Economic and biophysical impacts on agriculture under 1.5°C and 2°C warming

The goal of limiting global mean warming to well below 2 °C, and possibly to 1.5 °C, emerged in the Paris Agreement, motivated by the belief that achieving these targets ‘would significantly reduce the risks and impacts of climate change’. Understanding the climate impacts of relatively low levels of warming, in particular whether there are substantial benefits to reducing emissions to limit warming to 1.5 °C rather than 2 °C, is important for informing climate policy, but such studies are scarce. Here we evaluate the difference in global biophysical and economic impacts related to agriculture between 1.5 °C–2 °C warming. Given the small difference in global average temperature, accounting for uncertainties is important, and we include key uncertainties in three main components of the analysis: climate system (regional climate variability), biophysical system (crop response to CO2 fertilization), and economic system (trade responsiveness). We are unable to meaningfully distinguish the regional agricultural impacts occurring with 1.5 °C warming from those occurring with 2 °C warming when accounting for these uncertainties. Under some assumptions 1.5°C implies benefits relative to 2 °C, while under others it implies costs. Results are most sensitive to the uncertainty in the effect of CO2 fertilization on crop yield.

. Relative change (%) (2070-2099 vs. 1980-2010) in RCP8.5 decadal mean production (based on current agricultural lands and irrigation distribution) with (solid) and without (dashed) CO2 effects for wheat, maize and soy from AgMIP vs. CLM results without further CO2 fertilization beyond 2010 Figure SM9 Climate impacts on exogenous yield and economic outcomes (average 2081-2100), expressed relative to outcomes in SSP2 no-climate impact mitigation scenarios with ensemble member #5 Figure SM10 Climate impacts on exogenous yield and economic outcomes (average 2081-2100), expressed relative to outcomes in SSP2 no-climate impact mitigation scenarios without future CO2 fertilization effects (CO2 = 359.8 ppm) Figure SM11 Climate impacts on exogenous yield and economic outcomes (average 2081-2100), expressed relative to outcomes in SSP2 no-climate impact mitigation scenarios with Armington elasticity = 4.5

Figure SM12
Climate impacts on exogenous yield and economic outcomes (average 2081-2100), expressed relative to outcomes in SSP2 no-climate impact mitigation scenarios with Armington elasticity = 0.45

Figure SM1
Linkage between CLM and iPETS, with temporal and spatial scale provided in parentheses. CESM provides the climate scenarios (climate model output is originally monthly; we generated the 6hourly results using 6-hourly model output from a reference period combined with monthly anomaly signals from our climate scenarios); CLM produces the yield database that contains yield information at every grid cell for different climate and management scenarios; iPETS is the economic model used to evaluate the economic impacts.

