The Decline in Summer Fallow in the Northern Great Plains Cooled Near‐Surface Climate but had Minimal Impacts on Precipitation

Land management can moderate or intensify the impacts of a warming atmosphere. Since the early 1980s, nearly 116,000 km2 of cropland that was once held in fallow during the summer is now planted in the northern North American Great Plains. To simulate the impacts of this substantial land cover change on regional climate processes, convection‐permitting model experiments using the Weather Research and Forecasting model were performed to simulate modern and historical amounts of summer fallow. The control simulation was extensively validated using multiple observational data products as well as eddy covariance tower observations. Results of these simulations show that the transition from summer fallow to modern land cover led to ∼1.5°C cooler temperatures and decreased vapor pressure deficit by ∼0.15 kPa during the growing season across the study region, which is consistent with observed cooling trends. The cooler and wetter land surface with vegetation led to a shallower planetary boundary layer and lower lifted condensation level, creating conditions more conducive to convective cloud formation and precipitation. Our model simulations however show little widespread evidence of land surface changes effects on precipitation. The observed precipitation increase in this region is more likely related to increased moisture transport by way of the Great Plains Low Level Jet as revealed by the ERA5 reanalysis. Our results demonstrate that land cover change is consistent with observed regional cooling in the northern North American Great Plains but changes in precipitation cannot be explained by land management alone.


