A predictable prospect of the South Asian summer monsoon

The South Asian summer monsoon (SASM) displays strong interannual variability in rainfall and atmospheric circulation, affecting over a billion people and modulating large-scale circulations via diabatic heating. Operational seasonal prediction skill of SASM variability—often measured by the All India Rainfall Index (AIRI)—has been low in recent decades. AIRI focuses on rainfall over the Indian subcontinent and overlooks substantial SASM rainfall over the Bay of Bengal (BOB), the global maximum atmospheric heat source region. Numerous alternative indices, including dynamic shear-based and seasonality-based measures, have been proposed to quantify SASM variability, but many underrepresent variability associated with the Indian monsoon trough (IMT), a key feature extending from northern BOB to western India and responsible for much of South Asia’s rainfall. ENSO-related SST anomalies force SASM variability, but the ENSO–monsoon relationship has changed interdecadally. The leading Indian monsoon rainfall mode represented by AIRI tends to co-vary with equatorial Pacific SSTs, while the second mode relates to IMT over northern India and can be partially explained as a lagged ENSO response. Prediction based on AIRI is hindered by the ENSO spring predictability barrier. The study aims to better characterize and predict SASM by coupling rainfall and low-level circulation over all South Asia, emphasizing convection and the IMT over the northern BOB, and to develop a simple, physically grounded index representing the dominant, more predictable mode.

Prior work established monsoon variability’s societal and dynamical importance and the teleconnection with ENSO. Widely used SASM indices include: (1) thermal contrast/vertical shear metrics such as Webster–Yang index (WYI) and monsoon Hadley circulation index (MHI); (2) rainfall/convection metrics such as AIRI, extended Indian monsoon rainfall index (EIMRI), and convection index (CI); (3) shear vorticity-based Indian monsoon index (IMI); and (4) dynamically normalized seasonality indices (SASMI, SASMI1, SASMI2). These indices capture different aspects of the monsoon, often emphasizing broad-scale circulation or subcontinental rainfall and tending to neglect IMT variability over the BOB. ENSO influences SASM both via Indian Ocean thermal memory and simultaneous atmospheric forcing, but the strength and timing of these links vary. The literature documents limited predictability of AIRI-like measures due to ENSO’s spring barrier and highlights the IMT’s role in extreme rainfall and mesoscale systems. This study builds on these insights to target the coupled rainfall–circulation mode tied to the IMT.

Data: Monthly 850-hPa wind and geopotential height from ERA5 (0.25°×0.25°), monthly SST from HadISST v1.1 (1°×1°), and monthly rainfall from GPCP v2.3 (2.5°×2.5°) for 1979–2020. GPCC v2020 (1°×1°) monthly rainfall and APHRODITE daily rainfall (1979–2015) are used for validation. AIRI (1979–2019) and SASMI family indices (1979–2020) are obtained from their respective sources. All data are normalized and detrended as appropriate. Mode extraction: Singular value decomposition (SVD) is applied to coupled fields of June–August (JJA) 850-hPa wind over 65°–105°E, 10°–30°N and rainfall over 65°–105°E, 10°–40°N for 1979–2020 to isolate dominant modes of cross-covariance between circulation and rainfall. The first two modes, statistically separated via a Monte Carlo test, are analyzed. Principal components (PCs) from wind and rainfall counterparts are compared to assess coupling. New index definition: Based on SVD1’s physical structure (cyclonic low-level anomalies from northern BOB to south of the Tibetan Plateau and a NW–SE sandwich rainfall pattern), the Indian Monsoon Trough Index (IMTI) is defined as the area-mean 850-hPa zonal wind over 85°–100°E, 10°–18°N minus that over 83°–95°E, 23°–27°N, capturing IMT intensity and zonal position. Statistical prediction: Relationships between IMTI and global SSTs are assessed from preceding winter (DJF) through spring (MAM) to summer (JJA). Two spring predictors are selected based on persistence and correlation: TEIO_SST (eastern tropical Indian Ocean, 90°–130°E, 0°–20°N) and TA_SST (tropical Atlantic, 305°–345°E, 10°S–20°N). A LASSO regression is trained over 1979–2010 with 5-fold cross-validation to predict JJA IMTI one season ahead using MAM TEIO_SST and TA_SST. Validation is performed for 2011–2020. For comparison, a simple linear regression using Niño-3 DJF alone is also tested. Significance of correlations is assessed with Student’s t-test; for a 42-year record, r thresholds are 0.304 (95%), 0.393 (99%), 0.490 (99.9%). LASSO performance metrics reported include explained variance (training/prediction) and RMSE.

