Emergent Trends Complicate the Interpretation of the United States Drought Monitor (USDM)

Effective drought management must be informed by an understanding of whether and how current drought monitoring and assessment practices represent underlying nonstationary climate conditions, either naturally occurring or forced by climate change. Here we investigate the emerging climatology and associated trends in drought classes defined by the United States Drought Monitor (USDM), a weekly product that, since 2000, has been used to inform drought management in the United States. The USDM classifies drought intensity based in part on threshold percentiles in key hydroclimate quantities. Here we assess how those USDM‐defined drought threshold percentiles have changed over the last 23 years, examining precipitation, runoff, soil moisture (SM), terrestrial water storage (TWS), vapor pressure deficit (VPD), and near‐surface air temperature. We also assess underlying trends in the frequency of drought classifications across the U.S. Our analysis suggests that the frequency of drought class occurrence is exceeding the threshold percentiles defined by the USDM in a number of regions in the United States, particularly in the American West, where the last 23 years have emerged as a prolonged dry period. These trends are also reflected in percentile‐based thresholds in precipitation, runoff, SM, TWS, VPD, and temperature. Our results emphasize that while the USDM appears to be accurately reflecting observed nonstationarity in the physical climate, such trends raise critical questions about whether and how drought diagnosis, classification, and monitoring should address long‐term intervals of wet and dry periods or trends.


Introduction
Drought in the United States (U.S.) is costly.The hazard it presents to a range of sectors-from ecosystems to agriculture, from water management to the energy industry-has been documented extensively.Accounting by the National Oceanic and Atmospheric Administration (NOAA)'s National Centers for Environmental Information (NCEI), for example, estimates drought-induced economic losses from 1980 to 2023 total to some $349.8 billion (2023 CPI-adjusted estimate) over 30 different U.S. drought events (NCEI, 2023).These losses are considerably larger if you consider that costly wildfires often have their origins in droughts (Abatzoglou & Williams, 2016;Goss et al., 2020;Williams et al., 2019), as the last several years of drought in the American West makes clear (Mankin et al., 2021).Given the considerable costs and substantial risks to people and ecosystems drought poses, its accurate monitoring and prediction are vital to inform effective emergency management and mitigation.
Current drought assessments in the U.S. at the national scale are primarily based on the United States Drought Monitor (USDM; Svoboda et al., 2002).While many states have unique drought plans triggered by their own drought assessments, the USDM plays a crucial role in monitoring and classifying drought conditions nationwide and informing state declarations.Established in 2000, the USDM involves assimilating multiple sources of physical indicators and on-the-ground reports with expert interpretation to assess drought intensity for all U.S. states and territories.The assessment includes physical conditions such as soil moisture (SM) and runoff anomalies, which are validated through reports of drought impacts-fields left fallow, hydropower production curtailed, municipal water supply reduced, and the like.The product, which classifies drought into one of six categories, relies on a "convergence-of-evidence" approach, whereby USDM authors integrate myriad data sources by some 450 observers, all under a collaborative effort by the National Drought Mitigation Center (NDMC), NOAA, and the U.S. Department of Agriculture (USDA) (Hao et al., 2017;Leasor et al., 2020;Svoboda et al., 2002;Xia et al., 2014a).While there are long standing efforts to build independent and objective drought measures (e.g., Lorenz et al., 2017), the USDM has become an essential tool for drought management, policy, and decision making in spite of its classification process relying on subjective expert judgment.The product that holds significant political authority: the USDM informs the declarations of emergency made by states and the relief payments made by the USDA (Hao et al., 2017;Mankin et al., 2021;Svoboda et al., 2002).Because of this importance, it is essential to understand whether and how climate trends are expressed in the expertgenerated USDM.
Drought assessments, like those performed by the USDM, are challenging to interpret in a trending climate for at least two reasons.First, the USDM is not an objective drought classification approach.Instead, it is an expertgenerated product informed by a mix of objective measures of climate anomalies alongside subjective on-theground impacts and field reporting.In that sense, the USDM is more like an all-source intelligence assessment than an objective measure of drought.In particular, it is essential to understand how field reports are generated (Wilhite et al., 2007), aggregated by stakeholders and partners like State Climate Offices to support effective drought early warning, and ultimately passed to USDM author's desk, all under very tight weekly timelines (Hoylman et al., 2022;Otkin et al., 2022).It is easy to imagine that some communities are more effective at documenting on-the-ground evidence of drought impacts and navigating the bureaucratic process by which those reports are assimilated into a drought assessment (Parker et al., 2023).Yet, the genesis, number, and diversity of field reports, or how such reports are actually weighed by a USDM author faced with a difficult drought classification decision, is not documented.Moreover, whether and how such field reporting (and the economic, political, and social behaviors underpinning such reports) have evolved in time is essential to understand and should be a research priority.Together, this means that while some factors that USDM authors consider are transparent and reproducible because they are documented on the USDM website and in the peer-reviewed literature, other, essential elements are not.Here, we align our analysis in the physical science elements of the USDM that are documented in the product's publicly available methodologies and guidelines, while emphasizing that this is only part of the picture.
The second reason the USDM is challenging to interpret in a trending climate is that even for the objective elements of the USDM assessment, there is little in the way of standardized evaluation criteria.For example, the baseline against which anomalies are defined can evolve through time and with different authors, updated data products, and longer periods of record.While the USDM has guidelines to help authors assign drought classes, it does not have a fixed climatology or a period to which all climate "normals" are benchmarked.As such, it is difficult to know how to consider the USDM through time as climate warming and low-frequency variability alter regional drought characteristics.In particular, the USDM allows the period of record for any data product to grow in time, expanding the distribution against which hydrologic conditions are defined.The use of growing periods of record and the multitude of ever-evolving drought indicators implies that the USDM likely reflects climate trends or mean state changes, whether they are due to natural multidecadal fluctuations or trends forced by climate change.At the same time, the expert-generated approach of the product means that drought classification thresholds could implicitly change to accommodate such trends.However, such a practice can present challenges in drought management and the interpretation of drought assessments over longer periods of time, ultimately raising the question of whether a drought classification of "exceptional drought" is equally exceptional in the past, present, and future in terms of hydrologic conditions, stress, and thus, the management response required.
Given the emerging USDM climatology (Leeper et al., 2022) and the questions it raises about drought monitoring in a nonstationary climate (Hoylman et al., 2022), we compare trends in the USDM to those observed in a number of hydroclimate quantities like runoff and SM.Our investigation centers on examining the amount of time regions have spent in any one drought class over the 2000-2022 period, what we term "drought residence time" (Mankin et al., 2021).We take advantage of the emerging USDM climatology and ask: (a) What are the climatology and trends in the residence time of USDM drought classes?(b) How do USDM drought class assignments compare with anomalies in geophysical conditions, like precipitation, runoff, and SM?And (c) Are there trends in the residence time of geophysical variables that fall within USDM theoretical percentiles?We address these questions by examining the USDM and six geophysical variables related to drought and hydroclimate to assess their climatologies and coincident trends.We emphasize that this analysis is only focused on the roughly objective elements of the USDM.Expert assimilation means we cannot analyze how subjective decision practices have evolved through time to shape the product's representation of low-frequency variability, forced or not.Despite previous recognition of changes in drought characteristics due to low-frequency variability and anthropogenic climate change, particularly in the Southwest U.S. (Mankin et al., 2021;Williams et al., 2022), it is unknown whether these aridity changes are impacting the character, and thus, the interpretation, of the USDM.Our work, to the best of our knowledge, is the first to quantitatively link drought changes in a nonstationary climate to the USDM and its drought class frequency guidelines.Such an analysis is essential to ongoing discussions of drought assessment in the policy and stakeholder communities, such as that of the National Integrated Drought Information System, which is working to reliably assess drought in an era of climate change (NIDIS; Parker et al., 2023).Our results inform a discussion about drought monitoring tools like the USDM, with their undoubted utility in the assessment of conditions and impacts and forward planning, and how such tools need to be evaluated in light of nonstationary trends.

