L EVANTAMIENTO CUANTITATIVO DE INFORMACIÓN

Thesis Objective 1: Extreme Rainfall Regions (Chapter 3) 7.2.1

Classification of Extreme Rainfall Regions (ERRs) (Thesis Objective 1) reflecting similar extreme precipitation characteristics was needed to simplify local weather for improving physical interpretation of climate-flood links in Chapters 4 and 5. First, a station-based dataset of precipitation observations was compiled that was deemed suitable for examination of extremes; that is, balancing the highest spatial resolution available, against the longest temporal resolution available, without compromising data quality. A final set of 126 precipitation stations for both Ireland and Northern Ireland for use in classification of ERRs was identified (Thesis Data Objective 1). A set of 17 variables describing 5 key extreme precipitation characteristics (Magnitude, variance, extreme value tail behaviour, timing, and persistence) and physiographic properties (height and distance to coast) were reduced into four rotated principal components (PCs) using Principal Component Analysis (PCA). Cluster Analysis was undertaken using the k-means algorithm to objectively identify 3 ERRs with distinct extreme precipitation climatologies. The first PC accounted for 53 % of the variability in the original dataset with near uniform direction, mainly reflecting extreme precipitation magnitude, suggesting a relatively simple extreme precipitation regime. The remaining 3 PCs only explained ~ 10 % of the variance each but did have an influence in the final delineation of ERR boundaries, particularly the influence of elevation and coast, and extreme precipitation and wet-day spell length were considered, most notably in the north of the Island. After a rigorous sensitivity analysis, 3 ERRs where considered appropriate for IoI. These ERRs were then used to better interpret spatio-temporal changes in extreme precipitation in Chapter 4 and for grouping the developed flood-Index in Chapter 5 into more physically relevant regions, instead of political or agency boundaries which is often done.

Thesis Objective 2: Detection of Spatio-temporal changes in extremes 7.2.2

(Chapter 4)

Anthropogenic greenhouse gas induced climate change is expected to intensify the hydrological cycle leading to increased frequency and severity of precipitation extremes, and hence increasing the physical hazard component of flood risk. This

181

chapter, for the first time, assessed if there are statistically discernible changes in both extreme precipitation and flood time-series across the Island of Ireland (Thesis Objective 2). Much effort was dedicated to ensuring use of the best available station- based datasets and ensuring best possible spatio-temporal coverage. Only catchments that are considered ‘near natural’ were used so that any detected change can be attributed (with more confidence) to climatic drivers, rather than human modification within the catchment. A new flood Peaks-Over-Threshold (POT) database was developed to facilitate the assessment of flood frequency (Thesis Data Objective 2).

The Mann-Kendall test for monotonic trend was applied for exploring signatures of change in indices relating to the magnitude, duration, and intensity of extremes and logistic regression was employed for assessment of changes in frequency indices. Results show evidence of robust increasing changes in extreme precipitation indices for the periods investigated. Over 80 % of stations show increases in maximum 5-day precipitation (RX5day), a flood relevant indicator, with 22 % of those statistically significant at the 5 % level. The magnitude of increases in precipitation intensity (SDII) is within the range of what is expected with observed atmospheric warming (6.98 % with an interquartile range of 2.67 % and 16.99 % between 1956 and 2009). Despite the small size of the Island of Ireland in relation to the Atlantic Ocean, many extreme precipitation indices show most pronounced increases in the southwest (ERR-SW) and within the coastal and upland Extreme Rainfall Region (ERR-CU) indicating changes in maritime air masses that interact with the Atlantic facing topography.

For floods, while shorter time-series showed predominantly increasing trends, when longer records where considered, conclusions on floods are less certain than those for precipitation due to the more limited availability of long record stations. Caution is required in making inferences about the drivers (causes) of these changes. Although results are broadly in-line with climate change, time-series revealed large decadal- scale variability coherent across the entire Island for both precipitation and streamflow series. This provides compelling evidence that changes are indeed climate-driven but that the role of Decadal Climate Variability (DCV) is considerable.

