Effects of land cover change on moisture availability and potential crop yield in the world's breadbaskets

We used crop specific yield and harvested area data (Monfreda et al 2008) to define five major breadbaskets. By combining national, state and county level census statistics with several global satellite products, the data of Monfreda et al provided observed harvested area and yield for a multitude of crops on a global 5' × 5' grid (∼9 km × 9 km at the equator) around the year 2000. In this study we limited our focus to three prevalent primarily rainfed crops: maize, spring wheat and soybeans. Combined, these three crops make up approximately 40% of global cropland (Leff et al 2004) (supplemental figure 1 available at stacks.iop.org/ERL/7/014009/mmedia). The primary crop missing from this study was rice, due to its unique and generally irrigated growing conditions. Additionally, we used the composite fractional area of all croplands and pastures using data from Ramankutty et al (2008) (supplemental figure 1(d) available at stacks.iop.org/ERL/7/014009/mmedia), and a potential vegetation dataset for our reference vegetation (Ramankutty and Foley 1999).

The observed growing season for each crop was estimated using planting/harvesting data from observational data compiled in Sacks et al (2010). That study primarily used data from the United Nations Food and Agriculture Organization (FAO) and the United States Department of Agriculture (USDA) to construct global maps of planting/harvesting dates for 19 crops, which represented 71% of the globally cultivated area.

Using the crop data, we selected five regions as breadbaskets to be the focus of this study. These regions are shown in figure 1. They include maize in the Midwest United States (US), soybeans in Southeast South America (SA), maize in West Africa (WA), wheat in the Central Asian wheat belt (CAS) and wheat in East Asia (EA). The observed median planting and harvest dates for each region are shown in table 1.

Zoom In Zoom Out Reset image size

The breadbasket regions shown in figure 1 were chosen taking several factors into account. First, we selected regions that represented major crop producing regions. Second, regions were selected to be geographically distinct and represent unique biomes. Also, winter wheat growing regions were excluded as surface fluxes of moisture and energy tend to be relatively small for much of this crop's growing season, and modeling of winter wheat crop yields present unique challenges beyond the scope of this work. This prohibited investigation of the extensive Indian and Western European wheat growing regions observed in supplemental figure 1(b) (available at stacks.iop.org/ERL/7/014009/mmedia). Finally, we selected regions that represented a diverse set of climatological and meteorological conditions. The US and CAS regions represented agricultural regions where precipitation is largely influenced by midlatitude storm tracks. The SA region encompasses a relatively large range in latitude and is influenced by a combination of the tropical movement of the Intertropical Convergence Zone (ITCZ) and midlatitude systems. Precipitation in WA is strongly tied to seasonal movements of the ITCZ and tropical dynamics. Finally, EA represents a midlatitude crop growing region where the moisture for crop production is influenced by monsoonal systems (Wallace and Hobbs 1977, Hastenrath 1991).

The sources of water that precipitated over breadbasket regions were found using an evaporative source (ES) dataset developed in Dirmeyer and Brubaker (1999, 2007) and Brubaker et al (2001). For each terrestrial gridpoint on a global T62 Gaussian grid (∼1.9°×  1.9° at the surface on the equator), this source provided a 25 yr (1979–2004) mean monthly climatology of where the precipitation that fell over a given point last evaporated from Earth's surface (mm H₂O m⁻²). These data have been used in a variety of applications including studies of moisture recycling, hydrological feedbacks of floods in the Midwest United States, and estimates of atmospheric water vapor transports between countries (Dirmeyer and Brubaker 2007, Dirmeyer et al 2009, Dirmeyer and Kinter 2010). For more details of the ES data see supplementary information (SI1 available at stacks.iop.org/ERL/7/014009/mmedia). There are alternative methods for calculating ES such as those by Yoshimura et al (2004) and van der Ent et al (2010). While a full comparison of these methods is beyond the scope of this study, we do acknowledge that our results may vary to some degree with alternative methods for ES estimation. Here, we used the ES dataset to calculate the total ES footprint of moisture that precipitates over the fraction of each breadbasket region containing the specified crop. This was aggregated over the growing season of each of the breadbasket regions, with the growing season defined as the time interval between the observed median planting date and median harvest date (table 1). 30 yr mean NCEP reanalysis-II 850 mb fields were also used in this analysis (Kanamitsu et al 2002).

