A semi-structured questionnaire (Supplementary Material 1) was used to conduct a survey among small-scale farmers between November 2016 and June 2017, in two important potato growing areas in the central Peruvian Highlands: Paucartambo, in the department of Pasco, and Ulcumayo, in the department of Junin. Both areas are known to have high LB disease pressure. Farmers' houses and their potato plots were located between 2,422 and 4,209 m.a.s.l. and all participants had agriculture as the main economic activity and potato as the primary crop. A draft questionnaire was validated previously with socio-economists and pathologists from the CIP and subsequently with 10 farmers from each area. Some questions were focused on characteristics of respondents: age, gender, education level and altitude. Others focused on main potato constraints, the causes of LB, criteria of selection of fungicides, frequency of fungicide applications, fungicide dosages, number of cultivars grown, identification of cultivars resistant and susceptible to LB, and yield losses (Table 1).

In total, 1,198 farmers were interviewed with the final version of the questionnaire. Farmers involved in this study were randomly selected from each place to reflect potential differences in potato production technologies, late blight management, and pesticide use. All respondents came from different households.

Responses were first assessed individually or in pairs in tabular or graphical form. Where possible, associations between factors were tested with Chi-square or Wilcoxon rank sum tests. The majority of responses were also assessed in a multivariate correspondence analysis (CA, Nenadic and Greenacre, 2007). CA is based on a contingency table, and to create this for the survey results, many responses leading to continuous scales were made into discrete classes (Supplementary Material 2). For the contingency table, age-gender classes (young men, young women, mid aged men, mid aged women, senior men, and senior women) were used as rows so that these respondent characteristics could be compared against management perceptions, which formed the columns of the table. Management options were often modified.

Two rows and several variables were treated as supplementary in the CA, i.e., their location on a biplot is estimated but they are not used to form the plot dimensions. The two rows were young women and senior women, because they had relatively few representatives and would have thus had strong effects in the CA. Supplementary columns included education levels low, medium and high, altitude levels low, medium and high, resistant cultivar as Yungay, resistant cultivar as other, susceptible cultivar as Canchan, susceptible cultivar as other, spray interval 1–8 days and spray interval more than 8 days. Education related columns were made supplementary because of their very strong correlation with age. The other columns were made supplementary because they had overly strong effects in the analysis but were not deemed that important by the survey designers. Statistical and correspondence analyses were done in R (R Core Team, 2018).

There were 569 and 542 respondents participating in the surveys in Paucartambo and Ulcumayo, respectively (Table 2). There was an even distribution of ages, spanning a range from below 20 years to 80 or more years in both locations (Figures 1A,B). Both males and females were represented across the age distribution, except in Paucartambo where there were no women below 30 (Figure 1A). Women made up 37.1% of the respondents in Ulcumayo, but only 6.8% in Paucartambo (percentages calculated from Table 1).

Education was strongly associated with age in both locations (Table 1). Lack of education occurred only in the older age groups and secondary education was associated primarily with the youngest age class. Fisher's exact tests for independence of age/gender class and education were highly significant in both Paucartambo (p < 0.0001) and Ulcumayo (p < 0.0001).

Based on frequency histograms, both female and male respondents were farming at all altitudes in both locations (Figures 1C,D), however, in general, potato production was generally done at higher altitudes in Ulcumayo. At that location, 93% of production occurred above 3,000 masl, while in Paucartambo, only 51% of production occurred above 3,000 masl (calculated from Table 2). Age/gender classes were less clearly associated with altitude (Table 2); however, Chi-square tests of independence of age/gender classes and altitude were significant for both Paucartambo (p = 0.0065) and Ulcumayo (p = 0.0218).

Nearly all respondents in both locations identified LB as the primary constraint to potato production, while Andean weevil was considered the second most important constraint in both locations (Figure 2A). Other constraints were identified by 15% or fewer of respondents and a Fisher's exact test of independence between location and choice of constraint was highly significant (p < 0.0001). Respondents in both locations also considered that LB is primarily caused by a combination of sun, rain and fog, while other factors were considered less important (Figure 2B). As a slight exception, 25% of respondents in Paucartambo considered rain alone to be most important (Figure 2B). There were no significant gender differences in their perceptions on the primary constraint to potato production (Figures 3A,B), nor in the causes of LB (Figures 3C,D). In both locations, over 80% of respondents chose one of two sources of information: agrochemical vendor or self, with the former slightly more important in Paucartambo and the latter slightly more important in Ulcumayo (Figure 2C). None of the other sources of information were chosen by over 10% of respondents. Chi-square tests, based on the first four choices in each case, indicated that responses in the two locations were significantly different for primary constraint, causes of LB and source of information. There was a significant difference in information sources between men and women respondents in both locations (Figures 3E,F). Women respondents tended to rely on information from agrochemical vendors rather than self. Extension agents and institutions are not a major source of information for women as well as men.

