Technification in Dairy Farms May Reconcile Habitat Conservation in a Brazilian Savanna Region

: The assessment of the relationships between farm management systems and nature conservation may help in the design of more efficient strategies to uphold economic benefits and biodiversity conservation. To our knowledge, this is the first work in Brazil to study the relationship between farm conservation status and technification level. Here, we test the hypothesis that dairy farms with higher levels of technification have a higher percentage of natural vegetation and connectivity, and that differences in environment features between farms explain their conservation status. We obtained variables related to the level of technification such as feed, milking, sanitary control and breeding management systems. We show that farmers with a higher level of technification, such as artificial insemination in cattle breeding, tended to conserve a higher percentage of natural vegetation, as well as larger farms with a higher percentage of riparian forest. The adoption of artificial insemination is associated with other technification systems such as a forage diet, milking method and frequency and sanitary control. It is also significantly related to higher milk productivity. Our novel results point to a positive effect of technification on the conservation of natural vegetation, suggesting that economic incentives and programs aimed at increasing technification in cattle breeding may increase dairy production and conservation within the study area. Our findings also show an effect of larger areas of riparian forests, which are protected by Brazilian policy, in the conservation status of dairy farms.


Introduction
The trade-off between food production and biodiversity conservation is one of the major challenges for policymakers in the Anthropocene [1]. The expansion of agriculture and the use of unsustainable practices are the main threats to terrestrial natural habitats [2] and to biodiversity conservation in tropical and temperate regions [3]. The global agricultural land area was 4.9 billion hectares (Gha) in 2016-38% of the global land surface [4]. Two thirds (3.3 Gha) of the agricultural land were used as pastures for livestock.
Pastures are the main anthropic disturbances world-wide and have expanded more rapidly than cropland. While agriculture expanded from 265 ha to 1471 million ha from 1700 to 1990, pastures expanded from 524 ha to 3451 million ha in the same period [5] and environmental impacts such as soil degradation [37]. Because increasing the efficiency of livestock farming can reduce pasture expansion and deforestation [38], understanding how technification levels in farms can reduce deforestation is of utmost importance for Cerrado conservation and decreasing greenhouse gas emissions, as we investigate here.
Here we address the relationship between the level of technification and conservation in dairy farms in an intensive-farming landscape (Figure 1). We specifically test the hypothesis that dairy farms with higher levels of technification have higher amount of natural vegetation and connectivity. Alternatively, to account for the effects of environment, we test whether farm area and differences in environment features among farms, such as slope, percentage of riparian forest along water courses, percentage of agriculture and pasture explain conservation of natural vegetation in dairy farms. Riparian forests, i.e., forests adjacent to water courses, are Areas of Permanent Protection (APPs) meant to protect sensitive ecosystems by Brazilian environmental law. We hypothesize that farms with more water courses, and thus with higher amount of riparian forest and lower amount of agriculture or pasture have higher amount of natural vegetation and connectivity, despite the technification level. For this, we applied questionnaires to characterize dairy production and farm management system, and to obtain variables related to the level of technification in feed and milking management, sanitary control, and genetics and breeding systems. Using the Brazilian SICAR (Brazilian System of Rural Environment) database of rural properties, we obtained farm boundaries and mapped the area of natural vegetation remnants and estimated the connectivity among these remnants to predict the effects of technification on natural vegetation conservation.

Study Area
The study was carried out in a Brazilian intensive-farming landscape comprising a Long-Term Ecological Research (LTER) project called COFA-LTER (Functional Connectivity in an Agricultural Landscape, Figure 1), in the Central-West Brazil, one of the most important Brazilian agribusiness regions ( Figure 1). The COFA-LTER landscape comprises a reserve (Silvânia National Forest), the urban area of the city Silvânia, and the rural area with farms.
We mapped the land cover in COFA-LTER landscape using visual digitalization and manual classification of high-resolution images of Google Earth from 2019, freely available at the Geographic Information Systems QGis 2.4, with validation by field checking. The final map corresponded to 5 m spatial resolution, comprising 11 different land cover classes ( Figure 1): (i) water courses; (ii) savanna and open savanna; (iii) seasonal and riparian forests; (iv) wetland; (v) pasture; (vi) agriculture (corn or soybean); (vii) rural building; (viii) mining; (ix) urban area; (x) road and train rail, and (xi) Eucalyptus spp. plantation.
All dairy farms in the COFA-LTER landscape (25 farms) were identified and mapped ( Figure 1, Appendix S1, Table S1). Farm boundaries (Appendix S2 Figure S1) were obtained in the Brazilian SICAR (Brazilian System of Rural Environment) database of rural properties (http://www.car.gov.br/publico/imoveis/index, accessed in February 2020) from 2019, or were manually digitalized in Google Earth.

