The Queimadas program from the Instituto Nacional de Pesquisas Espaciais (INPE, Brazilian National Institute of Space Research; INPE, 2021) showed a total of 3,019 active fire spots in the state of SP during 2018. Even though this number is lower than the average for the period 1999–2020, July/2018 registered the highest record since the beginning of their monitoring (84% above the climatological mean). Several articles from local newspapers pointed out fires in different municipalities in SP state, mainly depicting the number of fire spots, the spread of fires over populated areas, damages in forests—threatening several animal species—and crops (especially sugarcane fields) and the poor air quality due to the fire smoke, leading to low visibility and respiratory diseases (INPE, 2021).

Despite all damages registered by the media, only a few civil defense municipalities officially reported the disasters to the federal government; following the Sistema Integrado de Informações sobre Desastres (S2iD, Integrated Disaster Information System; S2ID, 2021), in August/2018 the municipality of Taciba registered the drought as a disaster, where the whole population were somehow affected and the private agricultural sector had an expressive monetary loss (S2ID, 2021). In addition, the municipality of Rio Claro also reported disasters related to fires in May, July and August. The environmental company from SP state (Companhia Ambiental do Estado de São Paulo; CETESB, 2019) described the period between May and September as the most adverse for dispersion of primary pollutants in the SP state and that the 2018 winter was considered the most adverse when compared with the last 3 years. In the Metropolitan Region of São Paulo, most of the automatic stations registered a slight increase (compared to 2017) in the mean concentrations of the fine particulate matter 2.5 (PM2.5), which is extremely harmful to the respiratory system (Arbex et al., 2012).

With dry conditions persisting from 2017 to 2018, SP and southeastern Brazil faced a severe water crisis. The water shortage in reservoirs built up from the previous hydrological drought event of 2013–2014.

Figure 1 shows the useful volume (percentage of the water volume between the minimum pool level and the full capacity of the reservoir) stored in the Southeast/Midwest water subsystem, that includes the state of SP, in the last two decades. The drought of 2000–2001 lowered the subsystem's level drastically (Jardini et al., 2002; Filgueiras and Silva, 2003), but the rainy period of 2005–2014 brought the reservoirs back to their normal state. Severe dry conditions returned in 2014, decreasing the stored volume to 15.8% by the end of that year. The clear contrast between reservoir volumes in the first and second decades of the 2000's may be in part linked to the interannual variability of the AAO; this discussion is presented in Section Discussions and Conclusions. Rigid water use control, gradual return of rain in 2016 and near average rainfall in 2017 were not sufficient to bring the useful volume of the subsystem back to the levels it had during most of the previous decade (Figure 1). Thus, it reached January/2018 with 31.1%, a very low volume for the peak of austral summer. After the dry rainy season, the level was at 34.2% in July/2018—the lowest level for July in 17 years (ONS, 2022).

Though reaching such low reservoir levels, the 2018 water scarcity was not as harmful to the general population as during the 2013–2015 drought event. It was because the state government created a program of rational water use and adapted the water system by connecting reservoirs (Braga and Kelman, 2020). These actions highlight the importance of governmental mitigation plans during water shortage periods.

Monthly mean precipitation for the period 1979–2020 was obtained from the Climate Research Unit (CRU) TS4.05 dataset (Harris et al., 2020) with horizontal resolution of 0.5° latitude and longitude. Air temperature and horizontal wind components at pressure levels, 500-hPa geopotential height, mean sea level pressure (MSLP) and outgoing longwave radiation (OLR) were obtained from NCEP/NCAR monthly datasets (Liebmann and Smith, 1996; Kanamitsu et al., 2002) with horizontal resolution of 2.5° x 2.5°. Sea surface temperature (SST) anomalies were calculated using the 5th version of the NOAA Extended Reconstructed SST dataset (ERSSTv5; Huang et al., 2017).

For SPI and most of the anomalies computed throughout this work, the climatological reference period used is 1981–2010, as recommended by WMO - World Meteorological Organization (2017). Different reference periods (due to the use of available products or methodology needings) are informed and, when suitable, justified.

Regarding the climate indices, some of them are calculated using climatological periods that do not match the 1981–2010 period. It occurred because diverse datasets were used in this study, to acknowledge climate data publicly and readily available to the community. It is a way to show that these derived products can be assembled to help describe the climate state during drought events, as long as all datasets are reliable and all climatologies are drawn from approximately similar periods.

Figure 2 shows the SP state location in Brazil and South America, along with the area used for SPI calculations in this study, indicated by a red rectangle (53.2°W-44°W and 25.2°S-19.8°S). SPI was calculated from the areal average of mean monthly precipitation to represent droughts in a broad area, avoiding localized precipitation anomalies resulting from smaller scale processes. Correlation between SPI obtained from areal average and from weather stations is moderate to high, as shown by Gozzo et al. (2019), indicating that SPI areal average may be a good representation of the drought conditions over diverse locations in SP.