To determine changes in surface fluxes of moisture associated with LCC, and potential changes in crop yield associated with changing moisture availability, we used 25 yr runs of the Predicting Ecosystem Goods and Services Using Scenarios (PEGASUS) model described in Bagley et al (2011) and Deryng et al (2011). To simulate the effects of LCC, PEGASUS used a set of fifteen biomes and three crops (maize, soybean and spring wheat) to describe the vegetative state at each gridpoint. The biome assigned to any gridpoint was manually altered to represent a change in surface cover by LCC. Associated with each biome were a set of literature-derived parameters describing water, energy, and nutrient properties.

Potential crop yields in PEGASUS were calculated by integrating the effects of climate, planting dates, crop specific irrigated areas, cultivar choices and fertilizer choices for maize, wheat and soybeans. For this study, nutrients were assumed to be unlimited, planting and harvesting dates were set to those currently observed using the planting/harvesting data from Sacks et al (2010), and all crops were assumed to be rainfed, assumptions that we examine later in this letter. The sensitivity of potential crop yields in PEGASUS to changes in precipitation and moisture availability can be found in Deryng et al (2011). Further description of the PEGASUS model, and its simulation of surface energy balance and potential crop yield can be found in the supplemental information (SI2 available at stacks.iop.org/ERL/7/014009/mmedia).

To determine the potential impact of LCC on breadbasket moisture availability via alterations to a region's ES we developed a simple linear model. To begin, the total moisture availability (M) at gridpoint i (mm H₂O day⁻¹ m⁻²) can be written as:

$$\begin{eqnarray} &&{M}_{i}=\frac{\sum _{j=0}^{n}{S}_{i,j}{A}_{j}}{{A}_{i}} \end{eqnarray} \tag{ 1 }$$

where s is the ES (mm H₂O day⁻¹ m⁻²) of gridpoint i at gridpoint j,n represents the global number of gridpoints, and A (m²) is the area represented by a given gridpoint. Next we assumed that fractional change in ES for gridpoint i due to LCC at gridpoint j was equal to the fractional change in surface evapotranspiration E (mm H₂O day⁻¹):

$$\begin{eqnarray} &&\frac{{S ^{\prime}}_{\!i,j}}{{\bar {S}}_{i,j}}=\frac{{E ^{\prime}}_{\!j}}{{\bar {E}}_{j}} \end{eqnarray} \tag{ 2 }$$

where overbars indicate climatological means, and primes indicate perturbed values due to LCC. Thus the fractional change in moisture availability (F) due to LCC can be written as:

$$\begin{eqnarray} &&{F}_{i}=\frac{{M ^{\prime}}_{\!i}}{{\bar {M}}_{i}}=\frac{\sum _{j=1}^{n}{\bar {S}}_{i,j}{A}_{j}\frac{{E ^{\prime}}_{\!j}}{{\bar {E}}_{j}}}{\sum _{j=1}^{n}{\bar {S}}_{i,j}{A}_{j}}. \end{eqnarray} \tag{ 3 }$$

Equation (3) allowed us to estimate potential changes in moisture availability in breadbasket regions for LCC scenarios, and required only fractional changes in evapotranspiration from each scenario. These changes in evapotranspiration were modeled with PEGASUS and depended on soil and vegetative properties, as well as climatological rates of precipitation. Essentially, the linear model assumes that fractional changes in ES for any grid point proportionally change the precipitation and hence moisture availability at all downwind grid points, as determined by the ES dataset. It should be noted that although changes in atmospheric circulation and stability are key to fully encompass changes in precipitation due to LCC (Charney 1975, Eltahir 1996, De Ridder 1997, Nobre et al 2009), they are not accounted for in this study in order to focus exclusively on changes in moisture availability due to alterations in moisture supply from LCC, and results should be interpreted with this caveat in mind. Depending on whether the circulation and stability feedbacks on precipitation are negative or positive, our approach may overestimate or underestimate the impact of LCC on cropland moisture availability.