Introduction
Global temperatures are rising, primarily due to greenhouse gas emissions from anthropogenic activities (Stocker et al., 2014).Future temperature increases are exceedingly likely (IPCC, 2023), as is an increase in precipitation extremes in extratropical zones (O'Gorman & Schneider, 2009).Embedded within this global context are changes to regional temperatures and precipitation (Christensen et al., 2007) that often result from the impacts of land management and land cover change on regional energy balances (Luyssaert et al., 2014;Mahmood et al., 2014).Some of these regional climate changes may be beneficial to agricultural and ecosystem management objectives, such as dampening extreme temperatures (e.g., Juang et al., 2007;Mueller et al., 2015); others are not if they result in for example, decrease in rainfall or a greater likelihood of freezing in areas where this is unexpected (Mande et al., 2015;Marshall et al., 2003).It is important to understand how land management impacts climate processes to develop strategies to minimize the deleterious impacts of climate change and become effective stewards of the earth system.
A unique interaction between land management and climate may have emerged across the northern Great Plains of North America (hereafter the NNAGP).Beginning in the 1960s and 1970s, concerns over soil health and profitability lead to widespread changes in agricultural management away from wheat-fallow rotation agriculture and toward a more diverse agricultural system that avoids bare ground by rotating wheat with pulses, cover crops, and other crops (Long et al., 2014;Miller et al., 2002Miller et al., , 2003)).These changes appear to have unintentionally benefitted regional climate (Gameda et al., 2007).Agricultural areas of the Canadian Prairie Provinces have experienced a 6 W m 2 decrease in radiative forcing during summer across parts of this period of rapid land cover change (Betts, Desjardins, & Worth, 2013).Summer maximum temperatures in the Canadian Prairies decreased by nearly 2°C and extreme temperature events have become less frequent (Betts, Desjardins, & Worth, 2013;Betts, Desjardins, Worth, & Cerkowniak, 2013).These regional climate effects have been attributed to the widespread decline of summer fallow from ca. 110,000 km 2 (25% of Canada's cultivated lands) to some 35,000 km 2 (8%, Gameda et al., 2007;Vick et al., 2016).In the U.S. portion of the NNAGP and across a similar time period, near-surface air temperatures have cooled by nearly 0.2°C decade 1 during late spring and early summer, and near-surface atmospheric vapor pressure deficit (VPD)-which strongly depletes crop yields (López et al., 2021)-has decreased by 0.04 kPa decade 1 on average (Bromley et al., 2020).These changes to the regional climate are concurrent with a period of summer fallow decline on the order of 50,000 km 2 in the United States from a peak in 1987 until 2012 (Figure 1).
The observed climate changes are consistent with a transition of large expanses of land away from bare ground and toward crops which actively transport water from soil to atmosphere and increase latent heat flux and evaporative cooling (Vick et al., 2016).Changes to land management have likewise decreased the surfaceatmosphere flux of sensible heat to help create a moister, shallower atmospheric boundary layer during summer (Gameda et al., 2007;Vick et al., 2016) through decreases in the Bowen ratio.Combined, these changes in surface fluxes have enhanced cloud formation (Betts, Desjardins, Worth, et al., 2013) and increased the probability of convective precipitation (Gerken et al., 2018).Monthly mean precipitation has increased in the Canadian Prairies by 10 mm decade 1 (Betts, Desjardins, Worth, et al., 2013;Gameda et al., 2007) and the U.S. northern Great Plains by 8 mm decade 1 (Bromley et al., 2020).Given the multitude of factors that drive precipitation change, the full suite of mechanisms that underlie these observed increases in precipitation and the potential role of land cover change remain uncertain.
Empirical observations and modeling studies to date have made critical inroads into our understanding of landatmosphere-precipitation connections in the NNAGP.Planting crops at the expense of summer fallow decreases planetary boundary layer (PBL) height (Gameda et al., 2007;Gerken et al., 2018;Vick et al., 2016) and, coupled with increases in humidity, lowers the lifted condensation level (LCL) (Betts & Desjardins, 2018;Betts, Desjardins, Worth, et al., 2013).Shallow cumulus clouds can result when the PBL crosses the LCL, a "necessary but not sufficient condition" for the formation of convective precipitation (Juang et al., 2007).Calculations of PBL and LCL height based on eddy-covariance data and one-dimensional mixed-layer atmospheric models show that the likelihood of PBL-LCL crossings are maximized in May and June when "wet coupling" (Roundy et al., 2013) prevails in the NNAGP such that increased moisture increases the likelihood of convective events (Gerken et al., 2018).This contrasts the prevailing "dry coupled" conditions later in summer when convective precipitation is unlikely (Gerken et al., 2018).Mixed layer models of atmospheric boundary layer development and landatmosphere coupling have demonstrated an increase in PBL-LCL crossings and an increase in the likelihood of convective rain events across parts of the NNAGP (Gerken et al., 2018), but the mechanisms underlying potential changes in convective precipitation across larger regions have not been explored.
At the same time, changes to surface-atmosphere fluxes due to land cover change may play a minor role in key aspects of the hydroclimate of the NNAGP.Convective precipitation in the NNAGP is dominated by mesoscale convective systems (MCSs) that are responsible for as much as 60% of the warm season precipitation (Carbone & Tuttle, 2008).These systems form in the west and propagate eastward, often overnight, and mixed layer models are generally unable to account for these dynamics (Carbone & Tuttle, 2008;Gerken et al., 2018).The buildup of convective available potential energy (CAPE) that supports the development of MCSs comes primarily from the advection of warm, moist air into the region in addition to the diabatic heating of the boundary layer by way of sensible and latent heat fluxes from the surface (Agard & Emanuel, 2017).The reduction of summer fallow has the potential to influence the flux of heat and moisture into the boundary layer, leading to the buildup of CAPE.Several studies have posited that increased evapotranspiration from more continuous cropping-and less summer fallow-has led to more growing season convection and potentially stronger storms (Raddatz, 1998;Shrestha et al., 2012).However, since the boundary layer has also become cooler and moister, the cooling may act to reduce CAPE and perhaps balance the tendency of added moisture to increase CAPE.Increases in temperature and moisture aloft will also increase convective inhibition (CIN) which may balance or even dominate the effects of any increase in CAPE.The exact response of convective processes to such changes in near-surface conditions is unclear and requires a mechanistic modeling environment that can explicitly account for the dynamics of convective precipitation across regional scales.
How do changes to agricultural management impact regional atmospheric and climate processes?We seek to understand how land management impacts the regional climate and hydrometeorology of the NNAGP, a critical global breadbasket for wheat production.To do so, we modeled the regional impacts of the reduction of summer fallow across the NNAGP using the Weather Research and Forecasting model (WRF) at a 4 km spatial grid to explicitly model convective precipitation processes across multiple year periods.After validating the model using reanalysis data products and describing the major results of the modeling analysis, we explore reanalysis data sets for a more comprehensive view of the changing hydroclimate of the NNAGP.We then discuss results in the context of the local and large-scale patterns that are consistent with observed climate trends across the region.