Technification Variables and Farm Characteristics
We obtained variables related to the level of technification in the dairy farms and farm management practices using face to face interviews with farmers (25 farmers) in 2019. Before the interviews, we had several meetings with local stakeholders, including local government, non-governmental organizations (NGOs), researchers, farmers and technical assistants, to design a robust research approach in COFA-LTER project. All farmers signed an informed consent before interviewing.
We measured five different characteristics of technification in dairy production (Table 1), following the milk production systems prevalent in Brazil [39]: (i) productivity [40], measured by daily milk production; (ii) feed management [41,42], measured by primary forage diet, fertilizer pasture management, time of supplementary feed, criterium to supplement, and addition of vitamins to supplement; (iii) milking management [43], measured by milking method, milking frequency, and farm infrastructure for milking; (iv) sanitary control [44], measured by cleaning of udder before milking and CMT test (California Mastitis Test); (v) genetics and breeding [45], measured by breeding program for productivity improvement, selective breeding method and pregnancy diagnosis. We used daily milk production as a measure of productivity because farmers had no information on milk production per cow.
Besides technification variables, we obtained variables related to the overall milking management to describe dairy farms in the studied area (Table 1): (i) cattle breed composition; (ii) pasture restoration during the last decade (iii) milk cooling tank; (iv) milk quality-based payment; (v) management separating pregnant cows.

Environment Feature Variables
We obtained the slope map from the TOPODATA INPE (National Institute for Space Research, http://www.dsr.inpe.br/topodata/acesso.php) database with 30 × 30 m spatial resolution. For each farm, we used the mean slope of all pixels. Using farm boundaries (Appendix S2, Figure S1), we identified areas of natural vegetation in each farm and calculated the farm area, the percentage of riparian forest, and percentage of agriculture and pasture. The available soil maps for the studied area were performed in small scale with few details (1:250,000) and show that the region is comprised mostly by dystrophic red oxisol with low variation among farms; thus, soil type could not be used as explanatory variable.

Conservation Status of the Dairy Farms
We used percentage of natural vegetation and connectivity as proxies of the conservation status of each farm (response variables). To verify the conservation status of natural vegetation that exceeds riparian forests in farms, we calculated the legal reserve area, which are areas of natural vegetation that should be preserved following Brazilian environmental law, corresponding to 20% of the total farm area in the Cerrado biome. To calculate legal reserve area, which is independent of APPs (Areas of Permanent Protection) such as riparian forests, we obtained the total percentage of natural vegetation and extracted the percentage of riparian forest. The highest slope in the 25 farms analyzed was ~13%, lower than the slope considered as APP by Brazilian environmental law (> 45° in hillsides or 25° in hilltops). Thus, all APPs in our study area are riparian forests.
Connectivity is a measure of the landscape permeability to species movement among habitat patches [46,47]. To measure connectivity among remnant patches of natural vegetation in farms we used the Integral Index of Connectivity (IIC, [48]) implemented in Conefor [49] available in R version 3.6.1. [50]. The IIC, implemented in Conefor software [49], is based on the concept of graph theory and habitat availability, integrating habitat amount and the connectivity among patches of habitat [48]. IIC ranges from 0 (no connectivity among patches within the landscape) to 1 (high connectivity between remnant patches).
We calculated IIC considering each farm as a different landscape and habitat as all categories of natural vegetation. We used the mean value of IIC among four spatial distances as the response variable (100, 300, 500 and 1000 m), because most farms in the studied area were small and IIC values for all distances evaluated were very similar (Table  S1).