CLM crop model description and validation
The Community Land Model (CLM, version 4.5;Oleson et al., 2013) is a state of the art land surface model that simulates biogeophysical (radiation transfer, vegetation-soil-hydrology, surface energy fluxes, etc.) and biogeochemical (soil carbon and nitrogen cycle, vegetation photosynthesis, etc.) processes. CLM is a global grid-based model that is a component of the Community Earth System Model (CESM, Hurrell et al., 2013) and can be run either online (coupled with the rest of CESM) or offline (land model only, forced with climate model output) for multiple spatial extents (site, regional, and global) and different spatial resolutions. In this application, CLM was run globally at a resolution of 2 x 2 degrees. Each grid cell includes a subset of land units and plant functional types, as well as multiple layers of snow and soil. Levis et al. (2012) first added crop growth module from the AgroIBIS crop model (Kucharik and Brye, 2003). Since then, the crop module in CLM has been expanded to represent more crops types (maize, soybean, cotton, wheat, rice, sugarcane, tropical maize, tropical soybean) and processes, such as soybean nitrogen fixation (Drewniak et al., 2013). The AgroIBIS crop model within CLM has a relatively simple structure compared to other process-based crop models, but because it is integrated within CLM, it can interact more closely with biogeochemical cycles, water, and climate than is typically the case in standalone crop models.
As noted in the main text, CLM4.5 only allows for crop locations and management assumptions (fertilizer, irrigation) that are fixed over time. In order to estimate yield impacts of climate change in scenarios with changing assumptions over time, we take the approach of generating a pre-computed database of idealized CLM runs from which we interpolate yield outcomes. The details of this approach are fully described in Section 2 (Methodology) in the main text, SM8 and SM9 in Ren et al. (2018), and we summarize them here. In the current application, we assume fixed crop locations over time when estimating the climate effect on yield (SM5, below), so the approach does not have to account for that factor. In addition, we assume fixed ratios of rainfed to irrigated crops over time, since the changes in irrigated areas are small in the GCAM scenario to which we calibrate the iPETS model. We do, however, have to account for changes in fertilizer rates over time, which in our scenario follow an FAO projection of future fertilizer application (fertilizer use is not explicitly modeled in the GCAM scenario, but their base year yields are calibrated to FAO yield values). The essential feature of the method for accommodating this change is that we generate CLM database simulations assuming two different assumptions about nitrogen fertilizer: with or without. The scenario with nitrogen fertilizer assumes that each crop at every grid cell receives the current (crop-specific) North American nitrogen fertilizer application rate based on Potter and Ramankutty (2010). This is a relatively high benchmark rate; compared to actual fertilizer rates, it represents under-fertilization in China, EU and USA but over-fertilization in all other regions. To estimate yields consistent with the projected FAO fertilizer rates, we assume that the yields from CLM are linearly correlated with fertilizer application rates for application rates less than the CLM benchmark rate (current North American rate). For FAO application rates within this range, we linearly interpolate between yield without fertilizer and yield with the high benchmark rate to derive yield corresponding to the FAO fertilizer level. For FAO fertilizer levels above the CLM benchmark rate, we assume yield remains at the CLM maximum; that is, there is no further increase in yield with additional fertilizer application above the CLM benchmark rate.
To validate this approach to estimating yield, we compare the average yield values derived for the nine aggregated iPETS regions to current FAO yield data. After adjusting yields for observed fertilizer application rates, results match the FAO data reasonably well (Fig. SM1). However, there are still some crops in certain regions for which larger differences between CLM and observed yield data remain even after adjustment for fertilizer application rates, such as wheat in Latin America and Transition Countries, maize in India, Latin America, Other Developing Countries and Sub-Saharan Africa, and soy in Latin America, Other Developing Countries, Other Industrialized Countries and Sub-Saharan Africa. While more work is needed to investigate the reasons for these differences, we emphasize that what we employ in this study is not estimates of absolute yield, but estimates of the yield response to climate and CO2. In terms of yield responses to climate change, we first compare the CLM global average yield change to the changes found in the AgMIP multi-model study (Deryng et al. 2016), for the RCP8.5 climate change scenario. Yield changes are compared for individual crops in the decade centered around 2080 (relative to the decade centered on 2000) in Table SM1. With CO2 fertilization, global average yield responses from CLM are within the first and third quartiles of AgMIP results for maize and soybean, while the yield change for wheat is high relative to the AgMIP range. On the other hand, without (additional) CO2 fertilization, the yield change for wheat is within the range of AgMIP results, while negative effects on maize and soybean yield are not as large in CLM as in AgMIP. These two sets of results combined indicate that global average CLM yield responses are broadly similar to the AgMIP results, but that wheat, a C3 crop, has a substantially stronger response to CO2 fertilization in CLM than in the AgMIP models.  Deryng et al. (2016). Numbers in square brackets are the first and third quartiles across all  To evaluate the regional pattern of CLM responses to climate change and CO2, we compare the spatially explicit CLM yield change results to the AgMIP results of Müller et al. (2015) for RCP8.5 with CO2 fertilization. Figure SM3 shows results averaged over 2070 to 2099, relative to current conditions   (From Müller et al. 2015). The white spaces indicate no such crop grew in that region. The disagreement on white spaces indicates the differences between current land cover data used in CLM vs AgMIP.

Carbon capturing technology in iPETS
As described in the main text, a carbon capturing technology is introduced in the iPETS model to reproduce the global CO2 emissions pathway in the GCAM 1.  (Nordhaus, 2008). We assume that the cost starts at $300/ton of C and declines 1% annually after 2030. We assume that this technology uses inputs from the Materials industry to operate. Since energy is an input to the production of Materials, the technology creates a demand for energy indirectly.
The iPETS model does not explicitly represent bioenergy (and its associated land use), so the additional demand on the energy system has little impact on the agricultural system. Thus, while the carbon capturing technology affects the costs of reaching low emission levels, it has little effect on agriculturerelated variables and our impact analysis.
An advantage of the approach is that, in keeping with the philosophy of the iPETS model, it provides a relatively simple representation of mitigation that nonetheless imposes plausible costs on the economy for emissions reduction, so that factors like energy prices that are relevant to impact assessment are at appropriate levels. A drawback is that in this particular approach, mitigation does not directly affect land use, so it cannot represent mitigation strategies like bioenergy (with CCS) that could also influence agricultural impacts through competition for land. We believe this is an acceptable tradeoff in this application, since it makes our results conservative in terms of differences in agricultural impacts between 1.5 o C and 2 o C. That is, those differences would be smaller if a greater demand for land for bioenergy production in a 1.5 o C scenario drove agriculture to less productive land, exacerbating climate impacts on agriculture in that scenario.

