Simulated hydroclimatic impacts of projected Brazilian sugarcane expansion


  • M. Georgescu,

    Corresponding author
    1. School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, Arizona, USA
    2. School of Mathematical and Statistical Sciences, Global Institute of Sustainability, Arizona State University, Tempe, Arizona, USA
    • Corresponding author: M. Georgescu, School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ 85287–5302, USA. (

    Search for more papers by this author
  • D. B. Lobell,

    1. Department of Environmental Earth System Science and Center on Food Security and the Environment, Stanford University, Stanford, California, USA
    Search for more papers by this author
  • C. B. Field,

    1. Department of Global Ecology, Carnegie Institution for Science, Stanford, California, USA
    Search for more papers by this author
  • A. Mahalov

    1. School of Mathematical and Statistical Sciences, Global Institute of Sustainability, Arizona State University, Tempe, Arizona, USA
    Search for more papers by this author


[1] Sugarcane area is currently expanding in Brazil, largely in response to domestic and international demand for sugar-based ethanol. To investigate the potential hydroclimatic impacts of future expansion, a regional climate model is used to simulate 5 years of a scenario in which cerrado and cropland areas (~1.1E6 km2) within south-central Brazil are converted to sugarcane. Results indicate a cooling of up to ~1.0°C during the peak of the growing season, mainly as a result of increased albedo of sugarcane relative to the previous landscape. After harvest, warming of similar magnitude occurs from a significant decline in evapotranspiration and a repartitioning toward greater sensible heating. Overall, annual temperature changes from large-scale conversion are expected to be small because of offsetting reductions in net radiation absorption and evapotranspiration. The decline in net water flux from land to the atmosphere implies a reduction in regional precipitation, which is consistent with progressively decreasing simulated average rainfall for the study period, upon conversion to sugarcane. However, rainfall changes were not robust across three ensemble members. The results suggest that sugarcane expansion will not drastically alter the regional energy or water balance, but could result in important local and seasonal effects.

1 Introduction

[2] Energy security concerns and displacement of fossil fuels are principal motives for continued investment in bioenergy production. As the second largest global consumer and producer of bioethanol, Brazil is a key participant in the production and provision of bioenergy [Nass et al., 2007]. Brazil already exports about 15% of its current ethanol production but is in a unique position to supply additional biofuels to meet the increasing demand of both domestic and global markets [Nass et al., 2007; Leite et al., 2009]. Recent trade reforms are expected to encourage increased production of sugarcane-derived ethanol, with more than a tenfold rise anticipated within the coming decade [Cowie, 2011]. Intensification of Brazil's sugarcane industry will require future land-use change, which is expected to occur mainly in the country's native cerrado (i.e., tropical savannas) [Leite et al., 2009].

[3] Environmental assessment of Brazil's expanding sugarcane industry has largely focused on biogeochemical consequences [Lapola et al., 2010], a justifiable approach given the need to evaluate suitable land-use transition options that incur a minimum carbon debt [Gibbs et al., 2008], or none at all. This focus on biogeochemical consequences is partially based on the presumed lack of significant hydroclimatic impacts associated with replacement of agricultural and native lands with sugarcane plantations. However, additional direct impacts resulting from land-use change and the effect on the region's surface energy and water balance do require attention, in light of recent [Rudorff et al., 2010; Loarie et al., 2011] and anticipated [Leite et al., 2009] expansion of sugarcane plantations.

[4] Land use and land cover change (LULCC), via modification of biogeophysical properties that influence the manner in which energy is absorbed at the surface and repartitioned to the overlying atmosphere, acts as a first-order climate forcing agent on local to regional scales [Baidya Roy and Avissar, 2002; Wang et al., 2009; Beltran-Przekurat et al., 2011; Pielke et al., 2011; Boisier et al., 2012; de Noblet-Ducoudré et al., 2012; Lee and Berbery, 2012]. Recent work has emphasized the need to consider biogeophysically induced changes from LULCC associated with expanding bioenergy cropping systems in addition to the customary approach focused on biogeochemical assessment [Anderson-Teixeira et al., 2012]. For example, expansion of dedicated bioenergy crops within the US act as a dominant hydroclimatic forcing agent [Georgescu et al., 2009; Vanloocke et al., 2010]. For the United States, direct climate impacts resulting from conversion of annual to perennial bioenergy crops appear to be at least as important for regional climate as the carbon savings from offsetting fossil fuel use [Georgescu et al., 2011]. In Brazil, a satellite-based study of consequences of converting the region's natural vegetation to sugarcane indicates similar near-surface temperatures for the two land covers [Loarie et al., 2011]. However, the impacts of future sugarcane expansion, which may far exceed the scale of past expansion, on regional near-surface temperature and rainfall remain unclear.

