Modelled biophysical impacts of conservation agriculture on local climates

Abstract Including the parameterization of land management practices into Earth System Models has been shown to influence the simulation of regional climates, particularly for temperature extremes. However, recent model development has focused on implementing irrigation where other land management practices such as conservation agriculture (CA) has been limited due to the lack of global spatially explicit datasets describing where this form of management is practiced. Here, we implement a representation of CA into the Community Earth System Model and show that the quality of simulated surface energy fluxes improves when including more information on how agricultural land is managed. We also compare the climate response at the subgrid scale where CA is applied. We find that CA generally contributes to local cooling (~1°C) of hot temperature extremes in mid‐latitude regions where it is practiced, while over tropical locations CA contributes to local warming (~1°C) due to changes in evapotranspiration dominating the effects of enhanced surface albedo. In particular, changes in the partitioning of evapotranspiration between soil evaporation and transpiration are critical for the sign of the temperature change: a cooling occurs only when the soil moisture retention and associated enhanced transpiration is sufficient to offset the warming from reduced soil evaporation. Finally, we examine the climate change mitigation potential of CA by comparing a simulation with present‐day CA extent to a simulation where CA is expanded to all suitable crop areas. Here, our results indicate that while the local temperature response to CA is considerable cooling (>2°C), the grid‐scale changes in climate are counteractive due to negative atmospheric feedbacks. Overall, our results underline that CA has a nonnegligible impact on the local climate and that it should therefore be considered in future climate projections.

tion agriculture (CA) has been limited due to the lack of global spatially explicit datasets describing where this form of management is practiced. Here, we implement a representation of CA into the Community Earth System Model and show that the quality of simulated surface energy fluxes improves when including more information on how agricultural land is managed. We also compare the climate response at the subgrid scale where CA is applied. We find that CA generally contributes to local cooling (~1°C) of hot temperature extremes in mid-latitude regions where it is practiced, while over tropical locations CA contributes to local warming (~1°C) due to changes in evapotranspiration dominating the effects of enhanced surface albedo.
In particular, changes in the partitioning of evapotranspiration between soil evaporation and transpiration are critical for the sign of the temperature change: a cooling occurs only when the soil moisture retention and associated enhanced transpiration is sufficient to offset the warming from reduced soil evaporation. Finally, we examine the climate change mitigation potential of CA by comparing a simulation with present-day CA extent to a simulation where CA is expanded to all suitable crop areas. Here, our results indicate that while the local temperature response to CA is considerable cooling (>2°C), the grid-scale changes in climate are counteractive due to negative atmospheric feedbacks. Overall, our results underline that CA has a nonnegligible impact on the local climate and that it should therefore be considered in future climate projections.

