Cattle production in Southern Amazonia: implications for land and water management

The expansion of cattle in central western Brazil has been under scrutiny because of the region’s historic reliance on Amazon and Cerrado deforestation for cropland and pastureland expansion. In this study, we determined the volumetric water footprint (VWF) and the land footprint (LF) of cattle in Mato Grosso state for the years 2000, 2005, 2010 and 2014 using official statistics and remote sensing imagery. We found the average VWF of cattle for the time period to be 265–270 l kg−1 LW−1 (LW as live weight of cattle) and a LF which decreased from 71 to 47 m2 kg−1 LW−1. The largest contribution to VWF came from farm impoundments whose total area increased from roughly 46 000 to 51 000 ha between 2000 and 2014, leading to a total evaporation as high as 7.31 × 1011 l yr−1 in 2014. Analysis at the municipality level showed a tendency towards greater density of cattle with respect to both pasture area and impoundments. While cattle intensification on current pastureland is commonly viewed as a means to prevent further deforestation and greenhouse gas emissions, we stress the need to also consider the increasing demand for water associated with a growing cattle herd and the potential appropriation of additional resources for feed for feedlot finishing. Land and water resource management need to be considered together for future planning of cattle intensification at the Brazilian agricultural frontier as illustrated by the footprints reported here.


Introduction
South America has been the largest agricultural frontier on the planet for almost two decades. Since 2001, 96.9 Mha of pasture expansion took place on the continent, including large areas in Brazil and Paraguay (Graesser et al 2015). In 2012, Brazil was the largest producer of cattle globally with a population of 211 million, 13% of which were raised in the central western state of Mato Grosso (IBGE 2017). Recently, Mato Grosso's cattle production has been under scrutiny, particularly with respect to pastureland expansion into both the Cerrado (or savanna) and Amazon biomes (figure 1). Between 2000 and 2009, total pasture area in Mato Grosso only varied marginally between 22 and 24 Mha (Lathuillière et al 2012), similar to the national trend (Dias et al 2016). This apparent stagnation masks indirect land use change dynamics that took place during the 2000s, particularly related to soybean expansion. From 2000 to 2010, about half of the 5.5 Mha of forest clearings in Mato Grosso were utilized for pasture, while soybean expanded primarily into previously established pastures (Macedo et al 2012). The wave of soybean expansion may have displaced pasturelands farther into the Amazon agricultural frontier (Barona et al 2010, Arima et al 2011, indirectly causing more deforestation. The agricultural sector's role in deforestation in both the Amazon and Cerrado biomes has been the subject of much research (Barona et al 2010, Macedo et al 2012, Gollnow and Lakes 2014, Nepstad et al 2014, Zalles et al 2019. One proposed solution to avoid additional land use change is to enable cattle intensification through, for example, a set of complementary policies (Cohn et al 2014). Brazilian pastureland was found to sustain 32%-34% of the potential pasture carrying capacity, which indicates opportunities to increase productivity nationwide to meet the demand for cattle in 2040. While meeting demand through an increase in cattle density may help achieve deforestation targets, Mato Grosso would still need to consider questions regarding future water availability and usage.
Little information is available at present on water use for cattle in Mato Grosso. Research on the water consumption of production systems and supply chains has increased considerably with the development of the volumetric water footprint (VWF) . Many global and regional studies on animal products have attempted to quantify resource appropriation (Mekonnen and Hoekstra 2012, Gerbens-Leenes et al 2013, Vanham et al 2013 or environmental impacts of water consumption through life cycle assessments (e.g. Ridoutt et al 2012Ridoutt et al , 2014. The VWF exclusively quantifies the appropriation of water resources by combining direct (operations) and indirect (supply chain) water consumption, thus providing a comprehensive picture of the water use of a product or service over its entire life cycle . Unlike water withdrawals, water consumption refers to water that is removed from the watershed because of product integration, evaporation, inter-basin transfers, or direct release into the sea (Bayart et al 2010).
In Mato Grosso, cattle typically drink from small impoundments that are either rain-fed or created by damming firstor second-order streams. These water sources have to be carefully managed to guarantee year-long availability to cattle. Reduced precipitation and high evaporation rates can severely diminish water availability in impoundments during the dry season (May-November), implying vulnerability of the production system to extended droughts. This study's objectives are to (1) evaluate land and water resource appropriation for cattle production in Mato Grosso by quantifying its land footprint (LF) and VWF, (2) understand the combined evolution of pasture and impoundment area for cattle in Mato Grosso between 2000 and 2014, and (3) provide guidance on future strategies for land and water management. This information is paramount to understanding possible limits to future intensification of cattle in the region.

