Modeling precipitation occurrence in Northeast Brazil using a hidden Markov model: understanding the dynamic connections of rainy seasons

Abstract

The rainfall over Northeast Brazil (NEB) presents a large climate variability and is highly seasonal and intrinsically conditioned to changes in the large-scale atmospheric circulation. This work aims to employ the hidden Markov model (HMM) to describe the daily precipitation occurrence in the NEB for austral summer and autumn. We used a rainfall database to estimate the parameters of the model during the period from 1980 to 2015. The Bayesian Information Criterion (BIC) is applied to assess the fit of the model and then generate the Viterbi sequence of hidden states. We characterized the large-scale atmospheric circulation of the hidden states. Subsequently, the intraseasonal and interannual variability of the hidden states are described for the summer rainfall over the NEB. Whereas the interannual variability of the hidden states are assessed for the autumn rainfall and their potential predictors. Results for summer present a four-state HMM, States 1 (SS1), 2 (SS2), 3 (SS3), and 4 (SS4). The SS1 and SS3 correspond to the wet versus dry conditions associated with the interplay between the Intertropical Convergence Zone (ITCZ) and South American Monsoon System (SAMS). The SS2 and SS4 feature a southwest-northwest seesaw of precipitation occurrence, associated with antisymmetric conditions between the ITCZ and South Atlantic Convergence Zone (SACZ). The Madden Julian Oscillation (MJO) seems to affect the development of the SACZ in SS3 and SS4. In SS4, the MJO influenced the convection in the northwest NEB during phases 8-1-2. During Easterly Quasi-Biennial Oscillation (EQBO), the convective activity in the SACZ regions is intensified in SS3 and SS4 during MJO phase 1, and weakened in SS4 during phases 1, 3, and 4. SS1 (SS3) is preconditioned by a positive (negative) South Atlantic Ocean Dipole (SAOD), respectively. SS2 (SS4) is modulated by a positive (negative) El Niño-Southern Oscillation (ENSO). Results for autumn present a four-state HMM, States 1 (AS1), 2 (AS2), 3 (AS3), and 4 (AS4). The state sequence AS3 → AS2 → AS4 ↔ AS1 describes the autumn rainfall patterns in the NEB. The ZCIT and SAMS merge (separate) in AS2 (AS1) to a greater (minor) extent than climatology and NEB becomes largely wet (dry). AS3 is characterized by SACZ events, while AS4 is associated with an anomalous Eastern Wave Disturbances (EWD) event. AS3 (AS4) also experiences the influence of a southeast (northwest) shift of the South Atlantic Subtropical High (SASH). The four states are driven by the ENSO, which can act in co-occurrence with the Atlantic modes in AS1, AS2, and AS3 and with the Indian mode in AS1 and AS4. The extratropical atmospheric modes in response to the ENSO also impact the frequencies of AS1, AS3, and AS4. AS1 and AS2 exhibit high potential for predictability with the Atlantic and Pacific indices as their predictors. For AS3 (AS4), the Pacific (Atlantic) indices are the best predictors. The HMM effectively categorizes rainfall spatiotemporal characteristics and their association with large-scale atmospheric circulation and variability modes. These findings enhance the understanding of the dynamic processes of rainfall over the NEB and their association with potentially predictable teleconnections on seasonal timescales.