| INTRODUCTION
Agricultural land management has a substantial impact on regional climate (e.g. Davin, Seneviratne, Ciais, Olioso, & Wang, 2014;Hirsch, Wilhelm, Davin, Thiery, & Seneviratne, 2017;Luyssaert et al., 2014;Thiery et al., 2017), and influences local responses to projected climate change . However, historically climate model simulations contributing to the various Coupled Model Intercomparison Projects (CMIPs) have had limited or no representation of agricultural land management. Recently, considerable progress has been made to move away from basic representations of agricultural activity in Earth System Models (ESMs) where crops are typically modelled with the same physiological characteristics as C3 grasses.
However, conservation agriculture, involving minimal or no tillage, crop residue management and crop rotation (Kassam, Friedrich, Derpsch, & Kienzle, 2015), is generally not considered in the land surface component of ESMs. This is due to (a) a lack of global datasets characterizing where conservation agriculture is practiced and (b) the high uncertainty in soil carbon sequestration under minimal tillage regimes (Neufeldt, Kissinger, & Alcamo, 2015;Powlson et al., 2014). Nevertheless, the adoption of these land management practices is seen as a potential climate mitigation and adaptation strategy that also has climate impacts beyond its impact on carbon sequestration (Davin et al., 2014;Lobell, Bala, & Duffy, 2006). Here, we take the first step towards resolving the representation of conservation agriculture within ESMs by implementing a simple parameterization within a state-of-the-art ESM using a new spatially explicit dataset on conservation agriculture as input .
The adoption of conservation agriculture has increased from 45 Mha in 1999 to 157 Mha in 2013 (Derpsch, Friedrich, Kassam, & Hongwen, 2010;Kassam et al., 2015). This increase is partly associated with the need to increase productivity in response to various external economic and environmental pressures (Kassam et al., 2015). Furthermore, this increase is influenced by the growing awareness of how traditional tillage-based production systems have a negative impact on soil quality by increasing soil erosion rates and changing biogeochemical cycling due to soil disturbance (Govers, Quine, Desmet, & Walling, 1996;Govers, Vandaele, Desmet, Poesen, & Bunte, 1994;Quinton, Govers, Van Oost, & Bardgett, 2010;Van Oost et al., 2007;Wang et al., 2017) which has potential negative consequences for yield. In particular, yield levels with minimal tillage are comparable or even higher than conventional tillage systems (Kassam et al., 2015). Nonetheless, a recent meta-analysis does identify that yield gains from minimal tillage are mostly possible when combined with both crop residue management and crop rotation, particularly for drier climates rather than humid climates (Pittelkow et al., 2015). There are, however, some critical limitations on where conservation agriculture is possible. This includes practical knowledge about conservation agriculture and investment costs-including time and machinery-to establish and maintain the soil mulch, and finally the political and social support to move away from traditional farming practices (Kassam et al., 2015). Furthermore, conservation agriculture has biophysical feedbacks on climate associated with how the presence of a crop residue alters the surface energy balance through changes in surface albedo, roughness, and evapotranspiration (Davin et al., 2014).
The examination of the potential impacts of no-till farming, an essential part of conservation agriculture, on the climate system has been done in climate models. However, existing studies often use an idealized approach to evaluate the climate implications of a full conversion of global croplands (e.g. Davin et al., 2014;Hirsch et al., 2017;Lobell et al., 2006;Wilhelm, Davin, & Seneviratne, 2015). For example, Lobell et al. (2006) compare the impact of four different land management practices on present-day climate. In that study, no-till farming was represented by multiplying the soil albedo by 1.5 over the fractional area designated as croplands in the model. The results demonstrated the potential of increasing surface albedo to cool surface temperatures by reducing the available energy at the surface. Davin et al. (2014) apply more conservative changes over Europe in a regional climate model by increasing surface albedo over croplands by 0.1 and increasing the soil resistance by a factor of 4 to represent the effects of crop residue on evaporation. They found that the cooling potential of no-till farming was greater for temperature extremes than mean temperature. Wilhelm et al. (2015) also take an idealized approach to represent no-till via albedo changes and demonstrate that the temperature response scales linearly with the magnitude of the albedo change, the spatial extent, and the temporal extent. This was confirmed in Hirsch et al. (2017) who also demonstrate that the cooling potential from increased surface albedo associated with no tillage is comparable to that of irrigation, and therefore the effects of surface albedo changes associated with agricultural practices are important. Although these studies (i.e. Davin et al., 2014;Hirsch et al., 2017;Lobell et al., 2006;Wilhelm et al., 2015) use an idealized approach to examine the cooling potential of various land management practices, they all demonstrate that changes in albedo and evapotranspiration associated with no-till farming have implications for climate, particularly temperature extremes, and that therefore investment in further developing parameterizations to represent this land management practice, and more specifically conservation agriculture, within ESMs is worth pursuing.
In this study, we build upon previous research with ESMs to examine the climate implications of agricultural land management.
Deviating from the idealized approach used in previous studies, we use a new global conservation agriculture dataset  to constrain the application of albedo and evapotranspiration changes towards a more realistic distribution of this land HIRSCH ET AL. | 4759 management practice. We also limit albedo changes to the soil surface rather than the canopy albedo as in Wilhelm et al. (2015) and Hirsch et al. (2017) to emulate how the presence of crop residues alters the background surface albedo. Furthermore, we model this albedo change as a function of soil colour, recognizing that the albedo change from crop residue will be smaller over brighter soils than darker soils. We also aim to assess whether applying a more conservative approach of the biophysical effects of conservation agriculture within an ESM can improve the simulation of present-day climate and we explore the possible climate sensitivity to different conservation agriculture estimates.

| Model description and setup
We use the Community Earth System Model (CESM) version 1.2 (Hurrell, Holland, & Gent, 2013) with prescribed sea surface temperatures (SSTs) and sea ice fraction using a setup that closely follows  (Lawrence & Chase, 2007).
We use the same set of control simulations as those evaluated in Thiery et al. (2017). This consists of a 5-member control ensemble with a horizontal resolution of 0.9°latitude × 1.25°longitude, starting from 1976 to 2010 (35 years), where the first 5 years are discarded as spin-up. Ensemble members are generated by applying a random perturbation of 10 −14 to the atmospheric temperature initial conditions (Fischer, Beyerle, & Knutti, 2013). We prescribe SSTs and sea ice to focus on the influence of land-atmosphere interactions without the added complexity of ocean-atmosphere feedbacks on the climate system. In addition to the control ensemble, we run four 5-member ensemble experiments corresponding to the four different conservation agriculture estimates described in the following section: BASE, LOW, HIGH, and POT.