Anomaly forcing version of CLM
As described in the main text, we use the "anomaly forcing" version of CLM4.5, which allows for the use of monthly mean values of atmospheric forcing (the only outcomes available from the CESM initial condition ensembles) rather than the typical 6-hourly forcing. This is important because only one of the ensemble members for the 1.5 o C and 2 o C CESM scenarios saved 6-hourly output; others retained only monthly values (a practice common in other ensembles including the CMIP5 multi-model ensemble). The anomaly forcing version constructs 6-hourly future forcing data from monthly means by first defining a historical reference period over which 6-hourly atmospheric forcing is available, here taken to be 1996-2005. Next, future monthly anomaliesthat is, future monthly mean atmospheric conditions relative to the means over the reference periodare calculated for all future months. Finally, future monthly anomalies are combined with the six-hourly forcing from the reference period, with the reference period values being applied to the monthly anomalies in successive 10-year future periods, to construct future six-hourly atmospheric conditions. The values for the reference period were taken from a randomly chosen ensemble member. Future monthly anomalies were calculated either as a difference (temperature, specific humidity, wind, air pressure) or a ratio (solar radiation and precipitation) between monthly outputs from the future scenario and the mean values over the reference period. The anomaly forcing signal has both spatial variation and monthly variation, but assumes that sub-monthly variation is of second-order importance to crop yield outcomes.
As an example, the anomaly forcing simulation for January 2006 (the first month of the projection period) uses the January 1996 6-hourly data, and then either adds them to (for variables with anomalies calculated as a difference) or multiplies them by (for anomalies calculated as a ratio) the January 2006 anomaly signal. If the January 2006 temperature anomaly is 1K for a grid cell (relative to the mean January temperature over the reference period), all January 1996 6-hourly data will be increased by 1K for that grid cell. This approach will not generate exactly the same 6-hourly atmospheric conditions (and therefore crop yields) as the standard version of CLM, due to the difference in the sub-monthly variation in the reference forcing and the actual 6-hourly data (which is typically unavailable, but in principle was generated but not saved at the time of the original climate model run). For example, the anomaly forcing version does not produce changes in the timing of precipitation at the sub-monthly scale, although it does produce changes in intensity at that scale. However, we find that the anomaly forcing version of CLM reproduces well the differences in crop yield between scenarios with and without climate change obtained from the standard version of CLM, which is exactly the information required for this study.
We evaluated the performance of the anomaly forcing version of CLM versus the standard CLM by setting up three 70-year (2006-2075) paired simulations for 1.5 o C, 2 o C, and RCP4.5 scenarios with constant CO2 (359.8 ppm). The paired simulations include one simulation (the standard version of CLM) forced with 3-hourly atmospheric data from the Community Atmosphere Model (CAM) output (6-hourly forcing was unavailable), and the other simulation (the anomaly forcing version of CLM) forced with monthly anomaly forcing that we derived from the monthly output from the same CAM simulations. In all cases, cropland distribution, nitrogen fertilizer rates, and irrigation fractions were fixed at 2005 values.
We first compared absolute crop yields averaged over iPETS model regions and crop types (i.e., the type of aggregation of CLM results for input to iPETS applied in this study), and over the 70 year simulation period. Results showed that averaged over this time period, the anomaly forcing CLM underestimated global average crop yield by 5-8% across the three scenarios, while capturing well the regional yield variation ( Figure SM4). More relevant to the current application of climate change impacts, we also compared results for the change in yield over the 70-year time period, i.e. the difference between yields in 2066-2075 compared to yields in 2006-2015 (which are driven by climate differences over this period).
Results showed that the anomaly forcing CLM produced results for yield changes very similar to those produced by the standard CLM ( Figure SM5), explaining 85%, 84%, and 98% of the variation in change in regional yield in the standard CLM for the 1.5 o C, 2 o C, and RCP4.5 scenarios respectively. These results suggest that although the anomaly forcing CLM may have small systematic biases in absolute yield relative to the standard version, it produces a good approximation of the effect of climate change on yield over time, the key information needed for this study.

