Study area
Approximately 3/4 of cultivated areas in Brazil are occupied by pastures, amounting to ~ 178 Mha ± 2.5% in 2017 (Parente et al., 2019; Souza et al., 2020). About 35% of these pasture areas (~ 62 Mha) are in the Brazilian savannas, known locally as Cerrado (Parente et al., 2019). The Cerrado is the second largest biome in Brazil, occupying an area of approximately 2 million km², with climatic patterns characterized by strong seasonality, with well-defined dry and rainy seasons. The average annual temperature ranges from 18oC to 28oC, and rainfall, from 800 to 2,000mm, with a very strong dry around April to September (winter season) (Oliveira-Filho and Ratter, 2002).The four main soil groups occupy ~ 86.4% of the entire area of the biome, with Latosol (40.8%) being the most representative class, followed by Neosol (23.4%), Argisol (12.0%), and Plintosol (10.2%) (geoportal.cprm.gov.br/pronasolos).
Century Model
The ecosystem process-based model Century (Parton et al., 1987), version 4.5, was used to estimate C stocks in pastures in the Cerrado biome. The procedure involved three steps, (1) adjusting parameters to reduce differences between reference and simulated values, (2) utilization of these parameters to simulate C stocks in an independent dataset, a process also known as validation, and (3) application of adjustment parameters in estimating C stocks across the area classified as pastures in the Cerrado in 2017 (Parente et al., 2019).
Sites to model calibration and validation
The data used in the model calibration and validation process were provided by researchers who studied the effect of land use on soil carbon stocks in various regions of Brazil (Assad et al., 2013). The sites used were distributed along the Cerrado biome, covering the states of Goiás, Mato Grosso, Tocantins, Maranhão, Bahia and Piauí (Fig. 1), with field sampling carried out in 2010.
Pasture ages
The conversion year from native vegetation to pasture was defined by visual inspection of Landsat 5 satellite images from 1985 to 2010, using the TVI tool (Temporal Visual Inspection) (https:/ /github.com/lapig-ufg/tvi). Nine sites were converted to pastures in the period covered by the images. Their age was precisely determined, which was used in the model parameterization process, starting the simulation of pasture use in the year of conversion to this land use type. The other sites (n = 21) were already occupied by pastures before 1985 and were used to validate the parameterization.
The adjusted Century model parameters were used to estimate C stocks for all pasture areas in the Cerrado with a spatial detail of 1 km2. We used the annual series of maps of this land use class produced by Parente et al. (2019), covering the period from 1985 to 2017 and later extended to 2019 (available on the Brazilian Pasture Atlas platform: https://atlasdaspastagens.ufg.br) to identify the area and pasture age. As a reference, we adopted the areas classified in this land use cover in 2017, and we considered the first year in which the area was labeled as pasture to start the simulations of C stocks. The reference map of pastures, with a spatial resolution of 30 m, was resampled, using the mode criterion, where they remained classified as pastures only 1 km2 pixels with more than 50% of their area classified as pastures in the reference map, totaling ~ 50.7 Mha for the entire biome (Fig. 2). The same criterion was used to determine the pasture age, taking the year of conversion of the pixel as the one presenting the most extensive area as first-time classification.
Weather data
For the calibration and validation sites, monthly averages of historical data (1980–2017) of accumulated precipitation (mm) and temperature (minimum and maximum, °C) from the nearest meteorological stations were used. These were obtained through the Meteorological Database (BDMEP) of the Brazilian Institute of Meteorology (INMET). The correction of missing data was performed by linear regression using data from the nearest station, according to the methodology proposed by Pinto et al. (2016).
For spatialization, the values of monthly accumulated precipitation and temperatures - monthly average maximum and minimum - were obtained from the TerraClimate database (Abatzoglou et al., 2018). The TerraClimate base has a monthly temporal resolution and a spatial resolution of 4 km, covering the period from 1958 to the present, providing all the necessary variables for the Century-based modeling process, with the temporal and spatial coverage and frequency compatible with the objectives of this study, see S1 Fig. The TerraClimate data, based on models and interpolation, present systematic errors compared to local data obtained from weather stations. Therefore, correction factors were generated for the climatic variables based on data from 30 climatic stations in the Cerrado biome. Through these, the systematic errors of the Terraclimate dataset were corrected as proposed by Carvalho et al. (2015). Reflecting the climatic seasonality of the Cerrado biome, in 2017, the average annual precipitation was 1030 ± 204 mm, while the average monthly maximum temperature was 31 ± 1.4oC, and the monthly average minimum temperature was 18.9 ± 1.7 oC.
