In June 2026, GlobMaps published an open-access preprint (DOI: 10.5281/zenodo.20774311) analyzing 25 years of drought data across 156 provinces in Thailand, Vietnam, and Malaysia. The headline finding is counterintuitive: despite rising temperatures and widening climate anxieties, 89% of Southeast Asian provinces show no statistically significant trend toward worsening drought. What is actually driving regional drought is older, more cyclical, and in some ways more manageable than a monotonic warming story suggests.
The Dataset: ERA5, MODIS, and 25 Years of Province-Level Records
The analysis draws on ERA5 climate reanalysis from the European Centre for Medium-Range Weather Forecasts — the same data backbone powering the GlobMaps drought platform — combined with MODIS vegetation condition data and ALEXI evaporative stress satellite retrievals. The study period spans 2000 to 2025, covering all 77 provinces of Thailand, 63 provinces of Vietnam, and 16 states and federal territories of Malaysia.
ERA5 provides hourly estimates of global atmospheric and land-surface variables at 31-kilometer resolution, reconstructed from historical observations using data assimilation. For drought analysis, its temperature and precipitation outputs feed directly into SPEI calculations, producing a consistent physical baseline across the entire 25-year period — including years before dense rain gauge coverage existed in many rural provinces. That consistency is the foundation on which province-level comparisons are built.
MODIS vegetation indices add a biophysical check: when rainfall-based drought signals appear, greenness and canopy water content data from satellite should confirm crop and forest stress. ALEXI evaporative stress retrievals provide a third, independent signal. A province classified as D2 or higher in the composite index should show convergent evidence across all three streams.
Building Drought Confidence at Province Level
One structural challenge in province-level drought analysis is data heterogeneity. ERA5's 31-kilometer grid means that a geographically small province like Phuket (576 km²) may overlap only a handful of grid cells, while a large province like Chiang Rai (11,678 km²) captures dozens. The study addressed this through a confidence tier system: provinces with consistent multi-source signal agreement received high-confidence classifications (42 provinces), while those with ambiguous or limited spatial coverage received medium-confidence ratings (114 provinces).
No province was assigned a drought classification without at least two converging signal streams. This conservative approach means some provinces with real drought risk may be slightly under-classified, but it also means the high-confidence province set — 42 provinces — provides a solid benchmarking anchor for parametric insurance and financial risk models that require defensible data provenance.
The Main Finding: ENSO, Not a Warming Trend
Of the 156 provinces analyzed, 139 — 89% — showed no statistically significant monotonic trend in annual drought severity over the 25-year record. The data does not support a narrative in which Southeast Asia is uniformly and progressively drying out. Instead, drought variability tracks the El Niño–Southern Oscillation (ENSO) cycle: El Niño years push drought severity higher across the region; La Niña years provide relief.
The average ENSO correlation across all provinces — measured between annual drought severity and the Oceanic Niño Index (ONI) — is r ≈ −0.31. But the signal is far from uniform. Sabah, Malaysia, shows the strongest ENSO sensitivity in the dataset at r = −0.57. Lam Dong province in Vietnam's central highlands follows at r = −0.53. These numbers mean that the ONI explains roughly a third of annual drought variance region-wide, and over half in the most ENSO-exposed provinces — a leverage point for seasonal forecasting that trend-based models cannot provide.
The ENSO dominance has an important practical implication. ENSO forecasts now carry useful skill out to 9–12 months, issued by agencies including NOAA, the Japan Meteorological Agency, and Australia's Bureau of Meteorology. For provinces with the highest ENSO coupling (r above −0.45), a strong El Niño forecast in May translates directly into a quantifiable drought probability elevation by the following dry season — the kind of forward signal that can trigger pre-positioned relief supplies, crop switching decisions, or parametric insurance payouts before damage occurs.
Where Trends Do Exist
The 17 provinces with significant trends break down asymmetrically. Sixteen are in northern and north-eastern Thailand — the Isaan plateau region that was already among the driest in Southeast Asia. These provinces show statistically significant drying over the 25-year period: not just El Niño volatility, but a structural shift toward less rainfall per decade. For risk managers with exposure in Isaan specifically, historical drought frequency may understate the forward risk.
What sets the Isaan signal apart from ENSO noise is its persistence across ENSO phases. Even in La Niña years — which should bring relief to most of the region — several Isaan provinces continued to register below-average moisture conditions. This decoupling from the regional ENSO pattern is consistent with documented changes in the monsoon system affecting the Khorat Plateau, and suggests that Isaan's forward drought trajectory is driven by more than cyclical variability alone.
The remaining outlier is Phu Tho in northern Vietnam, where the data shows a significant wetting trend, consistent with documented shifts in the western monsoon track affecting Vietnam's northern highlands. This province represents the opposite tail of the distribution: a region that may face increased flood risk as rainfall intensifies, even as drought risk falls.
