Interannual and Intraseasonal Effects of Drought and Heatwaves on Expanding Soybean Production Regions in Brazil

Abstract

Climate extremes are major constraints on agricultural productivity, especially in tropical regions experiencing rapid expansion and intensification of soybean agriculture. This study analyzes spatiotemporal changes in soybean yields in response to droughts and heatwaves across highly productive municipalities in Brazil’s five macroregions from 1989 to 2020. By combining high-resolution meteorological data, satellite-based evapotranspiration estimates, and municipal-level crop yield data, we used standardized drought indices (Standardized Precipitation Index [SPI], Standardized Precipitation Evapotranspiration Index [SPEI]) and a heat index (Warm Spell Duration Index [WSDI]) with spatiotemporal linear regression analyses to explore the links between climate variability and soybean yields across Brazil’s diverse agroclimatic zones. The results show a clear rise in the frequency and severity of compound drought–heat events, especially in the Northeast and South frontiers, where yield sensitivity to hydroclimatic stress is highest. Municipal-level linear regression analyses and spatial patterns indicate that short-term dry events, rather than long-term climate trends, are the main drivers of recent yield variability, with notable spatial spillover effects observed across municipalities. Cristalina and Bom Jesus, for example, exhibit significant negative trends (p < 0.05) in both SPEI-6 (−0.04 and −0.03) and SPI-6 (0.04 and −0.03), indicating a consistent drying tendency over time. Over the 30-year period, municipalities accumulated total soybean yield losses of 3292.3 thousand tonnes (kt), corresponding to an average reduction of 3.7% relative to 5-year detrended yield. These findings highlight the increasing vulnerability of rainfed agriculture in Brazil and emphasize the critical role of seasonal timing, crop phenology, and regional climate patterns for effective climate risk management. This study provides empirical evidence linking combined extremes to agricultural performance and presents a scalable framework for early warning systems and for climate-resilient policy development.

1. Introduction

One of the most significant challenges of the 21st century is ensuring food security for a global population projected to exceed 10 billion people by 2050 [1]. As one of the most important crops worldwide, soybean (Glycine max) plays a critical role in addressing this challenge by serving as a major source of protein and oil [2,3]. However, soybean production is increasingly threatened by climate variability and the rising frequency, intensity, and duration of extreme hydrometeorological and climate events (EHCEs) [4]. Among these, drought and heatwave events (DHE) are significant constraints, particularly for tropical rainfed agricultural systems that are highly sensitive to climatic anomalies [5,6].

This challenge is especially relevant to Brazil, which has emerged as the world’s largest soybean producer, harvesting 169 million tonnes (~55% of global exports) in the 2024/25 season [7,8]. Although cultivated across a wide latitudinal range, from subtropical southern states to equatorial regions, Brazilian soybean yields remain highly vulnerable to climate extremes, particularly droughts [9,10,11]. Such conditions cause substantial interannual yield variability [7]. According to [12], in Southern Brazil, recurrent climatic anomalies have resulted in exceptional losses in grain production. Consecutive cropping seasons with diminished yields, particularly in 2020, 2022, and 2023, led to unharvested grains valued at approximately 8.2 billion U.S. dollars (USD), approximately 45 billion Brazilian reais (BRL).

The intensity and duration of droughts are rising amid increasing climate variability, underscoring the need for refined tools to monitor spatiotemporal patterns [13,14]. Process-based crop models have been widely used to simulate yield responses [15,16,17]. Nevertheless, their reliance on site-specific inputs (soil properties, cultivar traits, and management practices) often introduces calibration biases. More critically, their simplified treatment of extreme events tends to underestimate both the magnitude of climate risks and their spatial heterogeneity [18,19].

Among the available approaches to estimating drought, the Standardized Precipitation Index (SPI) quantifies precipitation over a given timescale as a standardized deviation from the long-term mean, and the Standardized Precipitation Evapotranspiration Index (SPEI) incorporates potential evapotranspiration alongside precipitation to assess available water resources. These indices provide valuable tools for characterizing drought [20,21]. They have been widely applied worldwide to assess drought trends, variability and impacts, including applications across Brazil [22,23,24].

In addition to water deficit indicators, the Warm Spell Duration Index (WSDI) offers a complementary perspective by capturing the frequency and duration of heat events, making it a valuable tool for detecting trends in heat waves and their intensification over time [25]. The WSDI is designed to measure prolonged periods of unusually warm days relative to baseline climatology. It quantifies the number of days in a year that are part of warm spells lasting at least six consecutive days, with daily maximum temperatures exceeding the 95th percentile of a reference period. Although brief, these events can overlap with droughts and occur during critical stages of plant growth, causing damage to yield and other crop phenotypes. Together, SPI, SPEI, and WSDI provide bias-free insights into the frequency, severity, and spatial extent of drought and heatwave events across multiple timescales.

This study addresses the following question: How do droughts and heatwaves affect soybean yields across Brazil’s soybean production regions? The objectives are: (a) integrate temporally variable and spatially distributed climate and soybean yield data at the municipal scale across Brazil’s soybean-producing areas; (b) identify DHE in Brazil’s soybean landscapes using relevant climate indices; and (c) quantify the spatiotemporal contributions of DHE to soybean yield losses, analyzing potential drivers and underlying causes in each municipality. This study focuses on Brazil’s key soybean-producing regions, combining high-resolution climate data with SPI, SPEI, and WSDI analyses to characterize regional patterns of drought and heatwaves.

2. Materials and Methods

2.1. Study Area

A wide range of climatic conditions is observed across Brazil’s extensive territory (Figure 1a). It spans from the subtropical southern states within the Pampa and Atlantic biomes, characterized by Cfa and Cfb Köppen–Geiger climate classes, to the low-latitude regions of the Cerrado biome, the country’s main agricultural frontier. In these northern regions, soybean “tropicalization” has enabled its expansion into areas with Am and Aw climate classes [26].

In this study, nine soybean-producing municipalities were selected for analyses of drought and heat indices (Figure 1 and Table 1). Each municipality represents a homogeneous climatic zone (CZ) as defined by [27,28]. This workflow is based on the climate and agricultural representativeness of characteristic soybean production regions at the municipal level across Brazil. Additionally, the workflow and associated analytics can be expanded to include new crops and their corresponding municipalities.

To ensure consistency with the period when the crop is biologically active and water use is agronomically relevant, we used only observations occurring between the local planting start and harvest end dates, as in [29]. The soybean growing season for each municipality was defined using the official crop calendar published by [30], which provides region-specific information on planting, vegetative development, and harvest periods. For each location, the planting window, the onset of vegetative growth, and harvest dates were converted into daily time series, allowing all climate and yield observations to be filtered to the months encompassed by the respective growing season.

