The study was conducted in two semiarid areas in the Pernambuco State, Northeastern Brazil (Figure 1). The first select area was the Catimbau National Park (Catimbau NAPAR), a Conservation Unit that covers 62,294.14 hectares, located between the “agreste” and “sertão” of the Pernambuco State in the domain of three municipalities in the São Francisco River Basin: Buíque, Tupanatinga, and Ibimirim (ICMBio, 2002) (Figure 2). In the Catimbau NAPAR, six rural communities were selected: Muquem (8.49414° S; 37.2999° W), Breus (8.49076° S; 37.29059° W), Açude Velho (8.4592° S; 37.30289° W), Tunel (8.46208° S; 37.31967° W), Dor de Dente (8.51355° S; 37.31413° W) and Igrejinha (8.49966° S; 37.25743° W). The total number of local adult residents in each of these communities, according to data obtained through the Family Health Program (FHP) of the Buíque Municipality, were: Muquem (n = 38), Breus (n = 32), Açude Velho (n = 17), Tunel (n = 6), Dor de Dente (n = 13), and Igrejinha (n = 112), totaling 218 people.

The second selected area was the Carão (08° 35′13.5″ S and 36° 05′ 34.6″ W), a rural community located in the Altinho Municipality, Pernambuco State, in the center of the “agreste,” 168 km from the state capital. According to information obtained through the Carão health center, the local population currently consists of 140 people (27 were under 18 years old) in 55 family units (households; Figure 3).

The semiarid tropical climate (BSh), according to the Köppen-Geiger Scale, typical of the semiarid region of Northeastern Brazil (Alvares et al., 2013), is predominant in both the Catimbau NAPAR (ME, 2015) and Carão (Cruz et al., 2014). In the Catimbau NAPAR, the mean annual temperature is 23°C, with great interannual irregularity in rainfall, varying from 650 to 1,100 mm annually (ME, 2015). All rural communities selected for the study in the Catimbau NAPAR are located in the driest strip of the park (PELD Catimbau, 2016). The annual temperature in Carão is around 25°C, with an average rainfall of around 622 mm between March and July (BDE, 2017).

Droughts are recurrent climatic events in Northeastern Brazil and are generally associated with the El Niño phenomenon, which results in lower rainfall than the historical average in this semiarid region, which is about 800 mm per year (Magalhães, 2016; Marengo et al., 2018). Additionally, the precipitation is concentrated in a few months and evapotranspiration levels exceed 2,000 mm per year. As the semiarid region is a climatic Frontier area, any decrease in relation to the average may have a large impact (Magalhães, 2016). Based on historical records, as illustrated in Table 1, in the last 50 years, there were at least nine periods of prolonged drought in the semiarid region of the Brazilian Northeast, which brought intensive losses to both the ecosystem and people’s livelihoods: 1979–1983, 1986–1987, 1992–1993, 1997–1999, 2002–2003, 2005, 2007–2008, 2010, and 2012–2017 (Marengo et al., 2017; Lima and Magalhães, 2018; Marengo et al., 2018; Cunha et al., 2019). Since 2010, the region has experienced the longest and most severe drought periods in its history, especially since 2015, due to the El Niño phenomenon (Marengo et al., 2018). According to data from the Instituto Nacional de Meteorologia (INMET, 2021), the Pernambuco State had lower precipitation and higher temperature averages in some months of an otherwise average year (Figures 4, 5).

In the Catimbau NAPAR and Carão, this research was carried out with small farmers whose main source of local income is derived from subsistence agriculture and livestock, with emphasis on the planting of corn and beans and the operation of small goat, cattle, pig, and poultry farms. However, both areas have suffered agricultural losses and water resource shortages, characterizing them as risk areas for extreme drought events (Martins et al., 2015; Marengo et al., 2018), which may imply the proliferation of diseases related to extreme events of climate change (WHO, 2009). In the Catimbau NAPAR, this situation is aggravated by socio-environmental conflicts that have distinguished the area since it was founded (in 2002), involving the park management body—the Chico Mendes Institute for Biodiversity Conservation (ICMBio) —and residents, who had never been compensated since the park’s foundation and continue to live within the area, which is not legally permitted, according to the National System of Conservation Units (Brasil, 2000). Such a scenario made it very difficult for us to access some people living in the region because our presence was interpreted as an intervention by the environmental agency responsible for the park’s administration. To alleviate this situation, we adopted a cautious approach to residents, based on informal conversations and by explaining our project. For this reason, we also started our data collection in Carão, as we wanted to obtain a larger sample and, for the reasons mentioned above, Catimbau NAPAR was no longer a viable alternative. We chose Carão because our team has been conducting research in the region for 13 years and, therefore, has extensive knowledge about the site and a good relationship with residents.

