Projected US drought extremes through the twenty-first century with vapor pressure deficit
Although VPD has become increasingly useful in drought research21,29,30,31, VPD itself may be more difficult to interpret compared to established drought indices (e. For this work, we compare three meteorological drought indices identified by the WMO in 2016: PDSI, SPEI and USDM, and a more recent drought index based on reference evapotranspiration: EDDI. June, July, August, and September 2003 PDSI, SPEI (1-month), EDDI, and SDVI_NLDAS (SVDI). Overall, SVDI_NLDAS captures the location, rapid onset, and duration of the 2003 Flash Drought event. Here, VPD has been utilized to produce SVDI, a simplified method for drought detection. Here, VPD is calculated with NLDAS data and compared to VPD calculated with data produced by three Global Climate models downscaled by the Weather Research and Forecasting (WRF) model44: WRF CCSM, WRF GFDL, and WRF HadGEM (See “Methods” section for model details). Spatial comparison of VPD statistics for 1995–2004: mean (left column), maximum (middle column), and standard deviation (right column). Figure 3 shows decadal averaged annual maximum VPD for WRF CCSM, WRF GFDL, and WRF HadGEM with current climate simulations (1995–2004; historic) and future climate projections based on RCP 8.
Spatial comparison of decadal averaged annual maximum VPD: 1995–2004 (Historic), 2045–2054 (Mid-Century) and 2085 -2094 (Late-Century). The large increases in maximum VPD by the end of the twenty-first century may be due to increases in modeled air temperatures using RCP8. June, July, and August averaged daily maximum temperatures for each timeframe with WRF CCSM, WRF GFDL and WRF HadGEM. Of the four focused locations described earlier, the inter-model differences in Tmax during the late-century timeframe shows the Midwest location with the most agreeability between models, a difference of only ~ 1. Although summer Tmax is expected to increase across the United States (Fig. June, July, and August averaged daily minimum relative humidity for each timeframe with WRF CCSM, WRF GFDL and WRF HadGEM. Overall, the projected increases in Tmax is likely driving the increases in daily VPD and may have serious repercussions on agricultural yield47 and hydrological resources in the western United States. Since many types of droughts are associated with high values of VPD, and no ground truth of identified droughts is available in projected climates, we further statistically quantify future extreme VPD via extreme value analysis.
A GEV statistical model is a three-parameter probability distribution model and has been used extensively to characterize extreme events such as extreme temperatures and precipitation48,49,50,51,52. Both stationary and non-stationary GEV models are fitted at each grid point and compared (see the Methods section for details). We fit GEV models to annual maxima of daily VPD and identify 5 return periods of interest (2, 5, 10, 25 and 50-year), and the associated return levels are computed with the fitted models. Ensemble return periods for the Northwest, Midwest, South and Los Angeles locations. Here we use the fitted GEV models to estimate return periods as a means of understanding the magnitude of future extreme events. The 2, 5, 10, 25 and 50-year return periods in the United States for the 5th, median (50th), and 95th percentile of the sampled model ensemble. While lower uncertainty in the 50-year return periods is found in many regions of the western and eastern United States, here we focus on regions of larger uncertainty. To better understand VPD return values in context, we compared locations with VPD values greater than 9 kPa with all three models for the historic and late-century timeframes.
Read full article at Nature.com