Agrarian Sciences Journal
Spatial distribution of coffee crop productivity in Minas Gerais over time
Renan Serenini
, Patrícia de Siqueira Ramos
, Lincoln Frias
Brazil is the world’s largest coffee producer and the state of Minas Gerais is responsible for half of the Brazilian pro-
duction. However, productivity is unevenly distributed throughout the state. Therefore, the purpose of this study is to
analyze the spatial distribution of coffee productivity in Minas Gerais from 2002 to 2017, a valuable information to
identify regions where coffee production may be more promising in the future. This paper investigates the existence
of spatial dependence of productivity between regions (using Moran’s I), its dynamics throughout the period and the
presence of clusters of high and low productivity (using local Moran’s I). The results show that the spatial dependence
of productivity was stronger from 2002 to 2009 than between 2010 and 2017. Some regions with small coffee areas
but high productivity have stopped producing the crop whereas some of those with large areas but low productivity
increased their productivity levels. Therefore, there is a tendency of homogenization of productivity in Minas Gerais,
with values close to 30 bags per hectare.
Key-words Moran’s I. Spatial autocorrelation. Spatial Statistics.
Distribuição espacial da produtividade de café em Minas Gerais ao longo do tempo
O Brasil é o maior produtor mundial de café e o estado de Minas Gerais é responsável por metade da produção brasi-
leira. Entretanto, a produtividade é distribuída de forma desigual no estado. Assim, o objetivo deste estudo é analisar
a distribuição espacial da produtividade de café em Minas Gerais entre 2002 e 2017, uma informação valiosa para
identificar regiões em que a produção de café possa ser mais promissora no futuro. O artigo em questão investiga
a existência de dependência espacial na produtividade entre regiões (usando o I de Moran), a dinâmica ao longo
do período estudado e a presença de agrupamentos de alta e baixa produtividade (usando o I de Moran local). Os
resultados mostram que a dependência espacial era mais forte de 2002 a 2009 do que entre 2010 e 2017. Algumas
regiões com pequenas áreas de plantio de café mas com alta produtividade deixaram de produzir enquanto regiões
com grandes áreas mas baixa produtividade aumentaram seus níveis de produtividade. Assim, existe uma tendência
de homogeneização da produtividade em Minas Gerais, com valores em torno de 30 sacas por hectare.
Palavras-chave: Autocorrelação espacial. Estatística Espacial. I de Moran
Sapienza University of Rome. Rome, Italy.
Universidade Federal de Alfenas. Varginha, MG. Brasil.
Universidade Federal de Alfenas. Varginha, MG. Brasil.
*Autor para correspondência:
Recebido para publicação em 10 de novembro de 2020. Aceito para publicação em 11 de dezembro de 2020.
e-ISSN: 2447-6218 / ISSN: 2447-6218 / © 2009, Universidade Federal de Minas Gerais, Todos os direitos reservados.
Serenini, R. et al.
Cad. Ciênc. Agrá., v. 12, p. 01–10,
In 2017, Brazil produced 32% of all the world’s
coffee, thus being the largest producer of this crop ac-
cording to the International Coffee Organization (ICO,
2018). In that year, the state of Minas Gerais was res-
ponsible for 53% of the Brazilian production (CONAB,
2018). Therefore, the study of coffee production in this
state is fundamental to understand coffee production in
Neither production nor productivity are evenly
distributed throughout the state. This is a recurring phe-
nomenon in agriculture, given that the development of
agricultural crops relies on factors that vary spatially.
These include both natural factors (soil fertility, tem-
perature, rainfall etc.) as well as logistic factors (roads,
warehouses, access to technical assistance, cost of inputs,
competition with other crops etc.) (Teixeira; Bertella,
2015; Souza; Perobelli, 2007).
In view of this, the purpose of this paper is to
analyze the dynamics of the spatial distribution of coffee
productivity in Minas Gerais from 2002 to 2017. The
aim is to verify whether coffee productivity is randomly
distributed in the state or whether there are patterns of
spatial distribution of regions with similar levels of pro-
ductivity. For example, the most efficient regions may tend
to be concentrated in the south of the state. The paper
also describes how this spatial distribution changed over
time, identifying regions which increased or decreased
their productivity levels.
