Modern anthropogenic drought in Central Brazil unprecedented during last 700 years

The study addresses whether the recent drying and hydrologic deficits in central-eastern Brazil are driven primarily by anthropogenic forcing or can be explained by natural variability and forcings. Since the 1970s, observations and models indicate widespread drying over land, with major societal impacts on agriculture, hydropower, and water supply. Central Brazil, with a strong wet-dry seasonal cycle and intensive agriculture, has experienced increased aridity strongly correlated with rising surface temperatures, raising concerns about the role of human-induced warming in altering hydrologic balance and future drought risk. Hydrologic drought is defined as periods when evapotranspiration exceeds effective precipitation (P−PET). At local scales, higher temperatures increase PET and decrease soil moisture; at larger scales, ocean SSTs modulate precipitation on interannual to multidecadal timescales. Reconstructing pre-instrumental hydroclimate is essential to quantify natural variability and external forcing contributions. Speleothems from ventilated cave settings can record changes in evaporative demand and precipitation via isotopic and trace-element proxies. The study aims to place recent hydrologic changes in long-term context and to detect and attribute the drivers of the observed drying, particularly assessing anthropogenic greenhouse gas contributions.

Prior work has documented global and regional drying trends and increased drought risk under warming, including impacts on Brazil’s water cycle and hydropower (e.g., Milly et al. 2005; Dai 2012; Ault 2020; Chagas et al. 2022; Cuartas et al. 2022). Drought indices sensitive to PET (e.g., SPEI) highlight the role of temperature-driven evaporative demand (Vicente-Serrano et al. 2010; Thornthwaite 1948). Hydroclimate variability in South America is influenced by tropical ocean SST patterns and atmospheric circulation (Seager et al. 2010; Kayano & Andreoli 2004; Jorgetti et al. 2014). Speleothems provide high-resolution archives with accurate chronologies; in ventilated caves, kinetic and PCP effects link δ18O, δ13C, and trace elements (Mg/Ca, Sr/Ca, Ba/Ca) to evaporative conditions and recharge (Fairchild et al. 2006, 2009; Deininger et al. 2012; Wong et al. 2011). Tropical speleothem δ18O has been widely used to reconstruct monsoon variability (Vuille et al. 2012), but ventilated cave settings can capture local evaporative demand. Previous calibrations in Brazil show relationships between rainfall and speleothem proxies (Moquet et al. 2016). Detection and attribution frameworks using CMIP6/DAMIP enable partitioning anthropogenic vs natural forcing contributions (Gillett et al. 2013, 2016; Hegerl & Zwiers 2011).