To meet the legal aspects of research involving humans, this study was submitted and approved by the Research Ethics Committee of the University of Pernambuco, according to the Certificate of Submission for Ethics Appreciation (CSEA) number 89890617.1.0000.5207. All residents who agreed to participate in the research were asked to sign a free and informed consent form, as governed by the current legislation of the National Health Council (Resolution No. 466/2012).

The initial contact with both rural communities was made through a previous survey on the number of households in each of the seven communities in the study (in the case of Carão, this step was not necessary due to long-term studies conducted by our team in the region for at least 10 years). Data regarding the number of people living in each rural community were obtained through the Family Health Program (FHP) of Catimbau Village, Buíque Municipality (PE), and the health center of Carão, Altinho Municipality (PE). Adults and older people (above 18 years old) living in rural communities were chosen.

Data collection for the study occurred between July 2018 and September 2019. Of the 358 residents in the two study areas (218 in the Catimbau NAPAR and 140 in Carão, as mentioned above), it is important to note that some of these residents are above 60 years old and have health problems. Additionally, some residents were not available at the time of the survey and refused to participate in the research (especially in the Catimbau NAPAR, a scenario of numerous socio-environmental conflicts, as previously mentioned). Thus, our sample was ultimately composed of 143 people who agreed to participate in the study, comprising 94 people living in the Catimbau NAPAR in the six selected rural communities: Muquem (n = 30), Breus (n = 20), Açude Velho (n = 9), Tunel (n = 6), Dor de Dente (n = 4), and Igrejinha (n = 25); and 49 people living in Carão.

Before our fieldwork, the study’s aims and methods were explained to informants to clarify doubts. After this step, people would be free to voluntarily accept (or refuse) to participate in the research. We also conducted a pilot study to ensure that the interviewers understood the questions.

The selection of the sample of the population that accepted to participate in the research was made from a census (Espinosa et al., 2014), as well as the collection of socioeconomic data from the selected informants (see Supplementary Table S1). Thus, residents who had relevant information for our study were interviewed, including people who lived in the same household. The information collected was carefully evaluated by removing inconsistencies or biases that compromised the validity of the information. Based on our observations and information from the scientific literature, people from the same family or who live in the same household do not necessarily have the same experiences or develop the same adaptive climate change strategies. This depends on other correlated factors, such as agriculture and risk experience (e.g., Weber, 2010; Bryan et al., 2013; Rodriguez et al., 2017).

To analyze perceptions on climate change-related diseases, semi-structured interviews were conducted with the participants due to the flexibility of the previously elaborated questions (Albuquerque et al., 2014; Magalhães et al., 2019). The field interviews were organized in two stages, which are described below. In the first stage, each participant was identified and asked about the years in which major extreme weather events (non-seasonal) occurred, such as droughts and rainfall. Without this design, the participants could have cited diseases related to environmental change in general and not those specifically related to climate change. Extreme droughts were the most characteristic climate change events faced in semiarid regions, with the period between 2012 and 2015 being pointed out as the most critical in the last 50 years (Martins et al., 2015). By comparing historical data on drought events in Northeastern Brazil (Table 1), the historical data cited by our informants were validated because they corresponded with the scientific data (such as the year 1993 and the period 2012–2014, the most mentioned by farmers in both study areas). Thereafter, using the free list technique (da Silva et al., 2014), the participants were asked to name or list diseases perceived as a result of the extreme drought and rainfall events in the region. We emphasize here that because the focus of the proposal of our experimental design was diseases “perceived by people” and not “scientifically designated” diseases, we do not use previously cataloged information regarding the incident diseases in the regions.

In the second stage, based on the list of mentioned diseases, each participant was asked to classify and organize them according to their severity i.e., the degree of impact on their lives (Smith et al., 2000). The severity of the diseases was represented by numbers following a decreasing ordinal classification (for example, number 1 represented the disease considered the most severe; number 2, the disease considered the second most severe; and so on). Additionally, based on each disease mentioned, the participants were asked about adaptive strategies used (if any) to mitigate them. In this study, the adaptive strategies, to paraphrase Smit and Wandel (2006), were considered responses developed by the participants to mitigate the potentially harmful effects arising from the diseases they perceived.

