The study combines daily precipitation observations for 1961–2022 with 25 retained CMIP6 model-variant simulations for 2015–2100. Its 0.1-degree observational dataset was interpolated from 11,473 rain gauges and 1,252 weather stations, then averaged within the boundaries of Brazil’s 82 official hydrographic microregions. Model precipitation was bias-corrected with multiplicative quantile delta mapping, using a fixed 1961–2014 reference period for calibration and for the principal SPI distributions. An independent test recalibrated the correction for 1961–1990 and evaluated it over 1991–2014, checking whether it improved precipitation and drought-related properties outside the calibration period. Because each source ID–variant combination received equal weight, the ensemble summaries describe the available models but are not independence-adjusted probabilities, and one realization per simulation prevents internal climate variability from being separated from inter-model spread.

Drought was measured with the 12-month Standardized Precipitation Index, or SPI-12, which transforms accumulated precipitation into a standardized measure of wetness and dryness. The analysis used the December value each year to represent approximately January–December precipitation. A negative-SPI-12 event was defined as consecutive years with values below zero; its duration was the number of years, its severity the absolute cumulative SPI deficit and its intensity the severity divided by duration. This definition captures prolonged sequences of mildly negative annual anomalies as well as sharply dry years. It also identifies an extreme-peak event as a negative sequence that reaches an annual SPI-12 value of −2.0 or lower. As a precipitation-only measure, SPI-12 does not capture temperature-driven evaporative demand, river flow, reservoir storage, groundwater recharge or socioeconomic impacts.

The projected distribution of annual SPI-12 was lower across most microregions under both emissions pathways. The median of model-specific temporal medians was −0.324 under SSP2-4.5 and −0.526 under SSP5-8.5. When each microregion was compared with its own observed median, the median local change was −0.305 SPI units under SSP2-4.5 and −0.449 under SSP5-8.5. The ensemble-median frequency of years with SPI-12 at or below −1 exceeded the observed frequency in 79 of 82 microregions under SSP2-4.5 and in all 82 under SSP5-8.5. Models nevertheless differed substantially: some produced more negative median SPI-12 values, while others remained closer to zero or slightly positive in certain locations. The regional result therefore describes a common distributional shift accompanied by substantial spatial and inter-model variation rather than an identical trajectory across Northeast Brazil.

Trend analysis showed broad agreement on the direction of change but much less evidence for a consistent linear decline at every location. For 2015–2100, the median proportion of simulations producing negative Sen slopes was 74.0% under SSP2-4.5 and 84.0% under SSP5-8.5. At least three-quarters of simulations agreed on drying in 41 and 58 microregions, respectively. Far fewer results remained statistically supported after a false-discovery-rate correction: the median fraction combining a negative slope with an adjusted p-value below 0.05 was 8.0% under SSP2-4.5 and 36.0% under SSP5-8.5. The horizon analysis also showed that the signal was not a simple progression. Under SSP2-4.5, the ensemble-median Sen slope was approximately zero in 2063–2100, whereas it remained negative in both horizons under SSP5-8.5, although its magnitude and agreement did not increase monotonically. These results distinguish a downward shift in annual dryness from widespread FDR-supported local monotonic trends.

Area-weighted results showed that drought conditions would coincide across larger portions of Northeast Brazil in more years, although severe and extreme coverage would remain less widespread than moderate coverage. In the second half of the century, mean ensemble-median coverage rose from 18.6% during 2015–2057 to 23.9% under SSP2-4.5 for SPI-12 at or below −1.0. Under SSP5-8.5, it increased from 21.0% to 34.3%. Severe coverage under SPI-12 at or below −1.5 reached 18.3% in the late-century SSP5-8.5 average, while extreme coverage under SPI-12 at or below −2.0 rose to 8.4%. During 2058–2100 under SSP5-8.5, 79% of years had at least 25% of the region under moderate drought according to the ensemble median, 28% exceeded 40% coverage, and severe coverage reached or exceeded 25% in 16% of years. Even in that late-century high-emissions case, moderate-drought coverage remained below 25% in about 16% of years, highlighting strong year-to-year and inter-model variability.