Study Area
We define the semi-arid NNAGP following Bromley et al. (2020) as the combination of the Canadian Prairie Ecozone and the U.S. National Ecological Observation Network Domain 9 (Figure S1 in Supporting Information S1).The NNAGP are dominated by grasslands, shrublands, and agriculture with minimal urban development and forests in isolated mountain ranges and river valleys (Stoy et al., 2018).The NNAGP is a critical region for the global production of wheat, pulses, and oilseeds, and corn-soy cropping is becoming increasingly common in its eastern portion (Maaz et al., 2018;Rosenzweig & Schipanski, 2019).Coupled with other land management pressures including biofuel production (e.g., Dolan et al., 2020), this area is a dynamic system characterized by notable recent changes in land management and widespread increases in vegetation greenness (Brookshire et al., 2020) and decreases in bare ground fraction (X.-P.Song et al., 2018).

Model Setup
The WRF model (Skamarock et al., 2008;Powers et al., 2017) is a state-of-the-art weather model that can explicitly simulate convective processes and has been increasingly used in high-resolution regional climate simulations (Liu et al., 2017;Powers et al., 2017;Wang et al., 2018).Our choice of model parameters relies heavily on the testing conducted by Liu et al. (2017).WRF version 4.1 was run on the Cheyenne computing cluster with a horizontal grid spacing of 4 km, 51 vertical levels up to 50 hPa, and a 20 s time step (Computational And Information Systems Laboratory, 2017).The study domain consists of 796 × 496 grid cells and encompasses the NNAGP with at least 40 grid cells between the study area and the edge of the domain which allows for features in boundary data to become fully realized within the domain (Brisson et al., 2016).Initial and lateral boundary conditions were provided by the European Center for Medium-range Weather Forecasting's ERA5 reanalysis at three-hourly intervals (Hersbach et al., 2020).The high resolution of ERA5 (∼31 km) allows us to directly downscale to 4 km grid spacing without an intermediate nest.The YSU (Yonsei University) boundary scheme (Hong et al., 2006), Thompson et al. (2008) microphysics scheme, and the RRTMG (Rapid Radiative Transfer Model for GCMs) radiation scheme (Iacono et al., 2008) were utilized in the simulations.We did not use spectral nudging due to the limited size of the domain.Analysis focuses on the warm season in 2011, allowing for 6 months of model spin up before the main analysis.Winter temperature and precipitation validation (Figures S2a and S3a in Supporting Information S1) includes data after 2 months of spin up.We used the Noah-MP (Noah-Multiparameterization) land surface model with dynamic vegetation options turned off and leaf area index prescribed by table values for each land cover category (dveg = 4, Niu et al., 2011).Vegetation fraction (fveg) was fixed at the annual maximum to facilitate land cover experiments.Full three-dimensional output was saved every 3 hr while surface and precipitation data were saved at hourly intervals.
The TOPMODEL groundwater option was turned on, as simulating groundwater and runoff helps reduce the Great Plains warm season bias that is common to many WRF simulations (Barlage et al., 2021).The simulations were run for 3 years coinciding with the water year beginning in October.The control simulation with lesser summer fallow fraction is simulated for October 2010-October 2013.It is difficult to avoid the effects of natural climate variability such as those induced by the El Nino Southern Oscillation (ENSO) in short simulations, and results are subject to forcing from ENSO and other climate modes.The simulation period starts during a positive ENSO phase and then shifts to more neutral conditions.Additional model setup details are provided in the Supporting Information.

