@PhDThesis{Viana:2015:SiRePa,
author = "Viana, Denilson Ribeiro",
title = "Sistema de reconhecimento de padr{\~o}es estat{\'{\i}}sticos
aplicado {\`a} previs{\~a}o clim{\'a}tica de temperatura e
precipita{\c{c}}{\~a}o no Centro-Sul do Brasil",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2015",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2015-09-15",
keywords = "precipita{\c{c}}{\~a}o, temperatura, previs{\~a}o
clim{\'a}tica, reconhecimento de padr{\~o}es, precipitation,
temperature, climate forecasting, pattern recognition.",
abstract = "Este trabalho tem por objetivo o desenvolvimento, a
aplica{\c{c}}{\~a}o e a avalia{\c{c}}{\~a}o de um modelo de
previs{\~a}o clim{\'a}tica mensal e sazonal de
precipita{\c{c}}{\~a}o e temperatura para o Centro-Sul do
Brasil, atrav{\'e}s da abordagem estat{\'{\i}}stica de
Reconhecimento de Padr{\~o}es. A metodologia consiste em um
conjunto de etapas que envolvem: 1) aquisi{\c{c}}{\~a}o dos
dados, 2) pr{\'e}-processamento, 3) extra{\c{c}}{\~a}o de
atributos, 4) sele{\c{c}}{\~a}o de atributos, 5)
classifica{\c{c}}{\~a}o e, 6) avalia{\c{c}}{\~a}o. Os
preditandos, precipita{\c{c}}{\~a}o e temperatura, foram obtidos
atrav{\'e}s das respectivas bases, GPCC (\emph{Global
Precipitation Climatology Centre}) e CRU (\emph{Climatic Research
Unit}). Os preditores atmosf{\'e}ricos [Altura Geopotencial (Z)
em 850, 700, 500 e 250 mb, Press{\~a}o ao N{\'{\i}}vel do Mar
(PNM), Temperatura do ar em 850 mb (T850), Conte{\'u}do de
{\'A}gua Precipit{\'a}vel (CAP)] e de superf{\'{\i}}cie
[Albedo (ALB) e Umidade do Solo (UMS)] foram obtidos da
Rean{\'a}lises \${'}\$S{\'e}culo 20\${'}\$ (\emph{20th
Century Reanalysis, V2 NCEP/NCAR}). Os preditores oce{\^a}nicos
[Temperatura da Superf{\'{\i}}cie do Mar (TSM) e
Concentra{\c{c}}{\~a}o de Gelo Marinho (CGM)] s{\~a}o oriundos
da base \emph{HadISST1 do Met Office Hadley Centre}. Os dados
compreendem o per{\'{\i}}odo entre 1951 e 2010, totalizando 60
anos de registros mensais. A extra{\c{c}}{\~a}o de atributos foi
realizada atrav{\'e}s de tr{\^e}s procedimentos distintos: 1)
defini{\c{c}}{\~a}o das Regi{\~o}es Homog{\^e}neas dos
preditandos (RH) utilizando An{\'a}lise de Agrupamentos
Hier{\'a}rquicos (AAH), 2) defini{\c{c}}{\~a}o das
Regi{\~o}es-Chave (RC) dos preditores atmosf{\'e}ricos e
oce{\^a}nicos por meio de An{\'a}lise de Componentes Principais
(ACP) em modo S e, 3) c{\'a}lculo das anomalias dos preditores de
superf{\'{\i}}cie para as RH identificadas. Com base nas
anomalias mensais dos preditandos, foram identificadas quatro
Regi{\~o}es Homog{\^e}neas de precipita{\c{c}}{\~a}o acumulada
e tr{\^e}s de temperatura m{\'e}dia. As an{\'a}lises,
conduzidas por ACP, mostraram que, em m{\'e}dia, 22 componentes
captam em torno de 70\% da vari{\^a}ncia acumulada para os
campos mais est{\'a}veis, relacionados {\`a} press{\~a}o
atmosf{\'e}rica (Z e PNM) e a TSM. Ao todo, foram analisadas 594
vari{\'a}veis, das quais 259 foram selecionadas para a
previs{\~a}o clim{\'a}tica. Nas fases de sele{\c{c}}{\~a}o de
atributos e de classifica{\c{c}}{\~a}o, as s{\'e}ries temporais
dos preditores atmosf{\'e}ricos, oce{\^a}nicos e de
superf{\'{\i}}cie foram correlacionadas com as s{\'e}ries dos
preditandos, por meio de tercis, utilizando An{\'a}lise
Discriminante Linear (ADL). Os resultados mostraram que, tanto
para precipita{\c{c}}{\~a}o, quanto para a temperatura, houve um
ganho m{\'e}dio de 29\% em rela{\c{c}}{\~a}o {\`a}
climatologia. Para a precipita{\c{c}}{\~a}o, destacam-se as
vari{\'a}veis relacionadas {\`a} press{\~a}o atmosf{\'e}rica
(Z e PNM), bem como o CAP, a TSM e os campos de
superf{\'{\i}}cie (ALB e UMS), e ainda, o papel da
circula{\c{c}}{\~a}o atmosf{\'e}rica na Ant{\'a}rtica e
adjac{\^e}ncias. Para a temperatura m{\'e}dia, destacam-se
novamente as vari{\'a}veis relacionadas {\`a} press{\~a}o,
juntamente com a TSM, T850, CAP e CGM. O campo de T850 nas
regi{\~o}es do Pac{\'{\i}}fico Equatorial, costa leste do
Brasil e no continente Ant{\'a}rtico e adjac{\^e}ncias, foram