Statistical Analyses
To analyze the effect of technification and environment features in conservation of dairy farms we first verified the multicollinearity among explanatory variables. For technification variables we used the generalized variance-inflation factor (Gvif), and for environment features we used the variance-inflation factor (Vif), excluding in this step the farm size which is a categorical variable. We used the car package [51] implemented in R version 3.6.1. For technification, we ran one model per variable group (Table 1). We used a stepwise approach eliminating the models with Gvif or Vif > 5.0 [52]. Gvif is equivalent to Vif for categorical variables, the inflation in size of the confidence ellipse or ellipsoid for the coefficients of the predictor variable in comparison with what would be obtained for orthogonal, uncorrelated data [53].
We then analyzed the effects of technification and environment features on farm conservation status using linear models. We defined all categorical variables as "factors," except daily milk production and environment variables, which correspond to continuous variables. To select the best variable explaining percentage of natural vegetation or IIC, we used the function drop1 in R version 3.6.1. [50] with stepwise backward selection and Chi-Square distribution to calculate the significance of each variable per group, considering technification and environment variables separately. We considered as significant all variables with p <= 0.05, and marginally significant variables with p <= 0.10.
We tested the significance of association between pairs of technification variables using Kendall's coefficient of concordance (Kendall's tau-b) to test for concordance or attribute agreement between variables. Analyses were performed using the software Minitab ® 18.

Dairy Farm Characteristics
Although farmers declared legal reserve area equal to the minimum required by Brazilian environmental law in Central-Brazil (20% of the farm area, Table S1), 84% of the farms had lower percentage of natural vegetation than the minimum required for legal reserve (Figure 2a, Supplementary Table S1). The total area of natural vegetation ( Figure  S2, Table S1) comprised 5.42 to 46.55% of the farm area (mean = 24.84%, SD = 11.45%), and only 16% of the dairy farms had proportion of natural vegetation > 20%, excluding riparian forests (Figure 2a).
Most dairy farms (80%) were family or small size, and only one was a large dairy farm (Figure 2b, Table S1, Figure S2). Mean IIC ranged from 0.0023 to 0.2141 ( Figure S3, Table S1), and the mean overall farms was very low, 0.0646 (SD = 0.0561), meaning that the connectivity among remaining patches of natural vegetation within farms was very low (Figure 2c).
Most dairy farms use free grazing (pasture-raised dairy) and semi-confinement as primary forage diet system (Table S2). They use fertilizers in pasture management and perform pasture restoration, and provide supplementary feed during the dry season (Table S2). Most farmers use productivity as the criterium to give supplement to the cattle. Machine with bucket is the most used milking method, and milking twice-a-day is more frequently used (Table S2). Most farmers milk in a milking pit or in a milking parlor with roof and cement floor, use farm milk cooling tank and can receive milk quality-based payment (Table S2). Most farmers use commercial sanitizer to cleaning udder before milking and do not perform CMT test or perform monthly (Table S2). Most farmers use selective breeding method with natural service with selected breeding bulls and pregnancy diagnosis with transrectal palpation (Table S2). Farmers have Holland or crossbreed dairy cattle and do not separate pregnant cows.

Technification and Conservation Status
We excluded the variables primary forage diet (Gvif = 8.974) for feed management (Table S3), and milking method (Gvif = 5.093) for milking management (Table S3) due to collinearity. For sanitary control and genetics and breeding we kept all variables (Gvif < 5.0, Table S3).
We found that selective breeding method (p = 0.046) significantly explained the variation in percentage of natural vegetation among dairy farms ( Table 2 and Table S4). Dairy farms applying artificial insemination (Figure 3a) had higher proportion of natural vegetation. Breeding program for productivity improvement was marginally significant (p = 0.059) and daily milk production did not explain variation in percentage of natural vegetation (p = 0.789). When analyzed in the final model (Table 2), feed and milking management, and sanitary control variables did not explain farm conservation status (all p > 0.05; Table S4). Connectivity (mean IIC) was not explained by feed and milking management or sanitary control variables, or daily milk production (all p > 0.05; Table S5).  We found significant association between selective breeding method and primary forage diet (p = 0.001, Table S6). In addition, daily milk production was significantly higher (t = 2.73, p = 0.026, Figure 4) in farms that used artificial insemination (mean production = 1023 l, SD = 933.0 l) than in farms that used natural service with selected breeding bulls (mean production = 166.0, SD = 165.0).

Figure 4.
Relationship between selective breeding method and daily milk production in the 25 dairy farms in the COFA-LTER long-term project landscape. 1, natural service with selected breeding bulls; 2, artificial insemination. The box plot represents the median (dark bar), the third quartiles (the box) and the minimum (lower bar) and maximum (above bar) values.

