Placing the east-west North American aridity gradient in a multi-century context

We used monthly 1/4° grids of 1901–2020 total precipitation (summed over all days), mean daily maximum (Tmax) and minimum temperatures (Tmin), relative humidity, wind speed, and solar and longwave radiation from a variety of climate data products. The methods used to develop these datasets are detailed in Williams et al (2017) and Williams et al (2020). Observations of long-term, spatially continuous soil moisture do not exist, so we incorporated an updated record of modeled gridded (1/4°) monthly 1901–2020 soil moisture anomalies from Williams et al (2020). The model was forced by the monthly climate data described above and calibrated to optimize agreement with monthly mean 0–200 cm soil moisture plus snow water equivalent as simulated by the Noah land-surface community model (Ek 2003). To avoid the effects of recent anthropogenic trends on baseline estimates, modeled soil moisture and climate data were standardized relative to 1910–1960 climatological means and variances.

We produced gridded (1°) summer (JJA) soil moisture reconstructions across central North America (24–50°N, 66–130°W), which included the contiguous United States, northern Mexico and southeastern Canada, using a network of 2800 chronologies of tree-ring width index values standardized to preserve low-to-medium frequency variability (Melvin and Briffa 2008), predominantly accessed from the International Tree-Ring Databank (www.ncdc.noaa.gov/paleo/treering.html). The tree-ring dataset is an updated version of that used by Stahle et al (2020).

We reconstructed gridded records of standardized JJA 0–200 cm soil moisture anomalies (SMz) using the standard point-by-point nested reconstruction methods used to develop the North American Drought Atlas (Cook et al 1999, 2004, 2010). The SMz data are updated from Williams et al (2020). We produced our own gridded soil-moisture reconstruction rather than the North American Drought Atlas to facilitate comparisons to idealized soil-moisture calculations covering the observational interval described in sections 2.3 and 2.4. Following Williams et al (2020) and Williams et al (2021), we produced an ensemble of reconstructions using varying calibration periods (1901–1978, −1983, −1990, −2000), search radii (50, 100, 150, 200, 300, 400, 500 km), spatial smoothing (1, 3, 5 degrees smoothing), and minimum chronology length (1850-, 1800-, 1750-, 1700-, 1650-, 1600–1978). Reconstruction models were cross-validated using an iterative leave-a-decade out approach (see Williams et al 2020, Bolles et al 2021 for methods), correlating out-of-sample reconstruction time series against the target time series (cross-validated R²).

Following Williams et al (2021), alternate reconstructions then replaced the primary reconstruction (calibration period: 1901–1978, search radius: 150 km, spatial smoothing: 3 grid-cells, minimum chronology length: 1850–1978) when either (a) the corrected Akaike Information Criterion (AIC_C; Akaike 1974) decreased by at least two without reducing the cross-validated R² or reconstruction length, or (b) the reconstruction length increased by at least 10 years without reducing R² or increasing AIC_C. Maps of the reconstruction parameters are provided in figure S1 (available online at stacks.iop.org/ERL/16/114043/mmedia).

Regionally-averaged SMz reconstructions were calculated for eastern (eastern NA; 24–50°N, 66–100°W; eastern United States and southeastern Canada) and western North America (western NA; 24–50°N, 100–130°W; western United States and northern Mexico), omitting grid cells with R² < 0.2. The reconstructions were then bias-corrected to match the calibration period variance of the target time series, as variances of regionally-averaged reconstructed time series can shift due to changes in available tree-ring records (Williams et al 2020). Next, regional observed and reconstructed SMz time series were standardized to the 1400–1960 mean and variance (calculated after replacing reconstructed values with observations during 1901–2020). The 'east-west North American aridity gradient' was calculated as eastern minus western NA SMz and is referred to as the 'aridity gradient' for the remainder of the paper. Positive aridity gradient values therefore represent years when SMz anomalies are more positive in eastern than in western NA and negative aridity gradient values represent the opposite. The location of the aridity gradient is dynamic and large historic shifts in the gradient's background mean location could reduce temporal variability in our calculations of east-west aridity differences. However, under a strong radiative forcing scenario, Seager et al (2018b) found the location of the aridity gradient to shift by approximately 2° longitude, small relative to the sizes of our eastern and western NA study regions. It is thus unlikely that our reconstruction of the aridity gradient's strength is strongly impacted by pre-observational shifts in the precise location of the east-west dividing line. Throughout the paper we focus on 20 year running means of regional SMz and the aridity gradient. The reasoning is that the 21st century has thus far been nearly continuously dry across much of western NA while, across much of the eastern NA, annual precipitation totals generally remained high following strong positive trends over the 20th century. As our observational climate data ended in 2020, this study focuses on 20 year running means specifically to compare the strong aridity gradient during 2001–2020 to all other observed and reconstructed 20 year periods.