Respondents in Paucartambo and Ulcumayo identified 21 and 26 potato cultivars, respectively (Supplementary Material 3). Yungay was the most important in both locations (the 15 most widely grown are shown in Figure 4A). Huayro, Canchan, Chaulina, Peruanita, Tumbay, and Andina were also important but to a varying degree in each location. Some less important cultivars were grown only in one location. Locations were significantly different for cultivars recalled (p < 0.0001) based on Fisher's exact test of independence (Figure 4A). The profiles of the number of cultivars grown per farmer were similar in both locations, but there was a slight yet significant (p < 0.0001) tendency for farmers in Ulcumayo to grow more cultivars (Figure 4B).

Respondents in both locations chose Canchan as the cultivar most susceptible to LB (Figure 4C). Otherwise, the locations differed in their perceptions of susceptible varieties. The cultivar Yungay was chosen by over 50% of respondents in both locations as the most resistant to LB (Figure 4D). Chaulina, more popular in Ulcumayo (Figure 4A), was also chosen by over 30% of respondents in that location. Other than the first choice, and similar to the choice of susceptible cultivars, respondents in the two locations often had different views of resistant cultivars. The cultivar Huayro was chosen by respondents in Paucartambo as both the second most resistant and second most susceptible to LB. The cultivar Tomasa was the second most susceptible and fifth most resistant for respondents in Ulcumayo (Figures 4C,D). Location differences for selection of resistant and susceptible cultivars were significant (p < 0.0001), based on Fisher's exact test of independence.

In both locations respondents recalled well over 30 trade products, but also in each location about half were named by fewer than 1% of participants and a few participants gave products that are not fungicides (e.g., the insecticide Furadan, Figure 5A), but this was rare. In both locations the order of importance of trade products were relatively similar (Figure 5A). The trade name of a fungicide most frequently given was Antracol, recalled by nearly 50% of respondents in both locations. Dithane, the second most common, was given by slightly over 45% of respondents in both locations and Curzate, the third most common, by over 30% of respondents in both locations. After that, recalled trade names differed between locations. Only those named by at least 1% of respondents are presented (Figure 5A), and products that appear to be unique to one location (e.g., Manzate in Ulcumayo, Figure 5A), may have been present in the other location but were mentioned by fewer than 1% of respondents.

Similar numbers per respondent of fungicide trade names were recalled in both locations, producing nearly identical box plots (Figure 5B). However, respondents in Ulcumayo on average recalled a slightly higher number of fungicides at 2.76 than did respondents in Paucartambo, with 2.63 (data not shown), driven in part by a small number of people in Ulcumayo who recalled 5 trade fungicide names. The difference between locations was significant based on a Wilcoxon rank sum test (p < 0.0011). Locations also had similar response patterns for reasons for initiating LB control sprays, but here respondents primarily chose one of two, emergence or 20–30 days post emergence, with the former being most important (Figure 5C). Environmental triggers such as rainfall, were not frequently chosen. A Chi-square test for independence of locations was significant at p = 0011 as, proportionally, fewer respondents in Ulcumayo chose emergence as the primary trigger for spraying.

Locational differences were slightly significant (p < 0.000) for dose of fungicide used per 20 l sprayer based on Wilcoxon rank sum tests (Figure 5D). Respondents in Paucartambo tended to spray more frequently than did those in Ulcumayo, with a larger percentage from the former choosing spray scale levels 3 and 4, representing between 11 and 20 sprays per season (Figure 5E). The interval between fungicide sprays was approximately 1 week in each location but farmers in Ulcumayo tended to have a slightly longer interval and there was more variation among farmers in this location (Figure 5F). Locational differences were highly significant (p < 0.000) for both number of sprays and spray interval, based on Wilcoxon rank sum tests.

Respondents in Paucartambo estimated potential yield to be slightly higher than did respondents in Ulcumayo, with median values of 25 and 20 bags per ha (bag = 70 kg), respectively (Figure 6A), but estimated slightly lower percentage loss due to LB, if fungicides were not used (Figure 6B). In both locations, potential loss was estimated to be above 50%. Locational differences were highly significant (p < 0.0001) for both potential yield and yield loss based on Wilcoxon rank sum tests.

In both locations, there was a relatively strong association of factors based on gender. Women of all ages generally clustered together. Mid aged men in Paucartambo (Figure 7) and young men in Ulcumayo (Figure 8) were projected on the biplot relatively distant from other respondents of the same gender, but it was nonetheless possible to make some inferences about potential gender-based perceptions of disease and crop management practices, as there were several associations among respondent characteristics and management practices that were similar in both locations. For example, in both locations, men of all ages tended to be associated with basing the first spray on a calendar, recalling a relatively high number (3–7) of fungicides, recalling systemic fungicides, and relying on personal experience as a source of information. In both locations women tended to observe others for first spray cues, recalled fewer fungicides, generally recalled contact fungicides and got information from outside sources. Some perceptions were not consistent across locations. For example, cultivar Yungay was considered resistant by women in Paucartambo and by men in Ulcumayo. A number of other perceptions were not clearly associated with gender. There was also no clear pattern associated with the ages of respondents.