Environment Features and Conservation Status
We excluded percentage of agriculture (Vif = 7.945) due to collinearity (Table S7), keeping slope, percentage of riparian forest and pasture in the models (Vif < 5.0, Table S7).
Percentage of riparian forest (p < 0.001) and farm size (p = 0.041) significantly explained the percentage of natural vegetation conserved in dairy farms ( Table 3, Table S8). Farms with higher percentage of riparian forest (Figure 3b Table S8). We found similar results for IIC (Table 3, Table S9). Farms with higher percentage of riparian forest (p < 0.001, Figure 3c) and larger farms (p = 0.019, Figure 2d) had significantly higher connectivity.

Discussion
Here we show for the first time that farmers using cutting edge technology such as artificial insemination in cattle breeding tend to conserve higher percentage of natural vegetation in the COFA-LTER landscape. However, larger dairy farms and with higher percentage of riparian forest also showed higher conservation status. Cattle raising farms in Brazil use predominantly extensive pasture-raised (free grazing) system with low productivity, which may hamper conservation of natural vegetation due to the constant need for new pasture areas [21,54]. In the Cerrado biome, ~39% of pastures are degraded [27]. In the COFA-LTER landscape, most dairy farmers use pasture-raised with free grazing or semi-confinement that includes free grazing and free-stall or Barn systems, which is similar to the most used feed management system in Brazilian dairy farms (e.g., [55]). The predominance of these feed management systems may contribute to the relatively low percentage of natural vegetation and connectivity among patches of vegetation in dairy farms due to the demand of pasture and grain plantation for cattle feeding. Few dairy farmers (16%) had higher proportion of vegetation remnants than 20%, which is the minimum determined by Brazilian environmental law (see Figure 2a). Moreover, larger farms with higher percentage of riparian forest preserved higher percentage of natural vegetation, and had higher connectivity (see Figure 2b,c). However, it is important to note that overall connectivity was very low, which may compromise biodiversity conservation in the landscape.
Indeed, this result was not a surprise, because farmers do not have technical support to design priority areas for conservation within the rural property that would maximize the trade-offs between ecological and economic benefits, such as improving connectivity. Usually, farmers in Brazil conserve unproductive areas, or with technical limitations for management that will not interfere in daily management practices (see [56]). The higher conservation status in larger farms with higher percentage of riparian forest suggests that in the COFA-LTER landscape farmers are not preserving vegetation above the required by Brazilian environmental law. Farmers may adopt conservation practices based mainly on short economic benefits and in other factors such as previous experience, familiarity with technologies, perceived risks, labor requirements, and interactions with peers and advisors [57]. In addition, public visibility and the influence of neighbor farmers that use conservation practices can drive the decision of surrounding farmers to adopt new or more ecological management practices [57]. To our knowledge, the local government has no economic incentives to guarantee conservation, restoration or sustainable practices to improve local and regional connectivity and conservation.
Furthermore, the adherence of farmers to the SICAR (Brazilian System of Rural Environment) seems to be more related to restrictions to credit and rural financing and insurance imposed by Brazilian policies than to conservation. SICAR declaration is related to the access to rural credit and adoption of practices to improve pasture management in cattle farms in Brazil [32]. In the COFA-LTER four dairy farms have no register in the Brazilian government database SICAR, that requires the information of legal reserve area or which area will be restore in the future to comprise the legal reserve and APPs. However, even with the SICAR declaration, most of the dairy farmers in the COFA-LTER landscape (84% of the dairy farms) do not conserve the minimum of natural vegetation established by law, suggesting that the current environmental policies provide no additional incentive to conservation in the study area.
However, it is important to highlight that positive outcomes can be identified at regional scales, as we found higher percentage of natural vegetation in a set of dairy farms that use higher levels of technification (see Figure 3). Furthermore, planned interventions with few interferences on the production systems can favor local biodiversity and ecosystem services [58,59]. Some of these interventions support the restoration of strips of natural vegetation between the crop fields [60] and the reduction of agrochemical inputs [61]. In the COFA-LTER landscape farmers are prone to restore areas of natural vegetation, particularly farmers with ecosystem service awareness and with higher number of springs in their properties [62].
Feed management may affect greenhouse gas emission directly due to both the forage production and feed conversion rates [63], and indirectly due to expansion of new areas of pasture and grain plantation for animal feed leading to deforestation [64,65]. However, we found no relationship between feed management and the percentage of natural vegetation, and also no relationship of percentage of pasture and conservation status. It is possible that farmers may have cleared their farms for pasture planting before the adoption of cutting-edge technology, and then turned to higher levels of technification [21,32,66] such as insemination, milking using automatic machines and confinement.
The increase in productivity of cattle raising and in general, the ecological intensification of agriculture [57] may improve not only direct economic incomes but also conservation of vegetation remnants [22]. In fact, our results evince that increasing technology and cattle raising efficiency may improve economic benefits. Artificial insemination was directly related to productivity in COFA-LTER landscape, i.e., dairy farms applying artificial insemination tended to have higher daily milk production (see Figure 4), which may increase economic benefits. The benefits of applying technification can also be detected in other important indicators of productivity, for instance, dairy farms adopting artificial insemination (mean = 48.6, SD = 30.5) have higher number of dairy cows (t = 3.25, p = 0.012) than those with natural service with selected breeding bulls (mean = 14.9, SD = 7.4).
However, although our novel findings indicate opportunities to improve economic and ecological benefits on dairy farms, these opportunities may vary throughout Brazilian territory, restricting our conclusions to the study area. Also, it is not easy to establish patterns in the production systems in Brazil. The country is very heterogeneous in a set of factors such as the environment, technology availability and adoption, which may establish patterns utterly different in the same production system such as dairy farms.
Sustainable intensification in Brazil can potentially increase 113% in pasture grazing beef and milk production without increasing land demand and sparing areas for biodiversity conservation [67]. Our findings point to important outcomes about technification and conservation in the COFA-LTER landscape, showing that additional economic incentives and programs aimed at improving technification in cattle breeding can increase dairy production and, consequently, may lead to additional incentives to the conservation and restoration of vegetation remnants. Besides artificial insemination, our results suggest that practices such as confinement and semi-confinement may contribute to less land demand. However, it is essential to mention that non-ecological intensification practices may reduce pasture area, but may cause a set of additional impacts on the environment. Farmers tend to adopt new management practices when they do not require high modification in the establish farming systems [57]. Furthermore, evidence-based new management systems have higher chances of adoption when the most relevant costs and benefits to the farmers are clear [57]. In this context, we are providing evidence that technification may bring positive income to dairy farms in the COFA-LTER region.