To compute SPI it is necessary to accumulate precipitation over monthly periods (1, 2, 3 months and so on) for a sufficiently long time (discussions on the “sufficient” length can be found in Wu et al., 2005). A statistical cumulative distribution function (CDF) is fitted to the time series of accumulated precipitation for each investigated monthly period. This CDF is then used to transform the accumulated precipitation values into standardized values (the SPI values) using a normal distribution having zero mean and unit variance by matching the percentiles of the two distributions. For this study, the CDF of the Gamma distribution and the 1981–2010 period are used for computing the SPI values for different timescales (SPI-1, SPI-6 and SPI-12 for accumulations over 1, 6, and 12 months, respectively). Further information on the SPI calculation, including equations, references and a more detailed description of the process' steps can be found in Lloyd-Hughes and Saunders (2002). For timescales larger than 1 month, the SPI month corresponds to the last month of the accumulated period (for example, the July SPI-6 represents the rainfall deviation for the February-March-April-May-June-July period).

SPI is calculated from the time series of mean average precipitation inside the red rectangle in Figure 3A, and from this value the dry condition is categorized in mild, moderate, severe and extreme, according to Table 1 from McKee et al. (1993) and WMO - World Meteorological Organization (2012). The intensity of the dry period is defined as the lowest SPI value registered during the event (Spinoni et al., 2014).

Blocking/cyclones/cold fronts frequency anomalies, intraseasonal/SST/stream function anomalies and wave activity flux (WAF) were calculated to discuss atmospheric patterns linked to the 2018 dry conditions identified via SPI.

Atmospheric blocking events are identified using the Tibaldi and Molteni (1990) index, adapted to the Southern Hemisphere (Tibaldi et al., 1994), based on the difference in 500-hPa geopotential height between two latitudes persisting for at least 3 days. Anomalies are presented as the percentage difference of blocked days from analyzed months with respect to the long term monthly climatology.

Extratropical cyclones anomalies (the difference in the number of cyclones by square degrees with respect to the climatology) are obtained using a tracking numerical scheme based on Murray and Simmonds (1991). Firstly, the cyclones are identified in the MSLP field from NCEP reanalysis with 6 h-frequency. Then, the climatology and anomalies are computed. Details on the cyclone density track product presented in this study can be found in Reboita and Marrafon (2021).

Cold fronts counting is performed using an objective criteria based on temperature decrease, MSLP increase and southward to northward shift of meridional wind. In each grid point, cold fronts are identified when these three changes occur simultaneously from 1 day to the next one (Pampuch and Ambrizzi, 2016).

The recursive Butterworth filter (Roberts and Roberts, 1978) with a cutoff period of 20–90 days is applied to daily OLR data in order to evaluate the intraseasonal signal of convection over SP. It is commonly employed in climate and hydro-climate analyses (Pichard et al., 2017; Afargan-Gerstman and Domeisen, 2020; Bhattacharya and Coats, 2020). To compute the anomalies, this filter needs to calculate its climatology using a period that contains the studied years—in our case, 2018—so it could not use the 1981–2010 period. Thus, the period 1981–2019 was used as the climate normal for our analyses.

Atmospheric teleconnections are assessed by stream function fields at upper and lower troposphere (200 and 850 hPa), calculated through a Poisson equation using the horizontal wind (Hawkins and Rosenthal, 1965). This quantity represents the rotational component of horizontal wind, with negative (positive) anomalies associated with anticyclonic (cyclonic) circulation anomalies in the Southern Hemisphere. With these maps, it is possible to identify patterns resembling the Pacific-South America—PSA Rossby wave trains (Mo and Paegle, 2001; O'Kane et al., 2017). The circulation patterns associated with these teleconnections are presented in Supplementary Figure 1, to facilitate interpretations in Section Results. To infer where stationary Rossby wave trains are emitted and/or absorbed, the upper-level wave activity flux (WAF) divergence field is calculated following Takaya and Nakamura (2001). Divergence/convergence of WAF vectors shows regions of Rossby wave emission/absorption. Table 2 shows a summary of atmospheric circulation analyses, with the used variables and climatological reference periods, and references to the computation methodology.

The relationship between dry conditions in SP and large-scale climate patterns was assessed through the use of climate indices. The Oceanic Niño Index (ONI), Southern Oscillation Index (SOI), AAO, PDO and AMO indices were obtained from the National Oceanic and Atmospheric Administration (NOAA), where further information about their computation can be found. The South Atlantic Subtropical Dipole Index (SASDI) was calculated following Morioka et al. (2011) and the PSA index was calculated as described in Souza and Reboita (2021). Table 3 shows the data and reference period used for the indices computation, and links to time series and detailed information on their methodology.