| Description of conservation agriculture dataset
We use the conservation agriculture dataset developed by Prestele et al. (2018) to prescribe the spatial distribution of conservation agriculture in the CESM simulations. This dataset builds on the national-level estimates of conservation agriculture published in Kassam et al. (2015) and additional regional datasets. The aggregated estimates of conservation agriculture were downscaled to a 5-arcminute regular grid using GIS-based multi-criteria analysis. The downscaling algorithm considers several spatial determinants of conservation agriculture adoption, including biophysical (aridity and soil degradation) and socio-economic (farm size, access to suitable equipment, and poverty level) variables. Uncertainties due to inconsistencies in the definition of conservation agriculture (e.g. Carmona et al., 2015;Hobbs, 2007) and the lack of systematic reporting schemes (Kassam et al., 2015) are represented by a range of spatially explicit maps (BASE, LOW, and HIGH). In particular, the baseline estimate

| Implementation of conservation agriculture
We implement conservation agriculture (CA) into CESM by splitting the existing CLM crop PFT into a fraction under conservation agriculture (C CA ) and a fraction under conventional management (C CM ).
Therefore, both forms of management are possible within a grid cell.
The fractions of cropland under conservation agriculture and under conventional management are determined as follows: where C ALL is the default CLM crop fraction, A CA is the area under conservation agriculture, and A CROP is the total cropland area. Both A CA and A CROP are obtained from the CA dataset (see previous section) and are conservatively aggregated from the original 5-arc-minute resolution to the CLM resolution used in this study (0.9°4 latitude × 1.25°longitude). By using this approach we avoid potential grid conflicts, as the CA dataset is based on the HYDE cropland extent for the year 2012, whereas the CLM land cover uses 2000 cropland extents from Ramankutty et al. (2008). The distribution of the CA crop PFT for the four CA estimates is illustrated in Figure 1.
Surface albedo (α) in CLM is calculated at the subgrid level for canopy and soil surfaces separately, which are then aggregated to a total surface albedo as a weighted combination of snow-free and snow-covered albedos. All albedo terms are modelled using a twostream approximation for radiative transfer to determine the direct and diffuse radiation contributions. To reflect the higher surface albedo of crop residue, we increase the soil albedo for the CA crop PFT using the following function: where N S is the number of soil classes (here we use 20) and s is the soil colour index (1-20 with 1 the brightest and 20 the darkest soil).
We limit the maximum change in soil albedo to 0.1, which is considered the maximum possible change in surface albedo by crop residue (Davin et al., 2014;Hirsch et al., 2017). Therefore, we constrain the albedo change by the soil colour, recognizing that the effective albedo change will be minimal on brighter soils compared to darker soils with the albedo change ranging from 0.005 for the brightest soil to 0.100 for the darkest soil. We assume that the crop residue is present all year, but our implementation ensures that the effect of the increased soil albedo on the total surface albedo is dampened during the growing season by the presence of canopy cover.
The presence of crop residue also has an impact on the amount of soil evaporation. Therefore, to reduce the soil evaporation, and mimic the effect of a crop residue layer for no tillage areas, we double the litter resistance of the soil column corresponding to the CA crop PFT. Note that the litter resistance is added to the aerodynamic resistance which together limit water vapour transfer from the ground to the atmosphere by scaling the soil latent heat flux. Due to a lack of global datasets that characterize tillage intensity, we do not implement partial residue cover. Furthermore, we do not modify the surface roughness or infiltration rates due to limitations in how these vary according to residue thickness. We restrict our implementation to changing the biophysical properties of the CA crop, but recognize that the presence of crop residues and absence of soil disturbance do influence the carbon stores in litter and the upper soil layers.

