Impacts of Climate Change on Coffee Production in Southern Minas Gerais, Brazil

1. Introduction

Coffee is a major global agricultural commodity that supports several strategic sectors, including cultivation, processing, transportation, marketing, and trade, with Brazil playing a leading role [1]. The genus Coffea comprises approximately 130 species, of which C. arabica L. and C. canephora Pierre ex A. Froehner are commercially important [2]. In Brazil, the state of Minas Gerais leads national coffee production, particularly in the South and Southwest mesoregion of Minas Gerais, where a network of cooperatives, exporters, and coffee brokers reinforces the region’s strategic position in domestic and international markets [3,4]. Arabica coffee thrives under favorable edaphoclimatic conditions [5]. Suitable elevations, annual rainfall of 1,200–1,500 mm and temperatures of 15–24 °C are considered ideal for Arabica coffee production, while regular rainfall during the reproductive period is essential [6,7,8]. However, climatic variability poses challenges to coffee cultivation, yield, and total production. High temperatures, drought, hail, and frost events can reduce photosynthetic rates and cause premature flowering, pollen tube desiccation, and flower abortion [9]. These meteorological irregularities may also favor pests as coffee leaf rust (Hemileia vastaris) and infestations by coffee berry borer (Hypothenemus hampei) that are considered the two most important and threatening coffee plants. The losses due to pests account up to 13% annual production worldwide [10,11]. The coffee berry borer is favored by drought periods and high temperatures conditions that have become increasingly frequent [12], whereas higher temperatures and less intense rainfall favor coffee leaf rust, which affects plants grown at higher elevations [13]. Continued warming also reduces land suitability for Arabica coffee [14], potentially making some current production areas unsuitable for cultivation or crop development and increasing the need for new growing areas at higher elevations [15,16].

Against this background, this study analyzed trends and rates of change in temperature and precipitation and their correlations with coffee yield and total production in southern Minas Gerais from 1984 to 2023. We used climate and harvest data provided by Ipanema Coffees (Ipanema Agrícola S.A.) for two production units located in the municipalities of Alfenas and Machado, southern Minas Gerais, Brazil. To this end, we applied the Mann–Kendall test and Sen’s slope estimator, widely used in agrometeorological and environmental analyses, to detect seasonal trends and quantify changes [17,18]. Based on the results of the Shapiro–Wilk normality test, we used Pearson’s or Spearman’s correlation coefficient, as appropriate, to assess the effects of climatic variations on yield and total production [19,20].

This study is warranted by the close relationship between climate and coffee cultivation and by the crop’s vulnerability to climatic fluctuations, which can compromise yield and have ecophysiological and economic consequences [21,22,23].

2. Materials and Methods

2.1. Study Area and Production Units

We conducted the study at the Conquista and Capoeirinha production units of Ipanema Agrícola S.A., located in the municipalities of Alfenas and Machado, respectively, in southern Minas Gerais, Brazil (Figure 1). The geological substrate of the region consists of gneisses overlain by soils and unconsolidated fluvial deposits of gravel, sand, and clay [24]. Both municipalities are located within the Guaxupé Massif [25]. The region lies within the Atlantic Forest biome, with transition zones to the Cerrado (Brazilian savanna) [26]. Under the Köppen climate classification [27], the region has a humid subtropical climate (Cwb), characterized by dry winters and mild summers, with a mean annual temperature of 21.2 °C and annual precipitation of 1,500–1,750 mm.

The Conquista production unit covers 2,045 ha, comprising 82.26% coffee plantations, 14.54% native forest cover, 1.00% eucalyptus plantations, 0.91% pasture, 0.88% built-up areas and 0.41% of water bodies. The Capoeirinha unit spans 1,772 ha, consisting of 68.07% coffee plantations, 23.08% native forest cover, 5.26% eucalyptus plantations, 1.80% water bodies, 0.93% pasture, and 0.86% built-up areas. Harvesting is fully mechanized at Conquista. At Capoeirinha, 98% of the harvest is mechanized, and the remaining 2% is harvested manually on steep slopes. At Conquista, rows are spaced 3.5–4.0 m apart, with 1.0 m between plants. At Capoeirinha, row spacing is 2.0–4.8 m, and in-row plant spacing is 0.5–2.0 m [28].