Soil physicochemical properties
The edaphic characteristics of the points used in the calibration and validation process are described in S1 Table. For each pixel, the model input data related to soil texture (sand, clay, and silt), pH, and soil density were obtained from the SoilsGrids database (Hengl et al., 2014; Batjes et al., 2019). It is a global database with a spatial resolution of 1 km2, which provides information on the physicochemical properties of the entire planet in six layers of depth, see S2 Fig. The average sand, clay, and silt concentrations were 61.3%, 23.6%, and 15.5%, respectively, while the average pH was 5.1 and the density was approximately 1.4 ton/m3 (Table 1).
Table 1
Soil texture, soil density, and Ph data in the Cerrado were obtained from the SoilGrids database and used to model carbon stocks in the pasture areas of the Biome.
Variable | Maximum | Minimum | Average | Standard deviation |
Sand (%) | 84.0 | 28.0 | 61.3 | 9.0 |
Clay (%) | 52.0 | 9.0 | 23.6 | 6.2 |
Silt (%) | 37.0 | 3.0 | 15.5 | 4.3 |
pH KCl | 6.5 | 4.3 | 5.1 | 1.6 |
Density (Kg/m3) | 1535.0 | 1137.0 | 1397.1 | 33.5 |
Century model parameterization
The model was run for 10 thousand years, using parameters pre-adjusted by Ferreira (2013) to represent the conditions of natural Cerrado vegetation without anthropic changes. This initial simulation aimed to achieve stabilization of C stocks in the soil and vegetation in the studied pasture sites, thus representing a system in balance concerning C inputs and outputs. Cerrado and pasture installation, according to parameter adjustments, were considered in such a way that the simulated soil C stock values could approach the values measured in the field.
The adjustment of parameters related to the pasture productivity efficiency (PRDX) was performed based on their effect on C output in biomass and soil, as well as by comparing the values with those obtained in the Rio Vermelho Watershed (Silva, 2019) and in the literature (Oliveira et al., 2015a). The allocation of C in roots or shoots of grass, concerning water or nutritional availability, was adjusted using the parameters CFRTCN(1), CFRTCN(2), CFRTCW(1), CFRTCW(2) to correspond to values found by Oliveira et al. (2015b). In addition, we adjusted the parameter value related to the optimal temperature of grass production (PPDF) (S2 Table) (Dias-Filho, 2011).
The simulated management for pastures was the conventional type practiced in Cerrado, as it is used in most rural properties in the biome (Dias-Filho, 2014). This is characterized by weed control through manual mowing or herbicide, continuous grazing throughout the year, low frequency of pasture fertilization, and cattle herd supplementation. As a result, there is a low stocking rate, depending on the quality and quantity of available grass, and the pastures generally show some degree of degradation (Dias-Filho, 2014).
Model performance analysis
The performance of the Century model to estimate C stock in soils under pasture in the Cerrado biome was evaluated through the response of the variable SOMSC (Carbon in soil organic matter) to the edaphoclimatic characteristics of the site. The mean error of the model was estimated by calculating the Root Mean Square Error (RMSE), considering the differences between the simulated and observed values squared. This result was expressed in proportional terms by dividing the value found by the average of observed soil C stocks (RMSE%). The association between simulated and observed soil C stocks was evaluated using the correlation coefficient (r) described in Smith et al. (1997). The calculation of the Nash-Sutcliffe efficiency coefficient (COE) was also performed, whose values vary between -∞ and 1, where 1 is considered the ideal fit (Nash & Sutcliffe, 1970). Another procedure for evaluating the model was to count the number of studied sites whose relative difference between the simulated and observed value divided by the observed value was within the range of ± 25% (Parton et al., 1993). Thus, the greater the number of sites within this range, the better the model's performance.
From a regional perspective, the responses of the Century model to soil types in areas covered by pasture in the Cerrado were investigated, having as reference the mapping of the Brazilian soil classes, provided by the Brazilian Institute of Geography and Statistics (available on geoportal.cprm.gov.br/pronasolos), and also the age of pastures. In the latter case, soil C stocks were compared between 1985 and 2017. The response of aboveground C biomass to water seasonality was assessed by monitoring the variables AGLIVC (carbon in live aboveground biomass) and STDEDC (standing dead carbon in aboveground biomass) throughout the year 2019.