To estimate the maximum potential impact of LCC on breadbasket moisture availability, we simulated two 25 yr scenarios using the PEGASUS model. In the first, we set the land surface to potential vegetation and precipitation to CRU 2.1 30 yr monthly means linearly interpolated to daily values. This scenario represented the climatological means (${\bar {s}}_{i,j},{\bar {M}}_{i},{\bar {E}}_{j})$ in equation (3). In the second scenario all land that was not agricultural land, as depicted in supplemental figure 1(d) (available at stacks.iop.org/ERL/7/014009/mmedia), was converted to bare soil. This represented the maximum potential impact of LCC on moisture flux (although other possible conversions such as vegetation to urban landscapes could have larger impacts on moisture fluxes). This scenario represented the perturbed values due to LCC in equation (3) (E'_j,M'_j). Using equation (3) we estimated the potential change in breadbasket moisture availability for points containing a given crop within each breadbasket region shown in figure 1. Finally, 25 yr PEGASUS model simulations with precipitation patterns altered to reflect changes in moisture availability as calculated above were integrated to calculate changes in potential crop yield relative to climatological conditions. Additionally, a series of LCC scenarios were considered by limiting LCC to encroaching areas around breadbasket regions excluding land already under agricultural usage and incrementally increasing the magnitude of LCC (SI3 available at stacks.iop.org/ERL/7/014009/mmedia).

To investigate the potential impacts of land cover change on the breadbasket regions, we determined the spatial patterns of ES for each region. The ES (shaded) results are shown in figure 2, along with the fractional area of the crop (contours) and NCEP reanalysis-II climatological 850 mb wind vectors.

Zoom In Zoom Out Reset image size

In East Asia (figure 2(a)), there was evidence of extensive moisture recycling within the region, with much of the moisture that precipitates over wheat last evaporating off the surface from the region itself. The East Asia region also had a long westward tail that followed the grassland and mixed forest biome to the north and west of the Gobi Desert, which contributed significant amounts of moisture to growing season moisture availability in the northern portion of the breadbasket region. Meanwhile only a small percentage of wheat's moisture source in the region originated over the ocean, and that evaporation was largely constrained to the Yellow Sea and South China Sea. This was related to circulation patterns associated with the East Asian summer monsoon, which is the climatologically dominate feature throughout much of the April–July growing season of wheat in the region (Lee et al 2008).

For the Central Asian wheat belt region, the ES pattern was primarily zonal for the May–August growing period, with large amounts of moisture recycling (figure 2(c)). This was anticipated given the large range of longitude (∼40°) encompassed by the region and the zonal flow of climatological winds in the region. In the extensive wheat crops of the Ukraine, Kazakhstan and Russia the moisture availability was largely dependent on water supplied by terrestrial evaporation from Eastern Europe and local sources. This region had a very limited oceanic influence, although the Black Sea and Caspian Sea did contribute to the ES.

In the Midwestern United States, moisture for maize growth is largely supplied by terrestrial evaporation to the southwest of the main growing region (figure 2(d)). Here, the regional recycling appeared to be small relative to other regions. There was also a clear influence in the US maize's ES from the Gulf of Mexico. Previous studies have found significant variations in precipitation in the United States to be correlated to changes in the nocturnal low-level jet of southerly flow from the Gulf of Mexico over the Great Plains, as well as changes in circulation associated with El Nino (Helfand and Schuber 1999, Hu and Feng 2001). Helfand and Schuber (1999) estimate that approximately one-third of the moisture entering the United States enters via the Great Plains low-level jet. This suggests that potential land cover changes which alter the ability of the surface to absorb and re-transpire moisture as it moves from the Gulf of Mexico to the maize crop in the Midwestern United States may significantly influence moisture availability and productivity of the region.