2.2. Data Sources and Methodological Overview

The company provided coffee yield (bags/ha), total production (bags), and climate data. Weather stations located at the production units recorded the climate data. The dataset covered 1984–2023 and included mean, maximum, and minimum air temperature; total precipitation; and rainfall during the dry and wet seasons. The methodological workflow used to process and analyze these data is summarized in Figure 2.

2.3. Data Analysis

For trend analyses, both parametric and nonparametric methods can be applied to climate time series. We selected nonparametric methods because they do not assume normally distributed data and are more robust to outliers. Specifically, we applied the Mann–Kendall test to detect trends and Sen’s slope estimator to quantify their magnitude [17]. We processed the data in Python using the pyMannKendall, scikit-learn, and SciPy packages in Visual Studio Code (VS Code).

We applied the Shapiro–Wilk test to assess normality and select the appropriate correlation coefficient for each variable pair [19]. We used Pearson’s correlation coefficient when both variables in a pair were normally distributed and Spearman’s rank correlation coefficient otherwise [20].

2.4. Mann–Kendall Test

The Mann–Kendall (MK) test is a nonparametric statistical method developed by Mann [29] and Kendall [30]. It has been widely used to identify increasing and decreasing trends in environmental time series, including temperature, precipitation, and humidity [17,18]. Because autocorrelation can affect the statistical significance of the test, we applied prewhitening to remove the temporal dependence from the data [31]. The MK test does not require normally distributed data and is less sensitive to outliers because it is based on observation ranks rather than absolute values [31,32].

For each time series, the MK test evaluates the null hypothesis ($H_0$) of no trend against the alternative hypothesis ($H_1$) of a monotonic trend. The test statistic S is calculated using Equation (1):

$$\mathrm{S}={\sum }_{i=1}^{n1}{\sum }_{j=i+1}^n\mathrm{s}\mathrm{g}\mathrm{n}\left(x_j{-x}_i\right)$$

(1)

Here, n is the number of observations; x_i and x_j are the observations at times i and j, respectively, with j > i; and sgn is the sign function defined in Equation (2) [33]:

$$\begin{gathered} \left\{1,(x_j-x_i)>0\right\} \\ \mathrm{s}\mathrm{g}\mathrm{n}=\left\{0,(x_j-x_i)=0\right\} \\ \left\{-1,(x_j-x_i)<0\right\} \end{gathered}$$

(2)

Positive values of S indicate increasing trends, whereas negative values indicate decreasing trends. The variance of S, corrected for tied observations, is calculated using Equation (3):

$$\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)={\displaystyle \frac{n(n-1)(2n+5)-{\sum }_{i=1}^m{t}_i(t_i-1)(2t_i+5)}{18}}$$

(3)

Here, m is the number of tied groups, and $t_i$ is the number of observations in the i-th tied group [30].

The standardized Z statistic is calculated from S and Var(S) using Equation (4):

$$\begin{gathered} \mathrm{Z}=\left\{{\displaystyle \frac{S-1}{\sqrt{ar\left(S\right)}}},S>0\right\} \\ \left\{{\displaystyle \frac{S+1}{\sqrt{ar\left(S\right)}}},S<0\right\} \end{gathered}$$

(4)

Under the null hypothesis of no trend (H₀), Z follows an approximately standard normal distribution. For the two-sided test, we adopted α = 0.05 and rejected H₀ when p < 0.05.

2.5. Sen’s Slope Estimator

Sen’s slope estimator quantifies trend magnitude as a rate of change in a time series [34]. We used it with the MK test because it is robust to outliers [35]. Pairwise slopes were calculated using Equation (5):

$$Q={\displaystyle \frac{(x_j-x_i)}{j-i}},=1,2,3,...,N$$

(5)

Here, x_j and x_i are the observations at times j and i, respectively, with j > i, and k indexes each of the N pairwise slopes. For a time series containing n observations, the number of pairwise slopes, N, is calculated using Equation (6):

N = n(n-1)/2

(6)

After sorting the N values of $Q$ in ascending order, we calculated Sen’s slope

($Q_{med}$) as their median and obtained a 95% confidence interval for the slope estimate. The calculation of Q_med is given by Equation (7):

$$\begin{gathered} Q_{med}=\left\{Q_{\left[{\displaystyle \frac{+1}{2}}\right]}\right\} \mathrm{if} \mathrm{N} \mathrm{is} \mathrm{odd}, \\ \mathrm{if} \mathrm{N} \mathrm{is} \mathrm{even}. \end{gathered}$$