182

Thesis Objective 3: Synoptic and Large-scale drivers of floods (Chapter 5) 7.2.3

Chapter 4 shows that the influence of natural large-scale climate variability can dominate signals in observed flood records expressed as periods of enhanced/reduced flooding. This is problematic for assessing trends in flood time-series as observed data are often too short and hence detected changes reflect a snap-shot of climate variability rather than evidence of long-term climate change. This Chapter used an objective weather classification scheme to reconstruct the atmospheric drivers of flood occurrence across the Island of Ireland since 1872, and explored further linkages with large-scale climate (Thesis Objective 3). Synoptic drivers show modest skill in reproducing observed Peaks-Over-Threshold (POT) floods at the annual-scale, but are not as skilful at the seasonal-scale. Four flood rich periods were identified: 1.) 1870s to 1890s; 2.) late-1900s to mid-1930s, 3.) a short spell in the 1980s, and 4.) late-1990s onwards. It was found that just 5 weather types could account for 75 % of annual floods, cyclonic being most dominant with important contributions from westerly (during winter) and southerly (during summer) types. Remarkably, the most widespread floods can affect up to 80 % of the network during a large event, highlighting the importance of considering the entire Island, rather than just small groups or individual catchments when assessing the large-scale drivers of floods.

The North Atlantic Oscillation was shown to be positively correlated with high flows and floods. For flood frequency, this relationship is only evident in the west and north west, but evidence of an NAO influence in other parts of the Island were more evident for high flow indices that describe catchment wetness. An exploratory analysis of the role of Atlantic sea surface temperature anomalies, expressed as the AMO, on modulating flood-producing weather types was also undertaken. The AMO was found to be positively correlated with anticyclones over the British-Irish Isles domain, meaning these blocking high pressure systems occur less frequently during AMO cool phases. This is supported by a negative correlation between AMO-NAO. Persistent periods of low anticyclonic frequency tend to coincided with prolonged periods of higher than normal cyclonic types, and are identified as a contributing factor to prolonged flood rich periods. However, much more research into this tentative link is needed, but if confirmed could have important implications for understanding long- term variability of flood propensity.

183

Thesis Objective 4: Attribution of detected changes in streamflow (Chapter 6) 7.2.4

Chapter 5 shows statistically significant links between large-scale climate indices and streamflow over decadal time-scales. However identifying the dominant driving mechanism(s) of detected changes in streamflow (i.e., attribution) at the catchment scale is a challenging task due to the confounding influence of human disturbance such as land-use changes, water abstractions, and river engineering.

This Chapter addresses this challenge by examining the utility of the multiple working hypotheses framework in moving towards more rigorous attribution of changes using the Boyne catchment in the east of Ireland as a case study (Thesis Objective 4). Previous research in this catchment found that a large upward change point in streamflow during the mid-1970s corresponded with a shift in the NAO towards a more positive phase, bringing increased precipitation, and hence increased risk of flooding. Here, the single driver analysis is extended to include multiple factors causing change, both climate driven and internal to the catchment, to establish relative contributions. Rainfall-runoff models were employed to reconstruct streamflow to isolate the effect of climate, taking account of both model structure and parameter uncertainty. The Mann-Kendall test for monotonic trend and Pettitt change point test were applied to explore signatures of change.

Results show that the detected increase in annual mean and high flows was not predominantly driven by changes in precipitation as a result of a shift in the NAO index. Rather it is asserted that the dominant driver of change was arterial drainage and the simultaneous onset of agricultural field drainage in the 1970s and early 1980s. It is also demonstrated that attribution can be more complex at different time-scales with multiple drivers acting simultaneously. This study emphasises the amount and range of data types needed for rigorous attribution, especially when substantial human modifications to catchments may be involved. In fact, it was shown that human disturbance can have larger impacts on the hydrological system than climate at the catchment scale. In the face of such complexity, the use of this systematic hypothesis testing framework, combined with hydrological modelling, helps to avoid confirmation bias and improves overall understanding of dominant drivers of change.

184