Cattle production in Mato Grosso
We analyzed land use and water consumption for cattle production in Mato Grosso using a suite of indicators. First, we estimated an average LF and VWF related to the total live cattle population (see supplemental material, equation (S1) is available online at stacks.iop.org/ERL/14/114025/mmedia). We considered cattle production involving the Nelore breed from the species Bos taurus indicus which represents 90% of the cattle breeds in Mato Grosso (FABOV 2007). We assumed a state-wide average cradle-to-farm gate cattle production system that follows a 46-51 month cycle according to practices described for Mato Grosso (FABOV 2007, Cardoso et al 2016, Cerri et al 2016, during which females (heifers, cows) and males (steers, bulls) reach 430 kg and 520 kg, respectively. This system does not include any potential breeding stock with a longer life cycle. Calves were assumed to weigh 30 kg at birth and to be weaned at 165 kg (Cardoso et al 2016) (table S1). In this system, cattle spend most of their lives on open pasturelands except in the finishing stage when they can either remain in pasture or be transferred to confinement (figure 2). Pasture finishing is a continuation of the adult development phase with pasture dry matter intakes (DMIs) continuing until slaughter. Confinement may occur in the last 6-8 months of the animal's life when cattle is generally fed a diet of 70% feed and 30% silage.
Secondly, we analyzed the evolution of pasture and water reservoir areas for 2000,2005,2010 and 2014 for all the municipalities of Mato Grosso derived using a combination of official statistics and remote sensing information. Given the change in size, shape, and number of municipalities in Mato Grosso during the study period, we combined the municipalities that had changed into greater municipal units (MUs) that remained constant over the study period, following methods outlined by Lathuillière et al (2012). This combination of political units resulted in 104 MUs for the entire state (figure 1).

Volumetric water footprint of cattle production
The water footprint has been used by distinct communities following separate guidelines and standards: (1) the water resources management community, which typically follows the Water Footprint Assessment Manual , and (2) the life cycle assessment community, which follows the ISO 14046 standard (ISO 2015). While these approaches are complementary (Boulay et al 2013, Lathuillière et al 2018, they differ in their objectives as they may emphasize either water resources appropriation or impact assessment, respectively. Here, we focus specifically on water resources appropriation, or the volume of freshwater consumed for cattle production, which we have designated as the VWF of cattle. The VWF highlights how water is consumed in the production system and, therefore, can identify water management strategies to reduce the water consumed per kg live weight (LW), as well as the overall water consumed by the sector between 2000 and 2014. This VWF can also undergo a characterization step in which the volume of freshwater consumed is translated into impacts according to local water scarcity conditions (Pfister et al 2009, Ridoutt and Pfister 2010, Boulay et al 2013, 2018, Berger et al 2014, with extension to impacts to human health and ecosystem quality (Boulay et al 2015, Núñez et al 2016 within a life cycle assessment. Our VWF is synonymous with water productivity (as its inverse) (Giordano et al 2017) or a water footprint inventory following ISO 14046 (ISO 2015). We do not carry out a life cycle assessment.