Data
Our analysis relies on data from the sources summarized in Table 1.We employ a gridded form of the USDM data set, which is a set of shapefiles that draw polygons to map drought categories at a weekly timescale from 2000 to
We analyze six geophysical variables to evaluate the hydroclimate conditions associated with each USDM week under the premise that drought is multivariate and one variable alone is insufficient to assess drought features (AghaKouchak, 2015;Wilhite, 2005): precipitation (P), runoff (Q), SM, terrestrial water storage (TWS), vapor pressure deficit (VPD), and temperature (T).P, Q, SM, and TWS characterize most of the dominant terms in the water budget, while VPD and T characterize radiative and land-atmosphere processes important to accelerating or intensifying drought impacts via evapotranspiration (Diffenbaugh et al., 2015).Precipitation (mm), with a spatial resolution of 0.25°, comes from the Climate Prediction Center (CPC)-Continental U.S. (CONUS) data set hosted by the World Meteorological Organization at their Climate Explorer website.The data set was developed with a suite of unified precipitation products at CPC and by using an optimal interpolation objective analysis technique (Chen et al., 2008;Xie et al., 2007).Runoff (kg/ m 2 ) and 1-m SM (kg/m 2 , SoilM_0_100cm) come from the North American Land Data Assimilation System (NLDAS-2) Noah land-surface model, provided at a spatial resolution of 0.125° (Xia et al., 2012a(Xia et al., , 2012b)).Runoff is the sum of surface runoff (Qs) and subsurface runoff (Qsb).We use the Noah version because it is one of the NLDAS land-surface models that have reasonably high correlation and low error statistics compared to in-situ SM at a daily timescale (Xia et al., 2014b).TWS comes from the Gravity Recovery and Climate Experiment (GRACE) Data Assimilation System (DAS) based on the Catchment Land Surface Model, provided at a spatial resolution of 0.125° (Houborg et al., 2012).VPD comes from a processed product based on the European Center for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) global data set with a spatial resolution of 0.25° (Montes et al., 2021).The ERA5-based VPD product is only available up to 2020.We extend the data for 2022 based on 2-m daily maximum air temperature (Tmax) and 2-m dewpoint (Tdew) at the hour when daily maximum air temperature occurs (Hersbach et al., 2023).Hourly Tmax and Tdew are from the ERA5 global data set.Surface air temperature (°C) comes from the CPC Global Daily Gridded Temperature with a spatial resolution of 0.5°.

Methods
Our general approach is to spatially and temporally align geophysical variables with the USDM.Spatially, we rasterize the weekly USDM vector maps to the spatial resolution of the hydroclimate data sets (ranging from 0.125°to 0.5°, Table 1).Rasterization assigns pixels the value of the overlapping feature that takes up the dominant percentage in that pixel.Temporally, we resample the geophysical variables to be at the weekly-scale to ensure consistency with the weekly-scale USDM.With the exception of TWS, all variables are aggregated from their native time resolution to the weekly scale, consistent with the USDM.Each weekly USDM map is valid from the Tuesday of each week.Daily scale data are aggregated to a 7-day week, defined as Wednesday to Tuesday to be as close to the USDM as possible.Precipitation and runoff were aggregated by taking weekly sums, while the remaining variables were aggregated by taking their weekly average.The weekly aggregations are then converted to standardized z-scores, with respect to the same week over a 44-year period from 1979 to 2022.The period from 1979 to 2022 is determined by the period of record of the NLDAS data set.Input drought indicators and indices used by the USDM are calculated based on their respective data sets and distinct periods of record (H.Wang et al., 2021).The period from 1979 to 2022 is likely different from various periods of records of regional products used by USDM authors in their weekly assessments.With the exception of the CPC precipitation and ERA5based VPD, which has data prior to the 1970s, we nevertheless have used the longest available period of record for the other data sets listed in Table 1, aligning with the method provided by the USDM website.The most up-to-date list of the inputs that the USDM uses is available at their website under the "Conditions & Outlooks" section, while Yatheendradas et al. (2023) list a total of 113 selected indicators and their associated period of records.We explore the implications of the different periods of records in the "Discussion" section.Considering that the USDM is a weekly expert-generated product that incorporates numerous short-term and long-term indicators with varying periods of record alongside social reporting, reconciling the periods of record of all variables is not feasible.Our purpose is not to reproduce the USDM; instead we want to understand both the hydrologic and radiative conditions associated with a particular drought class through time using products with reasonably long periods of record as a frame of reference.