This kind of information is useful in identifying
the most promising areas for coffee production. Also,
it could help in the improvement of the official harvest
forecasts (Neves; Luiz, 2006). The Companhia Nacional
de Abastecimento (CONAB) monitors the Brazilian coffee
crops, publishing regular forecasts of harvested area, pro-
duction and productivity, using, among other methods,
satellite images of statistically selected regions (CONAB,
2018). The results of the spatial analysis are helpful in
the sampling design of those regions.
The innovative contribution of this work is two-
fold. In contrast to previous studies, to be summarized
in the next section, this paper examines all the years
from 2002 to 2017, not only two or three points in time.
Secondly, it employs the new Brazilian regional division,
published by the Brazilian Institute of Geography and
Statistics (Instituto Brasileiro de Geografia e Estatística,
IBGE) in 2017. This revised division improves the old
micro and mesoregions division by taking into account
the urban network and the management flows. The in-
corporation of this new and more rigorous division in the
spatial analysis may reveal patterns previously masked.
Spatial analysis has been widely used in studies
of crop production and productivity. Souza and Perobelli
(2007) examined the spatial distribution of coffee pro-
ductivity in the 558 Brazilian microregions in 1991, 1997
and 2003. The values of Moran’s I were 0.30 in 1991,
0.34 in 1997 and again 0.34 in 2003, always statistically
significant. Therefore, the authors confirmed the existence
of positive spatial correlation, persisting throughout the
studied period.
Almeida et al. (2006) examined the spatial dis-
tribution of the coffee productivity in Minas Gerais for
the years 2000 and 2004 taking as units the 66 microre-
gions of the state. Productivity was measured as the ratio
between the sum of the production of all the cities in the
microregion and its total harvested area. The study not
only found spatial autocorrelation in coffee productivity
in both years but also that it increased between them, as
the Moran’s I went from 0.24 in 2000 to 0.32 in 2004.
The clusters with higher productivity were located in the
northwestern region of the state. According to the authors,
the rising concentration suggests that higher productivity
could be influencing the emergence of regions of equal
performance in their neighborhood.
The remainder of this paper is organized as follo-
ws. Section 2 presents the dataset, the kind of geographic
units under study and the central concepts in exploratory
spatial data analysis. Section 3 contains the results and
discussion of both the descriptive and the spatial analysis.
Finally, some concluding remarks are given in Section 4.
Materials and methods
This study uses data from the Municipal Agri-
cultural Production (Produção Agrícola Municipal, PAM)
survey (IBGE, 2019), ranging from 2002 to 2017. The
data begins in 2002 because this is the year when coffee
measuring unit changed from dry cherry coffee (café em
côco) to processed coffee or coffee beans (café beneficiado
ou em grão). The data is available through IBGE System of
Automatic Recovery System (SIDRA), table 1613 (“Area to
be harvested, harvested area, quantity produced, average
productivity and production value of permanent crops”).
In this study, productivity is specified as the ratio
between the total production volume (in 60kg bags) and
the total area harvested (in hectares) in a certain year.
High productivity regions are those that produce more
than 25 bags per hectare (1,500 kg), the median value
among the immediate geographic regions 2017.
In 2017, from the 853 Minas Gerais munici-
palities, 354 (42%) did not produce coffee. A variable
with a great number of zeros creates some problems for
the spatial analysis. Thus, instead of municipalities, the
adopted unit of analysis are the immediate geographic
regions, the updated equivalent of microregions. Consi-
dering the 70 immediate geographic regions in the state,
only 7 did not produce coffee.
Spatial distribution of coffee crop productivity in Minas Gerais over time
Cad. Ciênc. Agrá., v. 12, p. 01–10,
The division of Brazilian states into mesoregions
and microregions was used by IBGE from 1989 to 2017.
IBGE published in 2017 a new framework for the regional
division of Brazil, rearranging the former mesoregions
and microregions into intermediate and immediate geo-
graphical regions, respectively (IBGE, 2017). Comparing
the two divisions, Minas Gerais has 12 mesoregions, but
13 intermediate regions and, at the lower level, to the
66 microregions correspond 70 immediate regions.