Animal water consumptive use
Water for cattle is sourced from drinking water and moisture content in feed as well as additional water created by the animal's digestion system (metabolic water) (FAO 2019). We applied the animal water balance of Ridoutt et al (2012) to derive the volume of freshwater consumed by the animal based on its development cycle and finishing stage (open pasture or feedlot). The method also calls for the characterization of this volume considering local water stress conditions (Ridoutt et al 2012), which, as described above, we did not apply in this research. Here, the total water consumed by the animal (l d −1 month −1 ) is expressed as On the water input side of equation (1), W feed is the water contained in the feed that depends on DMI (kg d −1 ), the moisture content of the feed (MC, %), the amount of milk consumed by the calf until eight months of age (W milk , l d −1 ), and the amount of water needed to mix feed (W mix , l d −1 ) assumed to be 0.5 l kg DMI −1 according to Mekonnen and Hoekstra (2011) (equation (2)). W drink is the amount of water drunk by the animal, and W met is the water resulting from metabolic production as a function of feed digestibility (D, %) and described in equation (3) (Ridoutt et al 2012). The sum W feed +W drink represents the total daily water intake of the animal. On the water output side of equation (1), W evap is the amount of water lost by the animal to evaporation, W WG is the amount of water incorporated by the animal and represented by a weight gain, W feces is the water content in animal feces, while W urine is the water content in urine (Ridoutt et al 2012). Following the above definitions, W feed , W drink , W met , W feces and W urine represent water flows into and out of the animal, while W WG and W evap represent water consumptive uses We assumed that water contained in urine and feces are entirely evaporated, thereby simplifying W feces and W urine as water consumptive uses. This is a slight deviation from Ridoutt et al (2012) who considered W feces lost to evaporation, but W urine as a flow added to local discharge with potential water quality impacts (not considered in our study). We believe our assumption to be reasonable for Mato Grosso considering the region's high potential evapotranspiration (potential ET). On pastureland, moisture in cattle urine and feces can evaporate rapidly, as opposed to industrially confined cattle whose waste is typically collected in pits for subsequent spraying into fields.
The above assumption allows us to equate total animal water input to total water consumption in equation (1). Consequently, our calculation of the VWF of cattle relies exclusively on W feed , W drink and W met . All parameters used for these calculations are listed in table S2. We did not include water consumed for transport between stages of development as the production system in this study was assumed to be confined to one single property, nor do we include water consumed for the production of minerals, vitamins, and veterinarian services administered to the herd.

Feed water consumptive use
Water consumption from pasture and crops used for feed were considered as additional consumption indirectly attributed to cattle production. Diets depend on the animal, the production system considered, and whether confinement is involved at the finishing stage (figure 2). The VWF of pasture was estimated for each MU using spatial precipitation data from CHIRPS v.2.0 (Funk et al 2015) input into the model from Zhang et al (2001) to derive pasture ET (see equation (S2) in the supplemental material). Pasture ET was then related to two pasture productivity scenarios to derive the volume of water consumed per kg of dry matter (DM) corresponding to 3 tons of DM ha −1 (low productivity) and 5.3 tons of DM ha −1 (high productivity at 1.5 animal units ha −1 following Thiago and Silva (2006)) (table S3). Together, these scenarios provided a range of VWF of pasture that we used to represent potential VWF of feed in Mato Grosso. All feed for feedlot finishing (61% maize, 10% sorghum, 8% soy meal, 8% cottonseed, 8% soybean grain following Millen et al (2009)) was assumed to be sourced in Mato Grosso, with VWF for these crops from previous research (table S4) Hoekstra 2011, Lathuillière et al 2012), which resulted in a total feed VWF of 20.9 l kg −1 LW −1 d −1 , assuming the average feed composition. Given that there was minimal cropland or pastureland irrigation in Mato Grosso during the study period, all feed water use was assumed to be sourced exclusively from green water (i.e. soil moisture regenerated by precipitation (Lathuillière et al 2012)).

Water consumption from evaporation of small farm impoundments
Evaporation from small reservoirs used for cattle drinking water needs to be accounted as a blue water (surface water) consumptive use (Ridoutt et al 2012). Little information is available on such reservoirs in the region other than a recent estimate of size and depth by Rodrigues et al (2012) in central western Brazil, where reservoirs were typically less than 50 ha in size and up to 10 m deep (Rodrigues et al 2012). Impoundment area within each MU was determined using remote sensing (see supplemental material).
Total impoundment area was then multiplied by the average reference ET (ET 0 , mm yr −1 ) obtained from Xavier et al (2015) to be allocated to the total herd LW in each MU. We then obtained an average Mato Grosso small reservoir evaporation (in l kg −1 LW −1 ) from which we subtracted evaporation from fish tanks considering both the range of possible yields described above, and average ET 0 from all MUs in Mato Grosso. Since the Xavier et al (2015) time series ended in 2013, we assumed that ET 0 in 2014 was equal to that in 2013. Total animal LW was calculated for each MU assuming a 50:50 ratio of males and females in each MU (IBGE 2018, from 2006 census information) and their average LW in each of the calf (94.86 kg), mid-life (267.96 kg) and finishing stages (427.93 kg). Lastly, we combined our estimates of impoundment area with data on the number of live animals (see supplemental material) to calculate the reservoir cattle density (RCD, cattle heads per ha of water) for each MU and for each study year.

