Area of study

The Lake Patzcuaro region is located in the heart of the Trans-Mexican Volcanic Belt, in the State of Michoacan. The population and economic activities of this natural and cultural region are complex and heterogeneous. Indigenous localities, and mixed-raced cities and towns share space and primary production activities mix with tourism in the same landscape (^{Mapes et al., 1994}). The rural localities’ main economic activities are agriculture, cattle raising, handicrafts, and fishing (^{Castilleja, 1992}). The region is made up of the municipalities of Patzcuaro, Erongaricuaro, Quiroga, and Tzintzuntzan. The municipal capitals are the main population centres. The city of Patzcuaro stands out as the region’s economic center. The average human development index for the four municipalities (0.672) is lower than the national index (0.739). Patzcuaro’s index (0.695) is higher than the other three (^{PNUD, 2014}). Indigenous population accounts for 20 % of the overall population. Erongaricuaro is the municipality with the highest percentage (40 %) of indigenous population (^{CDI, 2010}).

The region is dominated by volcanic mountains, intermontane valleys, foothills, and lowhills. The lake surface is located at 2040 meters above sea level and the highest peak reaches 3400 meters above sea level (^{Barrera-Bassols, 1992}). The weather is temperate sub-humid and it rains in the summer. The overall average rainfall is 1100 mm per year. The average yearly temperature is 16.9 °C, with a minimum average of 8 °C and a maximum average of 25.7 °C (^{CICESE, 2015}). The vegetation in the lowhills is made of pine trees and holm oaks. Agriculture is carried out in the plains around the lake and in the valleys. The agricultural landscapes defined by ^{Mapes et al. (1994)} and ^{Astier et al. (2010)} are associated with the type of soil: a) andosols are residual moisture rainfeed agriculture landscapes where long-cycle maize varieties are sown early; b) acrisols and lithosols are rainfeed agriculture landscapes where short-cycle maize is sown when the rain seasons starts; c) vertisols are lake shore agriculture landscapes or ever humid lands on the bank of the lake; and, d) luvisols are irrigated agriculture landscapes (^{INEGI, 2013a}).

Agriculture is mainly carried out in small production units, with an average area of 3.7 ha (^{INEGI, 2007}). The main annual crops are maize (Zea mays subsp. mays), oat forage (Avena sativa), wheat (Triticum aestivum), beans (Phaseolus vulgaris), squashes (Cucurbita pepo), fig-leaf gourd (C. ficifolia), lentils (Len culinaris), and common vetch (Vicia sativa). Maize covers 8115 ha (71 %) of the cultivated area. In general, farmers sow their own seed; hybrid maize is hardly sown. The most important perennial crops are avocado (Persea americana) and alfalfa (Medicago sativa) (^{SIAP, 2015}).

Field work

From 2012 to 2015, 113 interviews were conducted with farmers from 29 localities in the Lake Patzcuaro region. The samples were selected using the method proposed by ^{Hernández-Xolocotzi (1972)}: maximizing the gathering of samples from different maize types with the lowest number of interviews. The interviews ended when every kind of maize recognized by each community had been collected. The interviews were specifically focused on those farmers that the community recognized as the ones who sow the highest quantity and quality of local maize races. Crop management and smallholding location were also recorded. In 2012, a minimum of six corncobs were collected; they were chosen by the farmers and their race was identified by José Alfredo Carrera, ScD. In 2015, farmers were interviewed, but no samples were collected. The technical team classified the maize samples at the farmers’ homes. Most of the local and native maize populations had characteristics from two or more races. In order to facilitate the analysis, the dominant race was used. Samples with mixed characteristics that the team was not able to classify were placed in the undetermined category. The GPSlogger Android app was used to georeferenced the smallholdings. Geographical data was processed using ArcMap 10.1 and were combined with the survey’s data.

Data analysis

Data were analyzed per region, locality, and production unit. In the first case, a geographical information system was designed. It included the following elements: 1) elevation digital model, developed based on the level curves of topographic maps (scale: 1:50 000; ^{INEGI, 2015}); 2) vector layer of soils, series II by (^{INEGI 2013a}) (scale: 1:250 000); 3) layer of villages and towns from the geostatistical framework, version 6.2 (^{INEGI, 2014}); 4) layer of land use based on series V (^{INEGI, 2013b}), which allowed us to define the moisture system (irrigation, rainfeed, moisture); 5) layer of water sources from the topographic maps (scale: 1:50 000; ^{INEGI, 2015}); and 6) vector layer of the location of the maize races, based on the surveys.