We considered the effects of observed variability in monthly precipitation, temperature, and relative humidity on the aridity gradient to account for the effects of evaporative demand and water limitations. We isolated the effect of potential evapotranspiration (PET; calculated as the Penman-Monteith reference evapotranspiration from Tmax, Tmin, solar radiation, relative humidity, and wind) and then the combined effects of temperature and humidity within PET on SMz as the change in SMz caused by allowing the variable of interest to vary and holding remaining variables at mean 1910–1960 climatology (e.g. Williams et al 2015). The precipitation effect was then isolated as the portion of SMz not attributed to PET.

The same approach was used to isolate the contributions of climate variables during specific seasons (Spring MAM, Summer JJA, Fall SON, Winter DJF). This leads to eight counterfactual seasonal records, where each season is assigned a recalculated version of precipitation or temperature and humidity. Temperature and humidity work together to affect evaporative demand. Contributions of all climate forcing variables to SMz were nearly additive, but nonlinearities prevent the contributions from being perfectly so. Seasonal contributions were therefore rescaled to sum to 100% of each annual contribution and annual contributions were rescaled to sum to the total SMz time series.

We characterized the effects of modeled anthropogenic climate trends on the aridity gradient as the difference between the observed aridity gradient and a counterfactual time series calculated with climate data after removing simulated anthropogenic trends. We characterized anthropogenic climate trends as the multi-model mean trends in precipitation, temperature, and humidity as simulated by 29 models that participated in the sixth phase of the Coupled Model Intercomparison Project (CMIP6) (Eyring et al 2016). Following Williams et al (2020), modeled trends were assessed as 50 year low-pass filtered time series of temperature and humidity and 100 year low-pass filtered precipitation totals from 1850 to 2100, where 1850–2014 data come from the CMIP6 Historical experiment and 2015–2100 data come from tier 1 of the shared socioeconomic pathway 2–4.5 (SSP2-4.5; medium-forcing pathway of +4.5 W m⁻²) scenario from ScenarioMIP (O'Neill 2016). For precipitation, CMIP6 typically projects smaller anthropogenic trends relative to interdecadal variability, necessitating a longer filter. To reduce effects of outliers, monthly precipitation values were log-transformed prior to filtering. Next, low-pass filtered time series are converted to change since 1901. For precipitation, change since 1901 is calculated as fractional change (each month's filtered precipitation is divided by the filtered value in 1901). Time series of change since 1901 were then subtracted (divided for precipitation) away from the observed time series in order to calculate counterfactual realizations of the observed climate history absent of modeled climate trends. Modeled uncertainty in anthropogenic trends was characterized by the range of trends simulated among the 29 models. The 29 models considered were selected based on availability of monthly precipitation, temperatures, and relative humidity data for Historical and SSP2-4.5 experiments. This approach is described in detail in the supplementary materials of Williams et al (2020), although that study used CMIP5 simulations.

During 1901–2020, 97% of eastern NA land area experienced positive precipitation trends (59% significant; p < 0.05) while 45% of western NA experienced negative precipitation trends (3% significant) (figure 1(a)). For summer Tmax, 37% of eastern NA experienced cooling (8% significant), while 96% of western NA warmed (82% significant) (figure 1(b)). These trends led to a positive trend in the aridity gradient since 1901, with 78% and 64% of eastern and western NA experiencing positive and negative SMz trends (31% and 18% significant), respectively (figure 1(c)). The divergence in eastern versus western hydroclimatic trends has been particularly strong since the 1980s, leading to 2001–2020 being the largest 20 year period of positive aridity gradient in the observed record (figures 1(a)–(c)). These climate trends are consistent with numerous regional studies across North America (Seager et al 2012, Easterling 2017, Mascioli et al 2017, Williams et al 2017, 2020, Bishop et al 2019a, 2019b).

Zoom In Zoom Out Reset image size

The observed aridity gradient has a consistent east-west contrast in correlations with tree-ring width records across North America (figure 2(a)). This indicates the tree-ring network's potential to contextualize the recent strengthening of the aridity gradient. Reconstructions of gridded SMz were generally more skillful in the West than in the East and many grid cells in the Northeast and Midwest were disqualified (R² < 0.2) (figure 2(b)). However, the gridded SMz reconstruction still overlaps with 85% of eastern NA land area that experienced 1901–2020 soil moisture increases (figure S2). Reconstructed western and eastern NA SMz (R² = 0.79 and 0.67, respectively) and the aridity gradient (R² = 0.68) were skillful back to 1400 CE (figure 2(c)).