Figure 1. Brazil’s soybean region and studied cities (black stars). (a) Brazil’s climatic zones (Adapted from [27,28]). (b) Spatial distribution of soybean production during the 2022/2023 crop season [30]. CZ: Climatic Zone [27,28]; RS: Rio Grande do Sul; PR: Paraná; GO: Goiás; MT: Mato Grosso; BA: Bahia; TO: Tocantins; PI: Piauí; MA: Maranhão.

Figure 1. Brazil’s soybean region and studied cities (black stars). (a) Brazil’s climatic zones (Adapted from [27,28]). (b) Spatial distribution of soybean production during the 2022/2023 crop season [30]. CZ: Climatic Zone [27,28]; RS: Rio Grande do Sul; PR: Paraná; GO: Goiás; MT: Mato Grosso; BA: Bahia; TO: Tocantins; PI: Piauí; MA: Maranhão.

Table 1. Characteristics of municipal host sites representing each climatic zone.

CZ	Municipality	State	Planting Date Window	Soil Profile		Latitude	Longitude	Elevation (m)
6801	Palmeira das Missões	RS	17-September	31-December	Oxisols	−27.92	−53.32	614
7801	Cascavel	PR	8-September	31-December	Ultisols	−24.88	−53.55	784
7601	Cristalina	GO	27-September	31-December	Oxisols	−16.79	−47.61	1211
7701	Primavera do Leste	MT	27-September	31-December	Ultisols	−15.58	−54.38	680
8701	Sorriso	MT	30-September	25-December	Oxisols	−12.56	−55.72	379
8401	Barreiras	BA	17-October	31-January	Entisols	−12.12	−45.03	474
9301	Bom Jesus	PI	6-November	9-February	Entisols	−9.08	−44.33	288
9401	Balsas	MA	17-October	20-January	Entisols	−7.46	−46.03	271
9701	Lagoa da Confusão	TO	8-October	1-March	Inceptisols	−10.83	−49.85	178

CZ: Climatic Zone [27,28]. Planting date window = typical soybean sowing period. Soil Profile = dominant soil orders in each municipality [31]. Latitude, Longitude, Elevation = geographic co-ordinates and altitude. RS: Rio Grande do Sul; PR: Paraná; GO: Goiás; MT: Mato Grosso; BA: Bahia; PI: Piauí; MA: Maranhão; TO: Tocantins.

Table 1. Characteristics of municipal host sites representing each climatic zone.

CZ	Municipality	State	Planting Date Window	Soil Profile		Latitude	Longitude	Elevation (m)
6801	Palmeira das Missões	RS	17-September	31-December	Oxisols	−27.92	−53.32	614
7801	Cascavel	PR	8-September	31-December	Ultisols	−24.88	−53.55	784
7601	Cristalina	GO	27-September	31-December	Oxisols	−16.79	−47.61	1211
7701	Primavera do Leste	MT	27-September	31-December	Ultisols	−15.58	−54.38	680
8701	Sorriso	MT	30-September	25-December	Oxisols	−12.56	−55.72	379
8401	Barreiras	BA	17-October	31-January	Entisols	−12.12	−45.03	474
9301	Bom Jesus	PI	6-November	9-February	Entisols	−9.08	−44.33	288
9401	Balsas	MA	17-October	20-January	Entisols	−7.46	−46.03	271
9701	Lagoa da Confusão	TO	8-October	1-March	Inceptisols	−10.83	−49.85	178

CZ: Climatic Zone [27,28]. Planting date window = typical soybean sowing period. Soil Profile = dominant soil orders in each municipality [31]. Latitude, Longitude, Elevation = geographic co-ordinates and altitude. RS: Rio Grande do Sul; PR: Paraná; GO: Goiás; MT: Mato Grosso; BA: Bahia; PI: Piauí; MA: Maranhão; TO: Tocantins.

2.2. Soybean Yield Dataset

We analyzed annual soybean yield ($Y_r$) data for selected Brazilian municipalities (Table 1) spanning the 1996/97 to 2019/20 crop seasons. For each municipality, the dataset included harvested area (ha) and observed yield ($Y_r$, t ha⁻¹), obtained from the Brazilian Institute of Geography and Statistics (IBGE) [32]. The municipality yield averages are calculated by IBGE using a methodological sample that reflects local diversity in soil types, cultivars, and planting practices [32].

The $Yr$ captures both interannual variability and long-term trends driven by technological improvements, management practices, and cultivar adoption. To isolate interannual deviations attributable to climate, $Yr$ was detrended using a five-year moving window, which effectively removes short- to medium-term trends while retaining year-to-year variation [11,33]. For periods at the start or end of the series, the window was adjusted to use all available data, with a minimum of one year. For municipality $i$ and year $t$, the detrended yield ${Ydetrended}_{,i,t}$ was computed as the centered moving average (Equation (1)):

$${Ydetrended}_{,i,t}={\displaystyle \frac{1}{5}}\sum _{k=-2}^2{Yr}_{i,t+k}^{obs}$$

(1)

where ${Yr}_{i,t+k}^{obs}$ represents the observed yield for years within the 5-year window.

The yield anomaly (${\Delta Y}_{i,t}$) quantifies the relative deviation of $Y_r$ from the $Ydetrended$ (Equation (2)):

$${\Delta Y}_{i,t}={\displaystyle \frac{{Yr}_{i,t}^{obs}-{Ydetrended}_{i,t}}{{Ydetrended}_{i,t}}}\times 100$$

(2)

Positive anomalies indicate yields above the expected trend, while negative anomalies indicate yield reductions relative to the local trend.

For each crop season, we calculated the negative yield anomalies or losses ($Y_{loss}$) individually, estimating that for the soybean area at the municipal level. It was calculated based on the $c$ and the $Y_{detrended}$ calculated by five-year moving windows, according to Equation (3):

$${Yloss}_{i,t}=\left\{\begin{array}{c}-{\displaystyle \frac{{\Delta Y}_{i,t}}{100}}\times {Ydetrended}_{i,t}\times A_{i,t}\\ 0\end{array}\right.if{\Delta Y}_{i,t}<0,otherwise$$

(3)

where $A_{i,t}$ is the soybean harvested area for each municipality $i$ in year $t$. This calculation ensures that only negative (less than zero) deviations contribute to losses, consistent with previous Brazilian soybean studies [11].

2.3. Weather Dataset

Daily precipitation ($P$), maximum temperature ($T_{max}$), minimum temperature ($T_{min}$), and reference evapotranspiration (${ET}_o$) were obtained from the Brazilian Daily Weather Gridded Data (BR-DWGD) dataset [34]. The dataset provides daily temporal resolution, a 0.1° × 0.1° spatial grid, and covers the period from 1989 to 2022 for the entire Brazilian territory. For each municipality, climate variables were extracted from the nearest grid cells based on geographic coordinates to capture spatial representativeness and local variability. To ensure consistency with crop development, only observations during the soybean growing season, defined as the period between the local planting start and harvest end dates, were retained [29]. This temporal filtering ensures that climatic variables are directly linked to periods of active crop growth and water demand.