Land Cover Experiments
The extent of summer fallow in Provinces and States that intersect the study area decreased from 151,900 km 2 in 1982 to 35,100 km 2 in 2012, the years closest to the study periods when data were available from the United States (data from Canada were available every year) (Figure 1).Some 45% of the total decline in fallow of 116,800 km 2 was attributable to Saskatchewan alone (53,000 km 2 ).These published agricultural statistics were used to nudge the Landsat fallow attribution analysis on a per-Province and per-State basis (Figure 2) to ensure that it simulated total fallow area, which was then used to adjust the bare ground fraction in Noah-MP.We note that the fallow attribution analysis retains some features of Landsat scenes; we explore the consequences of these features in WRF predictions.
Summer fallow was represented within the model by reducing the fveg parameter by the estimated fallow percentage for each grid cell.Noah-MP uses a split-cell calculation for land-atmosphere interactions, meaning fluxes are calculated for the bare ground fraction and the vegetated fraction separately and then combined to give the surface fluxes for the entire grid cell.The control simulation uses a fveg that matches the estimated summer fallow extent in 2011 (hereafter "C11"), while the simulation with modern climate and 1980s summer fallow extent is called F11.We chose 2011 to match data availability from the U.S. National Land Cover Data set (NLCD, Homer et al., 2015), noting that this year was subject to relatively large fallow areas in parts of Saskatchewan, Manitoba, North Dakota and Minnesota due in part to widespread spring flooding that limited planting (Figure 1, Stadnyk et al., 2016).From this perspective, our analysis represents a conservative interpretation of fallow change from the 1980s until the 2010s (Figure 1).We focus our analyses on the C11 and F11 simulations to isolate the role of land cover change apart from decadal global climate change on determining regional climate changes in the NNAGP.Statistical differences between simulations were assessed using the Mann-Whitney U test on daily data.We separate our analyses of convective environments between the early warm season (defined here as May and June) and late warm season (July and August) given differences in surface-atmosphere coupling in the NNAGP during these periods (Gerken et al., 2018).
The data processing workflow relied on the Climate Data Operators package (Schulzweida, 2019) as well as several analysis packages in Python, such as NumPy, xarray, Matplotlib, and scipy.stats(Harris, Millman, et al., 2020;Hoyer & Hamman, 2017;J. D. Hunter, 2007;Virtanen et al., 2020).Convective parameters, such as CAPE and CIN were calculated using the cape_2d function from the wrf-python package (Ladwig, 2017).Additional land cover details are provided in the Supporting Information.

Model Validation
The control simulations are extensively validated against observational data sets in Figures S2-S8 of Supporting Information S1 and validation will only briefly be discussed for completeness.Control simulations were compared to the Daymet data set (Thornton et al., 2016), the CRU (Climatic Research Unit) data set (Harris, Millman, et al., 2020;Harris, Osborn, et al., 2020), the Gridded Meteorological Ensemble Tool (Newman et al., 2015), and the Global Precipitation Climatology Centre precipitation data set (Rustemeier et al., 2020).Surface fluxes were compared against eddy covariance observations from Lethbridge, Alberta (Flanagan et al., 2002).The Lethbridge eddy covariance tower was the only tower in the NNAGP with observations that overlapped the simulation time period.The near-surface (2 m) temperature in the C11 simulation was well simulated during spring and fall with cold (warm) biases in winter (summer) that were similar in magnitude to other WRF simulations (Liu et al., 2017) in the NNAGP (Figure S2 in Supporting Information S1).Precipitation was well simulated in C11 for all seasons (Figure S3 in Supporting Information S1).The control simulation was compared to observational atmospheric soundings acquired from the Integrated Global Radiosonde Archive (Durre et al., 2006) for the locations of Edmonton, AB, Glasgow, MT, and Bismark, ND.C11 has a 2.5°C near surface warm bias at all locations for the late warm season (JAS), while the early warm season (AMJ) was well captured and was within 1°C at all levels (Figure S7 in Supporting Information S1).Surface energy fluxes largely matched eddy covariance measurements from a grassland site in Lethbridge, Alberta (CA-Let, Flanagan et al., 2002, Figure S8 in Supporting Information S1).
Investigating warm season (AMJJAS) climate trends is the focus of this work, and changes to other seasons are intermittently discussed for completeness.Seasonal changes are taken to be the average change across the 3-year simulation.