Local event sequences added detail to the projected changes in annual coverage. The median annual onset rate of all negative-SPI-12 events declined under every scenario and horizon, reaching a change of −0.064 events per year in the far future under SSP5-8.5. At the same time, median event duration increased by 1.0 year in most future conditions and by 1.5 years in the far future under SSP5-8.5. The proportion of microregions whose typical event lasted at least two years rose from 27% in the observations to between 89% and 96% across the four future cases. Median severity increased by 0.86 to 2.18 SPI-deficit units, while median intensity rose by 0.18 to 0.46 deficit units per year. The modeled increase in cumulative severity therefore reflected both longer continuity of negative annual conditions and greater average deficits within those sequences; it was not an estimate of hydrological or water-supply shortages.

Extreme annual conditions within those sequences were projected to become more common even as the onset rate of new negative-SPI-12 sequences declined. The median change in the annual onset rate of events reaching SPI-12 of −2.0 or lower was positive in all four scenario–horizon combinations. It ranged from 0.018 to 0.020 new extreme-peak events per year under SSP2-4.5 and from 0.018 to 0.034 under SSP5-8.5. Projected rates exceeded observed rates in 70 to 80 of the 82 microregions, depending on the scenario and horizon. Agreement was strongest under late-century SSP5-8.5, when at least 75% of simulations projected an increase in 51 microregions, although changes were generally smaller and less consistent in central areas. Together, the onset and sequence results indicate fewer interruptions of below-reference annual precipitation conditions, more persistent negative runs and more frequent extreme annual values within those runs.

A bivariate copula analysis tested the recurrence of events comparable to each microregion’s most severe observed event overlapping the 2012–2018 regional drought. One criterion required an event to equal or exceed both the reference duration and cumulative severity; another required it to equal or exceed either one. Under the stricter “both” criterion, the median projected-to-historical return-period ratio was 1.03 under SSP2-4.5 but 0.65 under SSP5-8.5, corresponding to a median return period about 35% shorter in the higher-emissions case. Under the “either” criterion, the median ratios were 0.75 and 0.60, respectively. Ratios below one occurred in 59 of 82 microregions under SSP2-4.5 and 63 under SSP5-8.5 for the inclusive criterion. Uncertainty remained substantial: the 95% bootstrap interval included one in 38 and 27 microregions for the “either” criterion under SSP2-4.5 and SSP5-8.5, and in 23 and 34 microregions for the “both” criterion. The inclusive results therefore gave the more extensive evidence of shorter return periods, while the stricter results retained greater spatial heterogeneity.

The findings provide a spatial basis for prioritizing local water-security assessment, but they are not direct forecasts of reservoir deficits, crop losses, hydrological drought or water-supply shortages. SPI-12 isolates precipitation-based meteorological drought and excludes temperature-driven evaporative demand, streamflow, soil moisture, groundwater, water use and social vulnerability. The comparison also uses different historical and projected catalogues rather than paired year-by-year realizations, and a univariate bias-correction method cannot guarantee that event persistence or duration–severity dependence is reproduced without bias. The projected return periods are conditional on each local reference event, selected copula family and catalogue-specific event rate, while bootstrap intervals exclude uncertainty in copula-family selection and event frequency. The central finding is therefore a heterogeneous, scenario-dependent shift toward more persistent and severe precipitation deficits, with more frequent extreme annual conditions and stronger evidence under SSP5-8.5, but limited FDR-supported local trend detection and no uniform region-wide trajectory of increasingly frequent independent droughts.