Geophysical Variables Data Processing
Using the daily-scale precipitation and hourly runoff, we calculate the backwards-looking rolling 3-month total for each week ending on Tuesday.This practice provides an indication of integrated precipitation/runoff at a weekly scale consistent with the USDM (no future information included), while also capturing 3-month hydroclimate deficits that could trigger droughts (AghaKouchak, 2015;Edwards & McKee, 1997;Viste et al., 2013;H. Wang et al., 2021).We then standardize that weekly 3-month rolling total precipitation and runoff.
We average hourly values of NLDAS-2 SM to a Tuesday-based week and also standardize them at a weekly scale to ensure consistency with the USDM.Considering that different accumulation periods may be more effective for different regions of the CONUS, and recognizing that 12-month precipitation is one of the most important indicators for the USDM, especially in the Western U.S. (Yatheendradas et al., 2023), we also aggregate and standardize precipitation, runoff, and SM data at a longer time scale of 12-month.
We calculate daily VPD from 2021 to 2022 at the daily maximum air temperature.The equations of the VPD calculation are as follows: where e s is the saturation vapor pressure in kPa estimated from air temperature and e is the actual vapor pressure in kPa estimated from dewpoint temperature.We define saturation vapor pressure as: where e 0 is 0.6113 kPa, L v is the latent heat of vapourization, 2.5 × 10 6 J/kg, R v is the water-vapor gas constant, 461 J/(K * kg), T 0 is 273.15K, and T is the 2-m daily maximum air temperature in K. Similarly, actual vapor pressure is formulated as: where T d is the 2-m dewpoint temperature at the hour when daily maximum air temperature occurs in K; the remaining variables are the same as defined in Equation 2. The daily mean temperature is calculated as the average of daily maximum and minimum temperature.
Lastly, the DAS-based TWS presents the groundwater percentile at a weekly scale on Mondays.The data set aims to combine the GRACE TWS observations with the high spatial and temporal resolutions of SM and groundwater estimates (Houborg et al., 2012).We analyze the week in the data set that ends with the Monday closest to the USDM Tuesday.For example, the USDM week that ended on Tuesday 21 June 2022, is compared with the DAS week that ended on Monday 20 June 2022.A challenge is that the TWS data are in percentiles relative to a 67-year period from 1948 to 2014 (Houborg et al., 2012), which is different from the relative period used in other data sets .We emphasize the importance of the base-period discrepancy when interpreting our results.

Percentile-Based Drought Thresholds
Drought categories in the USDM are based, in part, on percentiles of geophysical variables (Svoboda et al., 2002) that rely on functions to map particular magnitudes of a quantity (like precipitation) to their frequency of occurrence (Figure 1).Percentiles or their associated return periods allow analysts to identify how rare the event is, often framed as an event's chances of occurring in any given year out of 100 years, as in Slater et al. (2021), for example.In the USDM, percentiles of indicators like runoff or SM are standardized for time of the year for a given region; they are not meant to imply an average areal extent value for the United States at any given time (Svoboda et al., 2002).Percentile mapping ensures consistency through time and space by relying on climatological statistics of relevant variables.Such a percentile-based approach, which explicitly relates a magnitude to its frequency of occurrence, allows USDM classifications to consider shifts in the distribution of underlying hydroclimate variables, whether spatial, temporal, or both.Consider spatial shifts in climate: while a D4 "exceptional" drought is hydrologically different in Vermont than in California, percentile mapping means the likelihood and relative hydrologic anomalies associated with each event are similar, presenting similar stresses on agricultural and water management practices.Temporal changes are also theoretically controlled by percentile mapping.If average climate states change, a specific percentile, such as the 2nd percentile, would result in changed levels of precipitation or SM.For example, if the climate becomes drier, a fixed drought classification based in part on a quantity like the 2nd percentile in a hydrologic indicator, could actually correspond to drier and drier conditions over time.This implies that while a class of drought is unchanging, the actual hydrologic conditions could be.Consider this schematically for an example climatology of precipitation as presented in Figure 1.
As long-term precipitation declines, whether due to natural variability or forced by climate change (dotted line in Figure 1a), the absolute magnitude of the precipitation value, corresponding to a particular percentile, would be or "abnormally dry" conditions correspond to a 21st to 30th percentile; D1 or "moderate" conditions correspond to 11-20th percentile; D2 or "severe" conditions correspond to a 6-10th percentile; D3 or "extreme" conditions are between 3rd and 5th percentiles; and D4 or "exceptional" is a 0-2nd percentile event.For a particular location, the cumulative probability of precipitation in mm may vary depending on the period of time considered.Compare, for example, the first normal cumulative density function (CDF, solid line) to a drier second normal CDF (dotted line).In the first normal, a D4 drought corresponds to a precipitation value of "X," with a frequency maximum of 0.02, or 2% of the time.However, under an aridifying climate, the first normal shifts to the second normal, where that value of precipitation "X" occurs over 15% of the time (bar plots), changing the interpretation of drought intensity.
drier than under the earlier wetter climate (see colored arrows in Figure 1a).Because these magnitudes correspond to frequencies of occurrence defined in the first period of record, we expect the precipitation magnitudes corresponding to a percentile to occur more often in the second period of record.As such, what was formerly a rare magnitude of precipitation will occur more frequently or persist for longer periods of time relative to a previously established climate state.