Changes to Near-Surface Temperature, Energy Fluxes, and Humidity
Since the spatial amounts of fveg do not monotonically change between F11 and C11, there are areas that show increases and decreases in most variables analyzed here (Figure 2).We state changes in reference to the fallow simulation minus the control simulation (F11-C11).Two-meter air temperature (T2) showed a domain-averaged increase of about 0.18°C during the growing season in F11-C11 (Figures 1-3).The strongest simulated warming was limited to June, July, August (JJA) (Figure 4).T2 was cooler on average by 0.5°C in areas where fveg was higher in the F11 simulation compared to C11 such as southern Manitoba.The northern and western part of the NNAGP in Alberta experienced higher T2 on the order of 1.5°C.There is a linear relationship between simulated fallow and T2 (Figure 3b); T2 increases by almost 0.7°C for every 10% decrease in fveg.There is a modest cooling signal during winter in F11-C11 indicating the more vegetated C11 simulation is warmer by about 0.25°C on average (Figure 4) following the same spatial pattern as growing season temperature changes (not shown), but opposite in sign and of a lesser magnitude.
Changes in VPD in F11-C11 follow the same spatial pattern as T2; the near-surface atmosphere in areas that had lower fveg were drier and the areas that had higher fveg were moister (Figure 5b).The domain median VPD was lower by 0.045 kPa in F11-C11 with a wide distribution that encompasses regions with both higher and lower VPD (Figure 5a).VPD is higher by 0.15 kPa in F11-C11 over most of Alberta within the study area.The strongest change in VPD is nearly +0.3 kPa and occurs in central and eastern South Dakota.Fveg is higher in F11-C11 in Manitoba (Figure 1); VPD is subsequently lower there by 0.15 kPa, with areas nearly 0.2 kPa lower.Study area-averaged sensible heat flux (H) is higher by 10 W m 2 in F11-C11 in areas where fveg is lower compared to C11 (Figure 6a).The areas of largest change have magnitudes on the order of 20 W m 2 .Changes to latent heat flux (LE) follow the same pattern as H, but the magnitudes are larger (Figure 6b).Study area-averaged LE is 16 W m 2 lower in F11-C11.The magnitudes of LE differences are larger than H, with some areas in excess of 30 W m 2 in C11.Eastern Montana, western South Dakota and southern Saskatchewan show smaller  1).
Positive values indicate that the F11 simulation was warmer.magnitudes of change in both H and LE compared to the areas near the border of the study area that are considered "crop" land cover types in the land surface model.

Changes to Convective Environments
During May and June, changes to CAPE and CIN are not significantly different from background noise (Figures 7a and 7b).The height of the LCL is higher in F11-C11 for most of the study area.The mean change to LCL height is 13 m while some areas are more than 30 m (Figure 7c).Manitoba is the only area where the LCL heights are lower.Mean LCL heights in Manitoba decreased by 20-30 m.Differences in PBL heights closely follow the spatial pattern of differences in LCL heights (Figure 7d) and, consequently, the areas that have lower fveg in F11-C11 have higher PBL heights, while the opposite holds for areas that have a higher fveg in F11-C11 such as Manitoba.The mean change in PBL heights within the study area is 8 m but some areas change up to 30 m.
During July and August, CAPE is lower across most of the study area in the F11-C11 by 10 J kg 1 with minima in Alberta and the Dakotas (Figure 7a) where CAPE is lower by 20-40 J kg 1 .Changes to CIN in July and August weaker than the changes to CAPE (Figure 8b).CIN is lower in the F11-C11 simulation in Alberta and parts of Saskatchewan by 5 J kg 1 on average but this change is not statistically significant.Manitoba and north-eastern North Dakota exhibit higher CIN in F11-C11 by 20-30 J kg 1 .Differences in LCL heights show a similar pattern to May and June but much stronger with F11 simulating LCL heights that are 50 m higher than C11 across most of the study area (Figure 8c).The largest differences are located along the North Dakota-South Dakota border, northern Saskatchewan, and Alberta.LCL heights in these areas are over 100 m higher.PBL heights in July and August follow a similar spatial pattern as in May and June but with greater magnitude (Figure 8d).PBL heights in Alberta and northern Saskatchewan are higher in the F11 simulation than C11 by over 100 m on average but only 10-20 m higher in the central parts of the study area.The only area where PBL heights are lower in F11-C11 is in Manitoba with an average change of about 50 m, noting again that there was an increase in simulated fallow (decrease in fveg) extent in Manitoba between 1984 and 2011 (Figures 1 and 2).

Changes to Precipitation
Changes to precipitation were not appreciably different from background noise from May through August (Figure 9).When aggregated to 100 km to account for storm track differences in each simulation, a slight drying becomes apparent during July and August, but any effects are not statistically significant.

Discussion
We demonstrate evidence, using convection-permitting WRF model simulations, that historical land management change toward continuous cropping and away from summer fallow decreased near-surface air temperature and VPD.Precipitation did not appreciably change between the WRF simulations, indicating that simulated precipitation in the NNAGP are not very sensitive to the land surface changes of the magnitude experienced in recent decades.We believe the lack of precipitation change is likely because the increase in instability through increased boundary layer moisture is balanced by an increase in stability through a cooler boundary layer.Below, we elaborate on each of these findings to describe how land cover change has modified important aspects of the regional climate of the NNAGP, especially near the land surface.We then add to emerging evidence that observed changes in precipitation are likely due to moisture advection into our study domain rather than regional surfaceatmosphere interactions.