The database of maize races and their location was made up of 402 observations with the following fields: race, locality, municipality, grain color, use, type of soil (according to the farmer), moisture system, cultivation system, type of soil (^{INEGI, 2013a}), agri-environment, local variety (local name of the native maize population), altitude, and slope. A map was designed to represent maize race distribution and those races were qualitatively associated with the variables, by means of a visual exploration. The location of the collections was adjusted in the printed map, in order to reduce the superimposition of points. Labels were created for each point in order to make this possible; those points were moved in order to obtain minimum required distance. For each of the races with more records (Purhépecha, Cónico, Elotes Occidentales, Ancho, Chalqueño, and Elotes Cónicos), a general linear model was developed and those races were then associated with environmental variables. A binomial model with a logit link function was used. A separate model was made for each race (with R version 3.3.1; ^{Core Team R, 2016}); a point was marked in the map where a race of interest was located. The variables used to model the presence of a race were altitude, type of soil, and slope. The moisture system was not included, due to its association with the type of soil (X²=435.3, p=0.000). The residual moisture system matched andosols, the rainfeed agriculture system matched acrisols, and the irrigated agriculture system matched luvisols.

A local analysis was based on the information of eight variables that set each community apart (Table 1). Localities where only one or two farmers were interviewed were excluded, including the cities of Patzcuaro, Quiroga, and Erongaricuaro. Overall, 18 out of 29 localities in which interviews were carried out were included. Therefore, race abundance, local varieties abundance (number of local names of maize populations tilled in the community), total population, indigenous population, average altitude, abundance of agriculture soil types, and the Shannon diversity index were analyzed using the principal component analysis (PCA) with R (^{Core Team R, 2016}). The Simpson index was excluded from this analysis, because it was significantly correlated with the Shannon index (R=0.98; p=0.000). Ethnographic information was included in the PCA in order to characterize localities with greater abundance of maize types and to explain diversity variation per locality.

Table 1 Variables included in the analysis per locality and per production unit. 

^†Excluding localities with one or two interviews. ^¶Excluding (12) interviews with incomplete data. Overall, 101 interviews were used in the analysis.

Production unit analysis was carried out using the Poisson model, with a logarithm link function. The variable to be predicted was the number of local names for the native populations. This variable was chosen instead of race abundance, because the average number of races per production unit was low (mean: 1.96 races per production unit). Not only was the number of local names greater, but it also had a practical meaning for farmers, because each name is the basic unit they use to classify maize in their locality. The dependent variable was modelled with 19 variables of the production unit (Table 1). One of the evaluated models was chosen based on residual variance and Akaike’s Information Criterion (AIC). A subsample of units with more than five maize populations was chosen to complete the analysis of the production unit, and each production unit was described using the survey’s data.

Diversity distribution on a regional scale

The 11 maize races that were detected showed a typical distribution of species frequency: few races with high frequency and many races with low frequency. The dominant races were Purhépecha (170 records) and Cónico (72). Elotes Occidentales (34), Ancho (26), Chalqueño (20), and Elotes Cónicos (20) showed average frequency. The races with very low frequency were Mushito (9), Pepitilla (8), Cacahuacintle (5), Tabloncillo (3), and Palomero Toluqueño (1). A population that came from a hybrid variety was recorded, and 33 records were not associated with any particular race, due to their undefined characteristics.

The race distribution map showed several patterns (Figure 1). It seems that the Purhépecha races dominated a south-west strip in the highlands, which included the Casas Blancas, Opopeo, Zirahuén, Santa María Huiramangaro, and Pichátaro localities. This strip is dominated by residual moisture lands and its soils have high moisture retention capacity; therefore, they are sown before the rains start (^{Mapes et al., 1994}; ^{Astier et al., 2012}). Another pattern showed that the Cónico, Elotes Occidentales, and Elotes Cónicos races dominate in the lowlands of the northwestern bank of the lake, in the localities of Uricho, Napízaro, San Andrés Ziróndaro, and Jerónimo Purenchécuaro. These are rainfeed agriculture lands, with luvisol and acrisol type soils. Fast growth races are sown at the beginning of the rain season. Another agri-environment is irrigated lands, which showed a high diversity of long- and short-cycle races. The greater diversity seems to be located in the lowlands. Three lowland areas have high diversity: the first is located in San Jerónimo Purenchécuaro and San Andrés Ziróndaro, north of the lake; the second is located between Napízaro and Uricho, southeast of the lake; and the third is located in the irrigation zone between Tzurumútaro, El Jagüey, and Nuevo Rodeo, east of the lake.

Figure 1 Distribution map of maize races in the Patzcuaro region. 

Association with type of soil, altitude, slope, and agricultural environments

The analysis of the correlation between races and agro-environment only included Purhépecha, Cónico, Elotes Occidentales, Ancho, Chalqueño, and Elotes Cónicos, because they were the most frequent races. The other five races were not modelled because they had a significantly low frequency.