Characteristics of the 2018 drought over SP and its contextualization within the last decades are determined using precipitation data and the SPI time series on 1, 6, and 12 months timescales (denoted as SPI-1, SPI-6, SPI-12), suited to represent meteorological, agricultural and hydrological droughts, respectively (WMO - World Meteorological Organization, 2012).

The boxplot of monthly precipitation (Figure 3A) shows that accumulation for February 2018 (131.6 mm) was considerably lower than the climatological 1981–2010 mean (194.6 mm) and below the 25th percentile of the distribution. This also occurred in April, May and June 2018 (with accumulations of 42.4, 36.5, and 24.5 mm, respectively), and December 2018 (174.9 mm, and climatological mean of 219.8 mm). By the thresholds of SPI-1 (upper panel of Figure 3B), dry conditions were moderate in February, May and December (SPI-1 = −1.42, −1.46, and −1.33, respectively) and extreme in April (SPI-1 = −2.06). Given that the driest months were in summer (February and December) and mainly in autumn (April and May), this year had a quite different development in comparison to 2014, 2015, and 2017, when dry conditions were more severe during summer, with subsequent months near normal (Coelho et al., 2016b). In 2018, a not so dry rainy season was followed by a very dry autumn. This study will focus on these dry conditions from February to July, with the moderately wet break in March (Figure 3B). This break in dry conditions is an interesting feature of this 6-month drought, as it was not originated from just few sparse strong rain events, but from a month with persistent and widespread precipitation in central and eastern SP (as illustrated by Supplementary Figure 2)—or in other words, a month where climate drivers must have been somewhat different from the rest of the period.

We will not address conditions between November/2018-January/2019, since they were already discussed by do Nascimento Silva et al. (2020)—readers interested in this period are referred to this work.

Calculations of the spatial distribution of SPI-1 illustrate how the drought signal changes over the region along the period. February drought was more spatially localized, limited to the eastern portion of SP (Figure 4A). In March, central and eastern SP had above-average precipitation (SPI > 0), and the dry signal was restricted to the northwestern portion of the state (Figure 4B). Dry conditions intensified and spread spatially in April, when severe conditions occurred over the central and eastern SP and extreme drought was registered in western SP, along with northern Paraná, southern Mato Grosso do Sul and most of Paraguai (Figure 4C). Drought ceased to affect Bolivia and Paraguay in May, at the time when conditions also weakened in SP (Figure 4D). In June (Figure 4E), mild/moderate conditions prevailed over SP, while drought intensified over northern Argentina. In July, most of SP experienced mild conditions, with the westernmost area having a signal of extreme drought, linked to a drier signal affecting the western Brazil and eastern Bolivia (Figure 4F).

Persistent monthly dry conditions shown in Figure 4 resulted in an extreme drought in the February-July period (July SPI-6 = −2.15), the most intense 6-month drought registered in SP since 1984.

The remarkable SP dry conditions registered in the first semester of 2018 arose from a combination of several climate forcings at different temporal and spatial scales, such as the teleconnections in Table 4 and intraseasonal oscillations. This section describes the intraseasonal and synoptic forcings with special attention to their differences from February (moderate dry) to March (mild wet) to April (extreme dry). These months illustrate how climate patterns act to rapidly vary drought intensity and spatial distribution.

The 2018 drought climate drivers at intraseasonal timescales, discussed in the previous subsection, are superimposed on an atmospheric circulation variability of lower frequency. This section briefly addresses the conditions of PDO and AMO, two low-frequency climate oscillations known to influence the SP rainfall climate. Their impact occurs by changing the magnitude of horizontal moisture transport from equatorial regions toward southeastern South America (e.g., Silva et al., 2011; Jones and Carvalho, 2018; Reboita et al., 2021). But more importantly, these teleconnections act to enhance downward vertical motion over SP.

Kayano et al. (2019) show that the aforementioned teleconnection also occurs via a PSA wavetrain: during autumn, cold PDO and warm AMO (CPDO/WAMO) conditions favor a wavetrain pattern that tends to intensify the SASH. PDO and AMO signals were relatively weak during 2018 (neither reached the 1 standard-deviation magnitude in the considered months—Table 4), but they were still in their cold and warm phases, respectively. Indeed, in April/2018, the barotropic anticyclonic center of the PSA wavetrain over South Atlantic strengthened the SASH (Figure 5C) and displaced it toward the continent (Supplementary Figure 4), reinforcing subsidence in the state of SP. Besides PSA, CPDO/WAMO also enhances downward motion over SP by modulating the Hadley circulation cell (Kayano et al., 2019). In autumn, it promotes a more vigorous upward motion in the ITCZ region and reinforces anomalous subsidence in the subtropics (as seen in April and May—Figures 8C,D).