Temperature
To summarize findings on the impact of summer fallow on near-surface climate: areas that had a higher fveg from F11-C11 were cooler with lower VPD (Figures 3 and 6).Near-surface warming and drying occurred in areas where fveg was lower.These results lend evidence to the notion that a reduction in summer fallow is largely responsible for the near-surface cooling and moistening trend that is observed across the NNAGP.The changes to temperature in the simulations are stronger than the trends calculated by Bromley et al. (2020), noting that the trends in the latter are calculated from a 1970 starting point, whereas these experiments simulate fallow reduction from the 1980s-2010s.Temperature trends are stronger at nearly 0.5°C decade 1 (Bromley et al., 2020) when calculated using 1980 as a starting point, on the order of 1-1.5°C, similar to modeled changes in T2 associated with fveg differences from F11 to C11.The temperature difference simulated here spans from May until September, but given wheat is often harvested in August (if not sooner for the case of winter wheat), the September T2 difference is likely due to the prescribed seasonal cycle for each land use category.
The winter warming in the C11 simulation relative to the F11 simulation is likely due to the decrease in albedo from increased fveg in the model.Since the fveg does not change based on a seasonal cycle, areas with greater fveg are assumed to be lower in albedo since the vegetation is not covered in snow.The bare ground areas are covered in snow and thus are higher in albedo.This is similar to year-round cover cropping and the winter warming effect has been noted in global climate models (Lombardozzi et al., 2018).Simulating snow advection, especially the tendency of vegetation to trap blowing snow (Pomeroy & Li, 2000;Pomeroy et al., 1998), is a challenge to earth system models, which may overestimate the albedo effect of wintertime vegetation as a consequence (M.C. Hunter et al., 2019).Multiple modifications to the Noah-MP snow physics calculations were made by Liu et al. (2017), to create more realistic cold season surface-atmosphere interactions and spring melt profiles but wintertime processes are still an active area of land surface model research.

Vapor Pressure Deficit
Plant stomata respond strongly to VPD; if VPD is too high, stomata will close to avoid evaporative water losses, effectively shutting off carbon uptake by plants (Eamus et al., 2013;Grossiord et al., 2020;Novick et al., 2016) with important implications for surface-atmosphere fluxes (Rigden & Salvucci, 2017;Yuan et al., 2019).VPD is increasing on average across the United States, except for the U.S. portions of the NNAGP in which VPD is decreasing by an average of 0.5 kPa decade 1 (Bromley et al., 2020;Ficklin & Novick, 2017).The VPD change for the increase in fveg from F11 to C11 is on the order of 0.45 kPa which corresponds with the observed changes to VPD in the NNAGP.Our modeling analysis suggests that the impacts of simulated fallow reduction on near surface climate has acted to create more favorable conditions for crop growth by reducing growing season temperatures and VPD (Hsiao et al., 2019).Wheat yields differ in their sensitivity at different crop growth stages, and early-season days with mean temperature >28°C are especially detrimental (Asseng et al., 2015).It is interesting to note that the midwestern United States has experienced largely beneficial changes in near surface growing season climate as a result of agricultural intensification (Mueller et al., 2015), leading one to question if they can be sustained in the future as global climate change continues to stress water resources and production systems.

Planetary Boundary Layer
The PBL by definition is the near surface layer of the atmosphere that is strongly influenced by surface fluxes of water and energy, so it is not surprising that the systematic shift away from summer fallow affects PBL processes.The monthly mean boundary layer heights were 100 m higher in late summer in the F11 simulation where the fallow amounts were larger; the lowering of the PBL as fallow declines was proposed to be a consequence of the changes in energy partitioning from a fallow (bare) surface and a vegetated surface (Gameda et al., 2007) which was the case in these simulations (Figure 5).The change in PBL height was previously assessed using a simple slab model with inputs from eddy covariance observations of turbulent fluxes from wheat and fallow fields (Vick et al., 2016).These simulations suggested an increase in PBL height of about 200 m during the growing season over a fallow field versus a spring wheat field.This 200 m difference is larger than the 60 m difference in mean monthly PBL heights simulated here, which is perhaps not surprising given that WRF simulates a spatial mix of fallow and vegetated areas whereas Vick et al. (2016) modeled PBL impacts of fallow and vegetation separately.PBL growth is sensitive to heterogeneous landscapes and the model representation of seasonal and diurnal variations is improved if the heterogeneity of surface fluxes is captured (Beamesderfer et al., 2023;Rey-Sanchez et al., 2021).Using MM5, a model similar to WRF, Mahmood et al. (2011) found that simulations of bare soil were 1.4°C warmer than the control simulations (present day vegetation) and the seven-day average PBL heights were ∼550 m higher with a lower LCL under higher fveg fractions, which increased the probability of cloud development and convection (Mahmood et al., 2011).Our results are consistent with the notion that summer fallow changes PBL and LCL heights but its realized impact on precipitation was negligible and spatially variable (Figure 9).

