SSP-Based Land-Use Change Scenarios: A Critical Uncertainty in Future Regional Climate Change Projections

To better understand the role projected land-use changes (LUCs) may play in future regional climate projections, we assess the combined effects of greenhouse-gas (GHG)-forced climate change and LUCs in regional climate model (RCM) simulations. To do so, we produced RCM simulations that are complementary to the North-American Coordinated Regional Downscaling Experiment (NA-CORDEX) simulations, but with future LUCs that are consistent with particular Shared Socioeconomic Pathways (SSPs) and related to a specific Representative Concentration Pathway (RCP). We examine the state of the climate at the end of the 21st century with and without two urban and agricultural LUC scenarios that follow SSP3 and SSP5 using the Weather Research and Forecasting (WRF) model forced by one global climate model, the MPI-ESM, under the RCP8.5 scenario. We find that LUCs following different societal trends under the SSPs can significantly affect climate projections in different ways. In regions of significant cropland expansion over previously forested area, projected annual mean temperature increases are diminished by around 0.5°C–1.0°C. Across all seasons, where urbanization is high, projected temperature increases are magnified. In particular, summer mean temperature projections are up to 4°C–5°C greater and minimum and maximum temperature projections are increased by 2.5°C–6°C, amounts that are on par with the warming due to GHG-forced climate change. Warming is also enhanced in the urban surroundings. Future urbanization also has a large influence on precipitation projections during summer, increasing storm intensity, event length, and the overall amount over urbanized areas, and decreasing precipitation in surrounding areas.

or societally informed assessments, and examined the effect of future policy-driven land-use change scenarios and their combined effect on climate change in RCM projections.In one recent example, one of few that the authors know of, Berkmans et al. ( 2019) used a European policy-based LUC scenario in an RCM to examine the LUC effect on climate relative to greenhouse-gas (GHG) forced climate change for the near-future, and showed a clear influence of the LUC on temperature.Another, Yilmaz et al. (2019), used ongoing and near future infrastructure projects and their effect on local land-use to examine the influence of expanded irrigation on the upper Euphrates-Tigris basin water budget, finding a large climatological and potentially large societal impact.In some instances, RCM projections have been used to inform climate change impacts assessments including implied LUC using integrated assessment models, but have not incorporated the LUC into the RCMs (e.g., Harrison et al., 2019).These existing studies leave a critical gap in the assessment of plausible future LUCs and their effects on future climate in regional simulations.
The authors attempt to narrow this gap using LUC scenarios that are consistent with different Shared Socioeconomic Pathways (SSPs) in RCM simulations to assess the combined effects of GHG-induced climate change and scenario-based anthropogenic LUCs on regional climate projections.More specifically, the authors examine the influence of the LUCs that underlie the combined SSP + Representative Concentration Pathway (RCP) framework using simulations produced for the North-American branch of the international Coordinated Regional Climate Downscaling Experiment (NA-CORDEX) and complementary simulations produced for this assessment that incorporate SSP-based LUC.The authors aim to answer the question, "Does inclusion of SSP-based LUCs significantly modulate the RCM projections?," as the answer to this simple question may have broadly relevant implications for future regional modeling efforts, as we will discuss.For this initial effort, the authors focus on the conterminous United States (CONUS), and projections of near-surface temperature and precipitation, two of the most commonly used variables.
As global model simulations produced for Phase 6 of the Coupled Model Intercomparison Project (CMIP6) as a part of the Scenario Model Intercomparison Project (ScenarioMIP; O'Neill et al., 2016) incorporate SSP-based LUC scenarios related to RCP-based future emissions, exploring the effect of SSP-based LUCs in RCMs is highly relevant for informing future downscaling efforts that make use of ScenarioMIP simulations.This is particularly true for large-scale coordinated efforts like CORDEX, making our effort timely as well.Existing NA-CORDEX simulations hold land surface cover constant at present day conditions, which is typical in most, if not all, existing CORDEX simulations globally, while SSP-consistent projections anticipate potentially substantial changes in anthropogenic land-use amounts and patterns.For example, Gao and O'Neill (2020) found the global total amount of urban land can increase 6 fold by 2100, and economically developed regions (e.g., North America) experience comparable amounts of new urban land development to developing regions.All accentuate the need for investigations like ours.
Understanding the magnitude of the regional climate effects of LUC is additionally important to the SSP + RCP scenarios framework (O'Neill et al., 2019), in particular the assumption that climate model simulations that include a particular land-use scenario are a reasonable representation of climate outcomes in scenarios with the same greenhouse gas forcing but a different land-use scenario (O'Neill et al., 2016).Some results with global climate and land-use models (LUMs) challenge this assumption (Jones et al., 2013) and multi-model experiments are underway to further test it (Lawrence et al., 2016), but in general it is an understudied problem.This work helps address this question, and will help inform thinking about possible needed modifications to the scenarios framework to better account for climate-land-use interactions.

