Loading...
Publication
Downscaling climate with temporal correlation and spatial dependencies
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.08 MB | Adobe PDF |
Abstract(s)
Global Circulation Models have been improving significantly along with the computational power needed to run those models. Earths System Models (ESMs) are the current state-of-the-art. They expand on the coupled Atmosphere-Ocean General Circulation Models (AOGCMs) to include a representation of various biogeochemical cycles such as those involved in the carbon cycle, the sulphur cycle, or ozone. These models provide the most comprehensive tools available for simulating the past and future response of the climate system to external forcing, in which biogeochemical feedbacks play an important role. Regionalisation of climate data is the process that enables the representation of high spatial and temporal resolution information, usually derived from Global Circulation Models. This process can be quite complex and largely remains within the research domain of climate scenarios development. Although climate scenarios are becoming mandatory in impact studies at all levels, access to high-resolution climate projections is usually unavailable where the vulnerabilities tend to be higher, as in the case of developing countries. On top of that, the representation of extremes, often responsible for floods and droughts with high impact, is still a challenge recognised by the IPCC, especially when the climate is just one of the components included in a wide range of impact models. These challenges rise the questions if there is any way to: (i) run climate models faster; (ii) have higher temporal and spatial distribution; (iii) not to be limited by historical meteorological data for model calibration and validation; and finally (iv) to have a good performance on extremes. In short, this PhD proposal aims to explore recent advances in statistical methods for large datasets with Spatio-temporal structures, by exploring recent advances in Bayesian statistics, like the integrated nested Laplace approximations (INLA) algorithm, in combination with environmental data derived from satellite imagery, to provide a new approach that addresses some limitations of current downscaling methods to meet user needs. The methodology presented is structured in three phases. The first explores the application of a new downscaling method for precipitation based on data from meteorological stations, information about the atmosphere's synoptic conditions, and local effects such as distance to the coastline and altitude. The second phase aims to explore the ability of the presented downscaling method to project climate scenarios. For this purpose, twenty years of daily data from October to March were simulated for two intervals, a control period between 1991 and 2000, and a RCP8.5 climate scenario between 2021 and 2030. The 7300 simulations allowed to deepen the analysis on the frequency and intensity of extreme precipitation events, and to characterize possible generic trends resulting from the comparison between the control period and the respective climate scenario through the characterization of anomalies. In these first two phases the main goal was to test the proposed method to characterize the historical precipitation and use it to build daily climate scenarios, through its spatial-temporal characterization. Finally, in the third phase of this study, the obtained results were used in a practical application to characterize the risk associated with floods, based on the methodologies proposed by the IPCC since its fifth report (AR5). With the application of the previous results and methodologies it was possible to go beyond a national scale flood vulnerability analysis, to a true risk analysis where the probabilities of occurrence of extreme precipitation events in climate change scenarios are explicit. Results showed the ability of the proposed downscaling method to estimate about 7245 points in a regular grid, representative of the area of continental Portugal, in less than 20 minutes using a mid-range desktop computer for this purpose. The spatial resolution of 5 km on a daily basis, surpasses the best dynamic regionalization methods currently available at national and regional scales. Another important result was the ability of the proposed method to adequately represent the spatial distribution of precipitation using a limited number of meteorological stations. Although the choice of data was made regarding the north-south, coast-interior influences, and altitude variations, only 15 weather stations were used in order to evaluate the applicability of the proposed method in contexts where the available information is limited. In this case, the added value is clear for situations where impacts are assessed in island systems or in developing countries where baseline information is often limited or scattered. The analysis of precipitation extremes was also assessed and deserves to be highlighted. This has always been a difficult topic to address as its quantification remains in the scientific debate. It is important to note that when climate projections are made, the IPCC itself recommends the use of an average of models (ensemble), which makes it impossible to outset any approach that can be taken to assess the frequency and intensity of extreme weather events, and their possible trends in climate change scenarios. Even so, their analysis based on daily simulations is important and constitute relevant information for various sectors of activity related to risk management. As the methods become more robust the projections of extremes tend to follow these developments allowing new indicators to be explored and risk analysis to be improved. From the results it was possible to conclude that the proposed method tends to underestimate precipitation regimes above 100 mm in periods of up to 24 hours. Even so, the results are promising, allowing to assess changes in the spatial distribution of this type of events, and thus significantly improve the assessment of impacts, where the frequency of this type of occurrences are crucial to move from the characterization of vulnerability to the quantification of risk. Finally, the exercise of applying the results on the quantification of extremes for a reference period and in climate change scenarios allowed to obtain a comparative risk analysis on a national scale for floods. In this context, the results showed a trend towards an increase in frequency and intensity of severe precipitation events in the littoral-central region, with a consequent increase in the risk associated with floods. In summary, the presented method can open new doors of research on climate downscaling processes, and thus contribute to a better representation of the spatio-temporal component of precipitation, with important improvements in the application of its results in impact studies.
Description
Keywords
Climate change downscaling precipitation extremes INLA Alterações climáticas regionalização precipitação extremos