[5] Here we evaluate hydroclimatic impacts due to sugarcane expansion within south-central Brazil where the majority of sugarcane plantations are located [Nass et al., 2007] and where future expansion is anticipated [Leite et al., 2009]. Our work extends previous research by incorporating a biogeophysical representation of sugarcane—based on observed measurements [Cabral et al., 2012]—within the Advanced Research version of the Weather and Research Forecasting modeling system (WRF). Through use of a numerical modeling framework, we account for feedback effects between the altered landscape and overlying atmosphere, which cannot be estimated based on observations alone. Multimember, multiyear, simulations are conducted with WRF using current landscape and a hypothetical sugarcane expansion scenario.

2 Materials and Methods

2.1 Weather and Research Forecasting Modeling System

[6] We used WRF version 3.2.1 [Skamarock et al., 2008] to assess hydroclimatic consequences of sugarcane expansion across portions of south-central Brazil. A single domain is utilized, covering most of South America and portions of the Atlantic and Pacific Oceans (see Figure S1 in the Supporting Information). The domain is discretized by 250 and 205 points in the east-west and north-south directions, respectively, with a horizontal grid spacing of 32 km. In the vertical dimension, we utilize 29 levels, with 10 located within the first 2 km of the surface to better resolve planetary boundary layer and associated land-atmosphere processes. The Noah land surface model, widely used in the operational and regional climate modeling community [Ek et al., 2003], is used to update soil temperature and moisture following model initialization. A full accounting of options utilized for all numerical experiments is presented in the Supporting Information section (Table S1). The model's ability to reproduce the climate over South America and over areas undergoing conversion to sugarcane (hereafter Control experiments), evaluated against gridded temperature and precipitation observations, is presented in the Supporting Information (Figures S2S5). We note that although the model favorably reproduces the broad temperature and precipitation pattern of the South American continent (including the region of interest), a wet bias is evident across the simulated domain (Figures S4S5). Positive [Solman et al., 2008] and negative [Lee and Berbery, 2012] precipitation biases of similar magnitude have been simulated for the South American continent, highlighting ongoing concerns regarding precipitation simulation over this region. Future work aims to diagnose the origin of this precipitation excess.

Figure 1.

Biophysical representation of sugar and natural vegetation. Seasonally varying values for (a) albedo and (b) leaf area index are displayed.

Table 1. Naming Convention of All Experiments Performed. Control: Control Experiments, Utilizing Default MODIS Landscape Representation. Sugar: Experiments Utilizing Projected Sugar Cane Expansion. All Experiments Were Repeated Three Times With Staggered Initialization and Variable Spinup Time
Naming ConventionSpinupAnalysis
Control_1January 2003 to December 2003January 2004 to December 2008
Control_2April 2003 to December 2003January 2004 to December 2008
Control_3July 2003 to December 2003January 2004 to December 2008
SUGAR_1January 2003 to December 2003January 2004 to December 2008
SUGAR_2April 2003 to December 2003January 2004 to December 2008
SUGAR_3July 2003 to December 2003January 2004 to December 2008

2.2 Experiments Performed and Sugarcane Parameterization

[7] A suite of 5 year (2004–2008) simulations were conducted, and sensitivity to initial conditions was considered by performing experiments with three different initial start times (subsequently averaged to produce an ensemble mean), with varying model spinup ranging from 6 months to 1 year (Table 1). This approach reduces impacts of internal model noise, adding confidence in the robustness of simulated results.

[8] We use a Moderate Resolution Imaging Spectroradiometer (MODIS) based 20-category landscape classification for all experiments (Figure S1). The domain-wide representation of albedo was based on a monthly varying, 5 year, 0.144° climatological data set [Csiszar and Gutman, 1999; Skamarock et al., 2008]. The observed albedo variations implicitly include variable vegetation density effects, thereby accounting for both soil and vegetated contributions. We simulated sugarcane surface characteristics (hereafter SUGAR) by modifying the relevant biophysical parameters based on recent, 2 year, plantation observations made in southeast Brazil [Cabral et al., 2012]. Our modeling approach assumes periodic annual variability of albedo and leaf area index (LAI; these two parameters were used to parameterize sugarcane. They were the only variables modified relative to the Control experiments) for the 5 year length of experiments (Figure 1), because multiyear measurements of this crop do not currently exist. Given the lack of additional biogeophysical observational data associated with sugarcane crops over this region (e.g., root depth), our approach likely omits added hydroclimatic impacts whose influence could be important [Georgescu et al., 2009, 2011]. Also, given our focus on biogeophysical properties that modify surface absorbed radiation (e.g., albedo) and alter evapotranspiration based on vegetation phenology characteristics (e.g., LAI), aerodynamic properties (e.g., roughness length) were not considered in this work, but will be the focus of future research.