Climatology and Trends in Residence Time
Given the correspondence between drought classes in the USDM and specific percentile thresholds of hydrologic variables, we examine the climatology and trends in the occurrence of particular percentile thresholds; we calculate the latter as the total number of weeks per year a grid point spends in a given drought class over the 2000-2022 period.We call this frequency of occurrence "drought residence time."Note that residence time is not the same as "drought duration," which instead represents the length of time any one drought lasts (which can comprise a series of multiple drought classifications).Because drought classes are categories that do not lend themselves easily to trend analysis, we focus on trends in drought residence time.Our logic is that in a nonstationary climate, we might expect there to be temporal trends in the total number of weeks a region spends in any one drought class in a year, even if there are no trends in drought intensity denoted by trends in percentiles.A percentile estimate of a variable is a function of the length of the period of record.This means that the magnitude of a particular percentile can remain stationary, but its frequency of occurrence (i.e., residence time) may increase or decrease in time, meaning that USDM-based percentile thresholds could occur more or less frequently due to underlying climate nonstationarity.The implication is that the drought classes may occur with frequencies that no longer correspond to the theoretical percentiles that the USDM uses to evaluate hydrologic indicators.We evaluate this possibility by examining residence time in the USDM itself, and the hydrologic variables that inform USDM classifications.As such, our focus on estimating trends in residence times is twofold: first, a percentilebased anomaly should occur climatologically at a frequency consistent with its threshold; a 0-2nd percentile D4 drought, for example, should occur no more than two percent of the time.Second, the USDM uses differing periods of record for each data set, ranging from a few decades to over five decades, and these periods continue to expand as data sets are updated.This indicates that the nature of drought classes may be changing.A longerlasting drought in the present may be more impactful compared to the past.
We calculate the residence time (%) of the USDM by dividing the number of weeks in each drought class by the total number of weeks from 2000 to 2022 (1,200 weeks).We then compare the residence time estimate to the USDM theoretical percentile guidelines, that is, 30th, 20th, 10th, 5th, and 2nd percentiles as bins of D0, D1, D2, D3, and D4 drought classes.To assess whether there have been temporal trends in residence times in the USDM, we first calculate climatology in the total number of weeks that a USDM drought class occupies over the 23-year period.To estimate trends in the residence time of drought class, we fit linear regressions to the annualized residence time (in weeks) for each grid point and drought class, then estimate p-values, corrected for autocorrelation, to assess the statistical significance of these trends (Santer et al., 2000).To make the trend more interpretable, we multiply the trend (in weeks per year) by 70 to be in units of days per decade.A positive (negative) trend represents an increasing (decreasing) rate of the number of days over a decade.The p-value still reflects the statistical significance in its original unit of weeks per year.We estimate the composite hydroclimate anomalies associated with each drought class, and we also quantify trends in the residence time of hydroclimatic conditions.
Percentiles of geophysical variables, except TWS, are calculated using the method as described in Xia et al. (2014a), who calculated the percentile of SM, runoff, evapotranspiration, and snow water equivalent using NLDAS output.Percentiles are based on a 5-day moving window to improve the sampling density and to smooth out the record (Kumar et al., 2014).If data are hourly, such as SM and runoff, they are first aggregated to daily means or sums.The value is then ranked against the daily values from each day of the 5-day window from the entire period of record from 1979 to 2022 (n = 220).In this way, daily geophysical values are converted to daily percentiles.To calculate the climatology of percentile-based geophysical conditions associated with USDM drought classes, we aggregate daily percentiles to weekly averages following Ford et al. (2015).When analyzing trends, we calculate the annual residence time of the geophysical variables falling within a corresponding percentile range using the USDM guidelines.The trend of residence time (in days/year) is multiplied by 10 to be in units of days per decade.We assess trends in residence time for most geophysical variables in two periods

Hydroclimatic Drought Class Climatology
The USDM assimilates considerable information, including drought impacts, field observations and local expert insight.As such, we should not necessarily expect it to reflect a drought composite in a purely geophysical indicator, such as the runoff anomaly associated with all D4 drought events.We therefore assess how the USDM compares with temporal composites of hydroclimatic anomalies.Compositing hydroclimate anomalies during the time periods when USDM drought classes occur reveals a consistency among the USDM and the climate: the geophysical variables show a pattern of strong drier-than-normal conditions, as proxied by negative P, Q, SM, and TWS, and warmer-than-normal conditions, as proxied by positive VPD and T (Figure 2).We also examine the climatology of percentile-based geophysical conditions, along with P, Q, and SM at a 12-month timescale, associated with USDM drought classes, and the patterns are similar (Figures S1-S3 in Supporting Information S1).The results in Figure 2 indicate that the USDM is reflective of hydroclimate anomalies in geophysical variables, despite being a drought product created with some inevitable level of subjectivity.The magnitude of water deficits or anomalous heat increases as drought conditions intensify from D0 to D4, again consistent with expectation.There are notable spatial patterns as well: the western U.S. region is subject to exceptional drought classes under milder anomalies in precipitation, runoff, or SM than in the eastern region denoted by the gradient of climatological anomalies from west to east (Figure 2).One possible explanation for this phenomenon is that the standardized water deficit required to trigger the same drought category in the (more humid) East is greater than that in the West because each region is assessed relative to its local climate.Larger standardized anomalies translate into larger absolute anomalies given the more humid climate in the East than in the West.Furthermore, precipitation, runoff, and SM in the East require larger excursions (< 1 SD) than the West to achieve exceptional drought classes.The most extreme negative anomalies occur along the eastern side of the Great Plains.The association between lower SM anomalies and higher drought risks in the West may be attributed to a stronger correlation between drought risk and exposure rather than the drought hazard itself (Walker et al., 2022).The ambiguity of the USDM as a composite product in accurately depicting ground conditions is perceived in Ward et al. (2022).Our evaluation here, however, suggests the product is indeed reflecting physical processes.

USDM Drought Class Climatology
Between 2000 and 2022, all grid points in the contiguous U.S. experienced D0 and D1 drought conditions at some point in time (Figure 3).Some regions in the Northeast and Midwest never experienced drought conditions more severe than a D2 classification.Many others, including the entire Northeast U.S., Upper Midwest, coastal Northwest, and southern California never experienced exceptional drought (D4).Notably, Southern California has never experienced D4 drought, likely due to the essential role that Colorado River water imports via the Colorado River Aqueduct play in the wealthy state's water supply portfolio.Even in the 2020-2023 Western U.S. drought (Mankin et al., 2021), Southern California was spared owing to the way it managed to keep local reservoirs at average despite heavy losses upstream at Lake's Powell and Mead.Yet, the spatial pattern of drought frequency has a clear geographic pattern: the western U.S. has spent more time in severe or worse drought conditions than the eastern U.S. Stippled areas are where the percentage of the residence time is no greater than the incremental USDM percentiles, that is, 2% for D4, 3% (5% minus 2%) for D3, 5% (10% minus 5%) for D2, 10% (20% minus 10%) for D1, and 10% (30% minus 20%) for D0 (Figure 3).Many areas, especially the western U.S., experienced residence times in severe or worse drought conditions that exceeded the theoretical percentile intervals.
More than half of the D4 areas spent longer than 2% of the 23-year period (∼5.5 months) in D4, and these exceedances are primarily in Southwest, southern Great Plains, Alabama, Georgia, and parts of the Carolinas.Some parts of California have spent up to 18% of the last 23 years (∼4.1 years) in D4 drought, suggesting that an exceptional drought (D4) did not occur at a probability that is exceptional (<2%).This situation may arise during a multidecadal drought (Williams et al., 2022).Nevertheless, prolonged aridification poses a challenge to drought management.A D4 drought often triggers emergency declarations; however, as such emergencies become more frequent due to long-term aridification (Overpeck & Udall, 2020), it raises questions about how to effectively allocate resources.
Cumulative drought classes (i.e., D0-D4) also show higher residence time in the western U.S., with only a small portion of the CONUS meeting the theoretical guidelines for frequency of occurrence (Figure 3).While we expect D0-D4 drought conditions to occur at a percentile of no higher than 30%, numerous regions in the western U.  2022) found that the western U.S. experienced longer-lasting droughts than the eastern U.S., a finding we confirm.Beyond that important work, we identify the regions where drought residence times exceed the USDM theoretical guidelines (non-stippled regions in Figure 3).These results collectively indicate that the USDM drought classifications are representing climate nonstationarity with the implication that, in many regions, the 2000-2022 interval was trending toward drier conditions, resulting in an increased prevalence of drought classes in these locations.