Although age was one of the main differences between our two areas of study, as Carão has a higher proportion of older people (80.7%), we chose not to categorize the sample sizes according to the age for each place of study due to the diverse evidence from scientific literature demonstrating that age is not a good predictor of the number of extreme weather events or perceived risks related to these weather phenomena. In this sense, a person’s experience in agriculture explains much more than age (Slegers, 2008; Gbetibouo, 2009; Fosu-Mensah et al., 2012). For example, a 60-year-old person may have lived for 40 years outside the community and participated in activities unrelated to agriculture; and a 35-year-old person could have worked in agriculture for 30 years. In this case, the younger person has more experience in agriculture than the older person and, therefore, would tend to perceive a greater number of extreme weather events and risks related to these phenomena. We believe that studies that consider categorizing sampling units according to the length of experience in agriculture (which was not our study proposal), can provide interesting insights.

The incidence index (Ii) and the severity index (Sj) attributed to each perceived illness were initially calculated to test our hypothesis using the method suggested by Baird et al. (2009), adapted from Smith et al. (2000) (see Supplementary Table S2), and the objective was to understand and describe the local risk perception (in this case, diseases and risks to human health). Originally, this method—Participatory Risk Mapping (PRM)—was developed to access and quantitatively evaluate perceived environmental risks (see Smith et al., 2000; Baird et al., 2009). Compared to other methods used for this same purpose, the PRM stands out for being an extremely accurate and easily applicable method, which enables the researcher to assess the potential that a particular risk has of affecting people’s lives (da Silva et al., 2014; Magalhães et al., 2019). To the best of our knowledge, this is the first time the PRM has been applied to the perception of specific risks to human health (diseases).

The incidence index (Ii) corresponds to the ratio between the number of people who mentioned a certain disease (nr) and the total number of people in the sample (nj), represented by the equation: Ii = nr nj

The values assumed by Ii can vary between 0 and 1, which corresponds to the lowest and 1 corresponds to the highest incidence respectively.

Following the method proposed by Baird et al. (2009), the disease severity indexes (Sj) were calculated in two steps. Initially, the individual value of the risk severity index (Rij) was obtained by the equation: Rij = 1 − rij − 1 ni where ni represents the total number of risks mentioned by each informant (i), and rij represents the classification of each risk individually. Next, all the results found for Rij were summed and divided by the number of times the risk was mentioned as the most severe (Nj). Next, the severity index (Sj) was obtained, with values ranging between 0 (risks that are not considered severe) and 1 (risks considered extremely severe), using the following equation: Sj = ∑ i − 1 Nj = 1 − Rij Nj

Some of the obtained Ij and Sj results were represented using risk maps, graphical representations that allow the analysis of the space and distribution of incidence and severity of the mentioned diseases. As observed in the risk maps (Figures 6, 7), the upper part of the graph illustrates the first quadrant (where it is possible to observe the diseases classified as low incidence and very severe) and the second quadrant (where the diseases locally considered as high incidence and very severe appear); and the lower part displays the third quadrant (where the diseases perceived as low incidence with low severity are illustrated) and the fourth quadrant (where flu, a disease classified as not very severe, although with high local incidence is illustrated). Here, influenza is highlighted because it was the only disease mentioned in the fourth quadrant whose incidence was high. For this purpose, the PRM proposed by Baird et al. (2009) is the most recommended because the type of severity index (Sj) classification used facilitates both the elaboration and interpretation of risk maps (Baird et al., 2009) due to the similarity of the type of values presented by the incidence indices (Ii) and severity (Sj). For that reason, we opted for this model.

To calculate the incidence of adaptive strategies used to mitigate the diseases perceived by farmers, a similar logic to that used to calculate the incidence of risk (Ij), mentioned earlier, was adopted. The difference was to divide the number of adaptive strategies mentioned for each disease by the participants’ total number of adaptive strategies (see Supplementary Table S2). The indexes obtained in relation to the incidence of adaptive strategies (whose values could also vary from 0 to 1) were used in our analysis as dependent variables. The values of Ij and Sj were used as independent variables.