Description of SSPs and SSP-Consistent Land-Use Changes
The authors use SSP3: Regional Rivalry (A Rocky Road) and SSP5: Fossil-Fueled Development (Taking the Highway) in this study, because together they span the range of uncertainties in both urban and agricultural land-use in the US over the coming decades.Here, agricultural land includes crop and pasture, but not managed forest.Under SSP3, countries generally focus on domestic issues due to increasing nationalism.Economic development is slow, and countries focus on energy and food security.Population growth is low in industrialized countries but high in developing countries.As such, the US sees an increase in domestic cropland but low-population growth, which translates to low-urban land expansion.Under SSP5, the global economy grows quickly driven by material-intensive development and fossil fuel exploitation.Global population growth is low overall compared to many other SSPs, but in the US and other high-income countries, the population grows rapidly under a strong globalized economy.As a result, the US sees a large amount of urban land expansion and a minimal increase in domestic cropland.Pastureland area decreases slightly in both scenarios.For more detail on the SSP narratives see O'Neill et al. (2017).
Interestingly, these two scenarios also provide great contrast in our simulations.Because SSP3 experiences primarily cropland expansion and SSP5 primarily urban land expansion, our simulations can isolate the effects of these two different types of land-use change.Note that our future climate simulations follow RCP8.5 (the RCP that reaches 8.5 W/m 2 by 2100; Moss et al., 2010), and that SSP3 usually does not reach the radiative forcing of RCP8.5 in integrated assessment models, as SSP5 does (Riahi et al., 2017).SSP3 produces a radiative forcing of ∼7.2 W/m 2 (with a range of 6.7-8.0W/m 2 ), whereas SSP5 produces a radiative forcing of ∼8.7 W/m 2 (Riahi et al., 2017).Therefore, in this study, the authors use an agricultural land projection from a variant of SSP3 developed to ensure consistency with the radiative forcing levels in RCP8.5.This "High Growth" variant of SSP3 (SSP3HG) includes modestly higher GDP growth that increases emissions and also agricultural land-use relative to SSP3, without changing its basic nature (Ren et al., 2018).The urban land projection is based on the original SSP3; the effect of the higher GDP growth on this low urban land development scenario would be small.The LUCs consistent with the two SSPs were produced using two LUMs.For urban land change, the authors use a newly developed empirically grounded modeling framework (consisting of the Country-Level Urban Buildup Scenario [CLUBS] model and the Spatially Explicit, Long-term, Empirical City developmenT [SELECT] model; Gao & O'Neill, 2019, 2020) that produces realistic spatial and temporal patterns for longterm urban land change under different SSPs at a ⅛ degree resolution.For agricultural land change, the projections were produced using a spatially explicit agricultural LUM at a ½ degree resolution (Meiyappan et al., 2014;Ren et al., 2018).The agricultural LUM takes the urban land-use projections as input, and assumes if the two land uses (agricultural vs. urban) compete for the same land, urban land use would win.Both LUMs produce projections of LUC for the beginning of every decade, decade-by-decade.Each also provides the fraction of a grid box covered by the given land-use type as an end product.

WRF
This study leverages 25-km resolution Weather Research and Forecasting (WRF; Skamarock et al., 2005) model version 3.5.1 simulations that were produced for NA-CORDEX to save on computational costs (Mearns et al., 2017).Specifically, the authors use the simulations forced by the Max Planck Institute for Meteorology Earth System Model at Low Resolution (MPI-ESM-LR) GCM that follow the historical and RCP8.5 scenarios.The MPI-ESM-LR atmospheric component has a 1.8653° latitude × 1.875° longitude resolution, a mid-range equilibrium climate sensitivity relative to the full set of Coupled Model Intercomparison Program Phase 5 (CMIP5) simulations, and it provides relatively high-quality boundary conditions for WRF (Bukovsky & Mearns, 2020;Rendfrey et al., 2018).
The NA-CORDEX WRF configuration uses the United States Geological Survey (USGS) land-use categories listed in Figure 1.These are used in the Unified Noah land-surface model (LSM) parameterization within WRF.The Noah LSM represents the model's land-atmosphere interface and updates model land-surface variables related to, for instance, sensible and latent heat fluxes (SHF and LHF, respectively), soil temperature and moisture, and runoff while solving the surface energy and water balance per grid cell (e.g., Chen & Dudhia, 2001;Chen et al., 2001;Ek et al., 2003).Each USGS land-use type used in WRF has specific, assigned properties related to albedo, roughness, moisture, etc. Land-use is held constant throughout the entire simulation, and from the historical to the future climate in the NA-CORDEX configuration.The simulation domain with the USGS dominant land-use type for each grid box is shown in Figure 1.It is important to note that in this version of WRF, the Noah LSM only uses the dominant land-use category for each grid box when calculating the surface energy and water balance.This version does not have the ability to take into account multiple land types per grid box given their fractional grid coverage (this is an option in newer versions).Additionally, the urban environment is represented in this configuration of WRF, in the Noah LSM, simply as a type of surface cover with specific assigned properties, like the other land-use types in Figure 1.While these are the WRF settings that are most important in interpreting the results herein, a Earth's Future full list of parameterizations used and other configuration settings may be found on the NA-CORDEX website or in Table S1 (Mearns et al., 2017).
In order to assess the combined effects of the RCP8.5 GHG-induced climate change and future anthropogenic LUC, complementary simulations with the same WRF configuration as used in NA-CORDEX were produced for 2075-2100 with prescribed LUCs that are consistent with SSP3 and SSP5.Future LUCs for 2090 were prescribed and held constant for the entire 2075-2100 timeslice in the complementary simulations.As the NA-CORDEX simulations are transient simulations that cover 1950-2100, with the RCP scenario forcing starting in 2006, in order to guarantee an identical simulation initiation state, the new simulations were started using a restart file from the original NA-CORDEX simulation, but with modified land-use relevant variables, at July 1, 2073 (allowing a 1.5 years spin-up for the simulation to adjust to the new land-use state, which was removed for analysis).

