Resumo:
Brazil has faced one of the biggest water crises in recent years as a result of extreme
hydrological and meteorological events. An irregularity of rain events has a recorded effect or
water balance in the hydrographic basins, for example in the São Francisco River basin, a region
with high water vulnerability and with desertification trends. Due to the extreme national
relevance of this basin for the country, the present study had as main objective the analysis of
the predictability of extreme events that altered the water availability of the São Francisco River
basin, from the integration of a regional atmospheric model and a hydrological model. For this
purpose, a set of five members of seasonal rainfall forecasts was used for each month between
the periods from 2001 to 2018, these being generated by a downscaling of the atmospheric
model Eta of CPTEC/INPE. These data were provided as input variables in the hydrological
model MGB-IPH in the generation of seasonal streamflow forecasts in the São Francisco River
basin. As for the observed data of rainfall and streamflow, used in the comparison with the
predicted data, these were obtained from historical series of pluviometric and fluviometric
stations of the National Water Agency (ANA), respectively, in addition to series of natural
streamflows of the hydroelectric use in the case of the Três Marias, Sobradinho, Itaparica/Luiz
Gonzaga and Paulo Afonso/Moxotó hydroelectric stations. The accuracy of the forecasts was
analyzed both visually and statistically using the Relative Mean Error (EMR), Mean Absolute
Error (EMA) and Pearson's Linear Correlation Coefficient (r) indicators. The results of the
EMR and r indices indicated that, in general, the Eta/MGB models showed a good performance
of the seasonal streamflow forecasts for the basin, mostly for the sub-basins located in the Upper
and Lower São Francisco. Regarding the precedents, no significant differences were observed
between the horizons analyzed, although the results for one month were relatively better
considering the r index. However, it points to the need for corrections to the rainfall forecasts
and the bias in the streamflow forecasts in some regions, as in the case of Sobradinho, which
showed some systematic errors.
