This study involved a limited amount of field research that was approved by the University of Wisconsin Madison’s Education and Social/Behavioral Sciences Institutional Review Board under study number 20130044 (principal investigator: Holly Gibbs).
Land use patterns in Mato Grosso are atypical for Brazilian standards. Pasture area growth decelerated more rapidly than in other locations, while the opposite happened to crop areas. According to the agricultural census, pastures in Mato Grosso grew by 38.3% per decade in 1975–1995, and by 45.6% in other states in the Legal Amazon. In 1995–2006, however, Mato Grosso had an additional 2.8% pasture area, while the other Amazon states had a 31.8% increment [ 51 ]. The total crop area, on the other hand, grew by 117.7% in 1995–2006 [ 51 , 52 ], and by 116.1% in 2001–2014 [ 53 ]. In more recent years, while census data are unavailable, remote sensing data show that pasture areas continued expanding by a very small percentage [ 54 , 55 ]. This has come with increased cattle densities, as we show in the Results section.
With a small population (1.6% of Brazil’s total) but the largest cattle herd (13.9%), Mato Grosso has a dynamic slaughter industry that supplies to other states and overseas. It is the second biggest beef exporter after São Paulo [ 47 ], and only 18% of its production is consumed in-state [ 48 ]. An important part of the industrial agglomeration that took place in Brazil since approximately 2005, which led to the creation of the largest meatpacking conglomerate in the world [ 49 ], involved operations in the state of Mato Grosso [ 50 ].
Mato Grosso is the 3 rd largest state in Brazil (903,357 km 2 ) and a key conservation target for the Amazon and Cerrado biomes, which respectively occupy 54.5% and 38.3% of the state’s territory (the Pantanal wetlands cover 7.2%). It has been leading the modernization of Brazilian agriculture and cattle ranching for over a decade, especially in large-scale commercial agriculture [ 41 – 43 ]. Mato Grosso has the largest share of agriculture in its Gross Domestic Product (GDP) among all states, 21% in 2014, and the 5th largest agricultural GDP [ 44 ]. The expansion of mechanized agriculture, especially soy, has been pointed out as one reason for the state’s development process [ 45 ]. For example, Mato Grosso had the largest GDP growth relative to the country in 2002–2014, and the fourth largest per capita income growth in the same period [ 44 , 46 ].
Contents
Mapping the space-time signature of slaughterhouses in Mato Grosso
‘Slaughterhouse’ is a generic term that may refer to a physical plant or to a holding company depending on the context, so we start by clarifying the definitions. ‘Plants’ or ‘units’ are physical instances of individual slaughtering operations. In this study we chose to focus on plants that slaughter at least 300 head per year, which may be called abattoirs but not butcheries. This choice was made because very small facilities have a negligible impact on quantities but demand more data processing work as data are either unavailable or less transparent. Hence, where plants with slaughter volumes of <300 head appeared in the dataset without an address, we made no effort to retrieve the address from additional sources and instead grouped those CNPJs together into one ‘unidentified’ plant per municipality.
‘Holdings’ are the companies that own the physical plants. Each plant has a single holding company at each point in time, but multiple plants can be owned by a holding. We documented and report only the most recent holding company of each plant, and as such miss the dynamics of previous ownership changes. Our definition of holding comprises local businesses that own at least one small plant. ‘Legal persons’ are the formal business registrations that the Federal Revenue agency issues to companies in Brazil. These legal identities, known by the acronym CNPJ, are used to process corporate taxes and financial information. One plant will sometimes operate through two or more CNPJs.
To generate a map of all slaughterhouses in Mato Grosso, we triangulated across multiple data sources including a registry of companies, government records of cattle transactions and of the sanitation inspection system, data compilations by a think-tank and a research lab, open-access high-resolution satellite imagery, and others (S1 Table). provides a schematic view of the data sources used and the sequence of steps applied. The process was divided into five steps that we describe below: compiling a core dataset with key company identifiers, such as names and addresses; populating the dataset with attributes from multiple sources, especially opening and closing dates (of a CNPJ); grouping registered companies by physical plant and geocoding the addresses; documenting and inferring ownership changes and dates for the larger holding groups; and validating the data through comparisons with other sources of information.