Application of LUC in WRF
Crop, pasture, and urban fractional land-use fields from the historical period LUMs are not the same as their respective USGS/WRF counterparts in magnitude or spatial distribution, and in WRF, crop and pasture are represented by multiple land cover categories.Therefore, future changes in land-use from the LUMs could not be directly applied in WRF.In WRF using the USGS land categories, cropland is represented in categories 2-6 (Figure 1), and pasture, that is land that is suitable for grazing, could be seen as types 5, and 7-10.For this study, the authors applied the LUM changes as absolute fractional LUC deltas (as LUM future minus LUM historical period land cover fractions) to the USGS/WRF fractional land-use fields.The updated fractional fields were used to calculate new dominant land-use fields for the future LUC simulations in a later step.The LUC fractional deltas were added to types 2 or 3 for crop, using type 3 (irrigated crop) if it already existed as the predominant crop type in a grid box; pasture was applied to grassland category 7, and urban was applied to the urban land category 1.Total changes across the domain in the fractional land-use type fields for WRF were then adjusted to be within 5% of those projected by the LUMs.Next, new dominant land-use category fields, the fields used in WRF, and specifically the Noah land-surface parameterization, were calculated from the adjusted land-use fraction fields.Further details regarding the application of the LUM LUCs in WRF and the motivation behind some of the presented choices can be found in the supporting information S1.Historical, future, and individual change fields for crop, pasture, and urban land fraction from the LUMs and the modified fractional categories for WRF are also provided in the Figures S1-S3 for reference.Changes to the dominant land category for each grid box in WRF for SSP5LUC and SSP3LUC are shown in Figure 2. Changes in the crop, pasture, and urban fractional fields that were used to calculate those new dominant land-use maps in WRF for SSP5LUC and SSP3LUC are summarized in Figure 3. Additionally, the percent of the total area each land-use field represents over CONUS, as applied in WRF, is given in Table 1a.
Note that LUCs were only applied over the US, as plotted in Figure 3, as sub-country level crop and pasture projections could not be produced over the other countries in the domain due to the unavailability of historical crop and pasture data at subcountry level scales at the time of production.Land-use history is used to inform the spatial disaggregation of projections that are initially nation-scale within the LUMs.Therefore, the presentation of our results will focus on CONUS, where the results of the application of LUC on the climate are the most relevant.BUKOVSKY ET AL.

Analysis Methodology
Statistical significance of the climate change projections and the differences across the projections is tested at the 0.1 level using bootstrapping with bias correction and acceleration (Efron & Tibshirani, 1993;von Storch & Zwiers, 1999).Seasonal or annual means for every year within the historical and/or future scenario periods being differenced are pooled together, and from this pool, two lots of x number of years are randomly selected, with replacement, where x equals the number of years in one input period (where the two periods have the same length).The average of each lot is taken, and their difference is calculated.This is repeated 10,000 times to produce a distribution of differences from which the lower-and upper-tail critical values are estimated, bias corrected, and compared to the original difference between the two periods to determine if the original difference is outside of the critical values and, therefore, a significant difference.This method provides an estimate of where the differences are outside of the variability present in the range of years used in the analysis with 90% confidence.
The authors calculate several precipitation statistics: precipitation intensity, the percent of hours out of all hours that are either wet (%Wet, a.k.a., precipitation frequency) or dry (%Dry), the average number of consecutive wet hours per precipitation event (CWH), and the average number of consecutive dry hours between precipitation events (CDH).Intensity is the average precipitation rate across wet hours only.Wet hours are defined as hours with precipitation greater than or equal to 0.01 mm/h, and all others are considered dry.All precipitation statistics were calculated from hourly precipitation output.
Additionally, Figure 3 indicates urban-rural point pairs that are used for analysis in Section 3.Each pair of points represents an urban point and an eastward (or downwind, at least in winter) rural point (or at least less urban).Urban points in Figure 3 from west-to-east across the domain indicate the Dallas/Fort Worth, TX metroplex (DFW), the Minneapolis/ St. Paul, MN metropolitan area (MSP), the Chicago, IL metropolitan area (CHI), the central Florida megaregion centered on Tampa (FL), and the Northeast Megalopolis centered on New Jersey (NJ).