Exploratory spatial data analysis is the set of
techniques used to describe and examine spatial data
(Anselin, 1996). The central problem in this kind of study
is to check whether or not the data is randomly distributed
in space. In this context, the opposite of randomness is
spatial dependence (or spatial autocorrelation) between
the observations, meaning that neighboring regions are
more closely associated with each other regarding the
value of a determined variable than with distant regions,
either positively or negatively.
The fundamental statistics in an exploratory
spatial data analysis are global and local autocorrelation.
The first requirement in their calculation is the definition
of a spatial weights matrix (W) (Almeida, 2012). This
is a n-square matrix whose elements denote the spatial
connection (neighborhood) between the regions accor-
ding to some criterion of proximity, such as contiguity
or distance, representing for each i region its spatial
relationship to each other region j. The present paper
uses contiguity as the proximity criterion. In this case,
the matrix element w
has value 1 when the regions share
a border and 0 otherwise. This paper adopts the queen
criterion of contiguity, which defines as contiguous all
spatial units sharing a boundary, in any direction, with
the region in focus (Almeida, 2012).
Results and discussion
In 2002, 21.68 million bags of coffee were pro-
duced on 1.09 million hectares, resulting in 19.9 bags
per hectare. In 2017, 24.25 million bags were produced
in 925 thousand hectares, resulting in a productivity of
26.2 bags per hectare. The increase in production (11.9%
from 2002 to 2017) was, in relative terms, slightly less
relevant for the increase in productivity than the reduction
of the area (15.1% from 2002 to 2017).
Coffee productivity in the state has increased over
time, representing an improvement in efficiency in coffee
production, measured as the division of the production in
the state by the area harvested in each year. In 2002, the
average production was 20 bags per hectare, whereas in
2016, the last year with positive biannuality, 29.4 bags/ha
were produced (in 2017, 26.2 bags/ha). The increase in
productivity was not only due to a reduction in harvested
area (keeping production constant) or an increase in
production (keeping harvested area constant), but rather
a consequence of both factors: increased production and
reduced harvested area throughout the state of Minas
Moving from the entire state of Minas Gerais to
its subdivisions, Table 1 gives an overview of the distri-
bution of the harvested area, production and productivity
among the 70 immediate geographic regions of Minas
Gerais in 2002, 2010 and 2017.
Table 1 – Descriptive statistics of harvested area, production and productivity of the immediate geographic regions of
Minas Gerais (2002, 2010 e 2017)
year mean standard deviation minimum maximum
harvested area (ha)
2002 15,522.4 22,694.1 0.0 115,949.0
2010 14,665.9 21,564.9 0.0 117,440.0
2017 13,215.8 21,693.1 0.0 121,277.0
production (bags)
2002 309,768.8 483,054.0 0.0 2,419,200.0
2010 358,140.0 560,156.9 0.0 2,756,133.3
2017 346,376.0 548,525.4 0.0 2,648,583.3
productivity (bags/ha)
2002 17.1 9.0 0.0 47.8
2010 21.8 11.6 0.0 64.5
2017 23.0 10.6 0.0 42.5
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019).
Serenini, R. et al.
Cad. Ciênc. Agrá., v. 12, p. 01–10,
The average harvested area decreased 15% from
2002 to 2017. On the other hand, the average production
increased 12% in the same period. The minimum value
of both harvested area and production has remained at
zero since at least one region did not produce coffee in
each one of the years. The maximum value of harvested
area had a slight increase. However, production decreased
from 2010 to 2017.
Despite this recent fall in production, average
productivity shown a steady increase. In 2010, a redu-
ced harvested area and a larger production improved
productivity from 17.1 to 21.8 bags per hectare. A new
increase, although smaller, was seen in 2017 as average
productivity reached 23 bags per hectare. So, whereas
the first period saw a 27% improvement, it represented
only 5,5% in the second period.
In 2017, the distribution of productivity among
the immediate regions was as follows: 25% of the regions
had productivity of up to 18 bags/ha, 50% up to 24.7
bags/ha and 75% up to 29.7 bags per hectare.
