Seasonal predictability of the extreme Pakistani rainfall of 2022 possible contributions from the northern coastal Arabian Sea temperature

Pakistan and northwestern India experienced extremely wet conditions during June–August (JJA) of 2022, with the Pakistan rainfall (PR) anomaly over 20°N–35°N, 60°E–75°E reaching 2.76 mm day⁻¹—more than four standard deviations above the 40-year interannual variability (0.66 mm day⁻¹). The resulting floods caused major human, infrastructural, and agricultural losses. Seasonal predictability of regional rainfall is typically linked to tropical sea surface temperature (SST) anomalies and teleconnections (e.g., ENSO, IOD). Prior work has associated Pakistan’s summer rainfall variability with ENSO and the Indian Ocean Dipole (IOD), but relationships can be regionally heterogeneous within Pakistan and vary across eras and events. Despite successfully predicting the concurrent negative IOD and La Niña in 2022, major seasonal prediction systems (SINTEX-F2 and NMME) failed to forecast the extreme PR anomaly from May initialization. This study investigates the origin of that failure, focusing on inter-ensemble co-variability and targeted numerical sensitivity experiments, to identify potential avenues for improving seasonal prediction skill for Pakistan’s extreme rainfall.

The study builds on known sources of seasonal predictability from tropical SST anomalies and teleconnections. Previous research linked Pakistan rainfall to ENSO and the IOD: positive IOD phases can increase Pakistan rainfall (Ashok et al. 2004), and triple-dip La Niña was implicated in 2022 flooding (Jeong et al. 2023). Other studies reported regionally varying correlations within Pakistan, with limited IOD influence outside coastal/western areas (Hussain et al. 2017), and temporal shifts in flood drivers pre-/post-1970 (Iqbal and Hassan 2018). For 2010 flooding, roles of European blocking and tropical–extratropical interactions (Hong et al. 2011) and combined La Niña–negative IOD forcing (Priya et al. 2015) were highlighted. More generally, seasonal mean precipitation predictability is higher where SST–rainfall correlations are positive (Kumar et al. 2013). These mixed findings motivated a closer examination of regional SST influences, particularly in the northern coastal Arabian Sea (NAS), for 2022.

- Prediction systems and reforecasts: Analyzed 108-member SINTEX-F2 seasonal reforecasts initialized on May 1–9 for 1991–2022 and NMME outputs (hindcasts and forecasts; 53 and 82 members respectively) from four models (NCEP-CFSv2, NCAR-CCSM4, NASA-GEOS5, GFDL-SPEAR). Model anomalies were defined relative to each model’s 1991–2020 monthly climatology at corresponding lead times. - Regions and indices: Pakistan rainfall (PR) defined over 60°E–75°E, 20°N–35°N; northern coastal Arabian Sea (NAS) defined over 60°E–75°E, 20°N–30°N. IOD quantified by Dipole Mode Index (DMI) and ENSO by Niño3.4 index. - Observations/reanalyses: GPCP precipitation, NOAA OISSTv2 and OISSTv2 High Resolution SST, and NCEP/NCAR reanalysis for atmospheric fields; anomalies computed vs. 1991–2020 climatology. - Inter-ensemble correlation: Computed correlations across ensemble members between PR anomalies and gridded fields (SST; 850 hPa moisture convergence/divergence) to detect co-variability patterns. Significance assessed via paired t-tests, accounting for ensemble sizes (e.g., 108 for SINTEX-F2, 82 for NMME; reduced for individual NMME models). - Sensitivity (nudging) experiments from early May 2022 with SINTEX-F2 (108 members): 1) NAS_OBS: Strong daily SST nudging in NAS toward observed NOAA OISSTv2 High-Resolution SST (coefficient 2400 W m⁻² K⁻¹; ~1-day timescale in upper 50 m) from May–August with a linearly tapering 5° buffer (extending to ~15°N and 55°E). 2) NAS_ANOM: Nudged toward observed daily SST anomalies added to the model’s daily climatology (1991–2020). Anomalies defined relative to the original prediction’s mean climatology. Statistical significance of differences from the original experiment was assessed (e.g., 99% level). - Additional analyses for the 2010 event: Repeated inter-ensemble correlation and NAS nudging experiments for 2010 from early May initialization to test generality. - Diagnostics: Compared precipitation and low-level (850 hPa) moisture convergence/divergence, SST patterns, and evaluated model climatological SST biases in NAS. Assessed potential teleconnections via 200 hPa geopotential height and wave activity flux (WAF).