Impact of LUC on Temperature Projections
Mean temperature projections from our MPI-ESM-LR-driven WRF simulations for most of CONUS range from about 3°C-6°C in the annual mean, 3°C-7.5°C in winter (December-February; DJF), and 3.5°C-4.5°Cin summer (June-August; JJA) without LUC (Figure 4).Projected increases are greatest in the Upper Midwest, particularly in DJF, and the Interior West, particularly in JJA.Under SSP3LUC, projected warming decreases by 0.25°C-1.0°C,over a region stretching from the southern Texas-Louisiana border through Arkansas and into Missouri, regardless of season (Figures 4b,4e,4h,5a,5d,and 5g).Similar areas of noticeably cooler projections are also scattered throughout the rest of the Southeast US and occasionally in the Western US under SSP3LUC.These significantly cooler projections are strongly tied to locations where the dominant land-use category at a grid box in WRF changed from a forest type to cropland to accommodate the large increases in cropland in SSP3 (c.f.Figures 2a, 2c, 5a, 5d, and 5g).In JJA, the cooling effect of deforestation is most pronounced where deciduous broadleaf forest was replaced with cropland, and in DJF, where evergreen needleleaf forest was replaced.However, the average cooling effect on the projections over scenario-respective, dominant-cropland area over CONUS is only −0.18°C in JJA compared to −0.44°C in DJF (Table 1d).Conversely, the scattering of points across the Western US that are 0.25°C-0.75°Cwarmer in Figures 5a, 5d, and 5g (particularly from Northeast Oregon to Southwest Montana) are coincident with grid boxes that changed from dominantly grassland/pasture to a forest type in WRF due to the decrease in pasture in SSP3.The change in pastureland area, however, is small relative to the changes in cropland and urban land; therefore, the influence of pasture LUC is also small (less than 0.1°C) when considering the mean influence across all CONUS pasture area (Table 1d).The urban land increase in SSP3LUC is also small, and so is its overall effect on temperature (Tables 1b and 1c), but over some BUKOVSKY ET AL.Note: (a) Percent of CONUS that is classified as a given land-use type from the dominant land-use field and from the fractional land-use fields (labeled "total") for Hist, noLUC, SSP3LUC (SSP3), and SSP5LUC (SSP5).(b-e) Mean JJA and DJF projections or projection differences for precipitation (precip, %) and near-surface mean temperature (temp, °C) for CONUS (b) or land areas that are dominantly urban (c), crop (d), or pasture (e)."noLUC-Hist" provides the mean absolute (temp) or percent (precip) change from the historical period to the future noLUC scenario.The other columns provide the absolute differences between the noted projections.CONUS, conterminous United States; DJF, December-February; JJA, June-August; LUC, land-use changes; SSP, Shared Socioeconomic Pathway.

