@InProceedings{Coelho:2013:MeReLi,
author = "Coelho, Caio Augusto dos Santos",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)}",
title = "Uma metodologia de regress{\~a}o linear para as previs{\~o}es
clim{\'a}ticas sazonais probabil{\'{\i}}sticas de
precipita{\c{c}}{\~a}o para o Brasil em categorias
terc{\'{\i}}licas/ A linear regression approach for tercile
probability seasonal precipitation forecasts for Brazil",
booktitle = "Anais...",
year = "2013",
organization = "Simp{\'o}sio Internacional de Climatologia, 5. (SIC).",
keywords = "Seasonal forecast, precipitation, probability.",
abstract = "RESUMO: Previs{\~o}es clim{\'a}ticas sazonais de
precipita{\c{c}}{\~a}o para o Brasil v{\^e}m sendo produzidas
em modo operacional pelo CPTEC/INPE desde meados da d{\'e}cada de
noventa. A partir de 2004 a previs{\~a}o sazonal de
precipita{\c{c}}{\~a}o oficial para o Brasil passou a ser
elaborada em conjunto pelo CPTEC/INPE e Instituto Nacional de
Meteorologia (INMET). At{\'e} junho de 2012 o processo de
produ{\c{c}}{\~a}o dessas previs{\~o}es consistia de a)
diagnosticar as atuais condi{\c{c}}{\~o}es clim{\'a}ticas
regionais e globais; b) examinar as previs{\~o}es produzidas por
ambos modelos f{\'{\i}}sicos (din{\^a}micos) e
estat{\'{\i}}sticos; e c) estimar subjetivamente as
probabilidades para as previs{\~o}es de ocorr{\^e}ncia de
precipita{\c{c}}{\~a}o em tr{\^e}s categorias (abaixo da
normal, normal e acima da normal) ap{\'o}s avalia{\c{c}}{\~a}o
de especialistas de todas as informa{\c{c}}{\~o}es
dispon{\'{\i}}veis. Desde julho de 2012 um novo procedimento vem
sendo utilizado para definir objetivamente as probabilidades para
essas tr{\^e}s categorias no est{\'a}gio final do processo
imediatamente antes da dissemina{\c{c}}{\~a}o da previs{\~a}o.
O novo procedimento {\'e} baseado na regress{\~a}o linear entre
as observa{\c{c}}{\~o}es hist{\'o}ricas de
precipita{\c{c}}{\~a}o sazonal e as correspondentes
previs{\~o}es retrospectivas. Neste procedimento as
previs{\~o}es de probabilidade s{\~a}o estimadas levando-se em
considera{\c{c}}{\~a}o o sinal previsto (ou seja, a anomalia
prevista) por um sistema multi-modelos de previs{\~a}o e a
correspondente destreza retrospectiva das previs{\~o}es desse
sistema para o per{\'{\i}}odo 1989-2008. Este sistema
multi-modelos {\'e} atualmente composto por um conjunto de
previs{\~o}es f{\'{\i}}sicas (din{\^a}micas) produzidas pelo
modelo atmosf{\'e}rico do CPTEC/INPE, por um conjunto de
previs{\~o}es estat{\'{\i}}sticas produzidas pelo INMET, e um
conjunto de previs{\~o}es f{\'{\i}}sicas (din{\^a}micas)
produzidas pela Funda{\c{c}}{\~a}o Cearence de Meteorologia e
Recursos H{\'{\i}}dricos (FUNCEME). Este sistema multi-modelos
comp{\~o}em atualmente o sistema nacional de previs{\~a}o
clim{\'a}tica sazonal do Brasil. A anomalia prevista por este
sistema multi-modelo fornece uma estimativa de quanto a m{\'e}dia
da distribui{\c{c}}{\~a}o climatol{\'o}gica (hist{\'o}rica)
deve ser deslocada para a direita (caso o sistema multi-modelo
indique previs{\~a}o de excesso de precipita{\c{c}}{\~a}o) ou
para a esquerda (caso o sistema multimodelo indique previs{\~a}o
de d{\'e}ficit de precipita{\c{c}}{\~a}o). A destreza
hist{\'o}rica das previs{\~o}es desse sistema {\'e} usada para
definir a dispers{\~a}o da distribui{\c{c}}{\~a}o prevista pelo
sistema multimodelo. Para regi{\~o}es onde a destreza {\'e}
maior (menor) a distribui{\c{c}}{\~a}o prevista pelo sistema
multi-modelo ser{\'a} mais estreita (larga), apresentando
dispers{\~a}o reduzida (aumentada) quando comparada com a
distribui{\c{c}}{\~a}o climatol{\'o}gica. A
verifica{\c{c}}{\~a}o da consist{\^e}ncia f{\'{\i}}sica entre
a anomalia prevista pelo sistema multi-modelo e as predominantes
condi{\c{c}}{\~o}es for{\c{c}}antes de grande escala {\'e} uma
etapa fundamente do processo de previs{\~a}o. Este trabalho
apresentar{\'a} esta nova metodologia atualmente utilizada no
Brasil para definir as probabilidades das categorias
terc{\'{\i}}licas da previs{\~a}o sazonal de
precipita{\c{c}}{\~a}o para o Brasil. ABSTRACT: Seasonal
precipitation forecasts for Brazil have been operationally
produced since the mid-nineties by the Center for Weather Forecast
and Climate Studies (CPTEC), of the National Institute for Space
Research (INPE). In 2004 the official precipitation seasonal
forecast for Brazil started to be jointly elaborated by CPTEC/INPE
and the National Institute of Meteorology (INMET). Until June 2012
the process for producing these forecasts consisted in a)
diagnosing the current regional and global climate conditions; b)
examining the forecasts produced by both physically-based
dynamical models and statistical models; and c) subjectively
estimating forecast probabilities for three precipitation
categories (below normal, normal and above normal) after expert
assessment of all available information. Since July 2012 a new
procedure is being used for objectively defining forecast
probabilities for these three categories in the final stage prior
to issuing the forecast. The new procedure is based on the linear
regression of the historical seasonal precipitation observations
on the corresponding retrospective forecasts. In this procedure
forecast probabilities are estimated taking into account the
forecast signal (i.e. the forecast anomaly) of a multi-model
forecasting system and the corresponding retrospective skill of
this system during the period 1989-2008. This multi-model system
is currently composed by an ensemble of physically-based forecasts
produced by CPTEC/INPE atmospheric climate model, an ensemble of
statistical forecasts produced by INMET, and an ensemble of
physically-based forecasts produced by Cear{\'a} State
Meteorology and Water Resources Foundation (FUNCEME). Such
multi-model system composes the Brazilian national seasonal
prediction system. The multi-model forecast anomaly provides an
estimate of how much the mean of the climatological (historical)
distribution should be shifted to the right (if excess
precipitation is forecast by the multi-model system) or to the
felt (if deficit precipitation is forecast by the multi-model
system). The retrospective prediction skill of this system is used
for defining the spread of the multi-model forecast distribution.
For regions where the skill is greatest (smallest) the multi-model
forecast distribution is thinner (wider) presenting reduced
(increased) spread compared to the climatological distribution.
The check for physical consistency between the forecast anomaly
indicated by the multimodel system and the prevailing large scale
forcing is a fundamental part of the forecasting procedure. This
study will present this new methodology currently used in Brazil
for defining tercile probability seasonal precipitation forecasts
for Brazil, including a preliminary performance assessment of
forecasts produced by the new forecasting approach since its
implementation in 2012.",
conference-location = "Florian{\'o}polis, SC",
conference-year = "15-19, set.",
urlaccessdate = "29 mar. 2024"
}