Investigation of the Multidecadal Variability of the South Atlantic Region Hydroclimate since the Mid-Holocene Using a Non-Supervised Neural Network

• Ano de defesa: 2023
• Autor(a) principal: Gorenstein, Iuri
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Africa Ocidental; anti-correlação de chuva; decadal precipitation; EC-Earth; GISS,CESM; Nordeste Brasil; Northeast Brazil; precipitação decadal; precipitation anti-correlation; Self Organizing Maps; West Africa
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/21/21135/tde-01092023-101157/

Decadal precipitation (PPT) anomalies are related to water reservoirs, affect biota, and may interfere with higher frequency events such as floods and drought. Rainfall in the tropics is mostly associated with the Intertropical Convergence Zone (ITCZ), which in turn has its decadal displacements and anomalies controlled by Oceanic modes of variability. Populations from Northeast Brazil (NE) and West Africa (WAF), two regions adjacent to the Atlantic Ocean, have mainly agricultural economies dependent on ITCZ shifts and, consequently, on the Atlantic decadal variability cycle. Using Self Organizing Maps (SOM) non supervised neural networks, the tropical Atlantic region climate was analyzed and its sea surface temperature (SST) and PPT decadal variability was studied using: satellite observational data from 1979-2015; Reanalysis HadISST product from 1870-2019; and, finally, EC-Earth, CESM and GISS numerical climate models from Pre-industrial (PI) runs and different scenarios with mid-Holocene (MH) insolation and vegetation representing the Green-Sahara (GS) period. The SOM has successfully reduced the dimensionality of climate data. The Atlantic Ocean SST, pressure and wind anomalies are entangled at decadal scales. Together, they control the decadal PPT anomaly anti-correlation between NE and WAF regions, depicted by the standard precipitation index series from 1979 to 2015. The 1870-2019 HadISST reanalysis dataset has shown a 40 to 50 years periodicity, representing the full Atlantic decadal SST anomaly cycle in the 20th century. With the numerical model simulations from PI, MH and GS, SST anomaly structures that closely resemble the observational data cycles appeared at decadal scales. Using Shannon\'s Entropy as an analogue of the model runs\' decadal variability, the EC-Earth PPT variability showed a dependency in dust emission (GS with dust reduction had the lowest entropy of all), while the SST variability in this model seems to be affected more by the presence of vegetation in the Sahara (GS and GS with dust reduction shown lower entropy than the PI and the MH runs). The GISS model presented the lowest SST variability change between different scenarios and the CESM model point to a large PPT internal variability, with identical scenarios showing divergent entropies (GS full vegetation runs 1 and 2). Therefore, large ensembles are necessary if we want to attribute the uncertainties of internal variability from the models\' entropies and achieve more robust results.