1. Introduction
Drought is the world’s most economically disruptive, widespread, and long-lasting natural hazard (Wilhite 2000). Drought can cause adverse impacts on agriculture, ecosystems, and socioeconomic development (Hao et al. 2014). For example, according to the National Oceanic and Atmospheric Administration, the 2012 drought resulted in $34.2 billion in economic losses, 123 direct deaths, and a 26% decrease in corn yield in the United States (Rippey 2015). Ohio is greatly affected by agricultural drought. Corn yields in Ohio decreased from 160 bushels per acre (bu ac−1) in 2011 to about 120 bu ac−1 in 2012 because of the severe drought (Mallya et al. 2013). Therefore, it is important to monitor drought conditions to provide early warning and decision support to aid drought management and response.
Although there are hundreds of drought indices (Zargar et al. 2011; Svoboda and Fuchs 2016), this study focuses on four representative indices: standardized precipitation index (SPI; Guttman 1999; McKee et al. 1993), standardized precipitation evapotranspiration index (SPEI; Vicente-Serrano et al. 2010), Z index (Palmer 1965), and Palmer hydrological drought index (PHDI; Palmer 1965). These indices were chosen for three reasons. First, the four drought indices have been commonly used in previous drought studies. For example, the Z index and SPEI were found to be highly correlated with crop yields in the south-central United States (Tian et al. 2018). In the Canadian Prairies, the Z index was found to outperform the Palmer drought severity index (PDSI) and SPI in monitoring agricultural drought and predicting crop yield (Quiring and Papakryiakou 2003). Lorenzo-Lacruz et al. (2010) found that the SPEI is more appropriate for detecting the response of river discharge and reservoir storage to drought than the SPI in central Spain, because the SPEI considers the impacts from temperature. Similarly, Vicente-Serrano et al. (2012) showed that the SPEI had improved capability in identifying hydrological, agricultural, and ecological drought impacts as compared with the SPI. Second, the four indices are widely used in operational drought monitoring. They have been adopted by the “State of Ohio Emergency Operations Plan” (https://ema.ohio.gov/Documents/Ohio_EOP/EOP_Overview/IA_Drought_Incident_Response_Annex.pdf) for monitoring drought conditions and triggering drought response activities. They have been shown to be effective and complementary. Third, indices such as the SPI and SPEI can be calculated at various time scales to represent different types of drought, such as agricultural (e.g., shorter-term drought conditions) and hydrological drought (e.g., longer-term drought conditions) (Vicente-Serrano et al. 2012, 2010; Tian et al. 2018). Although Palmer drought indices (Z index and PHDI) do not provide the same temporal flexibility as the SPI, the Z index has been shown to be representative of agricultural and meteorological drought, and the PHDI has been shown to be representative of hydrological drought (Heim 2002). Therefore, the four indices that are used in this paper will allow us to assess agricultural and hydrological drought.
Drought intensity is the average value of a drought parameter below a defined level (Mishra and Singh 2010). Drought intensity thresholds are commonly used to monitor conditions and to trigger mitigation activities in various sectors (Quiring 2009; Beyene et al. 2014), such as agriculture, water utilities, energy, wildfire management, ecosystems, and recreation and tourism. Currently, many drought monitoring products in the United States at the national scale use fixed drought thresholds for determining drought intensity levels (e.g., evaporative demand drought index, SPI, and PDSI), while others, such as the U.S. Drought Monitor (USDM), use them as a theoretical guideline. These thresholds are generally determined using values from the tails of the distribution of observations during the reference period (Link et al. 2020; Leasor et al. 2020). Since these thresholds are fixed in time and space (the same thresholds for the SPI are used in all regions and seasons), they may not accurately reflect drought impacts in all regions, seasons, and sectors (Leasor et al. 2020). For example, Quiring (2009) found that SPI tends to have higher (i.e., less extreme) drought thresholds in drier regions than wetter regions. Leasor et al. (2020) confirmed that the fixed SPI thresholds for the USDM drought categories caused a systematic underestimation of drought severity in arid regions of the south-central United States. Therefore, it is important to develop regionalized impacts-based drought thresholds that are appropriate for the region and sector (e.g., agriculture or water resources) of interest.
The USDM is an important protocol developed by the National Drought Mitigation Center (NDMC), U.S. Department of Agriculture (USDA), and National Oceanic and Atmospheric (NOAA). The USDM uses a convergence-of-evidence approach to characterize drought intensity and development based on multiple types and sources of data (Svoboda et al. 2002; Leasor et al. 2020; Xia et al. 2014). The USDM approach implements nationally derived thresholds and targets county-scale drought classification through weekly, iterative stakeholder feedback processes. These thresholds are referred to as fixed because they do not vary over space or time. However, drought impacts can vary by region (Goodrich and Ellis 2006). Mishra and Singh (2010) also emphasized the need to choose appropriate drought thresholds for different hydroclimate regions. For example, USDM historical data show that Ohio has had no experience with exceptional drought (D4) conditions since the map’s inception in 2000, but Ohio did have great crop losses due to severe drought in 2012 (Mallya et al. 2013). Therefore, it would be helpful to develop impacts-based thresholds to guide local experts and further improve the USDM representation of drought conditions in Ohio.
Little guidance exists in the literature as to how states, communities, and others should develop local drought thresholds. Currently, efforts have been made in the following three categories: 1) using drought index thresholds commonly used in drought monitoring products or other plans (Steinemann and Cavalcanti 2006; Mizzell 2008; Steinemann 2014), 2) estimating drought index thresholds from regional probability distributions (Leasor et al. 2020; Steinemann et al. 2015; Steinemann 2003), and 3) estimating drought index thresholds based on drought impacts information to identify thresholds at which certain impacts are likely to occur (Bachmair et al. 2015). Our work contributes to the third category and develops impacts-based drought thresholds that are appropriate for monitoring and assessing drought intensity in Ohio.
Three objectives are accomplished in this study: 1) calculate four drought indices, including SPI and SPEI at 1-, 3-, 6-, and 9-month time scales, Z index, and PHDI; 2) identify the optimal time scale of drought indices based on the correlation between drought indices and agricultural (corn yield) and hydrological (streamflow) drought impacts in Ohio; and 3) develop impacts-based thresholds for each index in Ohio.
2. Data
Table 1 shows a summary of the datasets used in this study. The period of record for all datasets is from 1990 to 2019 (30 years). The datasets used to calculate the four drought indices (SPI, SPEI, Z index, and PHDI) include precipitation, potential evapotranspiration (PET), and available water capacity (AWC). Daily precipitation data are obtained from the PRISM (Parameter-Elevation Regressions on Independent Slopes Model) AN81d datasets (http://www.prism.oregonstate.edu/) with 4-km spatial resolution. Daily PET data are obtained from the gridMET dataset (http://www.climatologylab.org/gridmet.html) with a 4-km spatial resolution, which is estimated based on the Penman–Monteith equation. The input to the equation is a combination of meteorological forcings from the North American Land Data Assimilation System-2 (Xia et al. 2012) and PRISM (McEvoy et al. 2020). AWC is the amount of water that the soil can store for plant use, and it depends on the characteristics of soil for a given location. It is obtained from the Gridded National Soil Survey Geographic Database (gNATSGO; https://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/survey/geo/?cid=nrcseprd1464625) with 30-m spatial resolution. The database was produced in 2019 by the Natural Resources Conservation Service (NRCS). The information on soil properties was collected by field surveys and subsequent laboratory analysis. The USDM archive data are used to represent drought area percentage by category (https://droughtmonitor.unl.edu/DmData/DataDownload/ComprehensiveStatistics.aspx).
A summary of datasets used in this study.