Land footprint of the cattle production
We used official statistics combined to the inverse of product yields (ha ton −1 ) to obtain direct and indirect land use estimates for cattle. First, total pasture area was obtained for each MU using agricultural census information (1995,2006,2017 . We then combined the total pasture area with the live animal population in each MU to derive the pasture cattle density (PCD, cattle heads per ha of pasture, hencefourth as cattle per ha) assuming that the pasture area was used exclusively by cattle rather than other animals. MU specific pasture areas were validated against the remote sensing images of Graesser and Ramankutty (2017) used to derive reservoirs to ensure compatibility in the total area allocated to cattle within the political units (see figure S1). The direct LF was then determined as the inverse of the PCD by allocating all pasture to living cattle. In addition, we estimated the amount of land required to grow inputs for the feed composition by considering the Mato Grosso average inverse yields in 2000,2005,2010 and 2014 for maize (2.77×10 −4 ha kg −1 ), soybean (3.33×10 −4 ha kg −1 ), cotton (2.72×10 −4 ha kg −1 ), and sorghum (5.54×10 −4 ha kg −1 ) (IBGE 2017). When considering the average feed composition, we estimated the indirect LF of feed to be 7.35× 10 −6 ha d −1 kg −1 LW −1 .

Cattle volumetric water and land and footprints
Mato Grosso's average VWF for cattle at farm gate for the 2000-2014 period was 265-270 l kg −1 LW −1 , considering sex, and finishing stage (figure 3). In the case of pasture finishing, the VWF of cattle was comprised of 20%-24% green water (from W feed ) and 76%-80% blue water (from W milk , W drink , W met , and W res ), based on sex. Reservoir evaporation as allocated to cattle production (as reservoirs detected within pastureland) was the largest contributor to the VWF of cattle representing 47% of total water consumed (W res ), followed by drinking (30%-31% for W drink ) and water contained in feed (18%-21% for W feed based on finishing stage), the remainder being water from W met , W mix and W milk (figure 3). During the same period, the average LF of cattle was 71 m 2 kg −1 LW −1 in 2000 and 47 m 2 kg −1 LW −1 in 2014 following ranges of PCD across Mato Grosso (PCD, described below) (table 1). The use of feed in the finishing stage provided an additional 12.9 m 2 kg −1 LW −1 when considering agricultural products going in to the feed (13.5 m 2 kg −1 LW −1 for males and 12.3 m 2 kg −1 LW −1 for females) (table 1).
Confinement in the finishing stage slightly decreased the cattle VWF attributed to the animal (excluding W res ) from 132 to 126 l kg −1 LW −1 in males and 156 to 153 l kg −1 LW −1 in females due to Values of VWF animal are separated into water content of feed (W feed ), milk consumption (W milk ,<1 l kg −1 LW −1 not shown in the graph), animal drinking water (W drink ), water used to mix feed (W mix ), water evaporated by farm reservoirs detected within pastureland (W res ), and metabolic water (W met ). Values of W res represents total small farm reservoir evaporation minus evaporation allocated to fish tanks with fish tank area determined by mean fish production (3.5-7 ton ha −1 of water). Numbers on the graph represent individual contributions to the VWF animal for a total of 270 l kg −1 LW −1 and 265 l kg −1 LW −1 for pasture and feedlot finishing, respectively.