Although the El Niño-Southern Oscillation (ENSO) was not used as a predictor in the analysis, we reference its phases for interpreting interannual variability. El Niño and La Niña events were defined using the Oceanic Niño Index (ONI), based on 3-month running-mean SST anomalies in the Niño 3.4 region that exceed ±0.5 °C [35].

2.4. Data Quality Control

To ensure the reliability and consistency of both crop and climate datasets, separate quality control procedures were applied to each type of data. For the soybean yield dataset, physically implausible or negative values were removed, and the five-year moving window detrending was applied to isolate interannual variability from long-term technological and management trends. Yield anomalies ($\Delta Y$) were then computed relative to the detrended series, and municipal-level yield losses ($Y_{loss}$) were derived considering only negative deviations, weighted by harvested area. This approach ensures that analyses of yield variability and losses focus on climate-driven effects while maintaining consistency with previous studies in Brazilian soybean production.

For the climate data from the BR-DWGD dataset, daily variables were first temporally aligned and merged into a unified municipal-level dataset. Missing daily observations were filled using nearest-neighbor spatial interpolation from adjacent grid cells, and, when necessary, temporal aggregation to daily means was applied. Anomalous or extreme values were identified using Z-score normalization relative to the local climatology, and values exceeding physically plausible thresholds were constrained to prevent distortion of subsequent analyses [36]. The standardized dataset was used to calculate drought and heat metrics, derive climate indices such as SPI, SPEI, and WSDI, and support multivariate analyses, including PCA and municipal-level regression models, ensuring comparability across municipalities and robustness of the results.

2.5. Crop Evapotranspiration (ETc)

Crop coefficient ($K_c$) values were determined using the methodology proposed by [37] in FAO Irrigation and Drainage Paper 56. The FAO 56 procedure defines $K_c$ as a function of the crop’s phenological stages, subdivided into initial, mid, and late-season phases. For each municipality, phase boundaries were aligned with the region-specific crop calendar, and fixed $K_c$ values were assigned to each phase based on the mean coefficients recommended for soybean in FAO 56 and validated in recent Brazilian studies using lysimetry and remote sensing data [29,38,39].

Daily Crop Evapotranspiration (${ET}_c$) was calculated using the standard FAO 56 formulation [37] (Equation (4)):

$${ET}_c=K_c\times {ET}_o$$

(4)

where ${ET}_o$ is the reference evapotranspiration, estimated by the BR-DWGD dataset for each municipality, and $K_c$ is the stage-dependent crop coefficient assigned to each day based on the local phenological calendar [34]. This approach aligns with experimental and modeling studies that quantify soybean water use across Brazil using daily $K_c$ curves and FAO 56 methodology [9,40]. More details of the dataset preprocessing are explained in [41] at Zenodo (https://doi.org/10.5281/zenodo.19167200).

2.6. Principal Component Analysis

Climate regionalization was performed using Principal Component Analysis (PCA) applied to municipality-level climate data. Annual precipitation and mean temperature time series were first organized into municipality-by-year matrices. Missing values were handled through row-wise mean imputation, and the combined dataset was standardized to zero mean and unit variance to ensure comparability among variables with different scales. PCA was then applied to the standardized dataset to reduce dimensionality and identify the main modes of climate variability. The first two principal components (PC1 and PC2) were retained, explaining 53.9% and 33.8% of the total variance, respectively, and together accounting for 87.7% of the overall variability. This indicates that most of the climatic information is preserved in the reduced space. The resulting PCA scores were subsequently used as input for a k-means clustering algorithm (k = 4) to classify municipalities into homogeneous climate regions based on their hydroclimatic characteristics. This approach is widely used in climate regionalization studies as it enables the identification of spatial patterns while reducing data dimensionality and redundancy [42,43].

2.7. Drought Indices

Meteorological drought was also characterized using the Standardized Precipitation Index (SPI), which is based solely on precipitation and quantifies precipitation anomalies across multiple accumulation timescales [20]. Monthly precipitation totals were aggregated over 1-, 3-, 6-, and 12-month periods to represent short- to long-term water deficits relevant to hydrometeorological and agricultural processes. For each municipality, SPI was computed independently to preserve local precipitation climatology. Accumulated precipitation series were fitted to a two-parameter Gamma distribution using maximum-likelihood estimation, following the original definition proposed by [20]. The Gamma distribution is appropriate for modeling nonnegative, right-skewed precipitation data and has been extensively validated for SPI applications across diverse climatic regimes [44]. Zero precipitation values were handled in accordance with standard SPI conventions using a mixed-probability adjustment [45]. Cumulative probabilities derived from the fitted Gamma distribution were transformed into standardized normal variates, yielding SPI values with a mean of zero and unit variance. Negative SPI values indicate drier-than-normal conditions, while positive values indicate wetter-than-normal conditions, with larger magnitudes corresponding to greater anomaly intensity. SPI calculations were implemented in Python 3.10.0 using a reference implementation consistent with the original methodology.

In addition to SPI, we used the Standardized Precipitation Evapotranspiration Index (SPEI), which extends the SPI framework by incorporating atmospheric evaporative demand through the climatic water balance [46]. The monthly climatic water balance was calculated as the difference between precipitation ($P$) and potential evapotranspiration (${ET}_o$), and then accumulated across multiple temporal scales. SPEI was also calculated at two temporal scales, 6 and 12 months, to capture both short-term agricultural impacts and longer-term hydrological variability. The 6-month scale reflects water deficits during critical crop growth stages and is widely used to assess the effects of droughts on agricultural productivity, as documented in previous studies [46,47].

For each municipality and accumulation scale, the accumulated climatic water balance was fitted separately for each calendar month using a three-parameter Pearson Type III distribution. Monthly fitting ensures stationarity of the distribution parameters and allows for consistent standardization across seasons, as recommended for drought index calculation [21,45]. The Pearson Type III distribution was selected for its ability to handle skewed water balance data with both negative and positive values, maintaining the physical interpretation of dry and wet periods. Its applicability for hydroclimatic analyses in Brazil has been confirmed in studies focusing on monthly rainfall distributions in Northeastern Brazil [48], drought trend assessments across the country [49], and probability-based standardized drought indices [50]. Cumulative probabilities de-rived from the fitted distribution were transformed into standardized normal variates, producing SPEI series with zero mean and unit variance. Negative SPEI values indicate drier-than-normal conditions, reflecting combined precipitation deficits and increased evaporative demand, while positive values indicate wetter-than-normal conditions. All SPEI calculations were executed using a custom Python implementation based on the standard SPEI methodology, adapted to employ the Pearson Type III distribution for robust handling of the full range of water balance values [21,47]. Drought conditions were classified based on SPI and SPEI values and standardized categories (

Table 2. Classification of drought based on SPI/SPEI values. Adapted from [45].