Streamflow data are used as a measure of hydrological drought impacts, which is consistent with previously published studies (Vicente-Serrano et al. 2012; Dai 2011; Myronidis et al. 2018). In addition, streamflow data have a long period of record, high temporal resolution, and are measured at many locations. This allows us to examine a relatively large percentage of the state using that available network of gauges. It should be noted that streamflow represents an integrated measure across the entire watershed and does not represent within watershed variations in drought impacts. Streamflow data are provided by the United States Geological Survey (USGS) at waterdata.usgs.gov/nwis. A total of 318 stations provides streamflow data in Ohio. After removing the stations that have missing data for more than 365 days (from 1990 to 2019), a total of 65 stations are selected for this analysis. All the watersheds use the 8-digit USGS Hydrologic Unit Code (HUC) and are at the subbasin level. Their areas range from 33 to 1.64 × 104 km2 with a median value of 909 km2. The 65 watersheds include 12 natural watersheds and 53 watersheds with human activities such as artificial diversions and dam storage according to the classification in the GAGES-II dataset (Falcone et al. 2010). Therefore, streamflow in these watersheds may be influenced by human activities. However, we do not confine our results to the natural watersheds because there are relatively few of them in Ohio (n = 12). In addition, the watersheds that are influenced by human activities still have a strong Pearson’s correlation with drought conditions (median r = 0.80 for SPI9, 0.78 for SPEI9, and 0.79 for PHDI; results not shown). This indicates that hydroclimatic variability has a strong influence on streamflow in these watersheds. Finally, we are interested in quantifying drought impacts in Ohio in both natural and human-modified environments since the state’s drought mitigation plan is designed to mitigate impacts on people and their environment. While spatial autocorrelation may exist at nearby contributing stream gauges, it may not have a substantial influence on our analysis since the 65 watersheds cover the majority of the state. To make streamflow comparable across watersheds, the original unit (ft3 s−1) is converted into millimeters (mm) by dividing by the area of the watershed and multiplying by the time scale that flow is measured. Streamflow thereafter in the paper is expressed in the unit of millimeters. Watershed areas are obtained from the GAGES-II dataset (Falcone et al. 2010).
Corn yield data are used to evaluate agricultural drought impacts. Similar to what was noted above, corn yield represents county-level impacts but does not represent within county variations in drought impacts. The county-level corn yield (bu ac−1) data of Ohio are obtained from the National Agricultural Statistics Service (NASS) Quick Stats Database (https://quickstats.nass.usda.gov/), which dates back to 1918. In this study, corn is selected as the focus crop because about 31% of Ohio farmland is planted with corn, according to the 2017 Census of Agriculture from USDA NASS (www.nass.usda.gov/AgCensus). The annual corn yield data from 1990 to 2019 are extracted for 86 counties in Ohio that have records over the 30 years. A simple quality assurance/quality control is applied to the yield data by replacing the missing data with NaN (not a number). This is used to filter out the counties without yield records during the last 30 years and to ensure that missing data are ignored in the correlation analysis. Figure 1 shows a map of Ohio watershed boundaries, counties, and streamflow gauging stations.
A map of Ohio watershed boundaries, counties, and 65 USGS streamflow gauging stations.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
3. Methods
Figure 2 shows the flowchart of all the methods used in this study. Each method is then described in detail below.
Flowchart illustrating the data and methods that are used in this study.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
a. Calculation of drought indices
The calculation of weekly drought indices requires fixed time periods; thus, it was not possible to use calendar week as the calculation time step. Instead, an artificial week scale was adopted. Here, we divided each month into four arbitrary weeks. Specifically, week 1 is from day 1 to day 7 (7 days), week 2 is from day 8 to day 15 (8 days), week 3 is from day 16 to day 22 (7 days), and week 4 is from day 23 to the end of this month (varied from 6 to 9 days). A moving window method, which is the method adopted by USDM (Lorenz et al. 2017), was used here to calculate drought indices each week. The drought indices in the current week were calculated using the climate data from the current week and the previous three weeks. For example, the drought index in week 1 in month X is calculated using the data covering the last three weeks in month X − 1 to week 1 in month X. In this case, the drought indices are calculated on a 4-weekly time scale, but they are updated each week. Using 4-week data ensures the robustness of the calculation of drought indices. This is because frequent zero precipitation may negatively impact fitting the probability distribution for the SPI, and this can distort SPI values (Wu et al. 2007; Stagge et al. 2015). When using the weekly data, there is a higher chance of receiving zero precipitation, as compared to using monthly data. The moving window method leads to a time series of 1440 samples (30 years × 12 months × 4 weeks) for each drought index from 1990 to 2019.
The four drought indices evaluated in this study are SPI, SPEI, Z index, and PHDI. The SPI is a commonly used index for quantifying drought conditions, and the World Meteorological Organization (WMO) recommends using the SPI to enhance drought monitoring and early warning capabilities (World Meteorological Organization 2012). Precipitation data are the only input for SPI, and they are calculated by first fitting a two-parameter gamma distribution to the precipitation data. Then the maximum likelihood method is used to estimate the parameters of the gamma distribution. Finally, the cumulative probability of the gamma distribution is transformed using an equiprobability transformation into a standard normal deviate, which is the value of the SPI (McKee et al. 1993).
The SPEI incorporates both potential evapotranspiration and precipitation into a multiscalar drought index (Vicente-Serrano et al. 2010). The calculation of SPEI is similar to that of SPI, except that instead of using precipitation, it uses the difference (D) between precipitation and PET. A three-parameter log-logistic distribution is fitted to D, and the L-moment procedure is used to estimate the distribution parameter. The SPEI values are obtained after standardizing the log-logistic distribution to a normal distribution.
The Z index and PHDI are part of the family of Palmer drought indices, and they are defined by the departure from normal monthly soil moisture conditions. The input data for the Z index and PHDI include precipitation, PET, and AWC of the soil. It takes two steps to calculate the Z index and PHDI. The first step is to estimate the evapotranspiration, soil moisture loss and recharge, and runoff based on a two-layer bucket water balance model (Palmer 1965; Dai 2011). The surface layer has a storage capacity of 1 in., and the underlying layer has a storage capacity of AWC − 1 in. (Jacobi et al. 2013). The soil moisture storage is assumed to be full at the initial step of the calculation. The actual evapotranspiration (ET) equals PET when precipitation is greater than PET; otherwise, ET equals precipitation and soil moisture loss from the surface and underlying soil layers. The soil moisture loss depends on initial soil moisture storage, AWC, and difference between precipitation and PET. Runoff occurs only when the two layers of soil are saturated (Quiring and Papakryiakou 2003). The second step is to estimate the climatically appropriate for existing conditions (CAFEC) precipitation using the output from the water balance equation in the first step. The Z index is the product of the monthly difference d between actual precipitation and CAFEC precipitation and a weighting factor K. The weighting factor is used to adjust the d so that the d values are comparable across locations and months. The PHDI is a derivative index from the Z index. It is calculated from the Z index in the current month and the PHDI value from the previous month. Therefore, the Z index is used to track short-term drought because it is based on the water balance in a single month (Karl 1986), without consideration of conditions in previous months. The PHDI is for longer-term drought monitoring because it accounts for the time lag between the end of a dry period and the recovery of the environment from a drought (Jacobi et al. 2013; Vasiliades and Loukas 2009). The PHDI is commonly used to represent hydrological drought conditions (Heim 2002). Negative values of the Z index and PHDI represent below-normal soil moisture conditions.
The SPI and SPEI were calculated using the R package “SPEI,” while the Z index and PHDI were calculated using the MATLAB code from Jacobi et al. (2013).
b. Relationship between drought indices and drought impacts
In this study, streamflow and corn yield are used to represent drought impacts and to identify appropriate impacts-based thresholds for SPI, SPEI, PHDI, and Z index. Following the approach of Vicente-Serrano et al. (2012), we used the Pearson’s correlation coefficient (r) to characterize the relationship between the drought indices and the hydrological (streamflow) and agricultural (corn yield) drought impacts and to select the time scale when the relationship between impacts and drought indices is strongest (Table 1 in the online supplemental material). The 9-month SPI and SPEI and the PHDI are used for representing hydrological drought because past studies have demonstrated that these are typically the best indices (Lorenzo-Lacruz et al. 2010; Vicente-Serrano et al. 2012; Zhai et al. 2010; Heim 2002). In addition, our sensitivity analysis showed that SPI and SPEI had higher correlations with streamflow at a 9-month time scale than other time scales, and PHDI showed higher correlations than Z index (supplemental Table 1). Dikici (2020) also found that a time scale of ∼9 months is optimal for hydrological drought monitoring. The optimal time scale for monitoring hydrological drought tends to be much longer than meteorological drought because of the associated hydrological processes. For example, large watersheds have a longer concentration time (Van Loon 2015). Hydrological drought also requires a significant decrease in baseflow, which is dependent on groundwater levels and so it responds on seasonal-to-interannual time scales. Streamflow response to snowmelt will also lead to a longer propagation time from meteorological drought to hydrological drought (Huang et al. 2017). There are also seasonal variations in surface water and groundwater withdrawals (e.g., thermoelectric power, public and domestic water supply, and livestock) and surface water losses from evaporation. Therefore, a longer time scale accounts for these hydrological processes related to supply and demand (Lorenzo-Lacruz et al. 2013; Yuan et al. 2017), leading to a higher correlation between SPI9/SPEI9 and streamflow.