Study region: Central-eastern Brazil, northern Minas Gerais, a tropical savannah climate with ~950 mm annual precipitation, strong seasonality (wet Oct–Apr; dry May–Sep; dry season ~5 months with only ~3% of annual precipitation). Study cave: Onça Cave (15°05′19.86″S, 44°14′41.39″W) in the Peruaçu karst valley, a well-ventilated entrance room with large seasonal humidity (∼50–100%) and temperature (17–25 °C) variations reflecting outside conditions. Land-use changes near the cave are minimal due to protected karst terrain. Instrumental datasets: - Precipitation: monthly station data from INMET and ANA within ~4°×4° domain centered on the cave; compared against GPCC v2018 1° gridded data. - Temperature: gridded ground-based station data and CRU TS 4.06 for 13°–17°S, 42.5°–46.5°W. - PET: FAO-56 Penman–Monteith using BR-DWGD fields; potential evaporation from Piche evaporimeters at local stations to characterize atmospheric evaporative demand. - Hydrologic balance: P−PET index computed from regional precipitation and PET. - Streamflow: z-scored discharge from 13 ANA stations in the 4°×4° domain (criteria: ≥25 years, <5% missing, start before 1990, end after 2010), median aggregated by hydrologic year; São Francisco main stem excluded due to regulation. Change-point detection (Killick et al. 2012) applied to streamflow to identify trend regimes; linear trends assessed by F-tests. Proxy records: Two stalagmites from Onça Cave entrance room: Onça2 (1298–2016 CE combined coverage; annually laminated; chronology by layer counting validated with 19 U–Th ages spanning 1760±4 to 2015±3 CE using StalAge Monte Carlo age modeling) and Onça4 (22 U–Th ages spanning 1298±2 to 1852±2 CE; StalAge age model). High-resolution sampling along growth axes yielded near-annual resolution δ18O and δ13C (Finnigan Delta Plus, VPDB) and continuous trace-element ratios Mg/Ca, Sr/Ca, Ba/Ca via LA-ICP-MS/CRDS (Resonetics M-50 ablation, iCap-Q ICP-MS; NIST SRM 612 calibration). Composites and age-uncertainty reduction for overlapping intervals used the iscam algorithm (Fohlmeister 2012). Cave monitoring and calibrations: - Drip-water and rainfall isotopes: δ18O and δD measured at USP (PICARRO L2130i; precision 0.09‰ and 0.9‰) for five drip sites (monthly, Feb 2018–Nov 2019) and local rainfall (Januária-INMET, weekly sampling 2011–2017; monthly amount-weighted aggregation). Drip-water isotopes compared to local meteoric water line to diagnose evaporation (drip-water line slope ~5.4 vs rainfall LMWL slope ~7.4). - Calcite farming: watch-glass collectors at five drip sites (monthly, Jul 2018–Oct 2019) to assess seasonal effects of temperature and RH on calcite deposition and Mg/Ca; elemental analyses via ICP-OES at University of Campinas. Cave T and RH monitored in situ and compared with nearby station data. Statistical analyses: - Proxy–instrument relationships: z-scored time series (typically 1915–2016; extended to 1942–2016 using Thornthwaite PET with monthly coefficient corrections per Aschonitis 2022 to approximate Penman–Monteith) regressed via ordinary least squares with 5-year running means; change-point detection identified onset of common proxy trends (~1942). Scaling factors (β) and significance assessed by F-tests and t-tests. - Detection and Attribution (D&A): Compared z-scored, 5-year smoothed proxy series (Onça2 Mg/Ca and δ18O) with PET and P−PET derived from DAMIP/CMIP6 ensemble simulations over the 4°×4° region for 1900–2015 across three experiments: GHG+NAT (historical with natural + anthropogenic forcing), NAT (natural-only: solar, volcanic), and GHG (well-mixed greenhouse gases only). PET computed from annual mean temperature using Thornthwaite with monthly correction to approximate Penman–Monteith. Multi-model ensemble (8 models) medians used; regression scaling factors estimated with 20 Monte Carlo minimax regressions to account for temporal dependence, with significance testing. - Additional analyses: wavelet analysis of proxy series to characterize periodicities; comparison with PHYDA Common Era temperature reconstructions (12–16°S; 41–47°W), which correlate with local observations (r=0.73, p<0.05).

• Instrumental hydroclimate trends (1979–2016): Streamflow shows a strong negative trend (z-score slope −0.054 ± 0.007; ~20% per decade decrease in m³/s). Precipitation declines modestly (slope −0.039 ± 0.013; ~7% per decade or ~70 mm/decade). Hydrologic balance P−PET mirrors streamflow (slope −0.052 ± 0.011), with an ~18% decline (~125 mm/decade), indicating a major role of increasing evaporative demand (PET) driven by warming. All regressions are significant at P<0.001.
• Cave-isotope evidence of evaporation: Drip-water δD–δ18O defines an evaporation line with slope ~5.4, lower than the local meteoric water line slope ~7.4, confirming in-cave evaporation affects drip waters. Despite ~2 °C warming since 1970, speleothem δ18O increased by 0.7‰ (trend 0.015 ± 0.002‰/yr), far exceeding expected equilibrium temperature effects (−0.2‰/°C) and local rainfall-isotope trends (~0.007 ± 0.003‰/yr), implicating kinetic fractionation under enhanced evaporation. δ13C and trace elements (Mg/Ca, Sr/Ca, Ba/Ca) co-increase since the 1970s, consistent with intensified Prior Calcite Precipitation (PCP) under reduced recharge and lower cave RH.
• Proxy–instrument scaling: Z-scored regressions indicate δ18O, δ13C, and trace-element ratios share coherent multidecadal trends with evaporation/PET and temperature; approximately 70% of the increase in geochemical ratios can be attributed to increased evaporative demand alone (P−PET or evaporation). Regression coefficients are significant at P<0.001.
• Detection and Attribution: Scaling factors relating speleothem Mg/Ca and δ18O to PET and P−PET derived from CMIP6 DAMIP simulations are significantly different from zero for GHG and GHG+NAT experiments, but not for NAT-only runs. This indicates the observed proxy trends and increased aridity require anthropogenic forcing and cannot be explained by natural forcings alone.
• Long-term context: The Onça2/Onça4 composite shows that post-1970 aridity (high δ18O, δ13C, and Mg/Ca) is unprecedented over the last ~720 years. Pre-20th century variability is characterized by short (~10-year) excursions and multidecadal (~60-year) variability (1300–1800 CE), whereas the late-20th-century drying reflects a persistent anthropogenically forced trend.
• Implications: Central-eastern Brazil faces increased risk of long-term hydrologic drought driven by warming-enhanced PET and reduced precipitation, threatening savannah biomes, agriculture, and water resources. Future projections suggest further exacerbation with continued GHG emissions.