Binomial models for race distribution indicate that the Purhépecha race is associated with an average altitude of 2260 m, frequently located in andosol soils with minimal slope. The Cónico race was frequently associated with environments in the low part of the basin, but there’s a weak association between this race and altitude. We could not define an association with the plot’s slope, because it was located in valleys and low hills. Due to this race’s association with acrisols and andosols, it seems to be cultivated during the rainy season and with residual moisture (Tables 2 and 3).

Table 2 Results of the binomial model (link=logit) to describe the association of races with environmental factors^†. 

^†Races were modelled separately. *** Pr≤0.001; ** Pr≤0.01; * Pr≤0.05; ^¶Pr≤0.1.

Table 3 Distribution variable intervals of the maize races in the region of Patzcuaro, Michoacan, Mexico. 

The Elotes Occidental race was associated with the lowlands and seemingly with acrisols and luvisols; however, the effect of the type of soil was not significant in the model, probably because the number of records for this race (34) was relatively low. This race had a greater frequency in lands with more slope, like the low hills around the lake. The Ancho race was associated with the low hills in the lower part of the basin and was not directly associated with a specific type of soil. The Chalqueño race was associated with lands with low slope and average altitude, and it was not associated with any kind of soil. The Elotes Cónicos race was more common in low lands, but was not directly associated with the slope or the type of soil. The Ancho, Chalqueño, and Elotes Cónicos race models provided few explanations about overall variance, because few records were available. An increased number of cases can generate stronger models (Tables 2 and 3).

Model adjustment changed from one race to another. The best explanation of data variation (13 % of the total variance) was provided by the model that included the Purhépecha race. The Cónico race model accounted for only 3 % of the variance, because few records were available for it. Additionally, this variance allowed us to suggest that this race is distributed in most of the region’s environments. The analysis of molecular markers usually indicates a greater variation from one population to another, rather than within each population (^{Vigouroux et al., 2008}; ^{Orozco-Ramírez et al., 2017}). The high genetic variability between and within populations enables races to have greater adaptation scope.

The altitude distribution for some of the races identified in the Patzcuaro region exceeded the interval reported by ^{Ruiz et al. (2008)}. According to them, the Elotes Occidentales race was found between 700 and 2170 m; however, our study indicates that this race was distributed up to 2675 m in this region. Our report indicated that the Ancho and Pepitilla races could be found 360 and 400 m above the distribution reported in that study. Meanwhile, the Palomero Toluqueño race was found below the distribution reported by Ruiz and his collaborators.

The races showed great adaptability, because they were found in all agri-environments in the region and we were able to distinguish the localities and the socio-environmental conditions under which some races are most frequently found. Farmers distinguish mainly between maize types for rainfeed systems and for residual moisture systems. The main variation is the length of their cultivation cycles. The highland farmers usually use seeds from the lowlands for the second sowing, when the moisture crops fail, because they know that lowland seeds have shorter cycles.

Local diversity distribution

The abundance, and the diversity of races and local names for native maize did not have any significant association with the variables that were selected according to the locality. Poisson models were discarded because they did not explain the variance. In contrast, PCA per locality indicated that localities with five or seven races shared characteristics, such as altitude (medium to low) and human population size (low to medium). Pichátaro does not fit this pattern, as a result of its abundance of races, big population, and high elevation. The first and second main components (CP) accounted for 69.5 % of the variance (Figure 2). The variables that were more important in the definition of the first CP were Shannon diversity index (0.57 rotation), race abundance (0.55), number of types of agricultural soils (0.42), and medium altitude (-0.38). The most important variables to determine the second CP were the number of local names for native maize types (-0.55), indigenous population (-0.48), and total population (-0.42).

Figure 2 Biplot of the main component analysis of the localities. 

The communities with more races were El Zapote (7), San Andrés Tziróndaro (7), Arócutin (6), San Francisco Uricho (6), Tzurumútaro (6), Napízaro (5), and San Francisco Pichátaro (5). The seven communities showed partially different environmental and socio-economic characteristics; therefore, we were unable to define a high diversity profile for these localities (Table 4).

Table 4 Characteristics of the localities with greater maize race diversity in Patzcuaro, Michoacan, Mexico. 

Analysis per family unit

Among the models used to predict the number of native maize names per production unit, we chose one that only included 3 out of the 19 proposed variables. The variables that provided a better explanation for the dependent variable were total maize area, number of smallholdings, and head of family that speaks an indigenous language. The model’s AIC with the 19 and the 3 variables were 352 and 324 and accounted for 50.3 and 43.4 % of the data variability, respectively.