Monthly streamflow data are aggregated to the 9-month time scale, which was used as an indicator of hydrological drought impact. For each month, the value represents streamflow from the current month and previous 8 months. The gridded (4 km) drought indices were averaged to the watershed scale to match the spatial resolution of streamflow, because streamflow represents the water volume that passes at a gauging station for a watershed. Correlation coefficients were calculated between drought indices and streamflow.
For agricultural drought, SPI and SPEI at a shorter time scale (3 months) and Z index are considered. This is because our preliminary analysis showed that the SPI and SPEI had a higher correlation with annual corn yield at a 3-month time scale compared to other time scales, and Z index showed higher correlations than PHDI (supplemental Table 1). This agrees with previous findings that agricultural drought was best represented by the SPI at a time scale of 2–3 months (Wilhite 2000; Peña-Gallardo et al. 2019; Park et al. 2016; Lu et al. 2017). Annual corn yield was adopted as an indicator of agricultural drought impacts. Because of the increasing trend in corn yield due to technological development and improved crop management (Agnolucci and De Lipsis 2020; Kukal and Irmak 2018; Osborne and Wheeler 2013), detrended corn yield (Hafner 2003) is used in this study. A line was fit to the yield data to represent the linear trend in each county. The detrended yield was obtained by subtracting the linear trend from the raw yield data. Student’s t tests showed that the trend (slope of the line) was statistically significant (p < 0.05). Gridded drought indices (SPI, SPEI, and Z index) were aggregated to the county level to match the county-based corn yield. USDA corn yield statistics are published annually, while the drought indices are updated weekly. Correlation analysis was used to identify the week with the highest correlation between annual corn yield and the weekly updated drought indices.
c. Identification of impacts-based thresholds
Rather than using fixed drought thresholds, this study develops impacts-based thresholds for the state of Ohio. Streamflow was used as an indicator of hydrological drought impact, while corn yield was adopted as an indicator of agricultural drought impact. We adopt drought categories that are similar to the guidelines used by the USDM, where Dl is moderate drought, D2 is severe drought, D3 is extreme drought, and D4 is exceptional drought. We did not consider D0 because D0 areas are not in drought. D0 indicates abnormally dry conditions that may precede or follow a drought (Svoboda et al. 2002). Each category is identified using drought impacts defined by a percentile. The nth percentile is denoted as n% (n = 2, 5, 10, and 20). For example, for hydrological drought, the D1 drought is associated with 10%–20% of streamflow, and D2, D3, and D4 droughts are associated with 5%–10%, 2%–5%, and <2% of streamflow, respectively. Similarly, the 20%, 10%, 5%, and 2% were applied to the detrended corn yield to obtain the thresholds of impacts-based agricultural drought in Ohio.
The drought indices falling within each category of detrended yield and streamflow were extracted and displayed using boxplots [results are shown in sections 4a(1) and 4b(2)]. The notches of each box, which represent the 95% confidence intervals around the median, were used to identify the drought thresholds in this study. The nonoverlapping notches indicate that the thresholds are statistically significantly different from one drought level to another.
For application purposes, the drought area percentages are compared among USDM archive, fixed thresholds, and impacts-based thresholds. Because the streamflow is updated monthly, similar to Hao and Aghakouchak (2014), the area percentage of the USDM archive data in the week that is closest to the end of the month is extracted. The monthly drought area percentages are then averaged to an annual time scale for comparison between the USDM archive and hydrological indices. For agricultural drought, the week with the highest correlation between detrended corn yield and drought indices is extracted for each year. The area percentages based on the USDM archive and agricultural indices are compared in that week.
Finally, area percentages of D4 drought in Ohio are determined using fixed and impacts-based thresholds. The nonparametric Spearman correlation is used to quantify the relationship between drought impacts (streamflow and yield) and area percentages of D4 drought.
4. Results
a. Assessing hydrological drought using streamflow
1) Impacts-based thresholds for streamflow
The scatterplot in Fig. 3 shows the distribution of drought indices within each category of streamflow. Overall, lower values of drought indices correspond to a lower value in streamflow. The correlation is statistically significant at the 0.05 level, with coefficients of 0.46 for the 9-month SPI and 9-month SPEI, and 0.45 for PHDI.
Scatterplot of drought indices (9-month standardized precipitation index, 9-month standardized precipitation evapotranspiration index, and Palmer hydrological drought index) within each category of streamflow (SF) in millimeters. Color codes are the same as the USDM drought categories from D1 to D4. Each point represents streamflow in one watershed in a given month from 1999 to 2019 that is below the 20th percentile of all samples (n = 4561).
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
Based on the relationship between drought indices and streamflow, boxplots are used to determine the impacts-based hydrological drought thresholds. Drought indices falling in each category of streamflow (SF) are used to define a new set of hydrological drought thresholds (Table 2; gray regions in Fig. 4). The thresholds for 9-month SPI and 9-month SPEI are very similar, with differences of up to 0.1. The impacts-based drought thresholds for the 9-month SPI are from −0.6 to −0.7 for D1: moderate drought (SF 10%–20%), from −0.7 to −1.0 for D2: severe drought (SF 5%–10%), from −1.0 to −1.4 for D3: extreme drought (SF 2%–5%), and <−1.4 for D4: exceptional drought (SF < 2%). The impacts-based drought thresholds for the 9-month SPEI are from −0.6 to −0.8 for D1: moderate drought (SF 10%–20%), from −0.8 to −1.1 for D2: severe drought (SF 5%–10%), from −1.1 to −1.4 for D3: extreme drought (SF 2%–5%), and <−1.4 for D4: exceptional drought (SF < 2%). The impacts-based drought thresholds for the PHDI are from −1.9 to −2.1 for D1: moderate drought (SF 10%–20%), from −2.1 to −2.7 for D2: severe drought (SF 5%–10%), from −2.7 to −3.6 for D3: extreme drought (SF 2%–5%), and <−3.6 for D4: exceptional drought (SF < 2%). All the impacts-based hydrological drought thresholds are less severe (i.e., less negative) than the fixed thresholds. This indicates that the fixed thresholds may underestimate the impacts of drought on streamflow in Ohio. For example, the impacts-based D4 drought threshold for PHDI is −3.6, which falls within the D2 drought category when a fixed threshold is used. Therefore, when the PHDI is −3.6, the fixed drought thresholds indicate that drought is severe (D2); however, based on the state historical streamflow conditions, the associated drought impacts on streamflow are exceptional (D4). The differences between the impacts-based and fixed thresholds generally increase from D1 to D4 drought. Kumar et al. (2009) also found that SPI-based drought classification based on fixed thresholds sourced from McKee et al. (1993) tend to underestimate drought intensity in India. They compared SPI with rainfall and rainfall anomalies from 1969 to 2007 and found that SPI fell within the moderate dryness fixed threshold range during extreme drought years. Leasor et al. (2020) also found that there were substantial differences between the fixed SPI drought thresholds and those that were locally defined in the southern United States. These differences were a function of the aridity of the climate. Our findings agree with Leasor et al. (2020), who found that drought intensity identified by objective drought thresholds, which are defined based on the climatic conditions at each location, tended to be more severe than using the fixed drought thresholds. Although, Leasor et al. (2020) noted that the differences tend to be most pronounced when using shorter SPI time scales (1 month).