The study demonstrates that recent hydrologic drought in central-eastern Brazil is primarily driven by increased evaporative demand associated with anthropogenic warming, compounded by a negative precipitation trend. Instrumental data show streamflow declines outpace precipitation decreases, while P−PET trends closely track streamflow reductions, highlighting PET’s role. Speleothem proxies from a ventilated cave entrance, sensitive to evaporative conditions and recharge via δ18O, δ13C, and Mg/Ca, record pronounced positive excursions since the 1970s, consistent with increased evaporation, lower cave RH, and enhanced PCP during reduced recharge. The isotopic evidence (evaporation line slope ~5.4) and the magnitude of δ18O enrichment beyond rainfall-isotope trends point to kinetic evaporative effects rather than equilibrium temperature fractionation alone. Detection and attribution against CMIP6 DAMIP experiments confirms that anthropogenic forcing is required to reproduce the observed PET and P−PET-linked proxy trends; natural-only forcings fail the detection tests. In long-term context, the modern drying is unique over 720 years, differing from earlier multidecadal variability, reinforcing that recent aridity is not a manifestation of natural variability. These findings underscore the central role of temperature-driven PET in tropical-subtropical hydrologic balance and highlight escalating drought risk for eastern South America under continued greenhouse gas emissions.

The study provides an annually resolved, multi-proxy speleothem reconstruction from Onça Cave demonstrating that post-1970 aridification in central-eastern Brazil is unprecedented over the last ~720 years. Instrumental analyses reveal streamflow declines largely attributable to increased PET, and detection–attribution using CMIP6 DAMIP shows anthropogenic forcing is necessary to explain observed trends. The combined isotopic and trace-element proxies robustly indicate enhanced evaporative demand, reduced recharge, and persistent drying. These results imply heightened future drought risk in savannah biomes, threatening agriculture, ecosystems, and water resources, particularly given projected increases in temperature and rainfall seasonality. Future work should: expand ventilated-cave proxy networks regionally to improve spatial coverage; develop higher-resolution temperature reconstructions for eastern South America to refine pre-20th-century comparisons; better quantify land-use contributions to hydrologic changes; and integrate improved regional climate modeling and downscaling to capture precipitation processes and PET under varying emission scenarios.

• Potential confounding from land-use change: While the cave surroundings are protected, regional streamflow declines since ~2000 may include contributions from intensive agriculture and water management, which are difficult to fully separate from climatic drivers. - Precipitation modeling uncertainty: CMIP6 models exhibit large spread and biases in regional precipitation, affecting P−PET uncertainties in D&A, though PET/temperature signals are more robust. - Instrumental and station coverage: Limited fluviometric station density and records prior to 1970 constrain earlier hydrologic trend assessments. - Lack of long, precisely dated instrumental temperature series for eastern South America hampers direct pre-20th-century comparisons; reliance on PHYDA reconstructions mitigates but does not eliminate this issue. - Ventilated cave kinetics: Proxies are influenced by kinetic fractionation and PCP; while leveraged here to track evaporative demand, these processes can complicate quantitative partitioning of precipitation vs PET without extensive monitoring and calibration.