Zoom In Zoom Out Reset image size

We evaluated 20 year running means of reconstructions and observations of western NA SMz, eastern NA SMz, and the aridity gradient to contextualize the 2001–2020 extreme aridity gradient relative to the past six centuries. In western NA, there were four distinct multi-decade pluvial events during the early- and late-20th, early-17th, and mid-to-late 16th centuries, and three multi-decade drought events in the late-19th, late-16th to early-17th, and early-21st centuries (figure 2(d)). In eastern NA, multi-decadal variability was generally less pronounced than in the West, consistent with Herweijer et al (2007). The largest multi-decade swings in the east consisted of long-term drying in the mid-to-late 1500s, wetting in the late 1500s to early 1600s, and wetting in the 20th and early-21st centuries (figure 2(e)). Opposing SMz trends in eastern and western NA led to distinct peaks in the aridity gradient reconstruction, and the strongest positive (wet east, dry west) gradients occurred in 2001–2020, 1574–1593, and 1846–1865 (ordered from most to least positive). The strongest negative gradients (dry east, wet west) occurred in 1976–1995, 1549–1568, and 1906–1925 (ordered from most to least negative) (figure 2(f)). The latter period includes the noted North American pluvial centered in the west (Fye et al 2003, Woodhouse et al 2005, Cook et al 2011). Over the past few decades, the swing from the most negative aridity gradient of −0.91 σ in 1976–1995 to the most positive mean aridity gradient of +1.00 σ in 2001–2020 resulted in the most rapid aridity gradient reversal since 1400 CE (Δ = + 1.91 σ). This reversal was rivaled only by the late-1500s reversal (Δ = + 1.52 σ) from −0.73 σ in 1549–1568 to +0.79 σ in 1574–1593.

The 2001–2020 aridity gradient remained the most positive when we retained all gridded soil moisture reconstructions (including R² < 0.2; figure S3). We repeated the reconstruction over longer and shorter time frames, finding the magnitude of the 2001–2020 positive aridity gradient is highest for the shortest reconstructions, as these reconstructions represent more of eastern NA that experienced 20th-century wetting and exclude higher-amplitude hydroclimatic swings of the Medieval period (figure S4).

The second-most positive aridity gradient in our reconstruction (1574–1593) coincided with the well-documented late-16th century megadrought in western North America (Stahle 2000, Cook et al 2018, Williams et al 2020) although the East was not as anomalously wet relative to 2001–2020 (figure 3(a)). Interestingly, the late 16th-century positive gradient immediately followed the second-most extreme negative aridity gradient in the reconstruction (1549–1568), similar to the record-breaking swing from 1976–1995 to 2001–2020 (figure 2(f)). But in the case of 1549–1568, the negative gradient was dominated by dry conditions in the eastern Great Plains (figure 3(b)). The third-highest positive and negative aridity gradients respectively occurred in the mid-19th and early-20th centuries. These events were promoted by the mid-19th century Civil War drought in the West (Herweijer et al 2006, 2007) and one of the wettest periods in the West on record (Fye et al 2003, Woodhouse et al 2005, Cook et al 2011), respectively (figures 3(c) and (d)). The most extreme aridity gradient magnitudes were influenced by the Southwest drought and the Midwest and Northern Plains pluvial during 2001–2020 and a late-20th century Southwest pluvial during 1976–1995 (figures 3(e) and (f)).

Zoom In Zoom Out Reset image size

The observed precipitation increase in the Midwest and Northeast was the main driver of the 2001–2020 aridity gradient extreme. Annual temperature and humidity (T + H), primarily in western NA, positively forced the aridity gradient (+12% effect relative to 2001–2020 aridity gradient magnitude) (figure 4(a)), but the annual precipitation forcing (+100%), mostly driven by positive eastern NA precipitation anomalies, accounted for the vast majority of the 2001–2020 aridity gradient anomaly (figure 4(b)). Given the strong negative SMz anomalies observed in the Southwest, we expected stronger roles for T + H on the 2001–2020 aridity gradient. The drying effects of T + H in the Midwest, however, dampened the effects of these variables on the aridity gradient while the drying effects of precipitation in the Southwest strengthened the aridity gradient (figures 4(a) and (b)). The >100% forcing from precipitation and T + H on the 2001–2020 aridity gradient was balanced by a −12% forcing from wind and solar radiation. While the T + H forcing was most positive during winter (+7%), the precipitation forcing was most positive during summer (+62%) and spring (+20%) (figures 4(c) and (d)), with weaker forcings (<11%) from the remaining seasonal contributions. The summer precipitation forcing resulted from long-term wetting in the Midwest and Northeast and some recent drying across the intermountain West (figure 4(e)).