We started by stacking up the two main sources of raw data (step 1). The first is a Brazil-wide company registry compiled by Empresômetro using data from the Federal Revenue agency on CNPJs and their names, legal names, economic activities, dates of creation, and addresses. We filtered 21 million records to obtain a list of 235 businesses registered across Mato Grosso under economic activities related to cattle slaughter. We used CNAE codes, an official classification of economic activities, to select CNPJs registered as slaughterhouses. One (or more) CNAE code is assigned to every CNPJ. We selected all CNPJs with at least one of the following CNAE codes: cattle slaughterhouses (code 1011201) and cattle abattoirs (code 1011205), or with at least one of the following words: “frigorífico” (slaughterhouse), “matadouro” (abattoir), “carne” (meat), and “abate” (slaughter). The second source is a compilation of government records (GTA, the Portuguese acronym for ‘Animal Transportation Form’) on cattle slaughters from Indea-MT, the state livestock sanitation agency (S1 Box). The GTA records include 2,976,962 transactions, from which we identified 175 companies (slaughterhouses or abattoirs) responsible for slaughtering cattle between 2013 and 2016.
Next, we added the data from Imazon, a Brazilian environmental think-tank, and from Lapig, a remote sensing and geoprocessing research lab. The Imazon data include inspection codes and the geolocations of 49 plants under federal or state inspection, and the Lapig data have the same information for 37 plants with federal inspection. We dropped observations with repeated CNPJs to get a raw list of 360 registered companies.
In step 2, we populated the dataset with the attributes in S1 Table. We used three sources of information: Sintegra, a web gateway to information from the State Revenue agencies; MAPA, the Brazilian Ministry of Agriculture; and MPF, the Federal Prosecutors Office. Sintegra can be queried using a CNPJ number and it provides the most accurate opening and closing dates. The MAPA portal allows for queries by SIF code, and it is the only source for start date of inspection at SIF plants. The MPF portal is the only source for the signing dates of the TAC (Portuguese acronym for ‘Conduct Adjustment Term’) supply chain commitments–binding contracts signed between slaughterhouses and Federal Prosecutors for properties in lack of compliance with environmental/labor standards to be excluded from supply chains [34]–and it provides a list that we matched to our database using names and municipalities.
At this point, the units of observation were still the CNPJs, but these do not bear a one-to-one relationship with physical plants. Due to fiscal or otherwise managerial motivations, an active slaughterhouse plant operates, on average, through between one and five active CNPJs [mean = 1.5]. In step 3, we aggregated the CNPJs into plants. While many CNPJs can be associated with one plant, multiple plants were never associated with a single CNPJ in our dataset. We grouped CNPJs within plants only if they were registered in the same municipality. All CNPJs with name strings and/or addresses that we judged to be the same were given a unique plant identifier. All CNPJs for which no company name or address information was found at a single municipality were placed under one ‘unidentified’ plant. The final dataset has 133 plants.
For the spatial coordinates, we followed a hierarchical decision rule. First, we used GPS points obtained in field visits. Second, we used the coordinates provided by Google on its enterprise registry. Third, we used the coordinates provided by [23, 24], both of which were visually inspected using high resolution satellite imagery. Fourth, we used Google Maps to geocode the addresses provided in the company’s CNPJ registry.
Similar to plants operating through multiple CNPJs, a holding group may control multiple plants across the state. In step 4, we grouped plants into holding companies. Plants with at least one CNPJ whose name or legal name was that of a known holding group, defined as any group listed in [23], the most comprehensive list available, were allocated to the known holding. If the name of a known holding group did not appear in any of the plant’s CNPJs, we chose the shortest name between all CNPJs as the holding name. In less than 5% of the cases, one plant had CNPJs referring to different known holding groups, so we used expert consultation, online news portals and other local sources of information to identify the holding controlling the plant at present.
After the plants were aggregated into holdings, all the plants within each holding were coded as having signed a TAC commitment when at least one CNPJs associated to any of the plants within the holding had a TAC in place.
