ENSO Drought Onset Prediction in Northeast Brazil Using Satellite Recorded Index

ABSTRACT

ENSO indices and satellite recorded NDVI were used to construct drought onset prediction model for Northeast Brazil through multiple linear regression technique. Monthly NDVI and ENSO indices anomaly data for the period of January 1981 to December 1993 were used to develop model, while those of 1951 to 1998 were used to simulate the NDVI anomaly time series for model evaluation.

Three models were constructed using NDVI anomaly as dependent variable and Niño3.4, SOI, DIP2 and SATL anomalies as independent variables. Model 1 was constructed using 12-month NDVI data while Models 2 and 3 used only 4 months (September to December). The results showed that R2 (R-square) values of 0.3765, 0.6172 and 0.7945 at a significant level of 1% were obtained for Model 1, Model 2 and Model 3 respectively. Simulated NDVI anomaly values agreed quite well with observed values for all three models but model 3 had better intensity estimate. The NDVI anomaly dynamic evolution of 1951 to 1998 simulated by models showed that model predicted NDVI anomalies coincided quite well with the historical ENSO induced drought events reported in the literature. It is concluded that the use of satellite recorded NDVI in stead of rainfall data improved the correlation with ENSO indices. Drought onset Model 3, based on the data set with high anomaly values of NDVI and ENSO indices, predicted drought onset in NEB four months before its occurrence with reasonable success (74%). It is suggested that a combined use of ENSO indices and NDVI inferred drought may provide a better alternative to construct ENSO drought onset prediction model for other regions. Further study will be carry out to investigate the ENSO drought and flood onsets in the southeastern South America.