(7)

2.6. Spearman’s Rank Correlation Coefficient

Spearman’s rank correlation coefficient is a nonparametric measure of the strength and direction of the monotonic relationship between two variables. We used it when at least one variable in a pair did not meet the normality assumption. We used this coefficient to identify monotonic, but not necessarily linear, relationships between climatic variables and Arabica coffee yield. In the absence of tied ranks, the coefficient can be calculated from the ranks of paired observations using Equation (8):

$$\rho =1-{\displaystyle \frac{6{\sum }_{i=1}^n{d}_i^2}{n(n^2-1)}}$$

(8)

Here, $d_i$ is the difference between the ranks assigned to the $x_i$ and $y_i$ for observation i, and n is the number of paired observations. The coefficient ranges from -1 to 1; its sign indicates the direction of the monotonic relationship, and its absolute magnitude indicates its strength. We assessed statistical significance at α = 0.05 [36].

2.7. Pearson’s Correlation Coefficient

Pearson’s correlation coefficient measures the strength and direction of the linear association between two variables. In this study, we used it for variable pairs in which both variables met the normality criterion based on the Shapiro–Wilk test. We used the coefficient to assess whether coffee yield and total production varied linearly with climatic variables. Specifically, we examined whether gradual increases in climatic variables corresponded to increases or decreases in the agronomic variables. This coefficient is calculated as the ratio of the covariance of the variables to the product of their standard deviations, as shown in Equation (9):

$$R={\displaystyle \frac{{\sum }_{i=1}^n(x_i-\underset{\_}{x})(y_i-\underset{\_}{y})}{\sqrt{{\sum }_{i=1}^n(x_i-\underset{\_}{x}{)}^2}\sqrt{{\sum }_{i=1}^n}(y_i-\underset{\_}{y}{)}^2}}$$

(9)

Where n is the number of paired observations, $x_i$ and $y_i$ are the observed values of the climatic and agronomic variables, respectively; and $\underset{\_}{x}$ and $\underset{\_}{y}$ are the corresponding sample means. The coefficient ranges from -1 to 1; its sign indicates the direction of the linear relationship, whereas its absolute value indicates its strength. We assessed statistical significance at α = 0.05 [37].

3. Results

3.1. Trend Analysis of Temperature and Precipitation

We evaluated trends in maximum temperature ($T_{max}$), minimum temperature ($T_{min}$), and mean temperature ($T_{mean}$), and precipitation at both production units (Figure 3 and Figure 4) and compared their estimated magnitudes of change (Figure 5). The complete results are provided in Appendices A and B.

Maximum temperature showed significant increasing and decreasing trends, but their occurrence varied among months over the historical time series. At Conquista, T_max

decreased significantly in April and December, with estimated declines of 3.70 and 2.53 °C over the study period, respectively. In September, Tₘₐₓ increased by 1.84 °C. At Capoeirinha, T_max decreased significantly in April (2.39 °C), whereas September showed an increase of 2.92 °C, which exceeded the increase at Conquista.

Unlike $T_{max}$, $T_{min}$ generally increased at both production units, although the months and magnitudes differed. At Conquista, T_min increased in June and September, with estimated increases of 1.82 and 1.93 °C, respectively. At Capoeirinha, increases occurred in more months and were larger: 2.54 °C in June, 2.20 °C in September, and 2.06 °C in October.

$T_{mean}$ showed significant declining trends, except in September, when it increased, consistent with the patterns observed for Tₘₐₓ and Tₘᵢₙ. At Conquista, $T_{mean}$ decreased in April, May, and December, with estimated changes of -2.59, -2.00, and -1.36 °C, respectively. At Capoeirinha, the declines were smaller in April and May, with estimated changes of -1.14 and -1.38 °C, respectively, whereas September showed an estimated increase of 2.34 °C.

We characterized seasonal precipitation patterns using total precipitation accumulated during the dry and wet seasons and monthly precipitation over the historical time series. The complete dataset is provided in Appendix B. In the study region, the dry season extends from April through August, whereas the wet season extends from September through March. Total precipitation at Conquista showed no significant trend. At Capoeirinha, however, May precipitation decreased significantly, with an estimated decline of 34.9 mm over the study period.