Effects of aggregation over crop types
In this study, climate impacts on exogenous yield are evaluated in CLM for eight different crop types. However, the iPETS model contains a single sector that produces a single crop, which represents the average crop produced in each region. While some integrated assessment models contain a more

Effects of assuming constant spatial distribution of cropland
When the impact of climate on future crop yields is derived from the pre-computed database of CLM simulations, we assume that there are no changes in the spatial distribution of cropland, i.e., no expansion or contraction of cropland. This assumption was made for simplicity, and because our previous agricultural impacts work with this same modeling framework (Ren et al., 2018), which included the ability to change the spatial location of crop growth within each iPETS region, but showed that these adjustments had only a small effect on the aggregate impact on crop yield for the region. Furthermore, the assumption of static spatial cropland distribution is common in projecting future crop yields, e.g., the Agricultural Model Intercomparison and Improvement Project (AgMIP) (Rosenzweig et al. 2014). For analysis of economic impacts on yield changes, this assumption is also commonly used, such as in Nelson et al (2014) and Wiebe et al (2015).
We note, however, that the assumption of static cropland in this analysis applies only to the estimation of climate impacts on exogenous yield using CLM. In the iPETS economic model, the agricultural sector can (and does) increase or decrease the amount of land used in agriculture in response to changes in consumption, production and climate change impacts (see, e.g., land use changes due to climate impacts in the 1.5 o C and 2 o C scenarios in Figure 2 in the main text). The only restriction on cropland area for each region in iPETS is that it must be less than the total arable land area. The estimation of the percentage change in yield due to climate change, which is an input to the iPETS analysis, is the only aspect of the analysis that assumes a static distribution of cropland.
By keeping the distribution of cropland constant in CLM, we exclude the possibility of adaptation by moving crop production within a region to locations less impacted (or more favorably impacted) by climate change. We also exclude the effect of iPETS-driven expansion or contraction of cropland area on average yield. Adaptation by moving crop production to locations with higher yields and contraction of cropland will generally increase future yields, while expansion of cropland will normally reduce future regional yields since newly cultivated land is generally less productive than cropland that is already being used. Our iPETS results show cropland expansion under SSP2 for all regions comparing to current day, especially in Other Industrialized Countries and Sub-Saharan Africa. Thus allowing for changing cropland distributions in our scenarios would have competing effects: cropland expansion would tend to decrease average yields, while relocating crops to areas less affected by climate change would tend to increase them. As noted, in prevous work we have found these to be second-order effects on yield.