To begin to understand the spatial dimension
of the data, Figure 2 includes choropleth maps of coffee
productivity by immediate region in Minas Gerais from
2002 to 2017. These kind of maps, also called thema-
tic maps, color the region according to the level of the
variable under study. In the present case, the threshold
values of productivity was based on 2017 data, the last
of the series, selecting the values which divided the ob-
servations into six quantiles of approximately equal sizes
(containing 11 or 12 observations each).
According to Figure 1, the spatial pattern of
productivity per region has changed over time. In the
early years, few regions had average productivity above
25 bags per hectare and they were predominantly loca-
ted in the north and northwest of Minas Gerais, more
specifically in the regions of Montes Claros and Patos de
Minas. In the following years, other regions began also
to show higher levels of productivity, resulting in a more
dispersed distribution of regions with high productivity
throughout the territory.
Table 2 shows the evolution of the number of
immediate regions in the high productivity group in each
intermediate region in the years under consideration,
whereas Table 3 shows the number of immediate regions
in each reference interval by year.
The spread of high productivity among the inter-
mediate regions is evident in Table 2. In 2002, of the 13
intermediate regions, only 4 contained any immediate
regions of high productivity. In the following years, a
steady increase was observed: 8 in 2006, 9 in 2010, 10 in
2013, and 12 in 2017. Thus, in the final year of the series,
of the 13 intermediate regions, only one (Ipatinga) did
not contain any immediate region with high productivity.
Table 3 shows in greater detail how the number of im-
mediate regions in each productivity interval (the same
used in Figure 2) changed by year. An impressive increase
was observed in the higher intervals (fourth, fifth and
sixth), that together compose what this paper has been
calling high productivity. From 2002 to 2017, the num-
ber of regions in the fourth interval (between 25 and 29
bags/ha) went from 2 to 13, while in the fifth interval
(between 29 and 33 bags/ha) it went from 4 to 9. Finally,
the number of regions producing over 33 bags/ha went
from 5 in 2002 to 12 in 2017.
Therefore, there was a steady spreading of pro-
ductivity throughout the state. In addition, this occurred
by increasing the productivity level of regions with lower
levels while generally maintaining the levels of those
regions that were already at the higher levels. In this
way, coffee productivity became more homogeneous in
the state of Minas Gerais.
It is possible that this spreading process has as
one of its main determining factors the contiguity (the
sharing of a border) to a region that already had high
levels of productivity. As can be seen in Figure 2, each
year many of the regions that showed higher productivity
for the first time were the neighbors of those that already
were in that condition, generating a spreading effect of
high productivity throughout the state.
The following paragraphs describe how high
coffee productivity spread among the immediate regions
of Minas Gerais from 2002 to 2017, stressing the role of
contiguity. This movement is also described on the upper
geographic level, that of intermediate regions.
In 2002, of the 70 immediate regions of Minas
Gerais (the old microregions), only 8 showed high pro-
ductivity. These regions were distributed in only 4 inter-
mediate regions of the 13 that compose the state (the
old mesoregions). Four immediate regions were in the
intermediate region of Montes Claros (Salinas, Pirapora,
Espinosa and São Francisco), two in the region of Patos
de Minas (Patos de Minas and Unaí), one in Uberlândia
(Uberlândia) and one in Uberaba (Araxá).
In 2006, the number of immediate regions with
high productivity increased from 8 to 15. Of the 8 regions
that were at that level in 2002, six remained, and two
left, Espinosa and Araxá. In addition to these six, 9 other
regions joined the group. Of the nine, five were neighbors
of regions that already had high productivity: Janaúba
(neighbor of Salinas and Espinosa), Curvelo (neighbor of
Pirapora), Dores do Indaiá (neighbor of Patos de Minas
and Araxá), Abaeté (neighbor of Patos de Minas) and
Monte Carmelo (neighbor of Uberlândia and Araxá).
Spatial distribution of coffee crop productivity in Minas Gerais over time
Cad. Ciênc. Agrá., v. 12, p. 01–10,
Figure 1 – Coffee productivity in the immediate geographic regions of Minas Gerais, 2002 to 2017
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019).
Concerning the upper geographical level (the
intermediate regions, the equivalent of mesorregions), if
in 2002 only four of them contained immediate regions
of high productivity, in 2006 this number had doubled.