SPI/SPEI	Drought Class
>1.00	No drought/wet
1.00 to −0.49	Near normal
−0.50 to −0.99	Mild drought
−1.00 to −1.49	Moderate drought
−1.50 to −1.99	Severe drought
−2.0 and less	Extreme drought

), ranging from extreme drought (SPI/SPEI < −2.0) to non-drought or wet conditions (SPI/SPEI > 1.00).

Droughts were characterized using four key metrics derived from the SPI/SPEI time series, namely severity (cumulative SPI/SPEI during the event), duration (number of days the event persists), intensity (severity divided by duration), and peak (lowest SPI/SPEI value). These metrics provide a multidimensional understanding of drought behavior, encompassing magnitude, persistence, and extremity, which are essential for agricultural and climatological assessments [21,51].

2.8. Heatwave Index

The first step in calculating the WSDI was to determine the 90th percentile of daily maximum temperatures (Tmax) for a reference period, often using a 30-year baseline. For each day of the year, the 90th percentile (T90) was calculated from the historical temperature record (Equation (5)) [52,53,54].

$$T_{90}\left(d\right)={Percentile}_{90}\left(Tmax\right(d,referenceperiod\left)\right)$$

(5)

where T90(d) is the 90th percentile of the daily maximum temperature for day d, and $T_{max}(d,referenceperiod)$ represents the historical maximum temperatures on day d over the reference period [52].

$$WSDI={\sum }_{i=1}^{Nspells}D{urationofwarmspell}_i$$

(6)

where $N_{spells}$ is the number of warm spells in the year, and the ${Durationofwarmspell}_i$ is the number of consecutive days in that spell.

2.9. Spatial Processing

Spatial processing was carried out using Google Earth Engine (GEE) and Python. Gridded meteorological variables and evapotranspiration data were initially extracted and aggregated to the municipal level using Brazilian municipal boundaries represented as multi-polygon geometries, creating municipality-specific time series. These aggregated time series were then used to compute the standardized climate indices (SPI, SPEI, and WSDI) during preprocessing in Python 3.10.0. This workflow maintains a consistent link between gridded climate data and municipal-level yield records. Data pre-processing involved handling missing values, detecting and constraining outliers, and standardizing variables as needed [55]. The overall methodological framework, including spatial aggregation, index calculation, and event characterization, is summarized in Figure 2.

3. Results

The central thesis states that Brazil’s soybean production, the largest worldwide, has been affected by chronic and interannual drought and heatwaves, challenging its expansion beyond its current geographic limits. The implications of exposure to changing extreme events begin at the municipal level and extend across vast Brazilian biomes and agroclimatic regions, including the Cerrado, Caatinga, Pampa, Pantanal, Atlantic Forest, and Amazon. For example, the Cerrado region accounts for almost 50% of Brazil’s soybean production, and the Atlantic Forest and Amazon represent the geographic limits of soybean production in this study and the frontiers of its expansion and intensification in the years to come. The following sections provide evidence of the spatiotemporal attributions of soybean yield responses to droughts and heatwaves.

3.1. Climate Characterization

We grouped municipalities using Principal Component Analysis (PCA) based on estimated temperature and precipitation anomalies during the soybean growing season in each region to identify regional patterns in agroclimatic conditions and soybean productivity. The first two principal components (PC1 and PC2) explained 53.9% and 33.8% of the total variance, respectively, indicating that most climate variability is captured. PC1 mainly reflects a temperature gradient from cooler southern highlands to hotter northern regions, while PC2 represents a precipitation gradient separating wetter from drier areas. As shown in Figure 3, PCA revealed three climate clusters: Central-West, Northeast, and South (Primavera de Leste, MT, and Cristalina, GO, were included in the Central-West and South, respectively). These clusters align with the agroclimatic regions reported in [56,57].

Brazil’s tropical dry winter covers ~53% of MT’s and all TO’s surface in the Central-West region (Figure 1). Annual precipitation, with a mean of approximately 1600 mm, and chronic winter dryness are evidenced by the fact that 88% of the annual precipitation falls during the growing season (

Table 3. Mean annual and seasonal precipitation, effective rainfall, and crop evapotranspiration values across the municipalities.

Municipality	Mean Annual P (mm yr−1)	Mean Season P (mm yr−1)	Effective P 60% (mm yr−1) 1	Mean ETc (mm yr−1) 2	Effective P 60% vs. ETc (mm yr−1) 3
Palmeira das Missões—RS	1788.2	1245.1	747	481.9	265.1
Cascavel—PR	1746.3	1126.1	675.6	443.8	231.8
Cristalina—GO	1368.2	1136.8	682.1	400.7	281.4
Primavera do Leste—MT	1485.3	1313.9	788.4	406.0	382.4
Sorriso—MT	1664.7	1486.6	891.9	362.8	529.1
Lagoa da Confusão—TO	1581.4	1542.8	925.7	501.4	424.3
Barreiras—BA	926.6	914.7	548.8	539.5	9.3
Bom Jesus—PI	814.5	743.3	446.0	498.6	−52.6
Balsas—MA	1082.8	1058.0	634.8	465.7	169.1

¹ P denotes precipitation, and ETc denotes crop evapotranspiration. ² Effective P 60% represents effective precipitation estimated as 60 percent of total precipitation, following applications commonly adopted in Brazilian agroclimatic studies [67,68]. ³ The surplus or deficit was determined as: (P − ETc). Positive values indicate that precipitation exceeds crop water demand, while negative values indicate a water deficit relative to ETc.

). This hydroclimatic regime contributes to reduced yield variation in an agroclimatic region that also overlaps with the Cerrado, an area considered the site of the largest soybean agricultural expansion in Brazil [58]. Agricultural expansion frontiers in the region include the states of Maranhão (MA), Tocantins (TO), Piauí (PI) and Bahia (BA) (also known as MATOPIBA), accounting for approximately 73 million hectares, consistent with recent findings by [59,60]. The conspicuous expansion of agriculture in the region evidences the growing climate risks to crops due to increasing hydroclimatic instability. Yet agricultural practices may have secured soybean yields as the consistency of ETc spatiotemporal variability across Central-West Brazil illustrates (Table 3 and Figure 4). The ET changes also shed light on how cropping systems use the limiting water and energy availability to support plants’ water demands as the growing season evolves. To support this process [61] suggests that energy is a limiting factor during the early stages of the growing season when moisture is present. Then, as water vapor deficits increase in later stages, water released to the atmosphere gradually limits soil water content and, in turn, ET. While the mechanisms described above are relevant to other regions with similar rainfall patterns, Ref. [62] found that vapor pressure and solar radiation are among the most important drivers of ${ET}_o$. As noted by [63,64], high $T_{max}$ and low precipitation often coincide during the terminal stages of the crop cycle, creating conditions of elevated atmospheric demand (high ${ET}_c$), exacerbating soil moisture deficits, and increasing the risk of crop failure in non-irrigated systems [61,62]. In the Central-West and Northeast regions, ${ET}_o$ values are similar, but during long dry spells, when water becomes a limiting factor, radiation offsets the vapor pressure deficit, reducing $ET$. Thus, during wet and dry periods, chronic or interannual precipitation and temperature maxima and minima alternate as drivers of ${ET}_c$, highlighting crops’ exposure to local-to-regional hydrometeorological conditions associated with drought [65,66].