- Baseline prediction failure in 2022: - Observed PR anomaly (JJA 2022): +2.76 mm day⁻¹ (>4σ; σ=0.66 mm day⁻¹). - SINTEX-F2 ensemble mean: +0.19 mm day⁻¹ (~7–10% of observed); only two members exceeded +1.0 mm day⁻¹. - NMME ensemble mean: +0.52 mm day⁻¹ (<20% of observed). - Historical skill (1991–2021) for May-initialized PR: SINTEX-F2 r=0.31; NMME r=0.33 (both not significant at 95%). - Large-scale drivers were captured but insufficient: - Negative IOD and La Niña were predicted, though DMI magnitude was overestimated (SINTEX-F2 −1.40 °C; NMME −1.16 °C vs. observed −0.77 °C). Niño3.4 was −0.68 °C (SINTEX-F2) and −0.38 °C (NMME) vs. observed −0.74 °C. - Critical role of northern coastal Arabian Sea (NAS) SST: - Observed NAS SST anomaly: +0.59 °C; SINTEX-F2 underestimated (+0.22 °C), NMME closer (+0.53 °C). - SINTEX-F2 inter-ensemble correlation between PR and NAS SST: r=0.27 (99% significant); positive co-variability also with 850 hPa moisture convergence over NAS/Pakistan. - Other NMME models: NCEP-CFSv2 r=0.42 (95%); NASA-GEOS5 r=0.73 (95%); NCAR-CCSM4 and GFDL-SPEAR negative/non-significant. - Sensitivity (nudging) experiments for 2022: - NAS_ANOM: PR ensemble mean +0.42 mm day⁻¹ (~220% of original; ~15% of observed); significant difference vs. original at 99% level; more members >+1.0 mm day⁻¹ (nearly tripled). Enhanced 850 hPa moisture convergence over NAS; increased orographic rainfall over Western Ghats (comparable effectiveness when normalized by climatology). - NAS_OBS: PR ensemble mean +0.69 mm day⁻¹ (~25% of observed); further enhancement of moisture convergence; increased probability of PR >+1.0 mm day⁻¹. - Broader evidence across years: - Observational correlation (1991–2022) between PR and NAS SST anomalies: r=0.60 (significant), evident for both extreme and moderate PR events. - Composite of extreme PR years (1994, 2006, 2007, 2010, 2020) shows positive NAS SST anomalies; La Niña contributes in some years (2007, 2010, 2020); IOD signal cancels across mixed phases. - NAS SST anomalies show insignificant observed correlations with Niño3.4 and DMI (JJA 1991–2022), suggesting NAS variability not simply driven by ENSO/IOD. In SINTEX-F2 ensembles (2022), NAS SST correlates with DMI (r=0.35, 99%); NCEP-CFSv2 r=0.39 (95%). - 2010 case study: - Baseline predictions from May failed to capture magnitude of PR; NAS warming and 850 hPa moisture convergence associated with observed wet anomalies. - SINTEX-F2 inter-ensemble PR–NAS SST r=0.42 (99%); NCEP-CFSv2 r=0.44 (95%). - Nudging: NAS_ANOM PR mean +0.21 mm day⁻¹ vs. original +0.073 (~300%; ~16% of observed +1.30). NAS_OBS PR mean +0.41 (~30% of observed). - Model bias and teleconnections: - Negative SST climatology bias in NAS in SINTEX-F2 and several NMME models likely contributed to underprediction. - Teleconnection signals: Enhanced NAS convection appears to excite stationary Rossby waves, meandering the subtropical Asian jet and contributing to hotter-than-normal conditions in East Asia (supported by 200 hPa height and WAF diagnostics).

The study set out to understand why seasonal prediction systems failed to forecast the extreme 2022 Pakistan rainfall despite correctly predicting concurrent La Niña and negative IOD. Analyses reveal that regional SST anomalies in the northern coastal Arabian Sea (NAS) were a crucial, underrepresented driver. Positive NAS SST anomalies enhanced local convection and low-level moisture convergence, which in turn amplified rainfall over Pakistan. When the model SST in NAS was constrained toward observations (or observed anomalies), predicted PR increased substantially and more ensemble members captured large positive anomalies, demonstrating a causal influence and an actionable pathway to improve forecasts. The findings refine the understanding of predictability sources: while basin-scale ENSO/IOD states set a background, local/regional SST in NAS provides additional, independent predictability for Pakistan rainfall. Moreover, the identified teleconnection pathway to East Asia suggests NAS-forced convection can excite upper-tropospheric wave trains along the subtropical jet, linking Pakistan rainfall anomalies with East Asian summer heat. Addressing NAS SST biases and better initializing/predicting regional SSTs can thus improve both local rainfall forecasts and downstream teleconnection forecasts.

Positive SST anomalies in the northern coastal Arabian Sea were key contributors to the extreme Pakistan rainfall in summer 2022 and to the failure of seasonal systems that underestimated NAS warming. Targeted SST nudging experiments (NAS_ANOM, NAS_OBS) markedly improved predicted PR, capturing up to ~25% of observed anomalies in 2022 and ~30% in 2010, by enhancing moisture convergence and regional convection. Observational records (1991–2022) support a robust PR–NAS SST relationship. The results highlight that improving prediction and initialization of NAS SST—alongside reducing model SST biases, potentially via higher-resolution ocean modeling—can substantially enhance seasonal forecasts of Pakistan rainfall and related teleconnections to East Asia. Future work should include coordinated multi-model sensitivity experiments, deeper investigation of NAS SST biases and Western Boundary Upwelling processes, and exploration of interactions between seasonal-mean SST anomalies and high-frequency cyclogenesis to better explain extreme tail events.

- Despite improvements, sensitivity experiments still markedly underpredicted the 2022 PR anomaly; no ensemble member reached the observed +2.76 mm day⁻¹ in NAS_OBS or NMME, indicating unresolved drivers and/or inherent atmospheric unpredictability. - Synoptic to subseasonal cyclogenesis and high-frequency processes in the eastern Arabian Sea were outside the study’s scope; their interaction with seasonal-mean NAS SST could be crucial for extremes. - Model climatological SST biases in NAS (negative biases) persist across several systems and may reflect resolution and physics limitations; attribution of biases remains incomplete. - Skill assessments are from May initializations and may vary with lead time; differences in anomaly base periods, ensemble sizes, and model subsets for NMME could affect comparisons. - Inter-ensemble correlation signals vary across models; not all systems show significant PR–NAS SST co-variability, limiting multi-model consensus.