Seasonal precipitation trends differed between the production units. At Conquista, wet-season precipitation showed no significant trend, whereas dry-season precipitation declined significantly, with an estimated reduction of 94.7 mm. At Capoeirinha, neither season showed a significant precipitation trend over the historical time series.

3.2. Correlation Analysis

To assess the relationships between climatic and agronomic variables, we constructed a correlation matrix for each production unit. We correlated coffee yield (bags/ha) and total production (bags) with the climatic variables; the results are shown in Figure 6 and Figure 7. In both figures, panel (a) shows the complete correlation matrix, while panels (b) and (c) show selected correlations of climatic variables with coffee yield and total production, respectively.

At the Capoeirinha production unit, the standard deviation and amplitude of dry-season precipitation showed strong negative correlations with both agronomic variables (Figure 6). For coffee yield, the correlation coefficient for the standard deviation of dry-season precipitation was -0.323, while the coefficient for its amplitude was -0.326. The corresponding coefficients for total production were -0.416 and -0.402, respectively.

At the Conquista production unit, the standard deviation of mean temperature and wet-season precipitation were significantly correlated with coffee yield (Figure 7). Their correlation coefficients were 0.376 and 0.318, respectively. The standard deviations of maximum and minimum temperatures, together with the amplitudes of mean, minimum, and maximum temperatures, also showed positive correlations with yield but did not reach statistical significance, although their p-values were close to the significance threshold.

For total production, wet-season precipitation and the standard deviation of minimum temperature approached, but did not reach, statistical significance (p = 0.052 and 0.097, respectively), with corresponding correlation coefficients of 0.308 and 0.265, respectively.

4. Discussion

4.1. Climate Dynamics and Agronomic Variables

Our results reveal significant changes in the climatic profile of both production units over the historical record, characterized by uneven warming, altered rainfall seasonality, and effects on the agronomic variables evaluated. Rates of temperature change in the study areas exceed the mean decadal warming rates projected in IPCC reports. This amplified local climate signal can be attributed to interactions between climate change and local land-use and land-cover change [38,39].

In this context, $T_{min}$ generally increased, whereas $T_{max}$ and $T_{mean}$ followed contrasting seasonal patterns, decreasing in the months preceding winter (April and May) and at the onset of summer (December), but increasing markedly in September. Increases in $T_{min}$ in June and September suggest progressive nocturnal warming. This pattern is consistent with studies of tropical warming, in which increased cloud cover and greenhouse gas concentrations reduce nocturnal longwave heat loss, thereby raising minimum temperatures and narrowing the diurnal temperature range [40,41]. Conversely, the April decreases in $T_{max}$ (3.70 °C at Conquista and 2.39 °C at Capoeirinha) indicate lower temperature extremes during autumn.

Although the dry and wet seasons are well defined in southern Minas Gerais [42], precipitation showed no significant trends in most months. This overall stability was interrupted by declining rainfall, while the dry season was characterized by high variability. At Capoeirinha, this variability showed strong negative correlations with total production and yield. Together with increasing minimum temperatures in June, this pattern indicates increased evaporation during a period of water scarcity.

Despite being located in the same region, Capoeirinha exhibited greater thermal variability, whereas Conquista was more stable. The strong negative correlations found at Capoeirinha demonstrate the agronomic implications of these patterns. By contrast, proximity to the Furnas Reservoir buffers temperature extremes at Conquista, limiting the rates of change detected by the trend tests [43]. Capoeirinha, in turn, lies within a landscape matrix of diverse croplands, isolated Atlantic Forest fragments, and roads. These landscape features favor warming, unlike the moisture-related buffering effect observed at Conquista [44,45].

4.2. Impacts on Coffee Cultivation in Southern Minas Gerais

Edaphoclimatic conditions, including temperature variability, are key environmental controls on the phenological development and yield of coffee, particularly Arabica coffee. Annual rainfall of 1,200–1,800 mm is considered optimal. Against these requirements, the negative correlations of yield and total production with the standard deviation and amplitude of dry-season rainfall, together with increasing minimum temperatures, provide statistical evidence of adverse effects on both agronomic variables at Capoeirinha. The optimal temperature range is 15–24 °C. Although Arabica coffee can tolerate brief periods of temperature extremes, prolonged exposure to such stress compromises plant physiological performance and yield; temperatures above 34 °C can reduce the photosynthetic rate to near zero [21,46]. Taken together, these ecophysiological requirements and the trends and correlations observed here show that altered temperature and rainfall patterns challenge the phenological development of Arabica coffee and affect both yield and total production [47]. Temperature increases observed in September, immediately following a period of pronounced water stress, make coffee plants vulnerable to irregular flowering and induce flower abortion due to pollen tube desiccation, substantially reducing yield [21].