Evolution of land and water use for cattle in Mato Grosso
We observed few significant changes in pasture area within the study time period, with different rates of  4. Discussion 4.1. On-farm land and water appropriation for cattle By combining the above information with previously published results on the carbon footprint (CF) of cattle (table S7) we obtained ranges of VWF, LF and CF for cattle in Mato Grosso for the time steps considered, including both direct (animal) and indirect (feed) land and water appropriation. Reported values for the CF include emissions from the animal and inputs into the production system totaling 4.8-8.2 kg CO 2 -eq a A res values are given as the mean obtained considering the range of mean fish production (3.5-7 ton ha −1 of water). Emissions from land use change are typically the largest in the production process and can include legacy emissions (e.g. decomposition) with allocation schemes that extend decades following deforestation (Zaks et al 2009, Karstensen et al 2013. Intensification of cattle production on existing rangelands is a widely proposed strategy for curbing deforestation and reducing the CF of cattle (Cohn et al 2014, Gibbs et al 2016, Silva et al 2016. This approach has been met with some skepticism given difficulties with enforcement of the Brazilian Forest code, which does not preclude large swaths of Cerrado from being legally deforested for pasture or crop production (Phalan et al 2016, Azevedo et al 2017, Soares-Filho et al 2014. Intensification on already cleared land (or through confinement) would avert further greenhouse gas emissions from land use change; however, an increase in cattle herd (or total LW) would still increase total emissions of production at rates as high as 8.2 kg CO 2 -eq kg −1 LW −1 (from Cerri et al (2016) in 2011).
Identifying on-farm strategies and incentives to promote intensification has been challenging. Recent work indicates that individual producers were less likely to implement new practices to boost productivity unless the property was part of the Rural Environment Registry (CAR, Portuguese acronym) (Latawiec et al 2017). Latawiec et al (2017) report that, despite being considered a source of water on their property, 70% of farmers interviewed in northern Mato Grosso did not see any financial benefit to forests, which, along with constraints from access to qualified labour and capacity building, constitute important barriers to intensifying cattle production. At the state and federal levels, enforcement of the Forest Code and the 2009 'Cattle Agreement' have helped decrease deforestation for cattle (Nepstad et al 2014, Gibbs et al 2016, whereas taxation or subsidies could help further increase cattle density and spare natural vegetation (Cohn et al 2014).
The calculation of the VWF was largely reliant on the estimate of farm reservoirs, with different effects on RCD based on allocation of evaporation to fish tanks (see supplemental material). Small farm impoundments are a key source of drinking water for cattle in Mato Grosso and should be considered carefully when calculating on-farm water balances. Remote sensing can play an important role in identifying the extent of reservoir networks as a means to estimate reservoir capacity and evaporation. For example, in central Brazil, Rodrigues et al (2012) used remote sensing to map 147 small reservoirs at a volumetric capacity of 1.8×10 −3 hm 3 km −2 in the Preto River basin. Small farm dams and reservoirs are also common in northeastern Brazil, where the area of small reservoirs has increased by 1.95% yr −1 between 1970 and 2002 (Malveira et al 2012). Reservoir evaporation is known to be a major loss of water in Australia (Ridoutt et al 2012), estimated nationally at 1.3-2.9×10 12 l (Baillie 2008) for an average volumetric density of 0.01 hm 3 km −2 (maximum of 0.12 hm 3 km −2 ) (Malveira et al 2012). In Mato Grosso, small impoundments are often constructed in headwater streams, thereby creating up-stream and downstream trade-offs for human and ecosystem water use (Lathuillière et al 2016). The increase in surface area created by these impoundments can increase stream temperatures (Macedo et al 2013) and evaporation (Lathuillière et al 2016), while reducing runoff and stream connectivity (Callow and Smettem 2009), as well as sediment loads (Morais and Pinheiro 2011), all of which can alter stream habitats and negatively impact stream biota. In this study, annual evaporative rates were fully allocated to cattle, but were also expected to vary over the course of wet and dry seasons, thereby potentially decreasing the water supply for cattle. Open water evaporation of a small reservoir in northeastern Brazil was 3593 mm yr −1 with a maximum of 8.9 mm d −1 (December) and a minimum of 5.3 mm d −1 (July) with water levels dropping by 1 m between September and December (Antonino et al 2005). As such, we expect farmers to either supply additional water (surface or groundwater) to cattle or further concentrate animals near larger reservoirs when necessary, particularly during drought periods or during the finishing stages when cattle is confined.
From a water management perspective, a strategy focused on reducing reservoir evaporation would reduce water consumption of cattle from the impoundments. For instance, Baillie (2008) lists evaporation mitigation technologies that include floating covers or chemical barriers, but whose application is not anticipated to be widespread. Additional strategies to reduce evaporation (e.g. deeper reservoirs, allowing for partial regeneration of riparian vegetation) could also be considered. Reservoirs and other on-farm water management infrastructure may persist on the landscape, even following the discontinuation of cattle ranching on the property. For instance, the region that is now known as Tanguro ranch (Hayhoe et al 2011) was dedicated to cattle ranching for about 15 years prior to converting all of its pasture area to cropland, yet many of its reservoirs remain intact. The fate of impoundments following the removal of cattle on farm properties is an important question for future water resources management and use, given that these water bodies could be available for other agricultural purposes such as cropland irrigation or aquaculture production, with the latter showing substantial increases in the state in recent years.