relevantes para a temperatura. A avalia{\c{c}}{\~a}o das
previs{\~o}es, realizada por meio de um conjunto de escores
categ{\'o}ricos e probabil{\'{\i}}sticos, mostrou que os
resultados obtidos foram superiores aos modelos atuais. Tanto para
a precipita{\c{c}}{\~a}o, quanto para a temperatura, o melhor
desempenho do modelo ocorreu nas categorias extremas (acima/abaixo
da normal), sendo que uma determinada previs{\~a}o nessas
categorias tem maior probabilidade de acerto. ABSTRACT: This study
aims to develop, implement and evaluate a precipitation and
temperature seasonal and monthly climate forecasting model for the
Central-Southern regions of Brazil, using a Statistical Pattern
Recognition system. The methodology consists of a set of steps
involving: 1) data acquisition, 2) pre-processing, 3) attribute
extraction, 4) attribute selection, 5) classification, and 6)
validations of the results. The predictands, rainfall and
temperature, were obtained from GPCC (Global Precipitation
Climatology Centre) and CRU (Climatic Research Unit) data bases.
Atmospheric predictors [Geopotential height (Z) at 850, 700, 500
and 250 mb, Sea Level Pressure (PNM) Air temperature at 850 mb
(T850), Precipitable Water Content (CAP)] and surface predictors
[Albedo (ALB) and Soil Moisture (UMS)] were from 20th Century
Reanalysis V2 - NCEP/NCAR. Oceanic predictors [Sea Surface
Temperature (SST) and Sea Ice Concentration (CGM)] came from the
HadISST1 Met Office Hadley Centre data base. The data covers a
period between 1951 and 2010, totaling 60 years of monthly
records. The attribute extraction was performed by three distinct
procedures: 1) definition of Homogeneous Regions of predictands
(RH), using Hierarchical Cluster Analysis (AAH), 2) definition of
the Key Regions (RC) of atmospheric and oceanic predictors,
through Principal Component Analysis (ACP) in S mode, and 3)
calculation of surface anomalies of the RH identified predictors.
Based on the monthly anomalies of predictands, we have identified
four RH of accumulated rainfall and three RH for average
temperature. The ACP analysis showed that, on average, 22
components explain approximately 70\% of the accumulated
variance, for the more stable fields related to the atmospheric
pressure (Z and SLP) and TSM. In all, 594 variables were analyzed,
of which 259 were selected for climate prediction. In the
attribute selection and classification stages, the atmospheric,
ocean and surface predictor time series were correlated with the
terciles of forecasting series, using Linear discriminant analysis
(ADL). Results showed that for both precipitation and temperature,
there was an average gain of 29\% in relation to the climatology.
As for precipitation, atmospheric pressure (Z and PNM), CAP, SST
and surface fields (ALB and UMS) variables presented the best
results. Also highlighted, is the role of atmospheric circulation
in the Antarctic region and its surroundings. Considering the
average temperature, again, the pressure related variables, along
with TSM, T850, CAP and CGM presented the best results. The T850
field of the Equatorial Pacific, the eastern coastal region of
Brazil and the Antarctic continent and surrounding areas were
relevant to the temperature. The forecast evaluation was achieved
by a set of categorical and probabilistic scores, showing that
these results were superior to current models. The best model
performances were obtained for the extreme rainfall and
temperature categories (above/below normal).",
committee = "Kayano, Mary Toshie (presidente) and Sansigolo, Cl{\'o}vis Angeli
(orientador) and Coelho, Caio Augusto dos Santos and Fortes, Lauro
Tadeu Guimar{\~a}es and Lucio, Paulo S{\'e}rgio",
copyholder = "SID/SCD",
englishtitle = "Statistical pattern recognition system applied to climate
forecasts of Central-Southern Brazil",
language = "pt",
pages = "246",
ibi = "8JMKD3MGP8W/3K3LAH2",
url = "http://urlib.net/ibi/8JMKD3MGP8W/3K3LAH2",
targetfile = "publicacao.pdf",
urlaccessdate = "27 abr. 2024"
}