| Evaluation datasets
We

| Analysis
Our analysis uses the ensemble average daily output from CESM to examine changes in the extremes indices as defined by the Expert To examine changes in surface temperature (T S ) in response to conservation agriculture, we use the surface energy balance decomposition method applied in previous studies (e.g. Akkermans, Thiery, & Van Lipzig, 2014;Hirsch et al., 2017;Luyssaert et al., 2014;Thiery et al., 2015Thiery et al., , 2017. Here, we express the surface energy balance as: where ε is the surface emissivity, σ is the Stefan-Boltzmann constant , α is the surface albedo, SW i is the incoming shortwave radiation, and LW i is the incoming longwave radiation, LHF is the latent heat flux and SHF is the sensible heat flux. The residual term (R) includes the ground heat flux and changes in subsurface heat storage. The change between the experiment and control (denoted as Δ) is calculated by taking the derivative of Equation 4 with respect to T S and solving for ΔT S : To examine regional changes we focus our analysis over regions where conservation agriculture is extensive. We use the (i) to the CTL for locations where CA is applied, this increase is very small (Figure 2a). The corresponding response in the radiation balance indicates a general decrease in the bias over large areas, particularly over Asia and North America, for SW net (Figure 2c), LW net (Figure 2e), and T2M (Figure 2i). For South America, there is an increase in the SW net bias (~1 W/m 2 ), but a decrease in the LHF bias (Figures 2c,g). Note that these changes in bias extend beyond the regions where CA is applied and, therefore, despite changes in the bias indicating some improvement in the simulation skill this result may still be influenced by the internal variability in the model. Similarly, the percent change in RMSE indicates that there is some skill improvement in simulating the temporal variability, particularly for the northern mid-latitude regions with a 5%-10% error reduction for the surface albedo, SW net , LW net , LHF, and T2M (Figure 2b,d,f,h,j). However, at higher latitudes (50N-90N) there is a decrease in skill with a 6%-10% increase in RMSE. Simulation skill over South-eastern South America (SSA) also improves for T2M with a 10% reduction in error, despite some decrease in skill for LHF. Generally, Figure 2 demonstrates that there is improvement in simulation skill, but that this is not uniform across the global domain.
As we are interested in how the uncertainty between the different CA estimates influences simulation skill, we examine the added value of including CA for different climate variables over the regions where the CA extent is greatest (Figure 3). We evaluate the added value by calculating for each region the change (experiment minus control) in the spatiotemporal root mean square error. Accounting for CA generally improves the simulation skill over the Mediterranean for all variables and CA estimates. For other regions, including WNA, CNA, and CEU, we find enhanced skill for some variables. For SSA and SAU, the added value is limited for all CA estimates. The BASE (Figure 3a) and LOW (Figure 3b) CA estimates contribute the most to improved simulation skill, likely due to the more realistic distribution of CA in these cases. Finally, if we consider the grid cells where land fraction within CESM exceeds 50% ("all land") or just the grid cells that have a nonzero CA fraction ("CA Land") is present, there is added value for most variables over the grid cells where CA has been applied, particularly for the LOW estimate. Note that for precipitation the added value is sensitive to which observational precipitation product is used and is likely a result of the considerable uncertainty between these datasets (e.g. Adler, Kidd, Petty, Morissey, & Goodman, 2001). Therefore, we are confident that our implementation of the biophysical effects of CA on the regionalscale climate has added value.

| Effect of conservation agriculture on climate
Using the PFT-level outputs from CLM it is possible to examine the subgrid-scale differences between the CA and conventionally man-  (Figure 5b). Coincidently, the warming from CA (e.g. Figure 4h for T s ) often occurs when the decrease in soil evaporation exceeds the increase in canopy transpiration (e.g. South America and Eastern North America; Figure 5). Cooling occurs when the decreased soil evaporation is comparable to the increase in transpiration. Increases in transpiration arise from more moisture availability due to less soil evaporation, and therefore, the thickness of the crop residue is likely to have some influence modulating local temperatures.
Note that similar responses were found for the LOW and HIGH CA estimates for the climate variables presented in Figures 4 and 5 with the exception of POT where the application of CA is much greater and therefore, the corresponding subgrid-scale changes between the CA and CM crops were larger over Europe and Asia.
In the literature possible cooling of hot temperature extremes has been found for regions where tillage is limited (e.g. Davin et al., 2014). Therefore, we examine the subgrid-scale changes in both the annual maximum daytime 2 m air temperature (TXx) and the annual minimum night-time temperature (TNn) for each of the CA estimates ( Figure 6). Generally, CA induces cooling of TXx by more than 1°C for most regions where CA is applied for all estimates (Figure 6, left column TXx and TNn is also remarkably similar between the BASE, LOW, and HIGH CA estimates (Figure 6a-f), indicating that the sensitivity of CESM to these different CA estimates is low at this resolution.