Subsequently, the data regarding perceived diseases were categorized (into direct and indirect diseases), and the mentioned adaptive strategies were classified (into therapeutic and prophylactic behavioral strategies) by the resident farmers of the Catimbau NAPAR (see Tables 1, 3) and Carão (see Tables 2, 4) regions to facilitate the explanation and discussion of our results. Finally, our hypothesis was tested for both areas using generalized linear models (GLM) with Gaussian distribution, in which the data were processed by calculating the logarithm (base 10) of n + 1 value to reduce the possible effects of large variation amplitudes. The statistical test was performed using the software R, version 3.6.3 (R Development Core Team 2019).

A total of 42 diseases related to drought and prolonged rainfall events were mentioned by farmers living in the Catimbau NAPAR. The farmers in Carão mentioned 31 diseases related to extreme climate change events. These diseases were listed in two categories based on their sources of transmission: direct (or primary) and indirect (subcategorized into secondary and tertiary), when transmission occurs through climatic phenomena, such as rain and heatwaves, and mediation of natural processes or systems (such as contaminated water or vectors) or human systems (fatigue caused by the need to walk long distances in the intense heat to work) respectively.

In both Catimbau NAPAR and Carão, most of the diseases mentioned were from direct transmission (n = 34, 26), followed by secondary indirect transmission (n = 6,4) and tertiary indirect transmission (n = 2, 1). Influenza stands out in the first category as the most frequently mentioned disease among residents of the rural communities of the Catimbau NAPAR (NC (number of citations) = 77) and Carão (NC = 39). The most mentioned diseases in the second category in both areas were chikungunya (NC = 23,15) and dengue (NC = 11, 10) (Table 3).

It is evident that all diseases perceived by farmers living in the two study areas are considered to have a low local incidence (except for influenza, whose incidence is very high, although with low severity; Figures 6, 7). Among these, diseases of low severity, such as dengue and cold predominated. Diseases perceived as having high severity were also mentioned, such as pneumonia and chikungunya, a disease of indirect transmission perceived as more severe.

Adaptive strategies were listed in two categories referring to the behavior adopted for developing the strategy: therapeutic behavior, subcategorized according to the origin of the behavior; and prophylactic behavior (Tables 4, 5). In the Catimbau NAPAR, 19 adaptive strategies were mentioned, including therapeutic (n = 10) and prophylactic behavior strategies (n = 9). In Carão, 14 adaptive strategies were mentioned by local farmers, including therapeutic (n = 8) and prophylactic behavioral strategies (n = 6).

Among the therapeutic strategies, the subcategory that obtained the highest number of mentions, both in the Catimbau NAPAR (n = 4) and Carão (n = 3), is related to the use of pharmaceutical drugs: NC = 112 in the Catimbau NAPAR used for the treatment of 30 locally perceived diseases; and NC = 80 in Carão, used for the treatment of 21 locally perceived diseases. Among the adaptive therapeutic strategies related to the local medical system, the number of citations referring to treatment with local herbal medicines is noteworthy: NC = 89, 63 in the Catimbau NAPAR and Carão, used for the treatment of 19 and 17 locally perceived diseases respectively. It is also worth noting that adaptive strategies were also cited related to individual care, such as resting (NC = 5, two in the Catimbau NAPAR, and Carão); and taboo/beliefs in Carão such as using “magic” medicine (NC = 2).

Among the adaptive strategies of prophylactic behavior (n = 9 in the Catimbau NAPAR and n = 6 in Carão), the most frequently mentioned was hydration (NC = 4 in the Catimbau NAPAR and NC = 9 in Carão). This category demonstrates the significance of strategies related to escape/evasion behaviors to avoid sources or vectors that transmit diseases.

The incidence and severity attributed to perceived diseases related to climate change explained the frequency of adaptive strategies used among farmers in both the Catimbau NAPAR (z = 5.221; p = 6.22e - 06) and Carão (z = 5.617; p = 5.15e—06; Tables 6, 7). Among the independent variables tested in our models, the incidence of perceived diseases had great predictive power in both the Catimbau NAPAR (z = 5.151; p = 7.77e - 06) and Carão (z = 5.288; p = 1.26e - 05). Severity did not demonstrate predictive power in the models, suggesting that severity may not influence people’s perception of diseases related to extreme rainfall and drought events that affect them the most, rather than potentiating them, as expected.