Based on this model, these three variables have a positive effect on the number of native maize names (Table 5). According to the model, the effect of each factor -when the other factors remain constant- is the following: 1) the number of native maize names will increase by a factor of 1.03 as maize cultivation per hectare increases; 2) the number of names will increase by a factor of 1.11 for each smallholding that is added; and 3) the number of names will increase by 43 % if the farmer speaks Purhépecha.

Table 5 Poisson model results for the number of native maize names per production unit, in Patzcuaro, Michoacan, Mexico. 

*Significant p value.

Out of 113 farmers interviewed, only seven had five or more maize names and 3 production units were located in Pichátaro, 2 were located in Napizaro, 1 was located in Nuevo Rodeo, and 1 more was located in Uricho. Four out of this seven farmers speak Purhépecha and they were 51-66 years old. The farmer who mentioned 11 names or types of maize said that he sows them in 11 smallholdings (28 ha), that he had not gone to school, and that he also raised cattle. The others reported sowing 3-5 smallholdings in 2-11 ha. Six said they had gone 3 years to school and only one had higher education. All of them reported another activity (cattle-raising, handicraft production, or forest-related). The farmer with the highest education was a primary school teacher, he was the youngest, and had used his oldest seed for the shortest time (10 years). The other farmers had used the same seed for an average of 25 years. Some farmers have used the same seed for over 50 years.

The minor adjustment of the binomial model for the regional presence of the Purhépecha, Cónico, Elotes Occidentales, Ancho, Chalqueño, and Elotes Cónicos races may indicate that the altitude, slope, and type of soil variables are not very relevant to explain the races’ presence. Additionally, this can be the result of the close variation interval of the variables in the region under study. Therefore, the close environmental gradient does not affect the possibility of cultivating one race or another. Expanding the zone under study would be necessary to prove this hypothesis. It is also possible that farmers prefer the Purhépecha and Cónico populations, which were the most frequent races. Although the Purhépecha race was associated with humid lands and the Cónico race was associated with rainfeed agriculture lands, they were not exclusive to those environments. Therefore, these seem to be the most essential maize types consumed in the region. The other maize types were complementary and are sown in small areas, few records about them were obtained, and the models had a low adjustment. This dominant and secondary varieties pattern is common in México (^{Perales et al., 2003}).

The Poisson models used to explain race abundance and the names of local populations per race did not explain the variance. Although PCA accounted for 69.5 % of the variance, the joint results with the Poisson models showed that PCA did not reveal any pattern that allowed us to explain the community-scale abundance and did show which variables had a greater variation.

The family unit provided a better model of the abundance of names in local populations. The variables selected for the final model were: total maize area, number of smallholdings, and indigenous language; it accounted for 43.4 % of the maize abundance. One of the reasons to keep a diversity of local varieties is the diversity of cultivation environments, as expressed by soil diversity (^{Bellon and Taylor, 1993}). Farmers in the basin of Lake Patzcuaro who had sown maize in more smallholdings in greater areas had a greater environmental diversity. Therefore, they required more maize types.

The positive quantitative association between indigenous culture and agrobiodiversity in the production unit has been proven: such is the case of indigenous farmers compared with mixed-raced farmers in an altitude transect in Chiapas (^{Brush and Perales, 2007}). In the Andes, it became evident that farmers with indigenous cultural roots invariably keep a greater agrobiodiversity in their fields. In this case, such association was made clear by the use of indigenous clothing and language and consuming traditional elements (^{Skarbø, 2014}). There is a close bond in the region between maize diversity and community life: the farmers’ world view, the localities’ holiday calendar, and other cultural aspects proves this (^{Mapes et al., 1994}; ^{Orozco and Astier, 2017}). The dishes that are prepared during these festivities, which go back to ancient rituals, frequently include maize in various ways (^{Castilleja et al., 2003}).

It seems that the factors that influence diversity are not the same for all scales. Diversity is associated with different variables, at the regional, local, and production unit scales. On a regional scale, we can observe high and low maize abundance zones. The following zones were outstanding: one zone in the north and center-west had high density and most of the communities had high indigenous presence; the east zone also had high density and some of its towns had a majority of mixed-race population. Altitude, slope, and type of soil (or moisture system) explained the distribution of races in these zones. Some races are typical and dominant in certain agro-environments. Local scale provided information about the existence of communities with high diversity which share bio-physical and demographic conditions, except for one community where other patterns prevail. In the production unit scale, abundance/diversity and the type of races/varieties at home resulted from such variables as area, number of smallholdings, presence of animals, and significantly the socio-cultural and ethnic characteristics of the farmer and his or her family.