[9] The imposed sugarcane August harvest date (note rapid drop of both albedo and LAI at onset of month) was based on maximum sugar crush (which occurs shortly after harvest) during the months of July and August in Sao Paulo and south-central Brazil [Barros, 2009].

[10] Our simulated sugarcane expansion generally coincides with areas deemed suitable within south-central Brazil [Leite et al., 2009]. Encompassing portions of the states of Sao Paulo, Parana, Mato Grosso do Sul, Goias, and Minas Gerais, conversion to sugarcane within this region would not require irrigation and would avoid environmentally sensitive regions (e.g., the Amazon forest). The precise areas of simulated conversion to sugarcane as well as geographically explicit, seasonally varying biophysical differences of altered parameters between Control and SUGAR simulations are shown in Figure 2. The broad region undergoing conversion is outlined by the black rectangle, delineating the geographical area (~1.1E6 km2) for time series calculations presented in the following section.

Figure 2.

Seasonal albedo difference (Sugar – Control) for: (a) March–April–May, (b) June–July–August, (c) September–October–November, and (d) December–January–February. Black rectangle (latitude 25°S to 15°S, longitude 56°W to 47°W) outlines location of time series calculations, corresponding to areas converted to sugar cane. (e)–(h) Same as (a)–(d) but for LAI.

3 Results and Discussion

[11] Conversion to sugarcane causes opposing seasonal impacts on near-surface temperature, with maximum local cooling approaching 1°C during the growing season and maximum local warming approaching 1°C postharvest (Figure 3). Maximum cooling between SUGAR and Control simulations occurs during the time of peak SUGAR greenness (June–July–August), coinciding with peak reflectivity differences (Figures 2a–2d). Although LAI values are generally greater for SUGAR during this time of season, differences are small (Figures 2e–2h), and this biophysical property plays a minor role. After harvest, SUGAR LAI decreases considerably (Figures 2g–2h) and this biophysical property assumes a prominent role in affecting near-surface temperature change (note warming on conversion to sugarcane despite greater albedo relative to prior land use). Modification of SUGAR harvest timing (and consequently, translation of the entire growing season in time) did not alter the general conclusion of preharvest cooling transitioning to postharvest warming.

Figure 3.

Simulated ensemble averaged 2m air temperature difference (°C) between Sugar and Control simulations for: (a) March–April–May, (b) June–July–August, (c) September–October–November, and (d) December–January–February, for entire length of simulations (2004–2008).

[12] The impacts of sugarcane conversion on surface temperature were negligible outside the region undergoing land use change (Figure 3). We therefore focus our analysis on area-averaged differences over this subregion of the model domain. Five day means of near-surface temperature and evapotranspiration (ET) were calculated for all experiments (to remove high-frequency variability). Area-averaged 2m temperature differences illustrate the distinct cooling (~1°C) during the growing season followed by warming (~1°C) coinciding with harvest (Figure 4a). While variability is evident among individual ensemble members, therefore illustrating the degree of simulated uncertainty, the oscillatory nature of this effect is consistent and occurs for the extent of the simulation period.

Figure 4.

Simulated (a) 2m temperature difference (°C), (b) evapotranspiration difference (mm d–1), (c) accumulated precipitation difference [mm], between Sugar and Control simulations. Simulated volumetric soil moisture difference (m3 m–3) between Sugar and Control simulations for (d) shallow [surface to 40 cm depth] and (e) deep soil layers [40–200 cm depth]. Black lines depict ensemble mean difference and green lines (with 5 day running mean applied to remove high-frequency variability, for Figures 4a and 4b) show individual member differences. All calculations performed over the following subregion (see Figure 2 for geographical extent): latitude 25°S to 15°S, longitude 56°W to 47°W.