USDM Drought Class Trends
Having identified a climatology of both residence time and associated hydroclimatic conditions within and across each drought class, we investigate if drought residence times have trended over the 23-year period of record.If temporal trends exist in USDM drought residence times and underlying climate variables, then it is likely that climate nonstationarity is influencing drought classification, with implications for how the USDM should be interpreted.Residence time trends of individual and cumulative drought classes from 2000 to 2022 are shown in Figure 4.There are clear and statistically significant decreases in D2 and D3 residence times in Idaho, Wyoming, western Montana, Georgia, and the Carolinas (columns a and b in Figure 4).In contrast, there are statistically significant increasing trends in D4 drought residence times in much of the western U.S., including Utah, Arizona, Colorado, and New Mexico.The temporal trend in D4 drought residence time in some of the regions can reach up to 35 days/decade (p ≤ 0.05).Regions such as the northeastern part of Montana have significantly decreasing trends in D0 and D1, accompanied by significantly increasing trends in D2 and D3.Other regions such as Arizona, New Mexico, and the Four Corners also exhibit a decrease in the occurrence of D0 but an increase in the prevalence of D2, D3, and D4, suggesting a trend toward drier conditions.In contrast, places such as southern Idaho and parts of Wyoming have decreases in the occurrence of D3-D4 (Figure 4d), indicating that there is not a simple trading among drought classes occurring.
Because drought classes display offsetting increasing and decreasing trends, we also conduct trend tests on the cumulative drought classes (e.g., D0-D4), as shown in columns c and d in Figure 4.In western coastal areas, although individual drought classes (D0, D1, and D2) do not display significant trends, the cumulative drought classes, for example, D0-D4, have statistically significant increases.This indicates a more frequent occurrence of drought events in time, even after correcting for autocorrelation in the data.Similarly, in northeast Montana, while D0 and D1 exhibit a statistically significantly decreasing residence time (Figure 4b), the residence time of all drought classes (e.g., D0-D4) shows an increasing trend (Figure 4c), although it is not statistically significant (Figure 4d).This may be attributable to the significantly increasing trends in the D3 drought class, indicating a potential increase in drought intensity.Meanwhile, the southeastern U.S., particularly the Carolinas, has experienced statistically significant decreases in D0-D4, D1-D4, and D2-D4.Although the majority of the U.S. does not present statistically significant trends in drought class frequency (proxied by residence time), the increases in residence time, wherever they occur, have led to a percentage of the frequency greater than the USDM percentile guideline, consistent with results presented in Figure 3.

Hydroclimatic Drought Class Trends
Given the trends in the USDM residence times, certain drought classes are becoming more or less frequent in regions around the US.Because the USDM is not purely a geophysical product, we assess whether these trends are reflected in hydroclimatic quantities themselves by similarly calculating trends in the frequency with which geophysical variables have spent within the same class percentiles following the USDM recommendations (Figure 5).
We find that the coincidence of statistically significant trends in the frequency of geophysical quantities and the USDM drought classes from 2000 to 2022 is primarily in the western U.S. and parts of the Southeast, suggesting at least some of the USDM trends are climate driven.For example, parts of the mountain West and Southeast show decreasing residence times across individual drought classes (Figure 4b), with P, Q, and SM showing modest wetting (Figure 5).The patterns are similar when considering P, Q, and SM at a 12-month timescale (Figure S4 in Supporting Information S1).In contrast, the 95-97th percentiles and the 98-100th percentiles for VPD and T, corresponding to D3 and D4 drought, show statistically significant increases, particularly in VPD.
Most of the increasing trends in the residence time of VPD 95%-97% occur in the Southwest and parts of Southeast, with the trend in some areas in California reaching up to 20.5 days/decade (Figure 5).This is also reflected in increased residence in D3 drought in the West Coast (Figure 4).Comparing across variables, TWS, VPD and T generally present more areas of increasing trends (longer periods residing in a particular percentile) than the more variable hydrologic measures.For the places with statistically significant trends in the residence time (shaded areas), more than 80% of the areas show increasing trends in TWS 0%-2% , VPD 98%-100% and T 98%-100% , while only 46%, 51%, and 53% of the areas show increasing trends based on P 0%-2% , Q 0%-2% , and SM 0%-2% .Increasing trends indicate that the current period experiences more time characterized by extreme dry or hot conditions compared to earlier periods.
Overall, regions where there are statistically positive trends in the residence time in D4 (Nevada and Utah in Figure 4b) and D3 (along the West Coast in Figure 4b) are reflected in the increase in the residence time in the most extreme categories of Q, SM, TWS, VPD, and T (Figure 5).This is consistent with arguments that water deficits, high temperatures, and evaporative water demand have been amplifying droughts in the West.Although trends of USDM drought categories or geophysical variables do not consistently display statistical significance in the eastern United States (Figures 4 and 5), there is a widespread and statistically significant correlation between annual residence time of geophysical variables and USDM drought classes, especially for SM (Figure S5 in Supporting Information S1).Among all variables examined, VPD overall best matches the trends in USDM drought classes.VPD has the highest percentage of significantly increasing trends in places where there are coincident significantly increasing trends in most USDM drought classes, particularly in more severe drought classes like D3 and D4.
While a USDM climatology is emerging, the period over which it is defined remains short.We therefore also assess longer-term trends in geophysical residence times from 1979 to 2022 (Figure 6).We note that we cannot assess TWS over this longer period due to the short extent of the GRACE satellite record.Statistically significant residence time trends are far more widespread over the 44-year record from 1979 to 2022 than in the 23-year ).The increasing trends in the residence time in drought classifications (Figure 4) coincide with the increasing trends in the residence time in the driest and warmest percentiles of geophysical variables (Figures 5 and 6) in regions such as the western U.S.This observation implies that the USDM implicitly reflects climate nonstationarity, adding complexity to the interpretation of drought classification, and the responses such classifications initiate, through time.
Our results complement trend analysis of the same hydroclimatic variables at the mean annual timescale as shown in Jasinski et al. (2019).For 1980-2015 based on the National Climate Assessment Land Data Assimilation System data, the West has seen statistically decreasing trends in mean annual precipitation, mean annual SM, and mean annual runoff (Jasinski et al., 2019).Statistically increasing trends in mean annual surface temperature have also been observed in the Southern Great Plains, the eastern U.S., and parts of the Southwest (Jasinski et al., 2019).In this study, statistically significant increasing trends in the residence time in the dry percentiles (≤30%) are also observed in the West from 1979 to 2022, including 3-month precipitation, 3-month runoff, and SM (Figure 6).Statistically increasing trends in the residence time in the warm percentiles of temperature (≥70%) are also observed in the Southern Great Plains and parts of the Southwest and West (Figure 6).The less widespread trends in the 0%-2% category, relative to the 21%-30% category may come about because hydroclimate variables are bounded by zero (Figure 6); one cannot have less than zero water, for example.The point is also noted by Jasinski et al. (2019).