Fixed nationally derived and Ohio-derived impacts-based drought thresholds. The fixed drought thresholds are sourced from USDM thresholds (Svoboda et al. 2002). The USDM thresholds for SPI are used for fixed SPI and SPEI and the USDM thresholds for PDSI are used for PHDI and Z index.
Comparison of the impacts-based hydrological drought thresholds determined using streamflow percentiles (SF; gray regions) and the fixed drought thresholds (colored regions) for 9-month SPI (SPI9), 9-month SPEI (SPEI9), and Z index over 30 years (1999–2019): SF2% = <2nd percentile of the all streamflow samples (65 watersheds × 30 years ×12 months = 23 400 samples), SF5% = 2nd–5th percentile, SF10% = 5th–10th percentile, SF20% = 10th–20th percentile. Each box represents values of drought indices falling within a given category of streamflow.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
2) Application of impacts-based thresholds for hydrological drought
Figure 5 shows the percent area of each drought category from 2000 to 2019 at an annual time scale in Ohio using the USDM archive data, fixed thresholds, and the impacts-based thresholds based on streamflow. Based on USDM-archive data, there has not been an exceptional drought (D4) in Ohio during the last 20 years. It should be noted that the USDM uses a convergence-of-evidence approach to classify drought based on numerous inputs (Svoboda et al. 2002; Svoboda 2016). The USDM is based on measurements of meteorological, hydrological, and soil conditions as well as reported impacts. Therefore, differences are to be expected between the USDM archive and other products when representing hydrological drought impacts in Ohio. At the mean annual time scale from 2000 to 2019, when using fixed thresholds, 0.1% of Ohio has experienced D4 drought conditions based on the 9-month SPI, 0.4% based on the 9-month SPEI, and 0% based on PHDI. By comparison, the SF-based thresholds identified D4, with 2.5% for 9-month SPI, 4.7% for 9-month SPEI, and 3.1% for PHDI. Overall, the USDM-based depiction of drought conditions in Ohio indicates that drought intensity is less severe than the SF-based drought thresholds. Impacts-based and fixed drought thresholds have similar patterns, although the impacts-based drought thresholds indicate more severe drought conditions. This systematic difference in intensity is important since it has implications for triggering drought mitigation measures. The impacts-based thresholds can be used as an additional input to the USDM and assist drought assistance programs, such as the Livestock Forage Disaster Program. Our analysis demonstrates that the impacts-based thresholds may be better suited for informing drought response in Ohio. Further discussion is provided in section 5.
Area of Ohio (percentage) in D1–D4 drought categories from 2000 to 2019 at an annual time scale. The drought information is obtained from three sources: (left) USDM archive, (center) drought indices (9-month SPI, 9-month SPEI, and PHDI) using the fixed thresholds, and (right) drought indices (9-month SPI, 9-month SPEI, and PHDI) using the SF-based thresholds. Data prior to 2000 are not shown because the USDM archive is not available.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
Figure 6 shows the time series of 65-station mean streamflow and the area of Ohio in D4 drought at a monthly time scale. The correlations between streamflow and the D1–D4 drought area were also analyzed, and the patterns are similar to D4 (supplemental Table 2). Because D4 represents exceptional drought conditions, the analysis for D4 drought is shown. It is observed in Fig. 6 that when streamflow experiences large declines, drought indices based on both fixed and impacts-based thresholds correctly identify the occurrence of drought in most cases. However, the area in D4 drought based on the impacts-based thresholds is consistently greater than the fixed thresholds. For example, when the 9-month streamflow decreased by 229.5 mm (75.9%) compared with the mean value of 302.4 mm in February 1992, the percentage of Ohio in D4 drought based on the impacts-based thresholds was 50.9% for SPI, 82.8% for SPEI, and 63.0% for PHDI. However, the percentage of Ohio in D4 drought based on the fixed thresholds was only 35.3% for SPI, 31.8% for SPEI, and 27.4% for PHDI. The correlation between 9-month streamflow and the area in D4 drought using SF-based thresholds is higher for all three drought indices than that based on the fixed thresholds. The differences in the correlation range from 0.13 to 0.19 and are statistically significant (p ≤ 0.05) according to Fisher (1921). This suggests that the impacts-based thresholds accurately reflect hydrological drought conditions in Ohio.
Time series of the 9-month running mean streamflow at 65 USGS gauges that were selected for this study (solid blue line, left y axis) and the percentage of Ohio in D4 drought based on (top) 9-month SPI, (middle) 9-month SPEI, and (bottom) PHDI, applying fixed thresholds (solid orange line, right y axis) and impact-based thresholds (dashed orange line, right y axis) from January 1990 to December 2019. The correlation coefficients between the 9-month streamflow and D4 area percentage calculated at a monthly scale are listed above each graph. Note the left y axis is reversed. The peaks represent decreases in streamflow. The horizontal blue dashed line represents the mean 9-month streamflow.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
b. Assessing agricultural drought using corn yield
1) Correlation analysis
A correlation analysis is conducted between detrended yield and the three drought indices (SPI, SPEI, and Z index) to select the calendar week with the highest correlation. Figure 7 shows that the maximum correlation occurs in week 30 (mid-August) for Z index and in week 33 (early September) for SPI and SPEI at a 3-month time scale. It is important to note that the drought indices in this study are calculated using a moving window (1-month window) method and updated weekly. Therefore, the week identified here represents an accumulated period that ends in that week (e.g., a 3-month period for SPI and SPEI, and a 1-month period for Z index), rather than a 1-week period. In this sense, the correlation between the three indices and corn yield peaks from early June to early September in Ohio, covering the flowering period and grain filling period of corn. This agrees with the Food and Agriculture Organization Land and Water division report (FAO 2015) that corn yield is sensitive to water supply during these two periods.
Plot of 30-yr (1990–2019) Pearson’s correlation (r) between detrended Ohio county-averaged annual corn yield and Palmer Z index, 3-month SPI (SPI3) and 3-month SPEI (SPEI3). Only weeks with statistically significant correlations (p < 0.05) are plotted. At each week, the correlation coefficient was calculated based on 2580 (86 counties × 30 years) samples. The large filled circles indicate the maximum correlation.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
The 3-month SPEI has the highest correlation with corn yield, followed by the 3-month SPI and Z index. By using the procedure developed by Fisher (1921), it is found that the correlation between each pair of the three indices (SPEI, SPI, and the Z index) are significantly different (p ≤ 0.05). The results are consistent with previous studies that corn yield showed a higher correlation with the SPEI than with the SPI (Labudová et al. 2017; Lu et al. 2020). The higher correlation with SPEI may be due to its incorporation of temperature and PET, which makes it more sensitive to crop stress (Beguería et al. 2014; Lorenzo-Lacruz et al. 2010). Based on a literature review of previous studies, a statistically significant correlation ranging between 0.6 and 0.8 was observed between the 3-month SPEI and SPI and standardized corn yields in summer months in the Danubian Lowland in Slovakia (Labudová et al. 2017). Peña-Gallardo et al. (2018) reported that the mean correlations between the SPI, SPEI, and Z index and corn yields are about 0.44 across the United States, and the strongest response for SPI and SPEI is found at 1–3-month time scales. Liu et al. (2018) found the median correlation with corn yield ranges from 0.25 to 0.78 for SPEI and from 0.36 to 0.85 for SPI during the corn growing season (June–September) in North China Plain. The correlation coefficients were reported as the maximum correlation coefficients from 1- to 24-month time scales. Lu et al. (2020) found that corn yield anomalies in the United States had the highest correlation with 2-month SPEI (0.45) and 2-month SPI (0.42) in July. In this study, we found the correlations with corn yield are 0.55, 0.50, and 0.45 for SPEI, SPI, and the Z index, respectively, falling within the range reported by previous studies.
2) Impacts-based thresholds for yield
The detrended yield is used to develop impacts-based thresholds for agricultural drought monitoring in Ohio. Figure 8 shows the drought indices within four categories of detrended yield (DY), including DY20%, DY10%, DY5%, and DY2%. A shift in the median value of drought indices from 0 to the negative values is observed, moving from DY20% to DY2%, especially with SPI3 and SPEI3 (Fig. 8). This indicates that there is a positive relationship between detrended yield and the two drought indices. Some negative yields associated with wet conditions are probably because of nondrought factors such as plant diseases, damage by severe precipitation (e.g., water-logged crops), and when soils are saturated, this can prevent farmers from accessing their fields. Generally, a more severe drought results in larger declines in corn yield.