Zoom In Zoom Out Reset image size

Turning to the 1976–1995 negative aridity gradient, it received more balanced forcings from both annual T + H (−51%) and precipitation anomalies (−42%) (figures 4(a) and (b)), predominantly from above-average spring and winter precipitation across the West and negative spring T + H forcings in the Midwest, suggesting that climate anomalies in the West were more important in promoting the negative aridity gradient extreme of 1976–1995 than the positive extreme of 2001–2020. While both observed and reconstructed aridity gradient indices suggest SMz variability in the western NA was the primary driver of past aridity gradient extremes, the recent positive aridity gradient extreme was principally driven by a century-long increase in eastern precipitation.

The contribution of anthropogenic climate change was confounded by high inter-model uncertainty. CMIP6 ensemble-mean anthropogenic precipitation trends were simulated to account for 15% of the observed 2001–2020 aridity gradient anomaly, which was counteracted by a slight negative forcing of –2% from anthropogenic trends in T + H (figure 5(a)). That is, the multi-model mean precipitation increases positively forced SMz in the East more than in the West. Notably, there was considerable inter-model spread in the contribution of anthropogenic precipitation trends to the aridity gradient (figures 5(b) and (d)). Modeled effects of anthropogenic T + H trends were more consistent, with all models simulating a warming-driven drying effect across the study domain and 52% of models indicating a negative effect on the historical aridity gradient (figures 5(c) and (e)).

Zoom In Zoom Out Reset image size

The above results imply a large degree of uncertainty in the effect of anthropogenic climate trends on the aridity gradient to date. First, individual models exhibited a wide range of variability in terms of how anthropogenic climate change affects regional precipitation trends. Second, models consistently simulated more ubiquitous warming across North America than was observed. Both could indicate that the effect of anthropogenic climate trends on the aridity gradient are not accurately represented by CMIP6 climate models. It is also plausible, however, that the multi-model mean is an accurate representation of anthropogenic forcing and that the recent intensification of the aridity gradient was dominated by internal climate variability. Resolving uncertainties in modeled regional anthropogenic climate trends, especially precipitation trends, will be critical to understand the future evolution of the aridity gradient, improve annual-to-decadal prediction of regional-to-continental hydroclimate variability, and prepare the agricultural industry for shifts in soil moisture gradients over North America.

Last, to evaluate the impacts of atmospheric dynamics on aridity gradient variability, we calculated regional SMz and aridity gradient correlations with 1948–2020 National Center for Environmental Prediction and National Center for Atmospheric Research Reanalysis monthly 200 hPa geopotential height and skin temperature data (Kalnay 1996). Positive aridity gradients were associated with a winter-spring atmospheric wave train that sheltered the western NA from Pacific storms, likely linked to cool eastern tropical Pacific sea surface temperatures (figure 6). It is well-documented that 21st century drought conditions in western NA were associated with anomalously cool conditions in the tropical Pacific (Delworth et al 2015, Lehner et al 2018) and that similar conditions were likely responsible for some or all of the reconstructed western North American megadroughts of the last millennium (Seager et al 2007a, 2015, Huang and Xie 2015, Cook et al 2018, Steiger et al 2019, 2021).

Zoom In Zoom Out Reset image size

While western hydroclimatic variability was the dominant driver of decadal variability in the aridity gradient over the past 400 years, the exceptional 2001–2020 aridity gradient magnitude was exacerbated by increased summer precipitation in eastern NA. During the summer months, positive aridity gradients were associated with a midlatitude wave train over North America, which likely promoted increased storm track activity in the Great Plains and Southeast (figure 6). While past studies have dynamically linked eastern precipitation variability to the NASH and the Great Plains Low-Level Jet providing enhanced moisture transport from the Gulf of Mexico into eastern North America (Weaver and Nigam 2008, Seager et al 2012, Bishop et al 2019a), the results here do not establish a link between the NASH and the summer aridity gradient. Further research into recent trends in both Pacific and Atlantic teleconnections with North American hydroclimate will be needed to establish dynamic causes and accurately project future changes in this phenomenon, especially the long-term eastern precipitation trend. This work motivates further work to understand land-atmosphere interactions and the degree to which regional soil-moisture trends, or trends in soil-moisture gradients, may feedback to affect climate at regional-to-continental scales (e.g. Seneviratne et al 2010, Koster et al 2016).