Subject of Research: CMIP6 projections of precipitation-based meteorological drought across Northeast Brazil

Article Title: Past-to-future drought across hydrographic microregions in Northeast Brazil: SPI-based diagnostics and bivariate (duration–severity) risk under CMIP6 scenarios

Article References: Rolim, L. Z. R., de Assis de Souza Filho, F., & Estácio, Á. B. S. (2026). Past-to-future drought across hydrographic microregions in Northeast Brazil: SPI-based diagnostics and bivariate (duration–severity) risk under CMIP6 scenarios. Natural Hazards, 122(22), Article 694. https://doi.org/10.1007/s11069-026-08433-x

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08433-x

Keywords: Northeast Brazil, precipitation drought, CMIP6, SPI-12, drought duration, drought severity, return periods, copula analysis, hydrographic microregions, climate projections, Past-to-future, drought

The study rests on the Meteorological Composite Index, or MCI, the official drought-monitoring metric used by the China Meteorological Administration under the national meteorological drought grade standard. Unlike simpler indices that rely on precipitation alone, the MCI blends daily records of precipitation, air temperature, wind speed, relative humidity, and sunshine duration into a single value that captures how atmospheric demand for water stacks up against supply. The researchers applied this index to daily meteorological data spanning 1961 to 2020 from seven representative stations across Guangxi, giving them six decades of drought history to dissect. From the daily MCI series, they used run theory, a technique borrowed from the analysis of time-series extremes, to identify individual drought events and extract three characteristic variables for each one: duration, measured in days; severity, the cumulative deficit accumulated while the index remained below the drought threshold; and peak, the single most extreme value reached during the event.

Once these three variables were extracted, the team confronted the central limitation of traditional drought analysis. Univariate methods treat duration, severity, and peak as if they were independent, fitting a separate frequency curve to each and reporting return periods one variable at a time. But drought characteristics are strongly correlated, and ignoring that correlation can badly distort risk estimates. To capture the dependence structure, the researchers fitted five candidate marginal distribution functions to each variable and then linked them using five Copula functions, mathematical constructs that describe how multiple random variables co-vary without imposing restrictive assumptions about their individual distributions. The fitting results were themselves informative: drought duration and severity across the Guangxi stations were best described by the Lognormal distribution, while drought peak followed the Gamma distribution, a pattern consistent with the skewed, heavy-tailed nature of drought variables observed worldwide.

The correlation analysis revealed just how tightly these drought dimensions are bound together. Drought duration showed the strongest relationship with drought severity, with Kendall’s tau, a rank-based measure of association, ranging from 0.66 to 0.72 across the seven stations. In practical terms, longer droughts in Guangxi are almost reliably more severe droughts, which means that any risk assessment based on duration alone will implicitly carry much of the severity signal while missing the independent contribution of intensity. Peak intensity, by contrast, behaved more independently, which is precisely why the multivariate framework matters: two droughts of identical duration and severity can pose very different threats to reservoirs and crops depending on whether their worst deficit arrived as a brief, brutal spike or a slow, grinding accumulation.

The spatial patterns that emerged from the analysis are striking. Baise, in northwestern Guangxi, recorded the longest annual average drought duration and the highest average severity of any station, at 127 days and 129 respectively, while Laibin, in the central part of the region, posted the single most extreme event in the record, with a drought lasting 194 days and reaching a severity of 416. Together, these results point to western and central Guangxi as the epicenters of long-duration, high-severity drought, a finding that aligns with the region’s vulnerability to rainless stretches between monsoon pulses and the limited water-holding capacity of its karst terrain. For planners, the message is that drought preparedness in these areas must be designed around events that are not merely frequent but exceptionally persistent.

Seasonality added another layer to the picture. Autumn and winter accounted for the highest proportions of drought days across the region, while summer showed a comparatively lower share, reflecting the seasonal rhythm of the East Asian monsoon, which concentrates rainfall in the warm months and leaves the cooler half of the year exposed. The station-level extremes were telling: Laibin recorded the highest proportion of autumn drought days at 11.6 percent, and Baise topped the winter ranking at 13.2 percent. Because autumn is a critical period for late-season crop maturation and winter matters for overwintering vegetation and reservoir recharge, these seasonal hotspots translate directly into agricultural and water-supply exposure that a purely annual analysis would smooth over.