Land and water appropriation for feed
Several studies have quantified the water used to produce feed as an indirect contributor to the VWF associated with animal products (Mekonnen and Hoekstra 2012, Gerbens-Leenes et al 2013, Palhares et al 2017. Others have examined the competition between water use for crops and feed (Ran et al 2017), in addition to land resources and greenhouse gas emissions, as metrics of environmental performance for livestock and animal products (FAO 2016). Since the production of animal feed is generally the largest contributor to the indirect VWF of cattle, reducing the water required to produce feed is often presented as a strategy for improving the efficiency of cattle production overall. Both the LF and VWF can be reduced by increasing feed productivity while reducing its water requirements, a strategy known as increasing feed water productivity (Giordano et al 2017). Since cropland and pasture in Mato Grosso are almost entirely rain-fed, reductions in the VWF of feed could be achieved through a water vapor shift favouring transpiration over evaporation (Rockström 2003), with potential savings of nearly 2×10 4 l kg −1 LW −1 as shown from our pasture productivity scenarios.
Feed VWF can vary greatly based on the individual VWF of crops, pasture and roughage, and the mix used in diets and production systems (Gerbens-Leenes et al 2013, Palhares et al 2017). For instance, the VWF of Brazilian meat was found to more than double when moving from a confined to a grazing system (Gerbens -Leenes et al 2013). Similarly, Palhares et al (2017) show that the choice of feed composition can dramatically change the VWF based on the inputs used. In contrast, the LF of cattle in a grazing system is much larger than the feed used in the feedlot finishing stage, and could potentially carry additional greenhouse gas emissions if the feed were to be sourced from cropland that expanded into previously deforested areas (Macedo et al 2012, Spera et al 2016. For instance, the CF of soybean grown in Mato Grosso, which could be used as feed, was estimated at 12.2 ton CO 2 -eq ha −1 yr −1 (Novaes et al 2017) equivalent to 4.0 ton CO 2 -eq ton −1 soybean.
The use of feed to reduce water consumption or land use for cattle can be deceiving as this strategy often focuses on improving the efficiency per cattle. Feed type and efficiency can accelerate the cattle development cycle and reduce the time to slaughter, leading also to reduced enteric methane emissions per cattle (Talamini et al 2017). However, in absolute terms, a greater cattle population in feedlot finishing can increase water consumption and the indirect land use of the herd through feed; therefore, it is essential to understand the effects of product efficiency in the context of production output.

Conclusion
This study combined new estimates of VWF and LF of cattle production to show water appropriation and land use in Mato Grosso's cattle production system. The live animal population grew by 8 million additional cattle between 2000 and 2014 to reach a total population of 23.7 million heads requiring 51 000 ha of farm impoundments and 27.5 Mha of pasture in 2014. The largest water consumption activity for cattle came from small farm reservoirs, whose evaporative losses should be considered in on-farm water management. The evolution of cattle production suggests that the demand for water resources will increase if cattle ranching intensifies. This water will be sourced either through further development of farm reservoirs, or by pumping from surface water bodies or groundwater. A greater use of feed for animal feedlot finishing will call for greater land and water appropriation for feed with potential water savings based on more efficient water use for cropland and pasture. Future research should explore further the details of the water balance of small farm reservoirs such as the inter-annual water availability and its relationship to cattle production, particularly considering the warmer and drier climate conditions expected for the region. by the Belmont Forum and the G8 Research Councils Freshwater Security Grant (# G8PJ-437376-2012) through the Natural Sciences and Engineering Research Council (NSERC) to MSJ. Support by NSERC was provided to MJL through the Vanier Canada Graduate Scholarship (# 201411DVC-347484-257696). Support for MM and KS came from NASA (# NNX16AI55G) and NSF-DEB (# 1457602). The authors would like to thank Erasmus zu Ermgassen, Trent Biggs, Benjamin Geffroy, Michael Coe, Cécile Bulle, and Andrew Black for comments on the initial version of the manuscript.

Data availability
Any data that support the findings of this study are included within the article, in the accompanying supplemental material and supplemental data.