Scatterplot of drought indices (3-month SPI, 3-month SPEI, and Z index) within each category of detrended yield (DY; bu ac−1). Color codes are the same as the USDM drought categories from D1 to D4. Each point represents corn yield in one of the 86 counties in Ohio for one year from 1999 to 2019 that is below the 20th percentile of all samples (n = 493).
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
The drought indices within each category of detrended yield are used to define the impacts-based drought thresholds for SPI3, SPEI3, and Z index (Table 2; gray regions in Fig. 9). The impacts-based drought thresholds for SPI3 are from −0.6 to −0.9 for D1: moderate drought (DY 10%–20%), from −0.9 to −1.1 for D2: severe drought (DY 5%–10%), from −1.1 to −1.5 for D3: extreme drought (DY 2%–5%), and <−1.5 for D4: exceptional drought (DY < 2%). Using SPEI3, the impacts-based drought thresholds are from −0.7 to −1.0 for D1: moderate drought (DY 10%–20%), from −1.0 to −1.4 for D2: severe drought (DY 5%–10%), from −1.4 to −1.7 for D3: extreme drought (DY 2%–5%) and <−1.7 for D4: exceptional drought (DY < 2%). Using the Z index, the impacts-based thresholds are from −1.2 to −1.7 for D1: moderate drought (DY 10%–20%), from −1.7 to −2.2 for D2: severe drought (DY 5%–10%), from −2.2 to −2.4 for D3: extreme drought (DY 2%–5%) and <−2.4 for D4: exceptional drought (DY < 2%).
Comparison of the impacts-based agricultural drought thresholds determined using detrended yield (DY; gray regions) and the fixed drought thresholds (colored regions) for 3-month SPI (SPI3), 3-month SPEI (SPEI3), and Z index over 30 years (1999–2019). DY2% indicates smaller than 2nd percentile of all detrended yield samples (86 counties × 30 years =2580 samples), DY5% indicates 2nd–5th percentile, DY10% indicates 5th–10th percentile, DY20% indicates 10th–20th percentile of all samples. Each box represents values of drought indices falling within a given category of detrended yield.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
Figure 9 shows that the fixed thresholds (colored regions) are systematically lower than the DY-based thresholds (gray regions). This indicates a potential underestimation of the drought intensity when fixed drought thresholds are used in Ohio. By contrast, the DY-based thresholds are more representative of the regional agricultural drought conditions in Ohio, especially those associated with yield loss.
3) Application of impacts-based agricultural drought thresholds
Figure 10 compares the area in different drought categories observed from 2000 to 2019 using the USDM archive data, fixed thresholds, and the impacts-based agricultural drought thresholds (DY). The first column in Fig. 10 shows the drought area percentage from USDM archive data in the week 33 (for SPI3 and SPEI3) and 30 (for Z index). The weeks have the highest correlation between detrended corn yield and drought indices, according to section 4b(1). Based on USDM-archive data (first column), there were no exceptional drought (D4) conditions experienced in Ohio during the 20 years, and the D3 extreme drought was only observed in a small region of Ohio (11.5%) in 2007. During most years, Ohio only experienced moderate drought (D1) conditions. When using fixed thresholds (second column), D1 droughts are the most frequent, followed by a small proportion of D2 drought for all three drought indices. Although D3 and D4 drought conditions were not observed using Z index, they were observed using SPI3 and SPEI3. The mean annual percent area of D4 drought during the recent 20 years (2000–19) was 2.2% for SPI3, 1.1% for SPEI3, and 0% for Z index based on the fixed thresholds.
Area of Ohio (percentage) in D1–D4 drought categories from 2000 to 2019 at an annual time scale. The drought information is obtained from three sources: (left) USDM archive, (center) drought indices (3-month SPI, 3-month SPEI, and Z index) using the fixed thresholds, and (right) drought indices (3-month SPI, 3-month SPEI, and Z index) using the detrended yield (DY)-based thresholds. Data prior to 2000 are not shown because the USDM archive is not available.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
When using impacts-based agricultural drought thresholds, a larger proportion of D4 drought is detected than when using either the USDM archive or fixed thresholds. For example, the mean annual area of D4 drought in Ohio over the recent 20 years was 6.5% for SPI3, 8.4% for SPEI3, and 15.9% for Z index. The overall larger proportion of D4 drought identified using DY-based thresholds is due to the higher values of the DY-based D4 thresholds. The Z index indicates a larger proportion of D4 drought than using SPI and SPEI when DY-based thresholds are used, possibly because 1) the correlation between Z index and DY is the lowest among the three drought indices (Fig. 8); 2) the Z index is on a single month of data rather than the 3 months used for the SPI and SPEI. The Z index represents the moisture anomaly for the current month without considering the antecedent conditions (Keyantash and Dracup 2002).
To demonstrate that the DY-based thresholds are more sensitive to corn yield variations in Ohio, Fig. 11 compares DY to the area in the D4 drought. Similar to hydrological drought analysis, only D4 drought results are shown here. The results indicate that the percentage of Ohio in D4 drought using the DY-based thresholds covaries closely with the detrended yield, while the fixed thresholds show less sensitivity to yield variation. For example, in 1991, the detrended yield loss was −18.13 bu ac−1, and the area of D4 drought using DY-based thresholds was 26.28% for SPI3, 24.62% was SPEI3, and 54.27% for Z index. However, the area of D4 drought using the fixed thresholds was only 10.18% for SPI3, and 0% for both SPEI3 and Z index. The correlation between the detrended yield and area of D4 drought using DY-based thresholds is also higher than that based on the fixed thresholds. The differences in the correlation range from 0.11 to 0.39, but they are not statistically significant (p ≤ 0.05) according to Fisher (1921). However, the correlation differences are statistically significant if an out-of-sample validation is used (details can be referred to section 5). It demonstrates that the DY-based thresholds are more sensitive to agricultural drought at the time scales used in this study. Finally, a case study of the application of different thresholds is provided using a typical drought event from May 2007 to January 2008 in the supplemental material.
Time series of detrended annual corn yield (DY) (solid blue line, left y axis) and the percentage of Ohio in D4 drought based on fixed thresholds (solid orange line, right y axis) and impacts-based thresholds (dashed orange line, right y axis) for (top) 3-month SPI, (middle) 3-month SPEI, and (bottom) Z index from January 1990 to December 2019. The correlation coefficients between the DY and area in the D4 drought are listed above each graph. Note the left y axis is reversed. The peaks represent decreases in DY.
Citation: Journal of Hydrometeorology 24, 7; 10.1175/JHM-D-22-0054.1
5. Discussion
This paper developed impacts-based drought thresholds for SPI, SPEI, PHDI, and Z index. Corn yield and streamflow were used to quantify drought impacts for agriculture and hydrology, respectively, and to develop impacts-based thresholds. The impacts-based drought thresholds identified in this study are designed to provide locally relevant drought thresholds for monitoring agricultural and hydrological drought impacts in Ohio. This approach can also be used to develop impacts-based thresholds at other spatial scales and other impacts that are of interest to stakeholders, in regions where such impact data are available for a long enough period of time. This study does not intend to compare different drought indices. Instead, it provides impacts-based thresholds for multiple drought indices to demonstrate the utility of our methodology. Users can choose the most appropriate index at their own discretion.
Our analysis shows that the fixed drought thresholds fail to accurately reflect the extent of impacts that are experienced in Ohio because this region is wetter than the nation on average. This indicates that fixed drought thresholds may underestimate drought severity in both arid (Kansas, Oklahoma, and Texas; Leasor et al. 2020) and humid (Ohio) regions of the United States. This is because the fixed drought thresholds are derived and intended for application at the national scale.