The municipalities in the Hot-Dry Northeast cluster in BA, PI, and MA are characterized by high mean annual temperatures (mean ~32 °C) and low annual precipitation (mean ~1100 mm), along with an elevated annual evapotranspiration (mean ~1050 mm), reflecting semi-arid to sub-humid conditions [69]. The municipalities within the Hot-Dry Northeast cluster, which includes regions in BA, PI, and MA, experience high mean annual temperatures of around 32 °C and low annual precipitation of approximately 1100 mm. Additionally, these areas have elevated annual evapotranspiration, averaging about 1050 mm, which indicates semi-arid to sub-humid conditions. Despite these conditions, similar intraseasonal variations in crop evapotranspiration between the Northeast and Central-West Brazil indicate their proximity to the tropics. Consequently, solar radiation levels are a major factor influencing patterns. During the growing season in the Northeast, summer precipitation accounts for 91–98% of total annual rainfall, the highest proportion observed in our study. In times of drought, summer precipitation becomes the key factor influencing agricultural outcomes in these arid areas, particularly due to the reliance on rainfed agriculture [70,71]. Further, similar intraseasonal variations between the Northeast and Central-West Brazil suggest that locations near tropical Brazil and, consequently, the associated radiation levels are the key drivers of these interdependent hydroclimate intraseasonal variations.

Overlapping with the South cluster, the Paraná is the second-largest area for soybean production in Brazil. This PCA cluster includes the states of Rio Grande do Sul (RS) and Paraná (PR), along with the Cristalina municipality, part of Brazil’s humid subtropical region, characterized by hot–temperate summers. This area enjoys favorable agroclimatic conditions for soybean cultivation. In RS and PR, summer rainfall accounts for 64–69% of the annual precipitation, which totals approximately 1400 mm. These humid subtropical conditions, coupled with summer temperatures averaging around 23.5 °C, contribute to soybean yields that often exceed 2400 kg/ha. Subject to low temperatures, frost risk in the South becomes a factor in August, when the expected warmer and wet conditions help successful sowing, the onset, and the evolution of the growing season across this area [9,56,64,72]. Thus, average precipitation and temperature magnitudes and fluctuations indicate that water deficits are not the primary constraint since the autumn–winter months and mid-spring are typically wet and have low evaporative demand (Figure 4 and Figure 5). Temperature and precipitation values in the South show limited differences compared to those in the Central-West and Northeast. However, at municipal level, differences and similarities emerge suggesting that the sensitivity of municipalities such as Cristalina in the upper lands of Goiás (GO) and Primavera do Leste and Sorriso in Mato Grosso (MT) to moisture or energy limited conditions can be due to changes in atmospheric evaporative demands and the inherent water pressure deficits driven by natural climate variability (Table 3) [70,71,72,73].

3.2. Interannual and Spatial Variability of Drought and Heatwave Events

The spatiotemporal variability of rainfall is a key determinant of rainfed soybean productivity in Brazil. The drivers of rainfall distribution over the soybean production areas suggest that a series of intertwined processes fueled by the contrasting sea surface temperatures (SST) between the Pacific and Atlantic Oceans, land surface–atmosphere interactions, and other climate-driven teleconnections, regulate the precipitation deficits and surpluses in Central-West, Northeast, and South Brazil [12,61]. Syntheses of recent research studies [74,75,76,77,78] supporting some of the variations in precipitation in soybean production regions indicate that the coupling between SSTs in the tropical Atlantic circulation patterns across the equator, which consequently drive contrasting wet and dry conditions in South America’s North and Northeast and Southeast. Furthermore, Ref. [79] indicate that contrasting SSTs now between the Atlantic and Pacific Oceans, during the warm and cold phases of the El Niño Southern Oscillation, lead to contrasting droughts, extended wet periods, and floods between northern and southern Brazil.

Analysis of rainfall and temperature event frequencies in selected municipalities within Brazil’s primary soybean-producing regions, as illustrated in Figure 6, reveals a “climate divide” between the North (MT, TO, BA, PI, and MA) and South (RS, PR, and GO). This divide corresponds to drought conditions in the North and Northeast clusters, which are closely linked to El Niño events, and to wet conditions in the Southeast that occur during the opposite phase of the ENSO, specifically La Niña. Although these interannual climate variability modes offer important insights for crop and water management, intraseasonal changes remain uncertain, particularly when soybean planting, flowering, or grain filling are impacted by reduced residual soil moisture during the austral winter or by prolonged water deficits throughout the growing season [23,24,64,74,80,81].

3.2.1. SPI and SPEI Relationship to Soybean Yield Losses

A balance between precipitation and crop water requirements underscores how hydrometeorological and climate variability shape agricultural outputs. Even slight deviations from normal seasonal rainfall can stress soil moisture, directly influencing drought indices such as SPI and SPEI. These and other drought indices capture both precipitation anomalies and atmospheric evaporative demand, integrating climatic and water availability [20,21], and helping diagnose agricultural drought conditions and assess the sensitivity of yields to climate extremes [82]. Examining the temporal variability and spatial distribution of the SPI and SPEI in relation to soybean yield losses is essential for understanding how climatic fluctuations translate into agricultural vulnerability across the study regions.

Figure 7 shows the spatial distribution of SPI-6 for each growing season, with March as the reference month, coinciding with the end of the soybean cycle across most regions. The spatial patterns indicate a persistent drought signal in the North and Northeast, especially in the Northeast, where SPI-6 values frequently fall below −1.0. As mentioned in the Section 3.2., ENSO has contrasting effects between the Northeast and Central-West regions and the South. During El Niño events (e.g., 2001–2003, 2015–2016), the North and Northeast typically experience negative precipitation anomalies, increasing drought risk, while the South generally receives above-average rainfall [83].

