A Hybrid dynamic-statistical approach to climate forecast in Northeastern Brazil
Weather forecasts, NMME, Precipitation, Average Temperatures, ENOS.
The growing demand for accurate seasonal weather forecasts in the fields of energy, agriculture,
transport, leisure, among others, can save lives, support emergency management and improve the
planning of responses to the effects arising from the climate, avoiding economic losses of high impact on
society. Therefore, the main objective of the present study is to calibrate raw forecasts of direct outputs of
models from the North American MultiModel Ensemble (NMME) international project for Northeast
Brazil (NEB), reducing systematic biases and increasing assertiveness for forecasting purposes
operational rainfall and monthly/quarterly average temperatures. The justification for this study is given
by the questioning of the predictive capacity of climate prediction by models in specific areas, as is the
case of NEB, as it presents patterns of climate response associated with modes of variability such as
ENSO (El Niño Southern Oscillation) . Precipitation data from the National Institute of Meteorology
(INMET) and state Meteorology Centers in the Northeast will be used, along with average temperature
data from the INMET stations. Data from gridded analyzes provided by XAVIER, Delaware, CHIRPS
and ERA5Land will also be used. To correct the systematic errors in the NMME model, the Canonical
Correlation Analysis (ACC) method will be used, which consists of associating indices with each of the 2
sets of data in order to maximize the correlation between the two indices, retaining as much information
as possible contained in the original variables. Due to aspects of the non-linearity of the climate, the
representation by simple statistical models, that is, based on linear relationships, such as those that use
multiple linear regressions, is called canonical correlation analysis and that is why the combination of the
Statistical knowledge with dynamic modeling, resulting in the term called: hybrid coupled models.