Implementing climate impacts on yield in iPETS
Different approaches are used to apply exogenous yield changes to CGE models. The two most common are applying them as a productivity shifter (total factor productivity for production, TFP) or as a land productivity shifter (partial factor productivity of land in production, PFP). In this study, the exogenous yield changes derived from CLM are implemented in iPETS as a change in total factor productivity (TFP) in the crop sector and in partial factor productivity (PFP) in the animal products sector.
TFP is the most appropriate means of implementing yield changes in the crop sector because TFP represents the change in output with the ratio of all inputs held constant, and this is the best interpretation of the conditions under which CLM is run to estimate climate impacts on exogenous yield. In iPETS, crop production is determined by inputs of land, labor, capital, energy, and materials (implicitly including fertilizer and irrigation), as well as by PFP for each input and by TFP. In CLM, crop growth is determined by "inputs" of land, management (fixed levels of nitrogen fertilizer or irrigation), and climate conditions.
There is no explicit representation of labor, capital, or energy in CLM, and the best interpretation of the implicit assumption about these inputs is that they are held fixed relative to land and management inputs.
That is, if climate change affects crop growth in a particular grid cell in CLM, the effect on crop growth does not implicitly assume a change in labor or capital inputs to that crop to adjust to the climate impact.
Another way to understand the relationship between the CLM outcomes and iPETS is that the CLM outcome of interestcrop growthis analogous to the output of a sector (the crop sector) in iPETS, which leads to implementation of CLM-derived effects on crop growth in a parameter (TFP) that affects output of the sector as a whole. Therefore, a given CLM-derived yield change in percentage terms in a given year is applied to the crop sector TFP value in the iPETS model.
In the animal products sector, the same interpretation does not apply because the information on pasture growth (i.e. yield) is not analogous to the output of the animal products sector (which is livestock). Rather, the yield information is analogous only to one of the inputs (land) to the production of animal products. If iPETS contained an explicit sector whose output was pasture, the situation would be analogous to the crop sector and yield changes would be most appropriately implemented as TFP changes in the pasture sector. As it is, however, implementing changes in pasture yield in iPETS is most appropriately done as an adjustment to the PFP of land inputs to the animal products sector.
In terms of the magnitude of the pasture yield change, we note that in a previous application (Ren et al 2018), because CLM does not explicitly represent pasture, we assumed that there was no land productivity change due to climate change for animal products. This is an unrealistic assumption and it causes an artificially large shift between crop and animal product consumption. In this study, we incorporated climate impacts on pasture yield in the analysis. Since CLM doesn't produce yield values for pasture and there is no consensus in the literature on climate impacts on pasture yield, we made the following assumption about impacts on pasture yield. We assumed that pasture yield changes by the same percentage as crop yields, simply to avoid introducing artificial demand shifts between crop and pasture.
Clearly a future improvement in the impacts analysis would be to incorporate pasture production explicitly in CLM so that differentiated climate yield impacts on pasture (vs crops) could be modeled, but in the meantime we take this strategy to remove inadvertent effects on crop sector outcomes from an unbalanced treatment of pasture and animal products.
Nonetheless, applying the CLM yield changes as TFP changes in the crop sector and PFP changes in the animal products sector results in different responses between these two sectors in the iPETS model.
Changes in TFP do not induce much substitution among inputs to the crop sector since it essentially increases the productivity of all inputs at the same rate. This generally induces a magnification effect on production in the direction of change in TFP since all inputs are more (or less) efficient compared to their use in production of other sectors, which leads to more (or less) input to the crop sector. In contrast, changes in PFP of one particular input, such as pasture in the animal products sector, induces substitution among inputs because land is more (or less) efficient and the producer will generally use more (or less) of that particular input. This substitution will reduce (or increase) the use of other inputs to the animal product sector and drive down (or up) the prices of other inputs. Thus production of the animal products sector changes in the same direction as the PFP change, but with a smaller magnitude compared to an equal change in TFP; that is, the substitution among inputs partially offsets the effect of the PFP change on output.

Ranges of assumptions for uncertainty analysis
For the uncertainty analysis, we choose parameter values or assumptions for each uncertainty source in order to characterize a wide but plausible range of outcomes due to each, with a small number of model runs.
To represent uncertainty in regional climate due to internal climate variability, we use two out of the 11 members in each of the 1.5 o C and 2 o C initial condition CESM ensembles (Sanderson et al., 2017).
These ensembles reflect inherent uncertainty from interannual to decadal variability. Thus while different ensemble members have approximately similar global average warming responses, they have different climate outcomes across regions at any given time. We choose two members from each ensemble based on their future (2090-2100) temperature and precipitation changes relative to the present (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015). No single ensemble members can be expected to represent the maximum temperature and precipitation change for all nine regions. We rather choose two members that represent a reasonable range of outcomes over most regions, compared to the range for all 11 members. We selected #M9 as our default climate forcing and #M5 as an alternative climate scenario ( Figure SM7). At the regional level, these two ensemble members capture a reasonable spread in temperature and precipitation changes across most regions. With regard to the effects of CO2 fertilization, the yield response to CO2 in CLM appears to be stronger than in many other crop models (Ren et al, 2018). Thus we use the CLM yield response with CO2 fertilization included to represent the high end of the range for this assumption. To represent the low end of the range, we assume that CO2 fertilization operates in CLM, but does not increase beyond its current effect as CO2 concentration increase in the future. To implement this assumption, we run CLM for scenarios with changes in future climate according to the 1.5 o C and 2 o C scenarios, but keep atmospheric CO2 concentrations at their recent (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005) average) level of 359.8 ppm. We adopt this assumption, rather than weakening the CO2 fertilization process within the CLM model, because CO2 effects are builtin to the photosynthesis process and cannot be easily changed within CLM (see SM8 for details).
Estimates of the range of CO2 fertilization effects on crop yields in the literature are large. To validate our approach to representing the lower bound of the uncertainty range, we compared the yield changes at the end of the century from CLM when driven with constant current CO2 levels with the ranges of yield changes from crop model simulations in AgMIP with CO2 fertilization effects (Rosenzweig et al., 2014). As shown in Figure SM8, the CLM yield results with current CO2 levels is a reasonable representation of the lower end of the range of impacts on yield when considering CO2 fertilization effects.