2023: The Worst Year in 25 Years of Record
The 2023/24 El Niño produced the most severe drought in the 25-year satellite-era record. Region-wide, the mean SPEI-3 reached approximately −1.3, indicating severe drought conditions simultaneously across a majority of the 156 provinces. That event is now a concrete, data-grounded planning baseline: if your infrastructure, agricultural finance portfolio, or supply chain touches Southeast Asia, 2023 represents the worst observed drought stress in a quarter century.
The severity of 2023/24 relative to the 2015/16 event — a super El Niño with a peak ONI of +2.6 — is itself a finding worth dwelling on. The 2015/16 El Niño was, by ONI magnitude, the strongest in the instrumental record. The 2023/24 event was moderate by comparison, peaking near +2.0. Yet the drought footprint in 2023/24 was larger. Several factors contributed: pre-existing La Niña conditions in 2021–22 may have reduced soil moisture reserves going into the event; regional warming trends in sea surface temperatures amplified evapotranspiration demand independently of ENSO; and certain ENSO teleconnection patterns were particularly unfavorable for mainland Southeast Asia in 2023.
The practical lesson is that ONI magnitude alone does not determine drought severity in any given province. Province-level ENSO correlation coefficients, combined with pre-event soil moisture conditions, provide a better forward indicator than a single index reading.
The Mekong Delta Anomaly
Among the clearest structural surprises in the dataset is the Mekong Delta's anomalous drought relationship. Across most of the study region, drought severity correlates negatively with ONI: higher ONI (El Niño) means more drought. In the Mekong Delta, the relationship is partially inverted: certain La Niña configurations that bring rainfall to most of mainland Southeast Asia simultaneously suppress Mekong river flows arriving from the upper basin, reducing alluvial recharge and producing moisture deficits in lower Delta provinces even as rainfall totals are near-normal.
This is not unique to this analysis — hydrological studies of the Mekong have documented the upper-lower basin asymmetry for decades. But the province-level composite index makes the pattern visible in a form directly usable for risk modelling: specific provinces in Kiên Giang, Cà Mau, and adjacent areas show ENSO correlations that diverge from regional norms, with drought risk elevated in La Niña conditions that elsewhere produce relief.
For agricultural finance and infrastructure risk teams with exposure in Vietnam's southern provinces, this means that standard ENSO-based drought outlooks — which typically predict La Niña as a positive signal for mainland SEA — require a province-specific correction for the lower Mekong basin.
How Often Does Drought Actually Occur?
Across the 25-year period, moderate-or-worse drought conditions occurred in:
- Thailand: 6.7% of province-months
- Vietnam: 8.8% of province-months
- Malaysia: 10.9% of province-months
Malaysia's higher frequency is structural: a tropical rainforest climate adapted to near-year-round rainfall responds to even a moderate El Niño dry spell more acutely than mainland climates that expect a dry season. Sabah and peninsular Malaysia experience drought less often in absolute terms — but when it arrives, it arrives as an anomaly in a system not designed to manage it, which is why the ENSO signal there (r = −0.57) is the strongest in the dataset.
Conditional on an El Niño event occurring, drought frequency roughly doubles across all three countries. Conditional on a strong El Niño (ONI ≥ +1.5), severe drought probabilities in the most sensitive provinces increase by a factor of three or more. This ENSO-conditioned frequency structure is the basis for the seasonal drought alert system integrated into the GlobMaps platform.
What This Means for Risk Modelling
The ENSO-dominant, non-trend structure of Southeast Asian drought has a clear methodological implication: risk models that extrapolate a linear worsening trend from recent data will overstate forward drought risk in 89% of provinces, while potentially underweighting the episodic severity of El Niño stress events — exactly the dynamic that caught many portfolios exposed in 2023.
A well-calibrated drought risk model for this region should have at least three components that trend-only models omit:
- ENSO phase conditioning: drought probability should be updated quarterly as ENSO forecasts are issued, not only as a long-run annual average
- Province-specific ENSO sensitivity: the r = −0.57 provinces are categorically different from r = −0.15 provinces in their exposure profile
- Structural exceptions: Isaan's trend, and the Mekong Delta's inverse ENSO relationship, are not captured by regional averages and require province-level treatment
Province-specific ENSO correlation coefficients — available for all 156 provinces through the GlobMaps API — give parametric risk models a calibration anchor grounded in 25 years of observational data rather than theoretical assumptions.
Access the Data
The full methodology, confidence tier assignments, and province-level results are published in our open-access preprint at DOI: 10.5281/zenodo.20774311. The dataset includes annual drought severity scores, ENSO correlation coefficients, trend p-values, and confidence tier assignments for all 156 provinces.
Province drought scores are updated monthly via ERA5, with province-level MDI classifications, historical time series, and ENSO correlation metadata — structured for direct use in Excel, Python, or any risk analytics environment without data transformation.
More on how we build and validate these indices at globmaps.com.