able to generate synthetic time series of a specified length that preserve the chronology at different scales from the inter annual to the intra storm, which is especially relevant for estimating erosion in which the chronology of the events plays a very relevant role. Different applications of the emulator evidence the potential for probabilistic assessments at different time scales. Although wave bulk parameters are still widely used, this simplification results in the loss of significant amount of information associated with multimodal seas, thus, in the last years it is becoming more and more evident the need of using the characteristics of the full spectrum. For this reason, and with special focus on small Pacific atolls where an elevated number of concurrent swells approaching the area at the same time is the most common situation, a new climate-based emulator of seas and individual swell trains has been developed. The emulator has been implemented and validated in Majuro atoll, in the Marshall Islands, and is based on the analysis of all the spectral wave energy approaching the study area. Nevertheless, tropical cyclones are not always correctly reproduced by atmospheric circulation models, resulting in an underestimation of the most extreme waves. This effect is passed through the emulators proposed, and hence, if we want to improve the emulation of the tails of the distributions, a new model for estimating the waves from tropical cyclones is needed. Thus, an empirical model based on combining information of historical tracks from tropical cyclones and satellite wave data is presented. The model allows to better understand the signature of the wind waves produced by tropical cyclones and to give fast estimations needed for early warning systems and risk assessments. Nevertheless, the amount of existing data makes it difficult to statistically characterize the most extreme waves. Overall, the research provides useful insights and tools for the generation of synthetic time series of hydrodynamic conditions, as a way of overcoming the difficulty of characterizing low probability events with historical records, and facilitating the implementation of probabilistic assessments to reduce the risk faced by coastal communities both in the present and in the future.