| Effect of conservation agriculture on the surface energy balance
The have larger-scale impacts on the climate that may be a model dependent result. Previous research with CESM examining albedo enhancement  found that this model has a tendency to produce large cloud feedbacks when more energy is reflected at the surface. Given that Figure 7a includes all land grid cells, the effective area that the albedo and LHF change from CA influences temperatures is relatively smaller. Instead, the cloud feedbacks that propagate from the introduction of CA have a larger nonlocal influence reflecting the larger contribution of SW i and LW i on changes in surface temperature. Note that in Figure 7 we distinguish between the direct forcings (i.e. albedo change) and the indirect forcings (i.e. SWi, LWi, and SHF), which are modified indirectly by land-atmosphere feedbacks.
As evident in Figure 6, this global aggregation of the surface energy balance decomposition of surface temperature is likely to mask regional and seasonal differences and, therefore, we focus on the monthly time-scale changes for six regions (denoted in Figure 1) where the CA fraction is greatest (Figure 8). We also examine the total mitigation potential of CA by comparing the grid-scale climate in the POT experiment to the BASE experiment ( Figure 10). Here, the difference in available energy (e.g. SW net and LW net ) is largely within 1 W/m 2 except for the Northern United States and India where the difference in SW net exceeds 5 W/m 2 (Figure 10a,b). The LHF (Figure 10c) is also damped in POT with differences ranging from 1 to 5 W/m 2 dependent on the intensity of the CA expansion. However, when comparing the grid-scale temperature response to the BASE experiment (i.e. present-day level of CA adoption) there is a general warming in POT (Figure 10d-f) over North America, Europe, and Asia, suggesting that CA could be a counterproductive climate mitigation measure if implemented at such a large scale. The apparent contradiction between the subgrid-scale cooling and large-scale warming effect of CA is due to the role of atmospheric feedbacks. The decrease in evapotranspiration, both due to higher albedo and to the higher soil resistance, triggers a decrease in cloud cover in the model that increases incoming radiation and thus temperature as seen in other studies Wilhelm et al., 2015). However, we note that most of the differences in the grid-scale response between POT and BASE were found to be not statistically significant and therefore requires further investigation to understand the potential for atmospheric feedbacks to negate any local cooling potential from CA.

| DISCUSSION
Conservation agriculture is a form of land management that is extensively practiced in several of the major agricultural regions. In this study we present the first results of implementing a new spatially   | 4771 timing in our simulations to which soils are more exposed after crop would also be necessary and may be possible using a global dataset characterizing crop planting and harvest dates (Sacks et al., 2010).
Furthermore, there are already several ESM groups developing the parameterization of irrigation practices due to the known impacts of irrigation on local climate. Integrating both irrigation and CA into ESMs will require data to identify regions where only irrigation, only CA or both are applied. Other data requirements include CA extent per crop type (e.g. maize and temperate cereals) to enable integration of CA within a prognostic crop model that accounts for different crop types and their growing cycles (e.g. Levis et al., 2012;Sacks et al., 2010). Finally, information characterizing the tillage percentage in terms of no-till, minimal tillage, or conventional tillage and mulch depth would be ideal to split the managed crop further to examine the effects of, for example, soil disturbance depth and partial crop residue cover compared to the all or nothing approach examined here. Therefore, there is plenty of scope for further model development as data become available to characterize differences in crop types and their rotations (Erb et al., 2017). However, including all of these different crop types and management practices will require a more comprehensive approach to representing these individual management techniques within an ESM.

| Implications for climate mitigation claims and outlook
Future expansion of CA is likely, with two potential scenarios for this expansion discussed in Prestele et al. (2018). By comparing the impact of a future potential CA extent to the present-day distribution on climate, we present a first estimate of biophysical benefits and trade-offs of this cropland management strategy (Figures 6 and 10). Our results indicate that large-scale expansion of CA management could decrease temperature extremes over the mid and high latitudes on the local scale. There is, however, also a warming response both for mean and extreme temperatures over wide areas of the tropics, which would increase the vulnerability of CA managed systems to climate change at these locations. Indeed, further ESM-based studies, implementing CA management, are required to confirm these responses and receive a robust signal. Moreover, next to these biophysical implications, enhanced soil carbon sequestration in CA managed land is discussed as a potential contribution to future net negative emissions (Neufeldt et al., 2015;Powlson et al., 2014;Smith et al., 2008) and have not been quantified yet at the large scale using ESMs.
Developing the parameterization of CA within CESM further to include biogeochemical influences is therefore necessary. Nevertheless, our study provides an important step towards the explicit parameterization of CA management in ESMs and the quantification of related climate impacts and feedbacks.

ACKNOWLEDG EMENTS
We thank National Center for Atmospheric Research (NCAR) for the development and access to the Community Earth System Model.
We greatly thank Urs Beyerle and the ETH Euler cluster for support