The cold phase (La Niña) produces the opposite effect of El Niño, resulting in reduced precipitation in the South and increased rainfall in the North and Northeast. These distinct drought drivers establish ENSO as a critical factor in risk management for southern soybean production. During El Niño years, more abundant and regular rainfall supports higher yields and mitigates the irregularity of frontal rains in the South [84]. In contrast, La Niña events significantly elevate the risk of yield losses in the South, as reduced and unpredictable precipitation may coincide with critical stages of the soybean growth cycle, intensifying water deficits and severely impacting potential yields [72]. Additionally, municipalities in the South and Southeast exhibit moderate SPI-6 variability, indicating alternating wet and dry years. Although these regions have historically experienced relatively stable rainfall patterns, there has been an increase in moderate-to-severe drought episodes since the early 2000s. This trend is indicative of regional drying and increased evaporative demand, consistent with a shift in the South Atlantic Convergence Zone (SACZ) and a reduction in convective rainfall [80,85,86].

Figure 8 and Figure 9 show the frequency distributions of drought classes derived from the 6-month SPI and SPEI, respectively, along with their relationships to observed soybean yield losses across the studied municipalities. This timescale was chosen to better capture water deficits during the soybean growing season. The analysis reveals a consistent link between severe-to-extreme drought conditions (SPI/SPEI ≤ −1.5) and significant yield reductions, especially in municipalities in the Northeast. Under these conditions, yield losses frequently exceed 20% compared to the municipal trend-adjusted mean, emphasizing the strong reliance of soybean productivity on sufficient water availability during critical growth stages.

In contrast, municipalities in the South experienced a higher frequency of near-normal and moderately wet classes (−0.5 < SPI/SPEI < 0.5), resulting in lower yield losses (<10%). This spatial differentiation highlights the gradient in drought exposure across the soybean frontier and is consistent with previous findings of northward intensification of hydroclimatic risk [83,84,85]. Southern municipalities such as Cascavel—PR and Palmeira das Missões—RS exhibited robust yield trends but also persistent drought signals and rising temperatures, demonstrating that technological advances alone are insufficient to counteract intensifying climatic extremes. This observation supports earlier conclusions by [87], which found that technology can reduce but not fully offset the impacts of increasing climate variability on crop yields.

Comparison of SPI and SPEI distributions indicates that SPEI identified a greater proportion of drought-affected months in most municipalities, especially during the 2015–2016 and 2019–2020 growing seasons. The discrepancy between indices suggests that evaporative demand, rather than precipitation alone, was the primary driver of water deficits during these years, consistent with regional warming trends and increased atmospheric aridity [86]. The stronger correspondence between SPEI-defined droughts and yield anomalies further supports the consensus that SPEI provides a more comprehensive representation of agricultural drought under current climatic conditions [21].

3.2.2. The Warm Spell Duration Index (WSDI)

The Warm Spell Duration Index (WSDI), defined as the annual count of days with at least six consecutive days exceeding the 95th percentile of daily maximum temperature, serves as a robust indicator of heat events affecting agroecosystems [25]. WSDI is essential for assessing thermal stress episodes that directly influence crop phenology, water demand, and yield potential [80,88]. Analysis indicates a statistically significant upward trend in WSDI across all surveyed municipalities from 1996 to 2020 (p < 0.05), consistent with broader regional warming trends reported for Central-West Brazil [80]. Northeast municipalities exhibited the largest increases in warm spell duration, with average annual increases exceeding one day per decade (Figure 10). These prolonged warm spells intensify drought conditions by increasing crop evapotranspiration demand and accelerating soil moisture depletion, thereby compounding the effects of heat and water deficits on soybean productivity [59,83]. The observed WSDI trends are consistent with projections of increased frequency and intensity of heatwaves in tropical agricultural zones under future climate scenarios [4].

Compound drought and heat events are increasingly responsible for yield losses, and the alignment of the Standardized Precipitation Evapotranspiration Index (SPEI) and WSDI indicates that combined stress effects exceed those of individual factors [59,89]. The 2015–2016 El Niño event exemplified this phenomenon, resulting in yield reductions exceeding 30% across municipalities in Northeast and Central-West Brazil [83,90].

3.3. Quantification of Soybean Yield Losses at the Municipal Level

Soybean yield losses at the municipal level in Brazil showed significant spatial and temporal variation over the 30 crop seasons studied. During the first decade (1989/1990–1999/2000), total yield losses reached 252.2 thousand tonnes (kt), with an average reduction of 3.0% across municipalities. The most impacted areas in this period included Sorriso (74.9 kt; 2.6%) in the Central-West and Barreiras (66.0 kt; 4.3%) in the Northeast. Although losses were mainly concentrated in the northern part of the study region, several municipalities, such as Bom Jesus and Lagoa da Confusão in the Northeast and mountainous regions in Central Brazil, did not experience drought or yield reductions during the 1990s (

Table 4. Soybean yield losses ($Y_{loss}$), expressed in thousand tonnes (kt) and as percentages (%), at the municipal level in Brazil, across crop seasons from 1989/1990 to 2020/2021.

Municipality	1989/90 to 1999/00		2000/01 to 2010/11		2010/11 to 2020/21		Total 30 Crop Seasons
	kt	%	kt	%	kt	%	kt	%
Palmeira das Missões—RS	18.8	2.8	278.1	13.8	157.8	6.7	454.6	9.0
Cascavel—PR	24.7	3.2	97.3	4.1	115.2	4.1	237.2	4.0
Cristalina—GO	13.0	3.1	228.7	7.5	222.0	4.0	463.7	5.2
Primavera do Leste—MT	37.8	2.2	174.0	2.6	86.8	1.4	298.6	2.0
Sorriso—MT	74.9	2.6	270.3	1.6	379.4	2.3	724.5	2.0
Lagoa da Confusão—TO	0.0	0.0	5.6	2.2	17.0	1.9	22.6	2.0
Barreiras—BA	66.0	4.3	227.0	6.7	265.6	6.3	558.7	6.1
Bom Jesus—PI	0.0	0.0	67.4	11.2	146.8	12.0	214.2	11.7
Balsas—MA	17.1	5.6	73.4	2.7	227.5	6.0	318.0	4.7
Total	252.2	3.0	1421.9	3.8	1618.1	3.7	3292.3	3.7

). The spatial distribution of these municipalities suggests that local convective events and other teleconnections significantly influence the regional effects of ENSO and The Intertropical Convergence Zone (ITCZ) [77]. This pattern is also seen in areas where both sides of the “climate divide” previously faced drought and yield losses, including Palmeira das Missões (278.1 kt; 13.8%) in the South and Barreiras (227.0 kt; 6.7%) in the Northeast. As indicated in Table 4, later decades show both contrasting and consistent yield responses to drought.

In a broader international context, these losses are similar to drought-related soybean yield reductions reported in major producing regions, including the US Midwest and the Argentine Pampas, where decadal-scale yield penalties of 3 to 7 percent during periods of climate stress have been documented. These findings indicate that Brazilian municipalities face climate phenomena of similar intensity as those affecting other global soybean-producing regions [91,92].