Discussion
Analyzing the USDM through time is scientifically challenging because there are elements of the USDM that are physically based on anomalies of runoff and SM, but there are also elements of the USDM that are subjective in the sense that expert judgment, on-the-ground social reporting, changes in data sets and nominative assessments shape classifications.The practice of drought classification benefits from expert interpretation in various ways, including implicitly accommodating climate trends as mentioned earlier.However, expert judgment in a product also introduces a level of subjectivity (Lorenz et al., 2017) that makes it challenging to interpret classifications over time.This is an issue because the product is designed to inform resource triaging and management of drought by federal agencies, like the USDA.The key indices used by the USDM, for example, have evolved from relying on Palmer Drought Severity Index, SM percentile, U.S. Geological Survey (USGS) weekly streamflow percentile, percent of normal precipitation, Standard Precipitation Index, and objective drought-indicator blends (Svoboda et al., 2002;Xia et al., 2014a), to numerous new inputs, with the growing availability and development of new data products.The USDM also uses different data sets to represent the same physical quantity.As such, it makes it difficult to assess whether and how to consider drought assessments (and the urgency and resources attached to such assessments) evolve as climate changes.Drought classifications can be confounded by a number of things, including changes in data input, classifications tied to expert judgment, and the underlying physical climate system.While there is no standardization in drought classifications beyond the associated percentiles and we can not rule out the possibility of potential confounders, it is still worth asking whether or not there are emergent trends in the drought residence time as we have done in our analysis.
That said, the USDM is an important product to analyze given its authority and shaping drought assessments at the federal level, and how drought should be assessed in a trending climate.As such, several uncertainties need to be taken into account when interpreting the results of this study.First, it should be noted that the calculation of the standardized z-score and percentiles of most geophysical variables is based on a 44-year period from 1979 to 2022.This base period may be different from the periods spanned by the products assimilated in the USDM, which vary from climate region to climate region, based on state climatologist practices and the like.The USDM thus uses a combination of many drought indicators that are based on their respective periods of record, ranging from a few decades to a century-long (H.Wang et al., 2021).These inconsistencies may lead to discrepancies between the geophysical conditions shown in this study and the real-world ground truth when drought occurs.
Second, the USDM has incorporated new indicators and more impact reports as it develops (Leeper et al., 2022;Peters-Lidard et al., 2021).While the general methodology and the drought class generation process have been consistent over time, there is the potential for significant variation in drought classifications.The USDM maps made early in the 2000s were based on climate division-scale data and relied on a limited number of indices and indicators due to data availability constraints.The creation and assimilation of drought impacts reports into the USDM is becoming more comprehensive across a wide range of potential sectors and scales over time, as the NDMC is developing linkages with various stakeholders (Wilhite et al., 2007).The enhancements in the accuracy and diversity of input data have naturally led to inherent differences in how drought is represented compared to 23 years ago.It is thus challenging to isolate only the impacts of nonstationary climate on a particular drought class.
Third, the USDM converges evidence from a variety of sources, including impact reports and local expert assessments.Our study only revealed the conditions of physically relevant measures during drought.Yet, the regional systems of drought management, impacts reporting hierarchies, and USDM classification practices are functions of choices made by people.As such, they have undoubtedly evolved not just in response to the drought and aridity trends we show here, but also in response to vast changes in social, political, economic, and technological drivers over the last 23 years.Consider, for example, how agricultural production in California has shifted to increasingly rely on groundwater as a means to buffer drought impacts (Legislative Analyst's Office, 2016), while this increased groundwater extraction has led to a substantial decline in groundwater levels (Ojha et al., 2018).Because groundwater is increasingly a factor weighed in the USDM (Yatheendradas et al., 2023), such unsustainable human practices may influence current drought classifications in the region.Indeed, a global study has shown that increased human water consumption has exacerbated hydrological drought in various regions across the U.S. (Wada et al., 2013).In contrast, human activities such as irrigation and reservoir regulation have been shown to alleviate drought (Tang & McColl, 2023;Wan et al., 2017).Together, this example highlights that it is not clear how the balance of the two effects-unsustainable groundwater use on the one hand and mitigated yield impacts from groundwater use on the other-is assimilated into a drought assessment.Future studies should consider linking the USDM to human-related information, such as impact reports and the complex feedbacks between people and the physical factors of drought to provide a more comprehensive assessment of the characterization of USDM drought categories and their evolution through time.
While we do not find ubiquitous and statistically significant trends in drought residence times, our results indicate that nonstationarity in the climate system is reflected in U.S. operational drought monitoring in some regions.These findings are consistent with existing work that has identified changes in meteorological drought conditions in many of the same regions where we find trends in residence time, such as parts of Nevada, Utah, Arizona, and New Mexico (Apurv & Cai, 2019).Such nonstationarity has its origins in natural interannual (Baek et al., 2021;Seager et al., 2022) and lower-frequency variability (Goodrich, 2007;Kam et al., 2014;Seager et al., 2022Seager et al., , 2023)), as well as anthropogenic climate change (Diffenbaugh et al., 2015;Williams et al., 2015Williams et al., , 2020)).These influences are manifest in static measures of drought like percentile mapping of drought classes via changes in drought characteristics such as return period (Kam et al., 2014), frequency (McCabe et al., 2004), and duration (Vicente-Serrano et al., 2021).Challenges to drought monitoring practices brought by nonstationary climate are also noted in recent studies in different places around the world, including the U.S. (Hoylman et al., 2022), Europe (Cammalleri et al., 2022), China (He et al., 2021;Shao et al., 2022), and India (Das et al., 2020).
The fact that the development and implementation of the USDM coincide with an historic 23-year drought period in the western U.S. complicates the interpretation of the product.This period, and the low-frequency hydroclimate variability to which it is connected (which appears to be a combination of natural and anthropogenic forcing, e.g., Mankin et al., 2021;Seager et al., 2022Seager et al., , 2023;;Williams et al., 2020), pose potential risks to the reliability of the USDM: drought classification is fixed, indicated by a given drought class, but the climate is not, particularly as anthropogenic activities intensify drought risks.This dynamic can result in varying representations of the residence time spent in a specific drought class.
Depending on how much the mean climate states change, nonstationarity can make drought occurrence more or less likely, and can complicate the comparison of drought classes through time.A D4 drought in the year 2000, for example, may actually represent very different hydrologic conditions than a D4 drought in 2023.Are D4 droughts today somehow "worse" or more impactful-or in some cases, less impactful due to adaptation-than they were just a few decades ago?To assess this possibility, we conduct a trend test on the magnitude of geophysical conditions associated with drought classes (Figure S11 in Supporting Information S1).Less than 10% of the grids show statistically significant trends in the magnitude of geophysical variables associated with drought classes greater than D3 (Figure S11 in Supporting Information S1).This could be due to the small sample size of the ≥D3 drought classes over the 23-year period or the fact that the percentiles that correspond to such severe droughts are very small in terms of their absolute magnitudes-one cannot have less than zero water, for instance.For drought classes from D0 to D2, most of the statistically significant trends are decreasing for TWS and increasing for VPD and T. This suggests that drought classes ≤D2 tend to be associated with drier TWS and warmer conditions under the current period compared to earlier periods.On the other hand, for P, Q, and SM, most of the significant trends are increasing.This indicates that drought classes ≤D2 tend to be associated with wetter P, higher Q, and increased SM.A possible explanation is that the USDM drought classification process considers temporal autocorrelation.As such, even when P, Q, and SM recover, there may be lingering hydrologic impacts in agriculture, reservoirs, and other managed systems that cause the USDM to consistently classify the drought intensity as ≤D2.Additionally, our work identified the spatial heterogeneity in the climatology of geophysical anomalies associated with a given drought class (Figure 2).Future work should investigate the spatial nonstationarity of the representation of USDM drought classes across various climate regions to enhance effective localized drought monitoring, a direction emphasized by previous studies (Leasor et al., 2020;Zhang et al., 2023).
Regardless of the underlying explanations of nonstationarity, which is causing changes in drought occurrence and complicating comparison of drought classes over time, an open question is whether and how it should inform drought assessment, management, and mitigation strategies.Having regions in exceptional drought more often than by design presents a management challenge, even though this might be a true description of geophysical reality as assessed from a longer-term perspective.For instance, although the original intent of the USDM was not specifically to help triage emergency response efforts, it has been used to serve that function.As stated on the USDM website, the USDA, for example, uses the USDM to trigger disaster declarations and eligibility for lowinterest loans.Various other organizations, such the Farm Service Agency, the Internal Revenue Service, and decision makers at different levels (state, local, tribal, and basin), also make use of the USDM.Considering the extensive use of the USDM, if it is an emergency everywhere and all the time, it is not clear where resources should be invested to the greatest effect.
The above considerations and emerging trends that we have identified suggest that the USDM is at a crossroads and raises multiple questions that drought assessors are undoubtedly already wrestling with.If the USDM is intended to accurately reflect drought conditions, such as whether there are drought conditions as anomalies from a consistent hydroclimatic baseline, then the product could modify the periods of record used, ensuring that D4 droughts occur based on a drying or wetting 0-2nd percentile.Such a method has been used in updating the USDA Plant Hardiness Zone Map to accommodate changes in climate (Daly et al., 2012;Krakauer, 2012).This implicitly implies that stakeholders and drought-impacted communities are already adapting to the long-term aridifying climate even as a D4 drought becomes something more severe and persistent than it would have been in a historical climate.Conversely, if the USDM aims to reflect longer-term aridification, introducing an additional drought class may seem a reasonable action.How that changes the aid allocations, urgency, and management responses that were historically attached to the other drought classes is an important question.
Simply tacking a sixth drought class onto the USDM will undoubtedly alter the evaluation and meaning of the other drought classes as well.Either way, it is important to distinguish how assessment and monitoring of shortterm drought risks should proceed in a climate that is aridifying.This is a central priority of NIDIS, which is establishing the Drought Early Warning Systems throughout the country, which recognizes that "[d]rought assessment in a changing climate will require significant adjustments in approaches to address non-stationarity" (Parker et al., 2023, p. 9).From a management perspective, institutional drought mitigation practices are designed to be short-term interventions, like emergency water cuts or relief payments to farmers such as from the Livestock Forage Disaster Program, rather than taking on a structural reallocation of resources for long-term adaptation to an aridifying climate.These are important issues needing resolution for present and future drought assessment and emergency management and how that dovetails, or does not, with longer term adaptation investments.Development of a reformed USDM should better balance the tensions between being reflective of short-term, on-theground conditions and accommodating long-term changes in aridity, whether caused by low-frequency variability or anthropogenic climate change.Striking this balance will help decision makers consider how long-and shortterm factors should shape resource allocations for sustained, large adaptations versus those that are more temporary, like emergency aid provision.This requires joint efforts and engagement from USDM authors and the wider group of stakeholders across the drought research and practitioner communities to consider the goals of drought monitoring and the needs of decision makers.