- Two dominant coupled modes of SASM rainfall–circulation variability explain 75.89% of the total squared covariance. SVD1 explains 51.91% and exhibits an anomalous low-level cyclonic circulation over the northern Bay of Bengal and south of the Tibetan Plateau, with a NW–SE sandwich rainfall pattern: positive anomalies along central India–northeastern BOB and negative anomalies to the west coast of India and SE edge of the Tibetan Plateau. Wind–rainfall PCs for SVD1 correlate at R=0.937. - SVD2 explains 23.98%, with a cyclonic circulation centered over the Indian subcontinent and widespread positive rainfall anomalies over India, and wind–rainfall PCs correlate at R=0.870. SVD2 corresponds to a meridional IMT shift. - ENSO links: SVD1 correlates significantly with central–eastern tropical Pacific and Indian Ocean SSTs in preceding winter and spring (ENSO decaying), while SVD2 shows only simultaneous summer Pacific SST signals. Correlations for rainfall PCs: PC1 with Niño-3 is −0.581 (DJF) and 0.199 (JJA); with Niño-3.4 is −0.542 (DJF) and 0.340 (JJA). PC2 with Niño-3 is 0.001 (DJF) and −0.267 (JJA); with Niño-3.4 is 0.021 (DJF) and −0.343 (JJA). The 95% threshold is |r|≥0.304. - New index (IMTI) closely proxies SVD1: correlations with PC1s are 0.896 (wind) and 0.938 (rainfall), both exceeding 99.9% confidence. IMTI-associated rainfall/wind patterns replicate SVD1’s structure. - Index benchmarking (Table 2): Most existing indices (AIRI, EIMR, MHI, CI, IMI) correlate strongly with SVD2 rather than SVD1; WYI and SASMII show weak/regionally displaced links. SASMI and especially SASMI2 correlate with SVD1; SASMI2 shows high association with preceding ENSO, though with different physical meaning from SVD1. - Predictability: IMTI exhibits strong SST linkages characteristic of ENSO’s mature–decay evolution. Spring predictors TEIO_SST and TA_SST correlate with IMTI at −0.567 and −0.597, respectively, and with preceding winter ENSO at 0.681 and 0.399; inter-predictor correlation is 0.372. - LASSO prediction using MAM TEIO_SST and TA_SST yields overall correlation skill R=0.698 for 1979–2020 (99.9% confidence), R=0.719 during 1979–2010 training, and only two false hits (2013, 2016) in 2011–2020. Explained variance is 51.66% (training) and 34.37% (prediction); RMSE 1.19 and 1.24, respectively. A simpler Niño-3 DJF-based model attains R=0.472 overall (0.506 training). - ENSO decaying linkage: 11 of 16 extreme IMTI years coincide with ENSO decaying summers (7 negative IMTI with El Niño decay; 4 positive IMTI with La Niña decay), underscoring the lagged forcing. - The authors estimate at least 48.7% of SASM variance is predictable via the first-mode-based IMTI and SST precursors.

The study directly addresses the SASM predictability problem by separating two coupled modes of rainfall–circulation variability and demonstrating that most commonly used indices emphasize the less predictable second mode. The dominant first mode reflects IMT intensity and zonal extension, tightly coupled with rainfall over central India–northeastern BOB and the SE Tibetan Plateau margin, and shows robust lagged connections to persistent tropical SST anomalies during ENSO’s decaying phase (notably basin-wide Indian Ocean warming and Atlantic signals). Consequently, it is substantially more predictable one season ahead. Conversely, the second mode lacks significant pre-season ocean signals and is influenced more by internal atmospheric variability and mid–high-latitude wave trains, reducing its predictability. Physical mechanisms include: (i) Indian Ocean capacitor effect, where SE tropical Indian Ocean warming strengthens local Hadley circulation, suppressing convection over northern BOB/central India, weakening the IMT; (ii) suppressed western North Pacific convection during El Niño decaying summers inducing westward-propagating Rossby waves and South Asian anticyclonic anomalies; and (iii) the first mode’s stronger sensitivity to tropical forcing and partial shielding from midlatitude influences by the Tibetan Plateau. The new IMTI, as a simple, dynamically grounded proxy for SVD1, enables objective real-time monitoring and seasonal prediction, reframing perceptions of SASM predictability and rationalizing discrepancies among existing indices by aligning them with distinct physical modes.

By identifying two dominant coupled modes of SASM variability and introducing the Indian Monsoon Trough Index (IMTI) as a physically grounded proxy of the first mode, the study demonstrates that a large and predictable fraction of SASM interannual variability exists. IMTI captures IMT intensity and zonal position, correlates strongly with the leading coupled mode, and is predictable one season in advance using persistent spring SST anomalies in the eastern tropical Indian Ocean and tropical Atlantic (overall skill R=0.698 for 1979–2020). The work reclassifies existing monsoon indices according to their association with the two modes, explaining past prediction shortcomings when relying on second-mode-related indices. The authors recommend IMTI for real-time monitoring and seasonal prediction and suggest future research to enhance predictive skill, further investigate the physical commonalities with SASMI2, and extend multi-scale monitoring/prediction leveraging IMTI’s responsiveness to intraseasonal variability.