Over the past decade, total soybean yield loss reached 1618.1 kt (3.7%), remaining similar in relative magnitude to the previous period but showing different spatial distribution patterns. Throughout the entire 30-year span, municipalities experienced cumulative losses of 3292.3 kt, averaging 3.7% relative to the 5-year detrended yield. Municipality-level regression analyses were conducted using Ordinary Least Squares (OLS), with climate indices as predictors and $Y_r$ and $Y_{loss}$ as dependent variables (Table S1, Supplementary Material). These analyses highlight the differential effects of drought and heat stress on soybean yield across municipalities, providing a quantitative basis for interpreting observed spatial patterns.

3.4. Combined Climate Drivers of Droughts in Soybean Regions

Table 5. Results of the linear trend statistics of climate indices and agricultural yield indicators at the municipal level. For each variable and municipality, the slope, coefficient of determination (R²), and p-value are reported from a linear regression of the variable over time (year). The variable itself is the dependent variable, and the independent variable is year. Positive slopes indicate increasing trends, negative slopes indicate decreasing trends, and the p-value indicates the statistical significance of the trend.

Municipality	Metric	Yloss 1	Yr 2	SPEI 6M 3	SPI 6M 4	WSDI 5
Palmeira das Missões—RS	Slope	0.031	55.138	0.019	0.016	0.014
Palmeira das Missões—RS	R2 value 6	0.000	0.395	0.028	0.025	0.011
Palmeira das Missões—RS	p-value < 0.05 7	0.898	0.000	0.118	0.016	0.106
Cascavel—PR	Slope	−0.231	51.455	0.001	0.007	0.017
Cascavel—PR	R2 value	0.015	0.642	0.000	0.004	0.038
Cascavel—PR	p-value < 0.05	0.064	0.000	0.928	0.322	0.003
Cristalina—GO	Slope	−0.535	49.919	−0.039	−0.034	0.001
Cristalina—GO	R2 value	0.030	0.566	0.123	0.107	0.000
Cristalina—GO	p-value < 0.05	0.012	0.000	0.006	0.000	0.893
Primavera do Leste—MT	Slope	−0.003	22.562	0.007	0.007	0.016
Primavera do Leste—MT	R2 value	0.000	0.514	0.004	0.005	0.010
Primavera do Leste—MT	p-value < 0.05	0.986	0.000	0.553	0.284	0.118
Sorriso—MT	Slope	−1.078	34.607	−0.016	−0.002	0.018
Sorriso—MT	R2 value	0.036	0.661	0.022	0.000	0.009
Sorriso—MT	p-value < 0.05	0.006	0.000	0.163	0.807	0.173
Lagoa da Confusão—TO	Slope	−0.061	42.331	0.008	0.005	0.024
Lagoa da Confusão—TO	R2 value	0.043	0.781	0.006	0.002	0.008
Lagoa da Confusão—TO	p-value < 0.05	0.006	0.000	0.464	0.516	0.231
Barreiras—BA	Slope	−0.217	60.532	0.000	0.003	0.043
Barreiras—BA	R2 value	0.003	0.604	0.000	0.001	0.017
Barreiras—BA	p-value < 0.05	0.383	0.000	0.973	0.699	0.046
Bom Jesus—PI	Slope	−0.419	26.836	−0.030	−0.026	0.015
Bom Jesus—PI	R2 value	0.040	0.055	0.067	0.065	0.003
Bom Jesus—PI	p-value < 0.05	0.004	0.003	0.045	0.000	0.405
Balsas—MA	Slope	−0.680	36.588	0.019	0.019	0.045
Balsas—MA	R2 value	0.023	0.385	0.029	0.034	0.031
Balsas—MA	p-value < 0.05	0.021	0.000	0.109	0.005	0.006

¹ $Y_r$: observed (real) yield; ² Y_loss: yield loss relative to the 5-year detrended yield; ³ SPI-6M: Standardized Precipitation Index at a six-month timescale, indicating medium-term moisture conditions based on precipitation [20,21]; ⁴ SPEI-6M: Standardized Precipitation Evapotranspiration Index at a six-month timescale, indicating medium-term moisture conditions based on climatic water balance [20,21]. ⁵ WSDI: Warm Spell Duration Index, defined as the annual number of days belonging to warm spells according to the Expert Team on Climate Change Detection and Indices (ETCCDI) [25]. ⁶ R²: coefficient of determination of the linear regression model. ⁷ p-value: Statistical significance of the estimated slope coefficients at conventional confidence levels.

summarizes the linear trends of climate indices and soybean yield indicators at the municipal level over the 30-year study period. Each variable was regressed against year using OLS to quantify temporal trends. The slope represents the annual rate of change, R² indicates the proportion of variance explained, and the p-value tests whether the trend is statistically significant.

The results in Table 5, supported by Figure 11, show a clear trend of increasing hydroclimatic stress on soybean production across Brazil’s main producing regions. While yields have increased in eastern municipalities (e.g., BA, MA, and GO), reflecting improvements in technology and management practices [83,93], negative SPEI⁶ anomalies and positive WSDI trends indicate that water-related stress is growing, especially in municipalities in the Northeast (MA, TO, PI, and BA). In the second half of the 2000s, negative SPEI⁶ values became more consistent, aligning with higher average temperature anomalies and longer warm spells. This pattern suggests a shift from precipitation-driven droughts to those driven by temperature and evapotranspiration [75,76,77,78,79,80,81,82,83,84,85,86].

Positive WSDI trends, particularly in the Northeast, indicate longer heatwave durations during the soybean season, consistent with projections of increased heat stress under warming climate scenarios [85]. The inconsistent spatial and temporal patterns of SPI 6 across most sites underscore the complexity of precipitation variability and its implications for agricultural productivity. These findings suggest that, despite improvements in yield, climate stressors such as drought and heat may increasingly limit soybean production in rapidly expanding agricultural regions, highlighting the necessity for climate-adaptive strategies. In municipalities located in the Central-West and along the Cerrado region, robust and consistent trends imply that adaptation measures, including irrigation, enhanced genetics, and improved management practices, have achieved relative success (Table 5). Nonetheless, the significant WSDI and SPI trends observed in these areas indicate that even well-adapted systems remain susceptible to heat-driven evapotranspiration losses, especially under dry-season planting schedules [94].

In the Northeast, the concurrence of negative SPEI-6 values with prolonged heat spells (high WSDI) intensifies crop water stress beyond the effects of precipitation anomalies alone. Elevated temperatures accelerate evapotranspiration and reduce soil moisture availability during critical phenological phases [58,80]. The widening gap observed in these municipalities indicates increasing climatic constraints on productivity, particularly where irrigation capacity is limited and evaporative demand is high. In contrast, municipalities in the Cerrado region exhibit smaller negative SPEI-6 trends and only moderate yield reductions despite rising WSDI. This pattern suggests that adaptive capacity, including cultivar selection, management practices, and technological investment, can partially mitigate heat-related impacts [95,96].