Concluding Remarks
This is the first study in Brazil showing that dairy farms using cutting edge technology in cattle breeding, such as artificial insemination, significantly conserve higher proportion of natural vegetation in their farms. Therefore, technification affected positively the conservation of natural vegetation in the COFA-LTER landscape. Farmers adopting artificial insemination had higher daily milk production and tended to use other technification systems related to dairy farm management more frequently than the other farmers, such as confinement and semi-confinement, and milking with automatic machine.
Brazil harbors a diversity of social, environmental, and economic conditions with can provide different perspectives to other parts of the territory and other productive systems (see an example in [32]). Our dataset provides evidence that incentives to technification may promote economic benefits to dairy farmers and can provide opportunities to preserve or restore habitats in COFA-LTER landscape. Our dataset can be a starting point to design successful strategies aimed at the conservation and dairy production in the study area. However, the use of new practices or strategies can be seen as a barrier if enable conditions are not offered safely, like technical assistance and training for labor [32,67].
The significant relationships between farm size and percentage of riparian forest and farm conservation status highlight the importance of riparian forests to maintain the connectivity at farm level, and to favor conservation status in productive dairy regions. Ecological initiatives should be designed at local and regional levels to favor essential components such as connectivity and designing farms and landscapes more permeable to organism's movement. Additional incentives of local government are essential to support the dairy farmers to achieve economic benefits, sustainable intensification, maintain or restore habitats, and ensuring ecological benefits within farms. These initiatives are necessary independently of farm size, to promote a conservation status at landscape level. We acknowledge the limitations of our results due to the sampling size, i.e., the number of dairy farms analyzed, and the dairy farm sizes. Most farms in the COFA-LTER region are family, small or medium size, thus our results may be used with caution to large size and very large states. Also, the specific environmental conditions of the region may also influence the results, since riparian forests are important landscape components in the region due to the topography and geomorphology. Also, more studies are necessary at the farm level to understand the relationship between technification and conservation, identify barriers, and promote efficient strategies to fix it.

Institutional Review Board Statement:
The statistical analyses involved only analyses of the rural property data and no association between farmer's personal data and the rural property data. The wok was approved by the Universidade Federal de Goiás (UFG, Federal University of Goiás, Brazil) ethics committee.

Informed Consent Statement:
The present work involved face to face interviews with farmers. All farmers signed an informed consent for before being interviewed, as stated by Brazilian law.
Data Availability Statement: Data and additional supporting information may be found in the online version of this article as supporting information.