Table 1
Urban land, cropland, and pastureland coverage over CONUS and their influence on CONUS mean precipitation and temperature projections.
The most notable and significant differences in the projections from SSP5LUC versus noLUC are the regions of additional warming of 0.5°C up to about 4°C in the annual mean, to 1.5°C-2.75°Cdepending on the region in winter, to upwards of 4.5°C in summer (Figures 5b,5e,and 5h).This additional projected warming is strongly tied to areas of urbanization in SSP5, but unlike the most significant changes in SSP3LUC, the additional warming projected in SSP5LUC expands beyond just the grid boxes that change to a dominantly urban land-use category over a greater region, especially in JJA.This is most obvious in the differences that are between 0.25°C-1.0°C(in light to dark gray) in Figure 5, which surround the areas that have changed to dominantly urban land (c.f.Figures 2, 5b, 5e, and 5h).Overall in SSP5LUC, the LUCs, predominantly the larger urbanization effect, increase CONUS mean temperature projections by 0.16°C in DJF and 0.25°C in JJA (Table 1b).Whereas, in SSP3LUC the total LUC effect on CONUS mean temperature projections is only 0.03°C in DJF and −0.02°C in JJA (Table 1b), even though dominant urban land in SSP5LUC accounts for only 3.24% of CONUS land area (a 2.79% increase over Hist and noLUC; Table 1a) and dominant cropland in SSP3LUC accounts for 23.38% of CONUS land area (a 9.26% increase over Hist and noLUC).In the end, CO-NUS-average projections are warmer in SSP5LUC than in SSP3LUC (Table 1b), and the differences between the scenarios are greatest across the Eastern US (Figures 5c, 5f, and 5i).Some of the projection differences noted for SSP3LUC, where the climate change induced warming is decreased, also apply in SSP5LUC, but to a lesser extent, as the LUCs in crop and pasture are less extensive (Figure 5).For instance, a decrease in projected warming is still evident in SSP5LUC near the Texas-Louisiana border, where cropland has replaced forest as the dominant land-category in WRF.
The differences between the near-surface mean temperature projections from SSP3LUC and noLUC are likely predominantly due to albedo changes and changes in the partitioning of LHF and SHF.These were shown to be the predominant causes of warming due to afforestation in Davin et al. (2020)   studies is an important uncertainty source that causes different model responses to LUCs.Nonetheless, here the authors likely have similarly influential processes from the Texas-Louisiana border region into Missouri, where warming due to GHG-induced climate change is countered by cooling via deforestation for dryland cropland.Notably, cropland has a higher albedo than forest, which promotes cooler daytime temperatures.Additionally, in JJA in particular, maximum temperature (Tmax) is reduced most where the deciduous forest cover is reduced, and this is additionally coincident with where LHF is increased and SHF is decreased.Further south, where needleleaf forest is reduced and the effect on maximum and mean JJA temperature is smaller, SHF is slightly increased and LHF reduced (Figures 6, S4a-S4c).Surface roughness may also be playing a role in the cooler projections over the deforested land, as minimum temperature (Tmin) is reduced where forest is reduced for cropland as well (Figure 6).This may be because deforested, lower roughness length land cools more than forested land at night as the stable conditions trap more cool air at the surface, whereas the increased turbulence over forest causes more mixing (Lee et al., 2011).
Differences in projected temperatures due to urbanization, particularly in SSP5LUC are also likely predominantly due to albedo differences and changes in the partitioning of turbulent heat fluxes.Urbanization notably lowers albedo and causes increased SHF and decreased LHF, and warmer daytime and nighttime temperatures as a result, as noted in many previous studies of the urban heat island effect (e.g., Arnfield, 2003;Janković & Hebbert, 2012;Masson, 2006) and as seen here 6).Overall, the effect on minimum temperature is larger than the effect on maximum temperature in SSP5LUC, as illustrated in Figure 6 (right column) for DJF and JJA.This was also seen in Argüeso et al. (2014).For either minimum or maximum temperature in JJA, the additional warming over urban centers due to urbanization alone is on par with the warming due to GHG-induced climate change alone.The same is generally not true in DJF over much of the US, except with minimum temperature in FL.

Impact of LUC on Precipitation Projections
Annual mean precipitation is projected to increase over much of CONUS north of about 40°N and over parts of the Southeast US, while drying is projected for the Southwest US and Mexico (Figure 7a).The same pattern generally exists in DJF, but a greater magnitude increase is projected for the north, less drying is projected for the Southwest US, and a stronger decrease is projected in Mexico.In summer, precipitation is projected to strongly decrease over parts of the Southwest US and Mexico, and projections for an increase in precipitation are more limited to Northcentral and Northwest CONUS.Although these projections are from one RCM simulation driven by one GCM, they are consistent with the projections from the full collection of GCMs in CMIP5 (Wuebbles et al., 2017), and generally in agreement with the rest of the NA-CORDEX ensemble (Bukovsky & Mearns, 2020).In SSP3LUC, the precipitation projections change only slightly, regardless of season (Figures 7 and 8).For instance, the CONUS-average percent increase of 5.41% in JJA and 21.99% in DJF increase by only an additional 0.56% in SSP3LUC in both seasons (Table 1b).However, there are patterns in the projection difference field that do align with LUC in JJA that are worth noting.In the Northwest US, for instance, projections for increased precipitation in Southern Idaho and westward from there are enhanced in areas of strong irrigated and dryland crop increases in SSP3LUC that occur at the expense of shrubland.Additionally, the widespread region of deforestation for cropland that occurs from the southern Texas-Louisiana border north into Missouri in SSP3LUC has general, insignificant increases in precipitation projected in noLUC in JJA, but in SSP3LUC the projection switches to a general, insignificant decrease in precipitation (Figures 7g-7h, and 8g).Although statistically insignificant, the magnitude of this shift (5%-15%) is noteworthy and potentially of practical significance since the sign of the projection changed (from an increase to a decrease), and the spatial extent of the effect is widespread (Figure 8g).
The differences between the noLUC precipitation projections and the SSP5LUC projections are stronger and more noteworthy than those that occur between noLUC and SSP3LUC (Figures 7 and 8), particularly during the summer over the eastern half of CONUS (Figures 8g-8i).Significant increases in precipitation occur over areas that experience urbanization under SSP5LUC in the Eastern US in JJA, particularly in areas that become dominantly urban (c.f.Figures 2b, 2d, and 8h).Precipitation projections under SSP5LUC are decreased in the surrounding areas in JJA, especially downstream from the urbanized areas.The same does happen under SSP3LUC, but to a much lesser extent given the much smaller increase in urban coverage.
On the other hand, over areas of urban expansion on the West Coast (i.e., near San Francisco, Portland, and Seattle), there is a reduction in precipitation in the SSP5LUC scenario in JJA compared to noLUC.Overall, the CONUS mean precipitation change in JJA in SSP5LUC is drier than in SSP3LUC by about 1% in the absolute sense (Table 1b).SSP5LUC is drier than noLUC and SSP3LUC in the CONUS average because of the widespread drying around the urbanized areas, despite having considerably more precipitation over the dominantly urban points in JJA in this scenario (Table 1c).Specifically, JJA precipitation is projected to increase over scenario-relevant dominant urban points by 15.6%, 24.7%, and 31.0%from Hist to noLUC, SSP3LUC, and SSP5LUC, respectively.Considering that LUCs under SSP3LUC are primarily agricultural and SSP5LUC primarily urban, these results suggest that urban land expansion is potentially more influential than cropland expansion on future precipitation patterns in North America.