Precipitation
There is little evidence that mean precipitation changed appreciably between F11 and C11 (Figure 9), suggesting that a reduction in summer fallow had minimal impact on observed precipitation trends.July and August are 10-15 mm drier in the F11 simulation when the precipitation change is aggregated to 100 km × 100 km boxes, but these changes are not significant at the 95% level.Precipitation in the NNAGP increased by 8 mm decade 1 in May and June, but July and August precipitation also increased, primarily on the eastern side of the NNAGP (Bromley et al., 2020).If the land surface is not appreciably changing mean precipitation, what is the source for the observed warm season precipitation increases seen in Bromley et al. (2020)?
Global mean precipitation has been increasing due to anthropogenic warming of the atmosphere at about a rate of 2% K 1 (Held & Soden, 2006;Pendergrass & Hartmann, 2014).This rate comes from the thermodynamic change to precipitation but does not account for changes to the dynamic components such as changes to circulation.Precipitation in the NNAGP is largest in the early warm season and May through September is a convectivelyactive period (Gerken et al., 2018).Precipitation during this time period can take the form of stratiform rain, MCSs, and organized pre-frontal convection; July and August are quite dry compared to May and June and precipitation is primarily from MCSs.The Great Plains Low Level Jet (GPLLJ) is a nocturnal wind speed maximum, positioned at about 850 hPa, that transports moisture from the Gulf of Mexico into the Great Plains.
July and August MCS development in the NNAGP is usually accompanied by a strong northward-penetrating GPLLJ (Feng et al., 2016(Feng et al., , 2019;;F. Song et al., 2019).To investigate the possibility that the observed increase in precipitation in the NNAGP is consistent with additional moisture sources from the south, we investigated meridional wind and specific humidity trends in ERA5. Figure 10 shows a vertical cross-section along the 42°l atitude line of 1979-2020 trends in monthly mean meridional wind and specific humidity.Due to the lack of strong trends in meridional wind, meridional moisture transport trends are only slightly positive (Figure 10).There is not a clear signature of a strengthening GPLLJ, but the increase in surface specific humidity corroborates Bromley et al. (2020); near-surface conditions are moistening during May and June.Trends in the North American Regional Reanalysis data set shows that moisture transport northward has increased during AMJ, particularly during days with MCS initiation (Barandiaran et al., 2013;Feng et al., 2019).1), their absolute difference (Abs Diff), and percent difference.Precipitation was aggregated to 100 km × 100 km boxes to display regional trends.
Most of the MCSs in the NNAGP occur during July and August, and the northward extension of the GPLLJ over the past four decades is clear in monthly mean trends (Figure 11).Specific humidity has increased at 0.3 g kg 1 decade 1 while meridional wind has increased at 0.35 m s 1 decade 1 .These trends add moisture to the NNAGP and likely contribute to the observed increase in precipitation on the eastern and southern boundaries of the NNAGP during summer as well as the lower VPD during JJA (Bromley et al., 2020).A full accounting of this additional moisture from potential sources including the Gulf of Mexico and other ocean-atmosphere drivers (Eischeid et al., 2023), contributions from irrigation (Pei et al., 2016), or other upwind regions would improve mechanistic understanding of the larger scale hydroclimatology that has impacted the NNAGP.We note that the magnitude of CAPE change was on average larger in July and August, which could mean that increase in convective environments conducive to strong storms has been aided by the reduction of fallow (Brimelow et al., 2011).An analysis that tracks MCSs and looks at changes to convective environments, for example, Feng et al. (2016), could perhaps show how much the land surface impacts these processes and resolve how changes in land cover and regional circulation processes have impacted the unique climate trends of the NNAGP.