[13] SUGAR ET undergoes a 0.5–1.0 mm d–1 decrease relative to the preexisting landscape for several months after harvest (Figure 4b), indicating increased partitioning toward sensible heating (and hence rising near-surface temperatures). This is due to near-zero LAI, in agreement with previous work illustrating enhanced Bowen ratios postharvest [Cuadra et al., 2012]. The onset of the rainy season coincides with sugarcane crop maturation and as SUGAR LAI begins to increase, ET differences are reduced. Nearly negligible growing season ET differences suggest that the simulated cooling trend upon conversion to SUGAR (Figure 4a) is primarily a consequence of albedo changes relative to the preceding land use. Averaged over the extent of our 5 year, multimember simulations, the conversion to sugarcane leads to a 0.32 mm d–1 reduction in ET. Although our results do indicate a wet bias, the annual cycle of simulated SUGAR ET (Figure S6) agrees with field measurements illustrating an observed range between 0.1 mm d–1 (during initial regrowth phase) to 5.3 mm d–1 (fully developed crop) [Cabral et al., 2012]. When averaged across the full extent of our simulations, conversion from the initial landscape (ET: 4.16 mm d–1) to sugarcane (ET: 3.84 mm d–1) translates to a loss of ~8% of water flux to the atmosphere.

[14] Although ensemble mean differences between SUGAR and Control simulations suggest a progressively decreasing trend in total precipitation, changes are of the same order of magnitude as absolute differences among individual ensemble members (Figure 4c). The large uncertainty in simulated precipitation precludes conclusive statements about the impact on the region's rainfall. However, a decrease in precipitation is consistent with simulated reductions in annual ET.

[15] Changes in volumetric soil moisture are also largely consistent with ET changes (Figures 4d and 4e). In general, shallow SUGAR soil moisture is enhanced after harvest, coinciding with decreased ET changes during this time. During the course of green-up, SUGAR LAI rises, leading to reduced differences in ET (between SUGAR and the prior landscape) as the sugarcane's water use begins to increase. No significant trends in soil moisture were simulated.

4 Discussion and Conclusions

[16] We performed a suite of multiyear, ensemble-based simulations with WRF to investigate regional hydroclimatic impacts of sugarcane expansion across portions of native cerrado and agricultural lands within south-central Brazil. Our results show that landscape conversion to sugarcane can lead to significant seasonal effects on near-surface temperature, with preharvest cooling transitioning to postharvest warming of similar magnitude (~1°C). However, when annually averaged, these seasonal effects largely offset one another. We therefore do not expect biogeophysical impacts from the conversion of agricultural and native lands to sugarcane plantations to rival the carbon-saving effect of reduced fossil fuel use. We also find no significant trends in simulated soil moisture (shallow or deep).

[17] Our results are in agreement with recent remotely sensed measurements indicating that temperatures between the region's “natural vegetation and sugarcane should be about the same” [Loarie et al., 2011]. It is instructive to examine each study's approach to provide future research guidance on this topic. While Loarie et al. [2011] account for separate cerrado and crop/pasture as initial starting points for conversion to sugarcane, the greatest land use category undergoing conversion for our simulations is a crop and natural vegetation mixture (i.e., not a separate representation of crop and cerrado), a consequence of our 20-category MODIS landscape classification. The analyzed LULCC between the two studies is therefore different and highlights an important issue that stresses the need for a common framework of land-use change implementation [Boisier et al., 2012]. Although our simulations include all time periods (e.g., cloudy regimes, rainy periods) Loarie et al. [2011] performed their analysis for clear-sky daytime conditions only, and the effect of sugarcane expansion on cloud formation and rainfall is an important consideration. A direct comparison between the two approaches is therefore not possible, but it is encouraging to note that two independent methods reached a similar overall conclusion: the conversion of agricultural and native lands to sugarcane plantations imparts small annually averaged hydroclimatic impacts on near-surface temperature.

[18] In addition, our simulation results indicate an annual average ET reduction of 0.32 mm d–1 upon conversion to sugarcane (the signal is most apparent after crop harvest), qualitatively similar to Loarie et al. [2011] who indicate an ET reduction of 0.17 mm d–1 upon replacement of cerrado lands with sugarcane.

[19] Given the significance of simulated seasonal effects and the likelihood of land use heterogeneity introduced by small-scale sugarcane plantations scattered amongst cerrado and agricultural lands, it is conceivable that mesoscale circulations could develop [Baidya Roy and Avissar, 2002] and modify land surface feedbacks. Future research will require additional analysis that includes consequences for cloud climatology (both shallow and deep clouds) and incorporates sounding profiles that represent the landscape under investigation [Wang et al., 2009] to complement high-resolution, cloud-resolving modeling [Baidya Roy, 2009] focused on climate time scales.


[20] This work was funded by NSF Grant EAR-1204774 and Stanford University's Global Climate and Energy Project.