In the South American region (figure 2(b)) we found that the climatological ES footprint for soybean producing areas was largely terrestrial for the November–April growing season. With the Andes physically bounding the west side of the region, much of the moisture that contributes to soybean crops comes from the north and northwest, with a substantial fraction of the ES being located over the Brazilian Amazon. Nearly the entire terrestrial ES resides within the region, indicating large moisture recycling. This is of particular consequence for this region, as multiple studies have indicated that human modifications to the area have the potential to drastically reduce evapotranspiration (Dickinson and Henderson-Sellers 1988, Zhang and Henderson-Sellers 1996, Snyder et al 2004). Additionally, this is a critical region for investigation of water vapor transport, with ∼15% of the Brazilian rainforest already converted to agriculture, and modeling studies suggesting that vegetation in this region may be susceptible to diebacks due to positive feedbacks associated with reduced water vapor flows (Oyama and Nobre 2003).

Finally, for West African wheat, the June–September growing season is directly related to the Northward movement of the ITCZ, with peak rainfall typically lagging its passage by several months (Hastenrath 1991). Figure 2(e) shows that the terrestrial ES is relatively local, with significant contributions from the tropical rainforest and savanna regions of Central Africa, and is negligible to the north due to the presence of the Sahara Desert. There was also evidence that a large fraction of the West African wheat ES footprint is oceanic relative to other regions.

Figure 3 provides quantitative estimates of the ES contribution for relevant areal groupings, and shows the total potential impact of LCC on ES. We found that the impact of removing vegetation from regions not under agricultural usage typically reduced moisture availability for breadbasket crops between 6 and 17% (yellow bars in figure 3, table 2), and potential crop yields from 1 to 17% (table 2). The smallest impact on moisture availability was found to be for North American maize, and the largest were South American soybeans and Central Asian wheat. These changes impacted potential crop yields greatest in the East and Central Asian wheat and North American maize regions, and least in the West African maize region. Additionally, as described in supplemental information (SI3 available at stacks.iop.org/ERL/7/014009/mmedia), the rates at which breadbasket regions responded to different magnitudes of LCC around a region varied greatly. Large fractions of the potential changes in moisture availability in Central Asia (table 2; figure 3) were found to occur with the first 20% of maximum potential LCC (supplemental figure 2 available at stacks.iop.org/ERL/7/014009/mmedia), while other regions had a more gradual response.

Zoom In Zoom Out Reset image size

There were several factors that contributed to the large reduction of ES for Central Asian wheat and South American soybeans with the removal of natural vegetation from non-agricultural land (figures 3(b) and (c); table 2). A large contributor to this was that the moisture sources for these regions tended to be densely concentrated over heavily forested biomes, where LCC has been found to most strongly influence growing season evapotranspiration. This impact was especially evident for South American soybeans where approximately 60% of the evaporative source for region's soybeans was recycled within the region, and a significant portion of that was from tropical rainforests, where vegetation removal has been shown to significantly reduce moisture flux to the atmosphere (Gash and Nobre 1997, Fisch et al 2004, Bagley et al 2011).

In the case of Central Asian wheat, the relatively small magnitude of total precipitation over the growing season coupled with the impact of temperate deciduous forest encompassing a significant fraction of the ES footprint played a large role on the impact of the removal of vegetation on crop water availability. Temperate deciduous forests transpire large amounts of moisture during the summer growing season. When this terrestrial moisture source to the atmosphere is replaced exclusively by evaporation from Earth's surface, the soil must be relatively moist to maintain high evaporation rates. If this is not the case, water that precipitated over devegetated regions gets partitioned into soil and groundwater storage and runoff and does not immediately return to the atmosphere. As shown in table 2, the modeled growing season soil moisture fraction (actual water content/field capacity water content) for the Central Asian wheat region was significantly lower than that of the South American soybean region. As a result, when vegetation cover was removed, leaving behind bare soil, the evapotranspiration and hence ES over the non-agricultural land was significantly reduced for this region. This relationship between soil moisture fraction and potential crop yield impact was consistent across breadbaskets. Potential crop yield in regions that had modeled climatological soil moisture fractions >0.5 had limited susceptibility to reduced precipitation due to LCC.