Earth's Future
As the differences between SSP5LUC and noLUC in the JJA precipitation projections over Eastern US urbanized areas are larger than the differences produced by the agricultural LUCs, statistically significant, and would potentially affect many people, the rest of this section will be spent examining these projections further.
Differences in the JJA projections of precipitation characteristics between SSP5LUC and noLUC for a representative sample of points targeting five Eastern US urban areas that vary in size and location (marked in Figure 3) are summarized in Table 2.These differences indicate that the stronger increase in mean precipitation over Eastern US urbanized areas in the future in JJA is associated with, in all locations, a greater increase in precipitation intensity and longer precipitation events (see "Intensity" and "CWH" in Table 2a for the projection differences, or Table S2 for the separate noLUC and SSP5LUC projections).Specifically, under SSP5LUC intensity is projected to be about 32% stronger, and events are projected to be about 12% longer, on average, across the five locations.Differences in hourly precipitation frequency projections are mixed depending on location, with increases in frequency over DFW, FL, and NJ and decreases over CHI and MSP ("%Wet," Table 2a).Precipitation frequency is projected to decrease at all locations under noLUC, but the sign of the precipitation frequency projection switches between the noLUC projections and the SSP5LUC projections in DFW and FL (Table S2).Intensity projections in SSP5LUC are also greater than those in noLUC at the "rural" points east of the urbanized areas, but to a much lesser extent than at the "urban" points (12.92% vs. 31.90%,respectively, averaged across the locations: Table 2b vs. Table 2a).The rural points under SSP5LUC all have less frequent precipitation than the noLUC scenario though ("%Wet" and "%Dry," Table 2b), meaning that the projection for decreased precipitation frequency in the noLUC scenario for these points decreases further (Table S2).Additionally, many of the rural locations have shorter precipitation events under SSP5LUC and a corresponding increase in the number of dry hours between events (Table 2, "CWH" and "CDH").
Similar processes are at play in producing the different summer precipitation projections from the SSP5LUC scenario near Eastern US urban areas versus the noLUC scenario as are seen in observation-based studies and modeling studies (e.g., Argüeso et al., 2016;Bornstein & Lin, 2000;Niyogi et al., 2011;Shepherd, 2005;Shepherd & Burian, 2003;Wu et al., 2019).The warming, potentially aided by the increased surface roughness, over the large urbanized areas in the SSP5LUC simulations compared to the noLUC simulations induces low-level convergence and low-level upward motion (Figures 9a and 9b).Surface humidity may be lower in the SSP5LUC simulations over the urbanized areas, as expected due to decreased surface evaporation, but the enhanced low-level convergence in the near surface winds leads to increased moisture flux into the urbanized areas (Figures 9c and 9d).The enhanced surface warming over the urbanized areas also destabilizes the lower atmosphere, as suggested by the lower convective inhibition (Figure 9e).The lifting condensation level and level of free convection are also higher over the heavily urbanized regions, but so too is the boundary layer height, presumably allowing these levels to be reached more often (Figures 9f-9h).All of the above translates into enhanced precipitation over all of the urbanized areas in the form of higher intensity storms and storms that persist for longer (Table 2a).It does not translate to more frequent precipitation than in the no-LUC future in all cities though, despite a slight diurnal enhancement in the frequency of precipitation in the late-afternoon/early evening in the Eastern US cities examined (Figure S5).In the less urbanized surroundings, conditions are made less favorable for precipitation.This is generally best represented by the stronger low-level divergence of the near-surface winds outside of the heavily urbanized areas and broad areas of increased convective inhibition across the Eastern US, that then leads to fewer and often shorter precipitation events.Near coastal regions, the large urbanized areas and their intense heat island effect also interact with and enhance the sea-breeze.In Florida (FL), this effect is strong enough to lead to a much-enhanced and much more diurnally persistent sea-breeze throughout the future mean JJA diurnal cycle in SSP5LUC (illustrated using near-surface moisture flux in Figures S6 and S7).This supports enhanced precipitation frequency and intensity throughout most of the FL diurnal cycle (Figure S5).