Declines in $T_{mean}$ and $T_{max}$ in April and May, combined with lower precipitation during the same period, have contrasting implications for bean maturation and quality. Milder temperatures during ripening prolong the ripening period, favoring the accumulation of carbohydrates and phenolic compounds and improving the sensory attributes of the beverage [48]. This pattern is consistent with the positive correlation found between coffee yield and the standard deviation of $T_{mean}$ at Conquista. Conversely, if the water deficit intensifies, coffee plants may experience preharvest stress and produce malformed beans [49].

The observed increases in $T_{min}$ in June and September also raise nocturnal respiration in coffee plants. Warmer nights accelerate the consumption of carbohydrates that would otherwise be allocated to branch and fruit growth, leading to early depletion of plant reserves, premature ripening, and greater pressure from pests favored by dry conditions [5,50]. Regular development across phenological stages is critical for coffee yield. Excessive heat in September, coupled with progressively decreasing dry-season rainfall, is a limiting factor that may make coffee cultivation unviable during intense El Niño events, as in 2015 [51]. The ongoing 2026 event is expected to strengthen through the end of the year [51].

4.3. Mitigation Strategies

Reducing greenhouse gas emissions is fundamental to mitigating climate change and its impacts. On-farm activities in coffee production are substantial sources of nitrous oxide ($N_{2}O$), carbon dioxide (${CO}_2$), and methane (${CH}_4$) emissions resulting from fossil fuel consumption, electricity use, firewood combustion, and, particularly, nitrogen fertilizer application and liming [28,52,53]. In this context, incorporating trees into coffee plantations and maintaining forested areas within production units are effective management strategies for mitigating future impacts on coffee crops.

Trees sequester carbon, while shade provides microclimatic benefits for Arabica coffee [54]. The use of fertilizers containing urease and nitrification inhibitors is also an important mitigation measure [55].

Climate forecasting based on weather-station data supports temporal analyses of precipitation and temperature, informs strategic decision-making in response to climate change, and facilitates the development of models and projections [56]. Alongside these strategies, advances in genomics and transcriptional regulation are also important. Because small RNAs (sRNAs) respond to environmental signals and regulate gene expression, they may provide tools for developing coffee plants with greater tolerance to climate-related stress [57].

4.4. Limitations

First, we controlled serial autocorrelation by applying prewhitening before the MK test to ensure the robustness of the detected trends in this study. However, Sen’s slope estimator assumes a linear and monotonic trend over time. It may therefore not fully capture nonlinear climatic fluctuations or abrupt shifts in the historical time series.

Second, although the study provides significant insights into climate dynamics and their ecophysiological implications for Arabica coffee, the analysis relies on historical weather-station records. Although these records are highly reliable, they may not fully capture site-specific microclimatic variation at Conquista and Capoeirinha.

Finally, the physiological effects discussed for coffee plants, including flower abortion and the depletion of carbohydrate reserves, were inferred from physiological mechanisms described in the literature rather than measured directly in this study. Future studies integrating automated microclimate sensors with crop models could refine the projections presented here.

5. Conclusions

Overall, the trend analyses and their estimated rates of change revealed marked climate variability across the study area. Increasing and decreasing trends occurred throughout the historical record, and their magnitudes differed between the production units and exceeded the decadal warming rates reported by the IPCC. Capoeirinha showed greater variability, whereas Conquista was more stable. The correlation analyses also produced contrasting results that were consistent with the trend analyses and estimated rates of change.

Global-scale assessments can smooth climatic extremes because they integrate extensive oceanic and vegetated areas, whereas the landscape characteristics surrounding Capoeirinha act as local risk factors. By contrast, Conquista is adjacent to the Furnas Reservoir, whose thermal buffering effect was reflected in the smaller rates of change detected by the trend tests.

This study demonstrated that climate-related effects differed between the two production units. Therefore, management and mitigation strategies should be site-specific and tailored to the conditions at each production unit.