The USDM is developed through a convergence-of-evidence approach that considers numerous hydroclimatic datasets, impacts data, and expert assessment and weekly local stakeholder feedback. It was developed for the entire United States, and it is designed to represent meteorological, agricultural, and hydrological drought. In contrast, we used drought impact data from Ohio to develop thresholds for each drought index for each drought impact. These localized and specific drought thresholds provide more meaningful information at the local level, which is consistent with Steinemann et al. (2015). These thresholds can be utilized for drought monitoring from weekly to monthly time scales for the Ohio Emergency Management Agency (EMA) and the State Climate Office of Ohio. They are also useful for providing state-level input to the U.S. Drought Monitor.
Although we used all the data to determine impacts-based thresholds, an out-of-sample validation was conducted and shown in the supplemental material. A randomly selected 80% of counties/watersheds were used to fit the thresholds, and the remaining 20% were used to conduct correlation analysis between drought indices and D4 drought area percentage for validation. The process was repeated 100 times. Results show that data sampling has a modest impact on threshold values. The impacts-based thresholds determined using all data generally lined up with the median of thresholds from all replications. This indicates that the thresholds determined using the entire period of record are robust. Based on the out-of-sample validation, impacts-based thresholds always have a significantly stronger negative correlation with either streamflow or detrended yield than those using fixed thresholds. This is also consistent with results shown in Figs. 6 and 11.
Anthropogenic influence on hydrological drought have been noted worldwide (Wada et al. 2013) and in places such as the United States (Wan et al. 2017) and Western Cape (Otto et al. 2018). However, we found that the impacts-based thresholds do not differ substantially in natural and human-impacted watersheds (results not shown). This may be because streamflow in this study was evaluated at a 9-month time scale, while above-mentioned studies investigated monthly streamflow. A 9-month time scale will include multiple seasons, obscuring human influences such as reservoir operations (Wan et al. 2017; Tijdeman et al. 2018; Wanders and Wada 2015). Previous studies showed that climate was the dominant driver that caused mean annual changes in streamflow in most watersheds in Ohio (Wang and Hejazi 2011; Li and Quiring 2021). In contrast, human water consumption was shown to contribute primarily to major drought events using monthly streamflow in parts of Ohio (Wada et al. 2013). Future studies are encouraged to investigate the impacts of water management and human water use on streamflow at a 9-month time scale.
Caution should be taken when interpreting the results in this study due to several uncertainties.
-
We acknowledge that corn yield/streamflow records may be impacted by factors other than drought. For example, corn yield may be impacted by pests and management activities; streamflow may be impacted by dam construction and diversion. They both may be impacted by tile drainage (Valayamkunnath et al. 2020). However, they are still two representative datasets that are strongly influenced by drought conditions, and therefore they can be used to identify impacts-based thresholds.
-
The record length may impact the drought threshold values. Sensitivity analysis shows that drought thresholds do not have substantial changes when 30-yr (1990–2019) and 39-yr (1981–2019) data are used (results not shown). However, we did find improved correlations or large changes in threshold values when a longer record was used. The impacts-based thresholds should be routinely updated when a longer record is available.
-
Differences are expected between USDM and our results because our results are based on the most recent 30-yr record, while many of the indices that are used to create the USDM use a longer period of record. In addition, the accuracy, latency, and variety of drought-related data used in the USDM have improved dramatically since its inception. The gradual improvement of USDM maps over time may add uncertainties to the comparison results. In addition, streamflow data are used in developing our impacts-based thresholds and in the USDM products. Therefore, it is not a truly independent comparison. Finally, the USDM simultaneously considers multiple time scales to create the weekly maps, while we select an optimal time scale for each drought index (9 months for streamflow and 3 months for corn yield). The goal of this study is to identify the thresholds for different drought indices to monitor agricultural and hydrological drought in Ohio. The USDM has tremendous value for monitoring and quantifying national drought conditions. Our study proposes a new method (impacts-based thresholds) that can help guide local experts and further improve the USDM representation of drought conditions in Ohio.
There are several limitations in this study that can be further improved in future studies.
-
Values of PHDI and Z index may be less accurate in winter and spring when snow occurs, because they do not handle frozen soil and snow-covered conditions (Liu et al. 2016). Previous studies such as Liu et al. (2016) and Qiu (2013) have attempted to improve Palmer drought indices by including the effects of snow and frozen soil. However, PHDI was calculated at a 9-month time scale in this study, which covers multiple seasons and may obscure the effects of snowmelt. The Z index in mid-August was selected to correlate with corn yield when the impacts of snowmelt are negligible. We thus do not think inclusion of snow effects would have a substantial impact on our results. Ohio also does not have permanent frozen soil. Future studies are encouraged to recalculate the Palmer drought indices and identify the localized thresholds by properly handling the effects of snow and frozen soil.
-
We do not consider seasonality in this study. Since the correlation between streamflow/corn yield and drought indices varies depending on seasons, drought thresholds may also be varied over time. Future studies could develop time-varying drought thresholds.
-
In this study, the week with the maximum correlation between corn yield and drought indices is identified using 86 counties. In a future study, county-specific correlation analysis can be conducted to identify the week with maximum correlation in each county. This may provide an in-depth evaluation of drought indices at a finer spatial scale.
-
Agricultural and hydrological drought may also be represented by other observations such as natural vegetation, groundwater levels, and lake levels. Corn yield/streamflow is just one example of the representation of drought impacts. Other observations can be used to characterize drought impacts and identify impact-specific drought thresholds in future studies.
6. Conclusions
This study developed impacts-based (corn yield and streamflow-based) thresholds for agricultural and hydrological drought in Ohio. These new impacts-based drought thresholds were generated for four common drought indices (e.g., SPI, SPEI, Z index, and PHDI). Based on our analysis, we conclude the following:
-
Agricultural drought in Ohio is better characterized by Z index and 3-month SPI and 3-month SPEI than other time scales. Hydrological drought in Ohio is strongly associated with SPI and SPEI at a 9-month time scale and PHDI.
-
The correlation analysis shows that for hydrological drought, the three drought indices (SPI9, SPEI9, and PHDI) have similar correlations (r = ∼0.46) with streamflow below the 20th percentile. For agricultural drought, the correlation between corn yield and three indices (SPI3, SPEI3, and Z index) peaks in weeks 30–33 in Ohio (r = ∼0.50). SPI3 and SPEI3 have a higher correlation (r = 0.37, 0.41) with detrended corn yield below the 20th percentile than Z index (r = 0.25).
-
The USDM tends to indicate less severe drought conditions in Ohio from 2000 to 2019 than using impacts-based thresholds based on individual indicators. The drought indices using impacts-based thresholds are more sensitive at the time scales used in this analysis to exceptional drought conditions and are strongly correlated with corn yield and streamflow.
-
The differences between the USDM, fixed, and impacts-based drought thresholds are primarily due to their different focuses, such as spatial scale (nationwide versus local), temporal scale (multiple versus optimal), method (hybrid versus single index), and periods of record adopted (entire versus 30 years). The impacts-based drought thresholds developed in this study are suitable for evaluating local drought impacts in streamflow and corn yield in Ohio.
The methodology proposed by this study can be easily applied to other regions where long-term streamflow and crop yield records are available, which may greatly improve regional drought monitoring worldwide and provide useful information to local farmers and decision-makers for decision support.
Acknowledgments.
The authors thank the Ohio Emergency Management Agency for providing helpful suggestions for this project. Funding for this work was provided by NIDIS. We gratefully acknowledge the guidance and support provided by Molly Woloszyn. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Data availability statement.
Daily precipitation is obtained from the PRISM dataset at http://www.prism.oregonstate.edu/. Potential evapotranspiration is obtained from the gridMET dataset at http://www.climatologylab.org/gridmet.html. The available water capacity is obtained from the Gridded National Soil Survey Geographic Database (gNATSGO) at https://www.nrcs.usda.gov/wps/portal/nrcs/detail/soils/survey/geo/?cid=nrcseprd1464625. Streamflow data are obtained from the United States Geological Survey (USGS) at https://waterdata.usgs.gov/nwis. Crop yield data are obtained from the National Agricultural Statistics Service (NASS) Quick Stats Database at https://quickstats.nass.usda.gov/.
REFERENCES
Agnolucci, P., and V. De Lipsis, 2020: Long-run trend in agricultural yield and climatic factors in Europe. Climatic Change, 159, 385–405, https://doi.org/10.1007/s10584-019-02622-3.