Summary and Discussion
Simulations were performed to examine how not including the land-use change that underlies the SSP + RCP framework may affect RCM projections of future climate change performed for CORDEX to date, and to answer the broad, but critical question we posed in the introduction: "Does inclusion of SSP-based LUCs significantly modulate the RCM projections?"Focusing on the effects on mean temperature and precipitation for CONUS, we have found that regional climate change projections are sensitive to SSP-based urban and agricultural LUC, as evidenced by statistically significant differences in the projections in some regions.The authors have also shown that the type of land-use change that is assumed matters (i.e., SSP3 vs. SSP5), a conclusion relevant to the scenarios framework.
In regions of significant crop expansion like the Southeast US, particularly under SSP3LUC, projected annual mean temperature increases are dampened by 0.5°C-1.5°C.In localities with large future urbanization projections (SSP5LUC), projected mean temperature increases are substantially magnified in and beyond urban boundaries.Projections for mean temperature are up to 4°C-5°C greater in JJA in urban centers.This additional warming in summer is on par with the warming due to GHG-forced climate change alone.This is also the case for both minimum and maximum temperature in JJA under SSP5LUC.In SSP5LUC the additional warming is not limited to urban centers.Projected mean and maximum temperature increases are up to around 0.5°C greater between them in the eastern half of the US in JJA.While regional precipitation is not greatly influenced by land-use change in SSP3LUC, in SSP5LUC over urbanized areas, mean summer precipitation is considerably enhanced, mostly due to an increase in the intensity of the events, but also an increase in the length of the events.This has potential implications for projections of increased exposure to urban flooding.Precipitation is also suppressed around the urbanized areas in summer.Overall, the differences between the projections from the SSP-based LUC scenarios suggests that urban land expansion is potentially more influential than cropland expansion on CONUS temperature and precipitation projections, at least under RCP8.5.instance, here the urban environment is simply represented by differences in land surface cover properties, and not an urban canopy model.Therefore, the three-dimensional nature of cities is not represented.Using an urban canopy model would likely provide more realistic simulations.Additionally, the land-surface parameterization used had no option for considering sub-grid scale fractional land cover at the time the CORDEX simulations were produced.It does now, and so do other land-surface schemes, so SSP-based fractional LUCs could be applied in future simulations.Considering only the dominant land-use type in a grid box may have caused an under-and/or over-estimation of the effect of the LUC on the projections, depending on the location and LUC type, and it likely also altered the intended amount of LUC applied in WRF relative to that projected by the LUMs (e.g., compare total vs. dominant land area in Table 1a).The authors hope to examine how these modeling choices affected our results in future work.However, results from this study are broadly consistent with observational-based studies and other modeling studies that have examined the role of LUC on climate in terms of their trend and broad physical effect, as discussed previously in the context of the results.Nonetheless, the resolution in this study is also potentially too coarse for some urbanization effects with or without the use of an urban parameterization as well (e.g., the 50-75 km downstream influence of the urban canopy on precipitation seen, for instance, in Niyogi et al., 2011).A higher resolution would also likely provide a better representation of summer precipitation, in particular, regardless of proximity to an urban area.Furthermore, the effect of urbanization on the precipitation projections here does not include any changes in anthropogenic aerosols and, therefore, does not consider their effect on nucleation.The authors also do not consider added anthropogenic heat.Likewise, urban land change considers only the expansion of urban extent.An enhanced data set of changes in urban morphological characteristics would be more realistic, but is not currently available.
Our methods for applying the LUCs in WRF may also warrant additional study.For example, while the authors tested different methods for applying the crop projections from the LUM to the different crop types in WRF, the authors did not examine our application of pasture projections with as much scrutiny.In the future the authors will experiment with applying the changes to other categories that could be considered pasture, not just grassland.Pasture projections in this case though do not have as widespread an effect on climate as the crop and urban projections, as pastureland area change is small.
Ultimately, this work suggests that for a more complete exploration of uncertainty in future regional climate projections, the regional modeling community should consider the LUCs that underlie the SSP + RCP framework, and not just the GHG concentration scenarios.This is particularly true as the community looks forward to downscaling simulations from CMIP6 ScenarioMIP (O'Neill et al., 2016).Such analyses, however, would require that sub-national land-use change scenarios that are consistent with all relevant SSP + RCP scenarios be available at near the resolution of the models over, preferably, the full region of interest.As there are many different methods in which the LUC can be incorporated into the RCMs and many different ways in which the land surface can be represented in RCMs, additional sensitivity tests should be performed, like those being produced for LUCAS (Davin et al., 2020) in Europe, and groups which undertake LUC incorporation in their projections should fully document their methods.Finally, the international CORDEX community should discuss modeling strategies and methodology for the use of SSP-based LUC scenarios to establish best-practices.