Where a plant from a known holding group had CNPJs with different names, the plant was flagged as having gone through an ownership change. We used [48, 50] to document the timing of the last ownership change. Where the date was not documented in those sources, we used two alternative assumptions. First, when a holding buys a new plant, it will often create a new CNPJ and will discontinue one or more CNPJs under the previous name. We used a switch in CNPJs as an indicator of ownership change. If there was no temporal coincidence between the closing and the opening of CNPJs, then we took the opening date of the most recent CNPJ under the current holding’s name as the ownership change date.
Finally, in step 5 we validated the spatial coordinates by the visual inspection of high resolution imagery on Google Earth. We assessed whether the mapped location showed the typical structure of a slaughter facility, which includes cattle corrals, industrial buildings and waste storage lagoons. High resolution images are available for relatively recent years, so plants that closed before the early 2000s were more difficult to identify. Our data have one closing date earlier than 2005 and 20 earlier than 2010 (33% of the closed plants). This was a relatively minor problem, however, because even the plants that closed before satellite images were available could in some cases be identified since their structure remained visible years after the closing.
Inference of plant activity and inactivity
We used the company’s opening and closing dates along with the dates of the GTA records to establish whether a plant was active or inactive (these terms are used interchangeably with open / closed) at a given year. Activity is defined as an ongoing company registration (at least one open CNPJ) for the years when no GTA information is available (prior to 2013) or positive slaughter activity for other years. For example, if there was at least one active CNPJ but no slaughter activity in 2014 and after, the plant was coded as inactive in 2014 and after. If there was slaughter activity in 2014 but no legally active CNPJ, the plant was active in 2014. If there was at least one active CNPJ in 2007, the plant was active then. If there was no recorded slaughter activity in any year, the company was coded as inactive in the latest year when one of its CNPJs became inactive, even if one or more active CNPJs remained. In seven situations–which we coded as active–did a plant show a positive slaughter activity in years subsequent to the closing of its last CNPJ (94 plants had GTA slaughter activity).
CNPJ closing dates were taken from two sources. The company registry, which shows if the company registration was discontinued at the federal level, and Sintegra, that shows whether the company’s state fiscal registry was discontinued. If at least one of these sources showed a closing date, we coded the CNPJ as closed.
The opening date was defined as the earliest date between the SIF registration date and the earliest CNPJ registration date. The Ministry of Agriculture’s data on when SIFs were first registered is the most accurate historic information for the decades of the 1970s through the 1990s. For non-SIF slaughterhouses, we used only the company registry. For plants operating earlier than 2013, these are the only sources of information for the starting date as the public GTA slaughter records start in 2013. For plants that were operative in 2013 or after, the GTAs allowed us to estimate the starting date even if the plants had no legal registration or sanitation inspection. In these cases, we assigned the first appearance in the GTA as the opening year.
Estimating uninspected slaughter
The weakest part of the animal inspection system is the municipality [56]. According to [37], only 20% of SIM or SIE plants were compliant with sanitation standards in 2012. Assuming that SIMs are even less likely to be compliant than SIEs, it follows that SIM plants are unlikely to be fully inspected. This is the first reason why we code SIM plants as uninspected. The second reason is that, even if desirable, distinguishing SIMs from uninspected at a large scale is not possible due to the lack of consistent data on the location of SIM plants. So for simplicity, we refer to both as uninspected. In the past, even the plants with state inspection where classified as uninspected due to the lack of data [38]. Moreover, only seven plants in Mato Grosso were reported by IBGE as being under municipality inspection in 2016 [39]. Considering the 20% compliance rate, this could imply that two of the seven SIM plants in Mato Grosso are effectively inspected.
We used the following formula to calculate the slaughter volumes of uninspected plants at the municipality level:
Tot_SLtm=SIF_SLtm+SIE_SLtm+UN_SLtm,
(1)
Where the subscripts t and m indicate year and municipality, Tot_SL is the total slaughter volume, SIF_SL is the slaughter in plants inspected by the federal authorities, SIE_SL is the slaughter in plants inspected by the states, and UN_SL is the estimated uninspected slaughter (including clandestine and non-clandestine). By using the GTA slaughter records we have largely captured the non-clandestine market, as all GTA transactions must include a CNPJ number. The only GTA transactions that could refer to clandestine plants are those whose CNPJs do not match with any known CNPJ record, which amounts to only 1.74% of all the transactions in our data.