Similar to a number of other studies in highland tropical locations (Kromann et al., 2009; Monteros and Delgado, 2021), respondents in both locations considered LB to be the primary constraint to potato production. As mentioned, and discussed in more detail below, LB is particularly difficult to manage in the highland tropics where potato or other wild hosts of P. infestans are present year-round. The difficulty of LB in the survey regions is also implied by the large number of fungicide sprays per season (Figure 5E). Increased capacity building for LB management and/or new technologies such as resistant varieties could provide major benefits for resource poor farmers in these conditions. Participants were not asked directly about interactions with public or private extension services, but these institutions were not frequently cited as sources of information for LB management. Most respondents either relied on their own knowledge or consulted agrochemical vendors (Figures 3E,F). Thus, either there had been few or no previous interventions for improving LB management in the communities or they did not have lasting impact. This issue could be addressed in a future intervention by offering refresher training courses for extension educators, including female extension workers, with participatory farmer field school approaches (Ortiz et al., 2004).

The responses of participants to questions related to fungicide use (Figure 4) demonstrated some commonalities across the two locations. Most of the fungicides recalled were older chemistries. According to the fungicide table of the Euroblight Network (EuroBlight, 2021), the active ingredients used in the trade products that respondents recalled were all released before 2007, and many of the new active ingredients were not mentioned in the responses.

The most recalled trade product in both locations was Antracol, which contains the contact fungicide propineb. Beyond that, most of the commonly recalled products contained mancozeb, a broad-spectrum contact fungicide released in the 1940s, which, when all fungicides are considered, is certainly the most common active ingredient used in both locations. Environmental and human health risks of excessive use of mancozeb have led to increasing pressure from environmental groups to ban the fungicide (Reuveni, 2021) and the global potato industry is anticipating the loss of mancozeb in the near future (Anonymous, 2021). The third and fifth most frequently recalled products contain the translaminar fungicide cymoxanil, and overall, 30% of products recalled contain a translaminar or systemic ingredient (data not shown), indicating a relatively frequent use of these more effective and more expensive compounds. It is not known, however, whether respondents understood the difference between contact fungicides and those with more mobility in plant tissue. Reasons for the apparent gender association with contact and systemic/translaminar products are also not clear. Overall, the question remains whether newer fungicides are available in the communities, and if so, are they not used because of the high cost or because farmers are not aware of their efficacy.

Survey results gave indirect evidence of the difficulty of managing LB under highland tropical conditions. In both locations the majority of respondents said that they initiate fungicide spraying at emergence, with the second most important period between 20 and 30 days after planting (Figure 5C). This indicates that farmers consider that plants could be infected at emergence or at least in very early stages of growth and thus crops must be protected during the majority of the growing season. Other studies in the Andean region have also found that very early infection is a common threat due to year-round climate propitious to potato growth (Garrett et al., 2001). Research has also demonstrated that it is possible for potato sprouts to be infected prior to emergence by aerially deposited spores of P. infestans (Kromann et al., 2008). The predominance of fungicide application at or soon after emergence may also indicate a lack of knowledge or equipment that would permit farmers to use weather-based indicators. However, if inoculum is present at emergence, there would seem to be no alternative to early application of fungicides.

The gender effect evident in both CA's confirms previous studies that show that women and men farmers often have different perceptions of management issues (e.g., Kawarazuka et al., 2020). This survey identified important gender differences, such as men tended to rely on personal experience in their decisions on LB management. Men also tended to spray a relatively higher number of fungicides than did women farmers. Men's reliance on their experience might be associated with their limited training opportunities or little trust in extension services. Therefore, specific training and visible evidence through demonstration fields may induce men as well as women to change their practices to utilize scientific information such as DSS instead of following their own decisions. Reaching women farmers is another challenge. While women play significant roles in potato production in Peru, potato farmer organizations are dominated by male farmers with a small number of women who tend to be highly educated and hold land titles (Patel-Campillo and García, 2018). Therefore, providing LB training through potato farmer organizations may not be effective for reaching many women farmers. The literature in other regions shows that agricultural programs with a clear target for women farmers can significantly increase women's access to information on crop management (Akter et al., 2016; Mudege et al., 2016; Williams and Taron, 2020). Offering tailored training to meet women's specific needs is our priority. To effectively reach women, alternative approaches, such as engaging with women's social groups through NGOs can be tested in the Peruvian Andes.

The questionnaire used in this study gave a good overview of farmer perceptions related to LB management practices. The questionnaire could potentially be improved by linking it to multi-indicator surveys (e.g., Hammond et al., 2017) where social scientists, pathologists, agronomists and occupational health workers can include questions to capture information related to potato crop management. An improved questionnaire could eventually become a tool for simplifying and facilitating baseline studies where LB is important.

Any recommendations for future interventions should take into consideration that the occurrence of the COVID-19 pandemic has dramatically changed training for farmers. Sessions programmed in Farmers Field Schools (FFS), field days and local meetings have recently been canceled to reduce physical contact to a strict minimum to avoid spreading the virus SARS-CoV-2. In this context, and although there are still many challenges in remote areas to access and utilize internet for agriculture, digital apps are being prioritized have a great potential, especially with young farmers, to transform extension in areas where internet is available.