Figure 1 .
Figure 1.Simulation domain including the dominant land-use category from the baseline simulation for each WRF grid cell.Land-use index with corresponding land-use category description listed on the right.WRF, Weather Research and Forecasting model.

Figure 2 .
Figure 2. Dominant land-use category for only grid cells that end up changing land-use category under an SSP-based LUC scenario, all others remain white (see Figure 1 for land-use index definition).(a) Land-use category used in the Hist and noLUC simulations for cells that change under SSP3LUC; (b) as in (a), but for cells that change under SSP5LUC; (c) new land-use category under SSP3LUC; (d) new land-use category under SSP5LUC.LUC, land-use changes; SSP, Shared Socioeconomic Pathway.

Figure 3 .
Figure 3. Absolute change in fractional land-use from the baseline to the future in WRF under SSP3LUC (top) and SSP5LUC (bottom).Blue symbols indicate locations of point pairs used in our analysis, as described in Section 2.3.The pair abbreviations are given to the left of each set in the top panel.Fields are plotted at 70% opacity so strong changes in multiple fields at a given point can be identified.Urban change is plotted over crop change, which is plotted over pasture change.LUC, land-use changes; SSP, Shared Socioeconomic Pathway; WRF, Weather Research and Forecasting model.

Figure 4 .
Figure 4. Change in average near-surface mean temperature from 1980-2005 to 2075-2100 for the noLUC future scenario (a, d, and g), the SSP3LUC scenario (g, e, and h), and the SSP5LUC scenario (c, f, and i) versus Hist.(a-c) Annual mean change, (d-f) DJF mean change, (g-i) JJA mean change.Projections at all points are statistically significant at the 0.1 level, so no indicator of significance was used in this figure.DJF, December-February; LUC, land-use changes; SSP, Shared Socioeconomic Pathway.

Figure 5 .
Figure 5. Differences in the average near-surface mean temperature projections across the future scenarios.Left column: SSP3LUC-noLUC, center column: SSP5LUC-noLUC, right column: SSP5LUC-SSP3LUC.(a-c) Annual mean difference, (d-f) DJF mean difference, (g-i) JJA mean difference.Differences that are statistically significant at the 0.1 level follow the lower colorbar, points that are not significant follow the faded upper colorbar.DJF, December-February; JJA, June-August; LUC, land-use changes; SSP, Shared Socioeconomic Pathway.

Figure 6 .
Figure 6.Left column (a, d, g, and j): change in DJF and JJA average Tmax and Tmin from 1980-2005 to 2075-2100 for noLUC versus Hist (as labeled).Projections at all points are statistically significant at the 0.1 level in this column, so no indicator of significance was used.Center column (b, e, h, and k): Differences in average Tmax and Tmin projections between the SSP3LUC and noLUC future scenarios.Differences that are statistically significant follow the lower colorbar, points that are not significant follow the faded upper colorbar.Right column (c, f, i, and l): as in the center column, but for the SSP5LUC versus noLUC.(a-c) DJF Tmax; (d-f) DJF Tmin; (g-i) JJA Tmax; (j-l) JJA Tmin.DJF, December-February; JJA, June-August; LUC, land-use changes; SSP, Shared Socioeconomic Pathway.

Figure 7 .
Figure 7.As in Figure 4, but for the percent change in mean precipitation from 1980-2005 to 2075-2100.Differences that are statistically significant at the 0.1 level follow the lower colorbar, points that are not significant follow the upper colorbar.

Figure 8 .
Figure 8.As in Figure5, but for the absolute difference in the average precipitation percent change projections.Differences that are statistically significant at the 0.1 level and grid cells where the significance changed between the projections follow the lower colorbar, points that are not significant follow the upper colorbar.

Table 2
Absolute Differences Between the SSP5LUC and noLUC Projections of JJA-Average Percent Change in Different Precipitation Characteristics for the Points Indicated in Figure 3 and Defined in Section 2.3 These projections, however, only come from one RCM configuration forced by one GCM under one RCP and two future SSPs.They demonstrate that the CORDEX projections can be significantly affected by including LUCs underlying the SSP + RCP framework.However, more research is needed to document the effect RCM + LUC sensitivities and structural uncertainties have on the projections across different LUC scenarios and across different regions.
For example, as the authors leveraged NA-CORDEX simulations here, and were constrained by the existing WRF configuration, changing some relevant model options may be worth exploring in future studies.For BUKOVSKY ET AL. 10.1029/2020EF001782 15 of 18