Table 5 and Figure 11 further demonstrate that rising heat stress and water deficits interact with technological yield gains, resulting in heterogeneous effects across regions. In areas experiencing rapid agricultural expansion, such as the Northeast, the increasing frequency of compound drought–heat events may constrain future yield potential. This trend highlights the necessity for climate-adaptive strategies, especially in rainfed systems where thermal extremes and evapotranspiration-driven drought are becoming primary limitations.

4. Discussion

The findings demonstrate that high-resolution, spatially disaggregated gridded climate datasets, when validated at the municipal scale, accurately represent the spatial and temporal variability of soybean yield responses to DHE stress across Brazil’s principal soybean macroregions. These results substantiate the assertion that standardized hydroclimatic indices (SPI, SPEI, and WSDI) effectively characterize regional yield sensitivities to DHE. Integrated trend analyses of these indices and detrended yield data reveal that all studied municipalities experienced significant declines in water availability and increased heat stress during the soybean growing season from 1989 to 2020. The Northeast region exhibits the most pronounced yield losses, corresponding with recurrent rainfall deficits and intensifying temperature anomalies, while the Central-West displays emerging negative yield trends associated with prolonged heatwave durations despite technological advancements. In the South and Central-West highlands, previously marked by greater climatic stability, precipitation variability is now increasing, disrupting crop development, particularly during reproductive stages. These results further indicate that the compound effects of DHE are undermining yield stability across all CZ, aligning with projections of tropical and subtropical agricultural vulnerability [4,86].

The recent intensification of DHE has exerted a greater influence on soybean production trends over the past decade, as indicated by precipitation anomalies. Additionally, detrended yield analyses demonstrate that negative yield deviations increasingly coincide with periods of simultaneous negative SPI and SPEI anomalies and positive WSDI trends, confirming that concurrent heat and moisture stress dominate interannual variability [59,83]. Seasonal analyses identify water deficits during the reproductive and grain-filling stages as primary drivers of $Y_{loss}$, resulting from insufficient recovery following early-season rainfall deficits. This seasonal mismatch between water demand and availability amplifies reproductive stress and accelerates phenological progression, ultimately reducing grain mass and final yield. The integration of high-resolution climate data, standardized indices, and yield detrending represents a key strength of this study, facilitating the identification of underlying yield–climate mechanisms independent of technological progress. Nevertheless, the absence of management and socioeconomic variables constrains the ability to attribute yield trends exclusively to climatic factors. Technological and agronomic adaptations, such as irrigation scheduling and optimized planting calendars, partially mitigate climatic impacts but remain inadequate to reverse the persistent downward yield trend under intensifying compound extremes.

Yields across the three clusters remain similar (Figure 3e), indicating that soybean varieties and agricultural management practices have adapted to relatively stable intraseasonal climate variability in Brazil’s soybean production areas (Figure 4 and Figure 5). However, the sufficiency of these adaptations in addressing the increasing frequency, duration, and intensity of droughts remains uncertain [97]. The compounded nature of drought events presents significant challenges to water and food security in regions where soybean agriculture is expected to expand further. Soybean responses to anomalously dry seasons or years are often triggered by heatwaves and deficiencies in agronomic and water management. Ref. [70] propose several strategies to mitigate drought impacts in Brazil, including diversification of agriculture, selection of drought-resistant varieties, and the use of native species for forage and wood. These strategies are closely linked to plant exposure to climate variations during the growing season. Thus, the exploration of changes in atmospheric evapotranspirative demand (AED), vapor pressure deficits, and ETc is critical for adaptive soybean cropping systems, particularly during unexpected dry years, and involves complex interactions between natural and human environments [73,98,99]. AED influences evapotranspiration and the processes governing available energy in areas where land use, vegetation, and soil moisture are continuously changing. Therefore, scientific and technological advancements aimed at mitigating rising temperatures and associated droughts and heatwaves in regions such as Paraná and Cerrado should prioritize constraining the positive feedback loop, in which crop evapotranspiration and soil moisture decline as temperatures continue to increase.

In summary, the analysis confirms that DHE exerts an increasing influence on soybean yield trajectories across Brazil’s soybean regions. The combined application of SPI, SPEI, and WSDI demonstrates their complementary diagnostic capabilities in identifying both precipitation- and temperature-driven components of agricultural variability. These indices establish an empirical foundation for operational climate-risk surveillance and adaptive management within compound-risk frameworks, particularly in tropical cropping systems where warming intensifies atmospheric water demand. Future agricultural expansion or intensification will benefit from studies that integrate process-based crop models with high-frequency meteorological data, management records, and remote sensing to refine yield response functions under DHE, as well as socioeconomic indicators to enhance vulnerability diagnostics and inform resilient agricultural planning [100,101,102]. Furthermore, integrating genomic, climate, and management data to model phenotypes will improve the capacity to nowcast, forecast, and predict phenotypes across regions, accounting for inherent variability and exposure to extreme events [103,104,105]. These advancements are essential for developing predictive models that capture the multidimensional and interactive nature of climate risks affecting regional soybean systems in Brazil.

5. Conclusions

This study provides a spatiotemporal assessment of compound drought and heat extremes affecting Brazil’s main soybean producing regions from 1989 to 2020. By integrating high resolution meteorological data, yield records, and standardized hydroclimatic indices SPI, SPEI, and WSDI, the analysis demonstrates that climate stressors are intensifying and becoming increasingly compound in nature. Distinct CZ exhibit contrasting sensitivities, with southern regions showing relatively higher yield stability, while northeastern and frontier regions experience greater concurrence of rainfall deficits and heatwaves during critical phenological stages, leading to greater yield variability.

The results show that DHE has increased in frequency and intensity, especially during reproductive phases, intensifying stress on rainfed soybean systems. Drought-related indices, particularly SPI-6, show strong correlations with yield variability in vulnerable areas such as Maranhão, Piauí, and Bahia, highlighting their importance for early warning systems and climate risk monitoring. Temperature-driven increases in evapotranspiration and extended heatwaves further worsen drought severity, making the Northeast the most climate-sensitive soybean region, while the Central-West and South face growing risks from heat extremes and rainfall variability despite technological progress.

Overall the findings highlight the importance of revising agroclimatic risk frameworks to explicitly include compound hydroclimatic extremes. Regionally tailored adaptation strategies, backed by climate-informed planning, enhanced monitoring with compound indices, and investments in resilient cultivars and infrastructure, are crucial to maintaining soybean production amid increasing climatic variability. Future re-search should combine more detailed agronomic, genetic, phenological, and socioeconomic data, along with remote sensing and crop modeling, to better understand multi-dimensional vulnerability and develop targeted adaptation policies.