The intermediate region of Montes Claros had four im-
mediate regions with high productivity, while each of the
regions of Belo Horizonte, Patos de Minas, Uberlândia,
and Divinópolis comprised two immediate regions in this
situation and the regions of Varginha, Pouso Alegre and
Juiz de Fora contained only one each.
Returning to the level of the immediate regions,
in 2010, once again there was an increase in the number
of regions with high coffee productivity (fourth, fifth
and sixth reference ranges shown on the maps), from
15 to 21. Of the fifteen immediate regions of 2006 in
this situation, ten remained. Of the eleven that became
highly productive (Araxá, which was already in the group
in 2002, returned), eight were neighbors of regions that
already had high productivity in 2006 and/or 2002. Ex-
ceptions are the immediate regions of Poços de Caldas,
Alfenas and São João del Rei, which were not neighbors
of highly productive regions.
Serenini, R. et al.
Cad. Ciênc. Agrá., v. 12, p. 01–10,
Table 2 – Number of immediate geographic regions with high productivity (above 25 bags/ha) by intermediate region
Intermediate Region 2002 2006 2010 2013 2017
Barbacena 0 0 1 0 2
Belo Horizonte 0 2 0 1 2
Divinópolis 0 2 2 3 3
Governador Valadares 0 0 0 0 1
Ipatinga 0 0 0 0 0
Juiz de Fora 0 1 1 6 3
Montes Claros 4 3 3 3 4
Patos de Minas 2 3 3 3 2
Pouso Alegre 0 1 1 1 3
Teófilo Otoni 0 0 0 1 1
Uberaba 1 0 2 3 2
Uberlândia 1 2 3 3 3
Varginha 0 1 5 2 8
Total 8 15 21 26 34
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019).
Table 3 – Number of immediate geographic regions in each reference interval per year
> 33
Highly productive
( > 25 bags/ha)
2002 24 32 6 2 1 5 8
2006 13 29 13 5 5 5 15
2010 9 33 7 9 1 11 21
2013 7 14 23 12 7 7 26
2017 13 10 13 13 9 12 34
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019).
In 2010, the number of intermediate regions with
immediate regions of high productivity increased to 9,
with the inclusion of Barbacena. The 21 highly productive
immediate regions were distributed as follows: five in the
intermediate region of Varginha, three in the regions of
Uberlândia, Patos de Minas and Montes Claros, two in
the regions of Uberaba and Divinópolis and one in the
regions of Pouso Alegre, Juiz de Fora and Barbacena.
In 2013, the number of immediate regions with
high productivity increased again, from 21 to 26. Of
these, fifteen were already in the group in 2010 and one,
the immediate region of Abaeté, had already appeared
in 2006. Of the other ten, six were neighbors of imme-
diate regions that were in the high productivity group in
previous years. Although the other four (Viçosa, Ponte
Nova, Manhuaçu and Muriaé) did not share a border
with regions that already had high productivity, they are
neighbors to each other and belong to the intermediate
region of Juiz de Fora, that until then only had the im-
mediate region of Além Paraíba in the high productivity
Finally, in 2017, the number of immediate re-
gions in the high productivity group went from 26 to
34, a remarkable increase when compared to the eight
observed in 2002. Among the 34, 17 were in the group
in 2013 and six had already been present in one of the
other previous years under study, although not in 2013.
Of the remaining eleven, seven were neighbors of regions
that belonged to the high productivity group in 2013.
Therefore, the increase in productivity tended to
occur in the neighbors of those regions that were already
Spatial distribution of coffee crop productivity in Minas Gerais over time
Cad. Ciênc. Agrá., v. 12, p. 01–10,
highly productive, pointing to the presence of spatial
autocorrelation. As mentioned in the previous section,
Moran’s I is a statistical measure that quantifies this kind
of spatial association. Table 4 shows its values for each
year in the period under consideration.
It is possible to divide the data on Table 4 into
two periods: the first eight years, from 2002 to 2009, and
the last eight years, from 2010 to 2017. In the first period,
Moran’s I was statistically significant at a significance
level of 5% in all years, with a positive sign, indicating
the presence of positive autocorrelation between the im-
mediate regions. In other words, there was a tendency of
highly productive regions to share some border with other
regions in the same situation, the same being applied to
those with low productivity.