The heart of the study lies in its return-period analysis, where the Copula-based joint distributions were used to answer the question that water managers actually face: how likely is a drought that is simultaneously long and severe, or severe and intense? The results showed that univariate return periods at each station consistently fell between two multivariate bounds: the joint return period, which describes the probability of exceeding specified values of both variables together, and the co-occurrence return period, which describes the probability of exceeding either one. This sandwich structure means that single-variable statistics can either overstate or understate the true compound risk depending on which combination of characteristics a given water system is most sensitive to. Among the three pairwise combinations tested, the duration-severity pairing produced a larger joint return period but a smaller co-occurrence return period than the other two, while the duration-peak and severity-peak combinations yielded joint return periods that were relatively close to each other even as their co-occurrence return periods diverged considerably, underscoring how differently each pairing encodes compound risk.

Perhaps the most actionable outcome is the regional differentiation of joint drought risk. Wuzhou, in the east, emerged as more prone to droughts combining long duration with high severity, the profile most threatening to multi-season water storage. Guilin, in the northeast, showed elevated risk of droughts pairing high severity with high peaks, the signature of intense, rapidly intensifying water deficits. Laibin, in the center, was identified as prone to droughts combining long duration with high peaks, a particularly punishing combination in which a prolonged dry spell also contains episodes of acute intensity. These distinct risk fingerprints mean that a one-size-fits-all drought mitigation strategy for Guangxi would be poorly matched to the actual hazards: reservoir operation rules, crop insurance design, and emergency water allocation plans each need to be tuned to the specific compound drought profile of their locality.

Methodologically, the study demonstrates why Copula-based multivariate analysis is rapidly becoming the standard for drought risk assessment, both in China and internationally. The approach preserves the full dependence structure among drought characteristics, allows flexible choice of marginal distributions, and produces return-period estimates that map directly onto the engineering decisions that infrastructure and agriculture depend on. As climate change alters precipitation regimes and increases atmospheric evaporative demand across much of Asia, the frequency and compound character of drought events are expected to shift, making robust multivariate baselines like the one constructed here essential reference points for detecting and attributing those changes. The authors note that their findings provide a reference basis for regional drought risk assessment and drought mitigation management, and the six-decade MCI record they have assembled for Guangxi offers exactly the kind of long, physically grounded dataset needed to stress-test future projections.

For the wider scientific community, the Guangxi case study carries lessons well beyond its borders. Karst regions across southern China and Southeast Asia share the same combination of thin soils, fast-draining bedrock, and monsoon variability that makes droughts there simultaneously hard to predict and devastating when they arrive. By showing that drought duration, severity, and peak must be analyzed jointly rather than in isolation, and that the compound risk landscape varies sharply from one station to the next, the study provides a template that drought-prone regions worldwide can adapt. The work was supported by the Guangxi Key Research and Development Program and a key research project on water resource allocation in the Beibu Gulf, reflecting the practical urgency that provincial authorities attach to understanding and managing the region’s evolving drought hazard.

Subject of Research: Multivariate statistical analysis of meteorological drought characteristics using the MCI index and Copula functions in Guangxi, China

Article Title: Multivariate Joint Distribution Characteristics of Meteorological Drought Based on the Meteorological Composite Index (MCI): A Case Study of Guangxi, China

Article References: Shen, W., Qiao, C., Zhu, C., Lu, F., Tang, G., Zheng, X., Tao, C., & Zhang, W. (2026). Multivariate Joint Distribution Characteristics of Meteorological Drought Based on the Meteorological Composite Index (MCI): A Case Study of Guangxi, China. Theoretical and Applied Climatology, 157(10), Article 625. https://doi.org/10.1007/s00704-026-06562-8

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06562-8

Keywords: meteorological drought, MCI, Copula functions, Guangxi, drought severity, drought duration, return period, run theory, multivariate analysis, drought risk assessment, climate variability, China