Bachmair, S., I. Kohn, and K. Stahl, 2015: Exploring the link between drought indicators and impacts. Nat. Hazards Earth Syst. Sci., 15, 1381–1397, https://doi.org/10.5194/nhess-15-1381-2015.
Beguería, S., S. M. Vicente-Serrano, F. Reig, and B. Latorre, 2014: Standardized precipitation evapotranspiration index (SPEI) revisited: Parameter fitting, evapotranspiration models, tools, datasets and drought monitoring. Int. J. Climatol., 34, 3001–3023, https://doi.org/10.1002/joc.3887.
Beyene, B. S., A. F. Van Loon, H. A. J. Van Lanen, and P. Torfs, 2014: Investigation of variable threshold level approaches for hydrological drought identification. Hydrol. Earth Syst. Sci. Discuss., 11, 12 765–12 797, https://doi.org/10.5194/hessd-11-12765-2014.
Dai, A., 2011: Characteristics and trends in various forms of the Palmer Drought Severity Index during 1900–2008. J. Geophys. Res., 116, D12115, https://doi.org/10.1029/2010JD015541.
Dikici, M., 2020: Drought analysis with different indices for the Asi Basin (Turkey). Sci. Rep., 10, 20739, https://doi.org/10.1038/s41598-020-77827-z.
Falcone, J. A., D. M. Carlisle, D. M. Wolock, and M. R. Meador, 2010: GAGES: A stream gage database for evaluating natural and altered flow conditions in the conterminous United States: Ecological archives E091-045. Ecology, 91, 621–621, https://doi.org/10.1890/09-0889.1.
FAO, 2015: Crop Water Information: Maize. FAO Land and Water division, https://www.fao.org/land-water/databases-and-software/crop-information/maize/en/.
Fisher, R. A., 1921: 014: On the “probable error” of a coefficient of correlation deduced from a small sample. Metron, 1, 3–32.
Goodrich, G. B., and A. W. Ellis, 2006: Climatological drought in Arizona: An analysis of indicators for guiding the governor’s drought task force. Prof. Geogr., 58, 460–469, https://doi.org/10.1111/j.1467-9272.2006.00582.x.
Guttman, N. B., 1999: Accepting the standardized precipitation index: A calculation algorithm. J. Amer. Water Resour. Assoc., 35, 311–322, https://doi.org/10.1111/j.1752-1688.1999.tb03592.x.
Hafner, S., 2003: Trends in maize, rice, and wheat yields for 188 nations over the past 40 years: A prevalence of linear growth. Agric. Ecosyst. Environ., 97, 275–283, https://doi.org/10.1016/S0167-8809(03)00019-7.
Hao, Z., and A. AghaKouchak, 2014: A nonparametric multivariate multi-index drought monitoring framework. J. Hydrometeor., 15, 89–101, https://doi.org/10.1175/JHM-D-12-0160.1.
Heim, R. R., Jr., 2002: A review of twentieth-century drought indices used in the United States. Bull. Amer. Meteor. Soc., 83, 1149–1166, https://doi.org/10.1175/1520-0477-83.8.1149.
Huang, S., P. Li, Q. Huang, G. Leng, B. Hou, and L. Ma, 2017: The propagation from meteorological to hydrological drought and its potential influence factors. J. Hydrol., 547, 184–195, https://doi.org/10.1016/j.jhydrol.2017.01.041.
Jacobi, J., D. Perrone, L. L. Duncan, and G. Hornberger, 2013: A tool for calculating the Palmer drought indices. Water Resour. Res., 49, 6086–6089, https://doi.org/10.1002/wrcr.20342.
Karl, T. R., 1986: The sensitivity of the Palmer drought severity index and Palmer’s Z index to their calibration coefficients including potential evapotranspiration. J. Climate Appl. Meteor., 25, 77–86, https://doi.org/10.1175/1520-0450(1986)025<0077:TSOTPD>2.0.CO;2.
Keyantash, J., and J. A. Dracup, 2002: The quantification of drought: An evaluation of drought indices. Bull. Amer. Meteor. Soc., 83, 1167–1180, https://doi.org/10.1175/1520-0477-83.8.1167.
Kukal, M. S., and S. Irmak, 2018: Climate-driven crop yield and yield variability and climate change impacts on the US Great Plains agricultural production. Sci. Rep., 8, 3450, https://doi.org/10.1038/s41598-018-21848-2.
Kumar, M. N., C. S. Murthy, M. V. R. S. Sai, and P. S. Roy, 2009: On the use of Standardized Precipitation Index (SPI) for drought intensity assessment. Meteor. Appl., 16, 381–389, https://doi.org/10.1002/met.136.
Labudová, L., M. Labuda, and J. Takáč, 2017: Comparison of SPI and SPEI applicability for drought impact assessment on crop production in the Danubian Lowland and the East Slovakian Lowland. Theor. Appl. Climatol., 128, 491–506, https://doi.org/10.1007/s00704-016-1870-2.
Leasor, Z. T., S. M. Quiring, and M. D. Svoboda, 2020: Utilizing objective drought severity thresholds to improve drought monitoring. J. Appl. Meteor. Climatol., 59, 455–475, https://doi.org/10.1175/JAMC-D-19-0217.1.
Li, Z., and S. M. Quiring, 2021: Identifying the dominant drivers of hydrological change in the contiguous United States. Water Resour. Res., 57, e2021WR029738, https://doi.org/10.1029/2021WR029738.
Link, R., T. B. Wild, A. C. Snyder, M. I. Hejazi, and C. R. Vernon, 2020: 100 years of data is not enough to establish reliable drought thresholds. J. Hydrol. X, 7, 100052, https://doi.org/10.1016/j.hydroa.2020.100052.
Liu, Y., L. Ren, M. Ma, X. Yang, F. Yuan, and S. Jiang, 2016: An insight into the Palmer drought mechanism based indices: Comprehensive comparison of their strengths and limitations. Stochastic Environ. Res. Risk Assess., 30, 119–136, https://doi.org/10.1007/s00477-015-1042-4.
Liu, X., X. Zhu, Y. Pan, J. Bai, and S. Li, 2018: Performance of different drought indices for agriculture drought in the North China Plain. J. Arid Land, 10, 507–516, https://doi.org/10.1007/s40333-018-0005-2.
Lorenz, D. J., J. A. Otkin, M. Svoboda, C. R. Hain, M. C. Anderson, and Y. Zhong, 2017: Predicting US drought monitor states using precipitation, soil moisture, and evapotranspiration anomalies. Part I: Development of a nondiscrete USDM index. J. Hydrometeor., 18, 1943–1962, https://doi.org/10.1175/JHM-D-16-0066.1.
Lorenzo-Lacruz, J., S. M. Vicente-Serrano, J. I. López-Moreno, S. Beguería, J. M. García-Ruiz, and J. M. Cuadrat, 2010: The impact of droughts and water management on various hydrological systems in the headwaters of the Tagus River (central Spain). J. Hydrol., 386, 13–26, https://doi.org/10.1016/j.jhydrol.2010.01.001.
Lorenzo-Lacruz, J., S. M. Vicente-Serrano, J. C. González-Hidalgo, J. I. López-Moreno, and N. Cortesi, 2013: Hydrological drought response to meteorological drought in the Iberian Peninsula. Climate Res., 58, 117–131, https://doi.org/10.3354/cr01177.
Lu, J., G. J. Carbone, and P. Gao, 2017: Detrending crop yield data for spatial visualization of drought impacts in the United States, 1895–2014. Agric. For. Meteor., 237, 196–208, https://doi.org/10.1016/j.agrformet.2017.02.001.
Lu, J., G. J. Carbone, X. Huang, K. Lackstrom, and P. Gao, 2020: Mapping the sensitivity of agriculture to drought and estimating the effect of irrigation in the United States, 1950–2016. Agric. For. Meteor., 292, 108124, https://doi.org/10.1016/j.agrformet.2020.108124.
Mallya, G., L. Zhao, X. C. Song, D. Niyogi, and R. S. Govindaraju, 2013: 2012 Midwest drought in the United States. J. Hydrol. Eng., 18, 737–745, https://doi.org/10.1061/(ASCE)HE.1943-5584.0000786.