Conclusions
As a climatological record for the USDM emerges, it is essential to assess whether and how the drought classifications in the product is impacted by nonstationarity in the climate system, whether through natural internal variability or due to anthropogenic climate change.Our analysis centers on whether the residence time for USDM drought classes has trended over the 23-year period of record.Similarly, because drought classes are (in part) defined based on percentile thresholds in geophysical quantities, we have analyzed whether there have been trends in the residence time of these percentile thresholds in 3-month precipitation and runoff, 1-m SM, TWS, vapor pressure deficits, and temperature.
We report three conclusions: First, the USDM drought categories, reassuringly, are generally associated with drier-than-normal hydrologic conditions and warmer-than-normal climate conditions: anomalies in geophysical variables increase with drought intensity, implying that the USDM is well-reflecting observed hydrologic conditions and not being unduly influenced by subjectivity.Second, "exceptional" drought (D4) in many US regions does not occur at a likelihood that is exceptional (<2%), meaning its residence time from 2000 to 2022 is inconsistent with USDM guidelines.Higher rates of D3 and D4 droughts, the most severe drought classes, are associated with statistically significant increasing trends in drought residence time in a number of states in the western U.S. The total percentage of occurrence of D4 droughts in those places over the USDM climatology is considerably greater than 2%.Third, there are statistically significant increasing trends in the occurrence of geophysical variables falling within the percentile thresholds corresponding to the most severe drought classes, primarily in the western U.S.These regions spatially overlap with regions where USDM D4 residence times are increasing.Our results suggest that the USDM is recording climate variability and change, particularly the aridification of the Southwest.The indication is that USDM authors and users should carefully consider the implications for drought assessment, monitoring, and management as the impacts of global warming continue to manifest.
The increasing trends in drought classes have important implications for emergency management and resource allocation.In terms of emergency management, the rising occurrence of drought categories due to the nonstationary climate could significantly strain the capacity of existing infrastructure.For example, the declining water levels of Lake Powell in 2022 disrupted hydropower production and compelled the Bureau of Reclamation to discharge approximately 500,000 acre-feet from Flaming Gorge Reservoir in Utah and Wyoming to Lake Powell.However, such actions may not serve as a sustainable solution if the climate is trending toward a drier mean state.Indeed, federal officials are considering major structural modifications of the Glen Canyon dam, given the substantial decline in water levels in Lake Powell and Lake Mead over the past 23 years, during the most severe drought in centuries (Williams et al., 2022).
In addition to existing infrastructure, the allocation of finite and scarce resources for various sectors, such as agriculture, water management, and energy production, could become increasingly challenging as the occurrence of drought categories continues to trend upwards.While the USDM is one of the important tools used by decision makers, trends in drought class occurrence may pose a challenge to the decision makers seeking to mitigate adverse drought impacts.Our results show that D4 classes are happening at a residence time of up to 18% in some places in the western U.S., which is nine times more often than they should (2%).If left unaddressed, such upward trends are likely to exacerbate over time.It is important to take these nonstationary considerations into account in an operational drought monitoring system and implement strategies that are adaptive to changing climatic conditions.