Modeling CO2 fertilization in CLM
In CLM, CO2 fertilization is a dynamic response to climate scenarios, water and nitrogen conditions, which is embedded in the photosynthesis process, including the process of transport of CO2 into the leaf (Section 8.2 in Oleson et al., 2013). Plants regulate the size of their stomata to control the exchange of CO2 and water with the atmosphere. Elevated atmospheric CO2 increases the CO2 concentration at the leaf surface, and therefore in the interior of the leaf where photosynthesis takes place, increasing the production of plant biomass. The elevation of CO2 concentration at the leaf surface causes a reduction in the size of stomatal openings and restricts the conduction of CO2 into the leaf to participate in photosynthesis (Drake et al, 1997), so that a self-limiting feedback is built into the CO2 effect. The smaller stomatal conductance also restricts evapotranspiration, which allows plants to conserve more water (or grow more with the same amount of water, i.e. increase water use efficiency). As evapotranspiration decreases, it also increases leaf temperature and the partial pressure of water vapor inside the leaves, which increases leaf transpiration (Kimball et al, 1995).
The effect of elevated CO2 also depends on limitations on other resources for growth. The stimulation of photosynthesis by elevated CO2 is reduced in CLM when nitrogen is limited. The combined effects of atmospheric CO2, irrigation, and nitrogen fertilizers can lead to a wide range of sensitivities of crop growth to elevated CO2. Although CO2 fertilization generally increases crop growth, in some cases it can result in lower crop yields if, for example, leaf and stem growth deplete soil nitrogen and lead to nitrogen limitation during the grain fill stage.
CLM uses biochemical photosynthesis models for C3 (Farquhar et al., 1980) and C4 (Collatz et al., 1992) plants, coupled with a Ball-Berry type stomatal conductance model (Collatz et al., 1991). Elevated atmospheric CO2 directly affects the leaf surface CO2 partial pressure (Cs), which in turn affects stomatal conductance and photosynthesis. Stomatal conductance and photosynthesis are themselves interdependent. For example, increased stomatal conductance increases photosynthesis and the evaporation of water, which in turn can dry the soil, resulting in a decrease of stomatal conductance and photosynthesis.

Caveats
There are a number of caveats to this study. Most importantly, while we have examined three important sources of uncertainty, there are a number of additional sources that should be explored and could change results. Although we represent regional climate uncertainty using two different simulations for both of the 1.5 o C and 2 o C scenarios, we have used results from a single climate model; other models would produce different regional patterns of climate change. Although we represent crop model uncertainty by using two different assumptions for CO2 fertilization, a major source of such uncertainty, we have used a single crop model (in CLM); including other models could produce a wider range of impacts that differ in response to climate and management. Methodologically, although our economic model includes extensification or intensification of land use, when deriving the exogenous yield change from CLM we make the simplifying assumption of constant cropland over time, which we do not believe would change the basic conclusions but should be kept in mind (SM5).
The economic modeling is also subject to limitations. As discussed above, aggregation over crops for representation in the global economy may affect results. It also produces results for average outcomes over large regions that can miss the impacts on the most vulnerable countries or sub-national regions or populations. While iPETS has a typical structure for CGE-style economic models, the use of additional economic models would capture structural uncertainty in the representation of the economy. As shown in Ren et al (2018), different socioeconomic development pathways have more impacts on economic outcomes than we explored in this study. Here we focused on the uncertainties from models rather than socioeconomic scenarios. Finally, we have focused on the effect of climate change on the agricultural system, largely separate from the energy system and mitigation activities. Our simulation of the deep emissions reductions required to limit warming to 1.5 o C and 2 o C relies on the effects of a carbon tax on energy and on a backstop technology most analogous to air capture, which has a relatively low cost and small impacts on land use. A more comprehensive treatment of mitigation would allow for exploration of the effect of competition for land between the energy and agricultural systems, and for the relative effects of mitigation costs and climate change impacts on outcomes like consumption and food prices.