McEvoy, D. J., D. W. Pierce, J. F. Kalansky, D. R. Cayan, and J. T. Abatzoglou, 2020: Projected changes in reference evapotranspiration in California and Nevada: Implications for drought and wildland fire danger. Earths Future, 8, e2020EF001736, https://doi.org/10.1029/2020EF001736.
McKee, T. B., N. J. Doesken, and J. Kleist, 1993: The relationship of drought frequency and duration to time scales. Proc. Eighth Conf. on Applied Climatology, Boston, MA, Amer. Meteor. Soc., 179–183.
Mishra, A. K., and V. P. Singh, 2010: A review of drought concepts. J. Hydrol., 391, 202–216, https://doi.org/10.1016/j.jhydrol.2010.07.012.
Mizzell, E. H. P., 2008: Improving drought detection in the Carolinas: Evaluation of local, state, and federal drought indicators. Ph.D. dissertation, Department of Geology, University of South Carolina, 149 pp.
Myronidis, D., D. Fotakis, K. Ioannou, and K. Sgouropoulou, 2018: Comparison of ten notable meteorological drought indices on tracking the effect of drought on streamflow. Hydrol. Sci. J., 63, 2005–2019, https://doi.org/10.1080/02626667.2018.1554285.
Osborne, T. M., and T. R. Wheeler, 2013: Evidence for a climate signal in trends of global crop yield variability over the past 50 years. Environ. Res. Lett., 8, 024001, https://doi.org/10.1088/1748-9326/8/2/024001.
Otto, F. E., and Coauthors, 2018: Anthropogenic influence on the drivers of the Western Cape drought 2015–2017. Environ. Res. Lett., 13, 124010, https://doi.org/10.1088/1748-9326/aae9f9.
Palmer, W. C., 1965: Meteorological drought. U.S. Weather Bureau Research Paper 45, 58 pp., http://www.ncdc.noaa.gov/temp-and-precip/drought/docs/palmer.pdf.
Park, S., J. Im, E. Jang, and J. Rhee, 2016: Drought assessment and monitoring through blending of multi-sensor indices using machine learning approaches for different climate regions. Agric. For. Meteor., 216, 157–169, https://doi.org/10.1016/j.agrformet.2015.10.011.
Peña-Gallardo, M., S. M. Vicente-Serrano, F. Domínguez-Castro, S. Quiring, M. Svoboda, S. Beguería, and J. Hannaford, 2018: Effectiveness of drought indices in identifying impacts on major crops across the USA. Climate Res., 75, 221–240, https://doi.org/10.3354/cr01519.
Peña-Gallardo, M., and Coauthors, 2019: Response of crop yield to different time-scales of drought in the United States: Spatio-temporal patterns and climatic and environmental drivers. Agric. For. Meteor., 264, 40–55, https://doi.org/10.1016/j.agrformet.2018.09.019.
Qiu, S., 2013: Improving the Palmer drought severity index by incorporating snow and frozen ground. M.S. thesis, Dept. of Atmospheric Sciences, University of North Dakota, 82 pp., https://commons.und.edu/cgi/viewcontent.cgi?article=2471&context=theses.
Quiring, S. M., 2009: Developing objective operational definitions for monitoring drought. J. Appl. Meteor. Climatol., 48, 1217–1229, https://doi.org/10.1175/2009JAMC2088.1.
Quiring, S. M., and T. N. Papakryiakou, 2003: An evaluation of agricultural drought indices for the Canadian Prairies. Agric. For. Meteor., 118, 49–62, https://doi.org/10.1016/S0168-1923(03)00072-8.
Rippey, B. R., 2015: The US drought of 2012. Wea. Climate Extremes, 10, 57–64, https://doi.org/10.1016/j.wace.2015.10.004.
Stagge, J. H., L. M. Tallaksen, L. Gudmundsson, A. F. Van Loon, and K. Stahl, 2015: Candidate distributions for climatological drought indices (SPI and SPEI). Int. J. Climatol., 35, 4027–4040, https://doi.org/10.1002/joc.4267.
Steinemann, A., 2003: Drought indicators and triggers: A stochastic approach to evaluation. J. Amer. Water Resour. Assoc., 39, 1217–1233, https://doi.org/10.1111/j.1752-1688.2003.tb03704.x.
Steinemann, A., 2014: Drought information for improving preparedness in the western states. Bull. Amer. Meteor. Soc., 95, 843–847, https://doi.org/10.1175/BAMS-D-13-00067.1.
Steinemann, A. C., and L. F. Cavalcanti, 2006: Developing multiple indicators and triggers for drought plans. J. Water Resour. Plann. Manage., 132, 164–174, https://doi.org/10.1061/(ASCE)0733-9496(2006)132:3(164).
Steinemann, A., S. F. Iacobellis, and D. R. Cayan, 2015: Developing and evaluating drought indicators for decision-making. J. Hydrometeor., 16, 1793–1803, https://doi.org/10.1175/JHM-D-14-0234.1.
Svoboda, M., and Coauthors, 2002: The Drought Monitor. Bull. Amer. Meteor. Soc., 83, 1181–1190, https://doi.org/10.1175/1520-0477-83.8.1181.
Svoboda, M. D., 2016: Essays on decision support for drought mitigation planning: A tale of three tools. Ph.D. dissertation, University of Nebraska–Lincoln, 24 pp.
Svoboda, M. D., and B. A. Fuchs, 2016: Handbook of drought indicators and indices. WMO-1173, 45 pp., https://library.wmo.int/doc_num.php?explnum_id=3057.
Tian, L., S. Yuan, and S. M. Quiring, 2018: Evaluation of six indices for monitoring agricultural drought in the south-central United States. Agric. For. Meteor., 249, 107–119, https://doi.org/10.1016/j.agrformet.2017.11.024.
Tijdeman, E., J. Hannaford, and K. Stahl, 2018: Human influences on streamflow drought characteristics in England and Wales. Hydrol. Earth Syst. Sci., 22, 1051–1064, https://doi.org/10.5194/hess-22-1051-2018.
Valayamkunnath, P., M. Barlage, F. Chen, D. J. Gochis, and K. J. Franz, 2020: Mapping of 30-meter resolution tile-drained croplands using a geospatial modeling approach. Sci. Data, 7, 257, https://doi.org/10.1038/s41597-020-00596-x.
Van Loon, A. F., 2015: Hydrological drought explained. Wiley Interdiscip. Rev. Water, 2, 359–392, https://doi.org/10.1002/wat2.1085.
Vasiliades, L., and A. Loukas, 2009: Hydrological response to meteorological drought using the Palmer drought indices in Thessaly, Greece. Desalination, 237, 3–21, https://doi.org/10.1016/j.desal.2007.12.019.
Vicente-Serrano, S. M., S. Beguería, and J. I. López-Moreno, 2010: A multiscalar drought index sensitive to global warming: The standardized precipitation evapotranspiration index. J. Climate, 23, 1696–1718, https://doi.org/10.1175/2009JCLI2909.1.
Vicente-Serrano, S. M., and Coauthors, 2012: Performance of drought indices for ecological, agricultural, and hydrological applications. Earth Interact., 16, https://doi.org/10.1175/2012EI000434.1.
Wada, Y., L. P. H. van Beek, N. Wanders, and M. F. P. Bierkens, 2013: Human water consumption intensifies hydrological drought worldwide. Environ. Res. Lett., 8, 034036, https://doi.org/10.1088/1748-9326/8/3/034036.
Wan, W., and Coauthors, 2017: Hydrological drought in the Anthropocene: Impacts of local water extraction and reservoir regulation in the U.S. J. Geophys. Res. Atmos., 122, 11 313–11 328, https://doi.org/10.1002/2017JD026899.
Wanders, N., and Y. Wada, 2015: Human and climate impacts on the 21st century hydrological drought. J. Hydrol., 526, 208–220, https://doi.org/10.1016/j.jhydrol.2014.10.047.
Wang, D., and M. Hejazi, 2011: Quantifying the relative contribution of the climate and direct human impacts on mean annual streamflow in the contiguous United States. