Estudio de las causas sinópticas de la precipitación en Canarias mediante modelos climáticos. Presente y futuro

Abstract

Estudio estadístico de bases de datos de predicciones y datos reales de precipitación de las Islas Canarias mediante Análisis de Componentes Principales (PCA), técnicas de Clustering y conocimientos de meteorología para predecir el cambio de los patrones de lluvia en las Islas Canarias en el futuro.

This is a statistical study of prediction databases and observational rainfall data of the Canary Islands using Principal Component Analysis (PCA), clustering techniques and meteorological knowledge to predict the change of rainfall patterns in the Canary Islands in the future. First of all, we have to know that we are going to work with databases that represent the observations of precipitation and sea level pressure in the Canary Islands or their surroundings, and predictions of these same variables. These data covers three periods of time, one for the recent past (1980-2009) and two for the future (2030-2059 and 2060-2099). The data for the future periods cover two scenarios related to global emissions paths (RCP-4.5 and RCP-8.5). The data gathered from the past, corresponds to predictions and real data, and the data from the future, obviously corresponds to climate projections. Specifically, the observational precipitation data for the recent past are directly extracted from the SPREAD database (R. Serrano-Notivoli, Beguerıa y col., 2017 ). This is a high-resolution gridded precipitation dataset covering Spain. This was constructed by estimating precipitation amounts and their corresponding uncertainty at each node on a 5x5 km grid. Sea level pressure data around the islands were extracted from the ERA5 reanalysis (Hersbach y col., 2020). Apart from this, other databases are used, that correspond to regional climate models, which predict sea level pressure and precipitations, i.e., they are not observational data, but simulations of these variables. Specifically, three databases are used, each one associated to the global climate model used for its generation: GFDL, IPSL and MIROC. Both, past and a future simulations, were provided. These models, which are regional climate simulations, have been performed with the WRF model (Non Hidrostatic Weather and Research Forecasting- WRF/ARW v3.4.1) using a unidirectional triple nesting configuration with grid resolutions of 27x27 km, 9x9 km and 3x3 km. These simulations were carried out by the Group of Earth and Atmospheric Observation (GOTA) of the University of La Laguna (ULL). The used domain is centered in the Northeast Atlantic region and covers a large area to capture the main mesoscale processes affecting the Canary climate, while other more internal domains are centered in the Canary archipelago. The WRF version and the physical parameterizations that they used to represent the different subgrid-scale atmospheric processes were selected by GOTA according to previous work in the same study area (Pérez y col., 2014) (Expósito y col., 2015). Now that the data used in this study have been explained, the methodology is outlined. First, some statistical methods are applied to the aforementioned databases to extract some features and information. In this study, among other methods, we use Principal Component Analysis (PCA), which is a mathematical technique to summarize the information contained in a set of data by means of other independent parameters; more specifically, it is a rotation of the coordinate axis of the original variables to new orthogonal axes, so that these axes coincide with the direction of maximum variance of the data. In this case, the data to which we apply this method are the daily rainfall values, and the axes correspond to each of the land pixels of the Canary Islands of the SPREAD database. In this way, we manage to group the pixels of the islands in different groups in which rainfall is correlated. Although with this method we could already have a grouping of pixels with a certain correlation in terms of rainfall, what we do now is, with the axes rotated by the PCA performed, to apply some Clustering technique to group the pixels in different regions. This should give us a coherence of the regions a little higher than the groupings that were made with the PCA. Specifically, we use the K-means method to divide the pixels of the Canary Islands in 6 groups. The weather types for each day are determined from the sea level pressure values measured at certain points of a grid located over the Canary Islands. We use the formulas proposed by Jones y col., 2013. Once we have defined the type of weather (WT) for each day, and the amount of daily precipitation related to each of our pixel groups (regions), we can elaborate heatmaps representing the percentage of rain and annual mean precipitation or heavy precipitation days related to each WT and region. Once we have each heatmap related to the past and to every RCP scenario of the future, we discuss them and extract some features of these heatmaps that we obtained from the aforementioned databases. These heatmaps could throw some light on how the patterns of rain in the Canary Islands could evolve from now to the next decades. Lastly, we mention some starting points on what next studies related to this subject could be based on.