Figure 1 .
Figure1.Schematic of how climate nonstationarity would affect a location's normals of water-year (WY) precipitation, and thus drought residence times, as classified by a static product like the United States Drought Monitor (USDM).Per USDM objective indicators, D0 or "abnormally dry" conditions correspond to a 21st to 30th percentile; D1 or "moderate" conditions correspond to 11-20th percentile; D2 or "severe" conditions correspond to a 6-10th percentile; D3 or "extreme" conditions are between 3rd and 5th percentiles; and D4 or "exceptional" is a 0-2nd percentile event.For a particular location, the cumulative probability of precipitation in mm may vary depending on the period of time considered.Compare, for example, the first normal cumulative density function (CDF, solid line) to a drier second normal CDF (dotted line).In the first normal, a D4 drought corresponds to a precipitation value of "X," with a frequency maximum of 0.02, or 2% of the time.However, under an aridifying climate, the first normal shifts to the second normal, where that value of precipitation "X" occurs over 15% of the time (bar plots), changing the interpretation of drought intensity.
: 2000-2022 (n = 23) and 1979-2022 (n = 44).The 2000-2022 period is to maintain consistency with the availability of USDM data, while the 1979-2022 represents the longest available record of geophysical variables.However, TWS is only available since April 2002.We thus calculate the trend for TWS over 2003-2022 (n = 20) to ensure a consistent sample size in each year.

Figure 2 .
Figure 2. Composite values of geophysical variables in standardized z-score (P, Q, soil moisture [SM], vapor pressure deficit [VPD]), anomalies (T), or percentile (terrestrial water storage [TWS]) associated with weeks in each drought class as defined by the United States Drought Monitor.Green or blue colors represent wetter or cooler than normal conditions, while brown or red colors represent drier or warmer conditions.The white colors in the CONUS represent "no data," meaning that those areas do not experience a particular drought class from 2000 to 2022.

Figure 3 .
Figure 3. Spatial maps of the residence time of individual drought classes (left) and cumulative drought classes (right) in the number of days and percentage over 2000-2022.The maps have a resolution of 0.125°.Stippled regions in the left column represent areas where the percentage of residence time is no greater than the incremental United States Drought Monitor (USDM) percentiles, that is, 2% for D4, 3% for D3, 5% for D2, and 10% for D1 and D0.Stippled regions in the right column represent areas where the percentage of residence time is no greater than the upper bound of USDM percentiles, that is, 2% for D4, 5% for D3-D4, 10% for D2-D4, 20% for D1-D4, and 30% for D0-D4.The white colors in the CONUS represent "no data," meaning that those areas do not experience a particular drought class from 2000 to 2022.
S. have shown D0-D4 occurrence exceeding 50% between 2000 and 2022.The occurrence of D0-D4 is particularly high in several Southwestern states, including Arizona, New Mexico, and Nevada.Some places in Arizona experienced up to 89.5% of time spent in D0-D4 from 2000 to 2022, which is approximately three times higher than the expected 30%.The patterns are similar to results from H.Wang et al. (2021) based on the period from 2000 to 2018 andLeeper et al. (2022) from 2000 to 2019.Leeper et al. (

Figure 4 .
Figure 4. Trends in individual and cumulative drought class residence times in number of days per decade from 2000 to 2022.The red colors represent increasing trends, while the blue colors represent decreasing trends.The (a, c) columns show the trends in all the regions in the U.S. The (b, d) columns show the regions where the trends are statistically significant based on autocorrelation corrections ( p ≤ 0.05), with gray colors representing insignificant trends ( p > 0.05).The p-values represent the statistical significance of the trend in its original unit of weeks per year.The white colors in the CONUS represent "no data," meaning that those areas do not experience a particular drought class from 2000 to 2022.

Figure 5 .
Figure 5. Trends of the residence time of geophysical variables in days per decade from 2000 to 2022 except terrestrial water storage [TWS], which is from 2003 to 2022.The values of geophysical variables fall within the percentiles that United States Drought Monitor uses for its D0-D4 classifications.Blue or red colors represent statistically significant trends ( p ≤ 0.05), while gray colors represent insignificant trends ( p > 0.05).The p-values represent the statistical significance of the trend in its original unit of days per year.

Figure 6 .
Figure 6.As in Figure 5 but for the period from 1979 to 2022 except terrestrial water storage [TWS], which is from 2003 to 2022.

Table 1 A
Summary of Data Sets Used in This Study