Table 4 – Spatial autocorrelation (Moran’s I) of coffee productivity in immediate geographic regions in Minas Gerais
by year
Year Moran’s I p-value
2002 +0.347* 0.001
2003 +0.370* 0.001
2004 +0.304* 0.001
2005 +0.308* 0.001
2006 +0.380* 0.001
2007 +0.202* 0.010
2008 +0.168* 0.007
2009 +0.182* 0.015
2010 +0.097
2011 +0.132* 0.031
2012 +0.106
2013 +0.009
2014 +0.056
2015 -0.043
2016 +0.099
2017 +0.068
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019). Pseudo p-value obtained from 999 random
In the second period, only one year had statis-
tically significant Moran’s I, 2011, and even in this case
the magnitude was smaller than those observed in the
previous period. This means that the spatial dependence
of productivity has been reduced over the years due to
the spread of productivity throughout the state, with the
ensuing greater homogeneity of the regions. This can be
checked using Moran’s scatterplot for each year, shown in
Figure 3. In this kind of plot, with each point representing
a region, the horizontal axis represents the value of the
variable in that region and the vertical axis represents the
spatially lagged variable, that is, its average value in the
contiguous regions. The adjustment line slope coefficient
corresponds to the value of Moran’s I.
The scatterplots of the early years show the dis-
tinctive pattern of positive autocorrelation, in which
most regions are in the high-high and low-low quadrants.
In 2002, 57 of the 70 regions were in one of these two
quadrants. On the other hand, in the last years the dis-
tribution is more homogeneous between the quadrants,
with the adjustment line slope becoming flatter.
To determine the presence of spatial clusters and
outliers, local Moran’s I for each region were calculated.
These are local indicators of spatial autocorrelation (LISA)
and are the building blocks for the LISA maps, such as
those shown in Figure 4. The maps show the names of
the immediate regions with significant LISA values at a
5% level of significance.
Confirming the descriptive and the global analy-
sis, the LISA maps in Figure 3 show a reduction in the
significant clusters and outliers over the period under
consideration. In the first eight years, large clusters are
noticeable, mainly the cluster of high productivity in
the Northwest region of Minas Gerais. However, over
the following years, this cluster gradually disappeared
Serenini, R. et al.
Cad. Ciênc. Agrá., v. 12, p. 01–10,
in response to two movements: the increase in the pro-
ductivity of other regions turned the difference between
them not statistically significant and the abandonment
of coffee production in some regions.
On one hand, some regions with low productivity
increased their productivity levels, as exemplified by the
largest cluster of low productivity seen over the period.
In 2002, there was a low-low cluster consisting of five
regions: Santa Bárbara - Ouro Preto, Guanhães, Ipatinga,
Itabira and João Monlevade. All of them had productivity
below 13 bags/ha. In 2017, of the five regions, three had
increased appreciably their productivity levels. Santa
Bárbara-Ouro Preto increased productivity by approxi-
mately 334%, Ipatinga, 210% and Guanhães, 128%. The
other two regions also increased their productivity, but
at lower levels, 4 and 15%.
Figure 2 – Moran’s scatterplots for productivity, 2002 to 2017
Source: Prepared by the authors based on data from the Produção Agrícola Municipal (IBGE, 2019).
In 2006, a low-low cluster comprised almost all
of the eastern area of the state, composed by 10 imme-
diate regions: Governador Valadares, Guanhães, Ponte
Nova, Capelinha, Ipatinga, Almenara, Manhuaçu, Ubá,
João Monlevade and Teófilo Otoni. In that year, of these
regions only two had productivity above 16 bags/ha. No-
netheless, in 2017, the situation was reversed: only two
of the same ten regions had productivity below 16 bags
per hectare. Nine of them (the exception being Teófilo
Otoni) increased their average coffee productivity. Seven
regions had an increase above 20%, of which three had
an increase above 70%.
In addition to the regions already mentioned,
others that were not included in low-low clusters also had
significant increases in productivity. Conselheiro Lafaiete,
Além Paraíba, Janaúba, São João del Rei, Montes Claros
and São João Nepomuceno-Bicas increased productivity
by more than 100% from 2002 to 2017.