Conclusions
Summer fallow in the NNAGP has declined from an estimated 151,900 km 2 in the 1980s to 35,100 km 2 in the 2010s, a decline of 116,800 km 2 which is approximately the land area of Pennsylvania, USA.To investigate the climate impacts of this reduction in summer fallow, two 3-year convection permitting WRF simulations were performed using ERA5 as the initial and lateral boundary conditions.The vegetation fraction of each simulation was adjusted using Landsat-estimated summer fallow and nudged to match published agricultural statistics for 2011 and 1984.The intention of these simulations is to understand how the near surface climate and precipitation processes have been impacted by these substantial changes in land cover.The summary of the results are: • Two-meter air temperatures were 1-1.5°C cooler and VPD was 0.15 kPa lower in areas where fveg increased between the fallow simulation and the control simulation.• The PBL and LCL were lower by 60 m, due to the cooler and more humid land surface.
• CAPE increased by 20-30 J kg 1 but there were minimal changes to CIN.
• Precipitation did not change appreciably between the simulations.
The results of these simulations suggest that observed near-surface cooling and moistening trends in the NNAGP are largely a result of the reduction in summer fallow.The lack of evidence for a land-surface induced change to precipitation stands in contrast to other observational studies focused on the same region; however, this is the first modeling study looking at summer fallow reduction on a regional scale.Further work is needed to better understand the precipitation processes, perhaps tracking the evolution of precipitating storm systems as they move over the heterogeneous and changing landscape of the NNAGP.
CA-Let research site (Flanagan, 2018).We acknowledge the contributions of data providers and curators from all data sets used here.Code used in the analysis is available at https://zenodo.org/records/11127644.

Figure 1 .
Figure 1.The area of land held in summer fallow in the (a) Canadian Prairie Provinces and (b) U.S. States of the northern North American Great Plains for the 1982-2012 period using data from Statistics Canada and the United States Department of Agriculture Economic Research Service following Vick et al. (2016).The primary study years, 1984 and 2011, are indicated with vertical dotted lines.

Figure 2 .
Figure 2. Differences in vegetation fraction (fveg) between the 2010 fallow and 1984 for the Noah-MP land surface model in the Weather Research and Forecasting model estimated using Landsat and adjusted to match published agricultural statistics (Figure 1).Brown areas indicate areas that have lower fveg in F11, and green areas have higher fveg in F11.

Figure 4 .
Figure 4. Monthly differences in T2 between the F11 and C11 Weather Research and Forecasting simulations (Table1).Positive values indicate that the F11 simulation was warmer.

Figure 5 .
Figure 5. Two-meter vapor pressure deficit difference between the modern fallow (F11) and control (C11) simulations during the 3-year simulation period for MJJA.Positive VPD values indicate that the F11 simulation is drier, while negative values indicate that the F11 simulation is moister.Stippling indicates significant differences at the 95% level.

Figure 6 .
Figure 6.(a) The difference in sensible and (b) latent heat flux for MJJA between the F11 simulation and the C11 simulation.Stippling indicates significant differences at the 95% level.

Figure 7 .
Figure 7. Changes to (a) convective available potential energy, (b) convective inhibition, (c) lifted condensation level, and (d) planetary boundary layer between 2011 control (C11) and fallow (F11) Weather Research and Forecasting simulations during May and June for the three-year simulation period.Stippling indicates significant differences at the 95% level.

Figure 8 .
Figure 8.The same as Figure 7 but for July and August for the 3-year simulation period.

Figure 9 .
Figure 9. Changes to precipitation for May and June ("Early Warm," Row 1) and July and August ("Late Warm," Row 2) for the F11 and C11 Weather Research and Forecasting simulations (Table1), their absolute difference (Abs Diff), and percent difference.Precipitation was aggregated to 100 km × 100 km boxes to display regional trends.

Figure 10 .
Figure 10.Vertical cross-section of 1979-2020 May and June meridional wind trends (black contours) and specific humidity trends (filled contours) for the levels between 925 hPa and 800 hPa from the ERA5 reanalysis.Inset axes show trends in meridional moisture transport (qv) for 1979-2020 and the location of the crosssection.Brown contour shows the pressure where topography is located along the cross section.

Table 1
Abbreviations and Explanations for Each Model Simulation STOY ET AL.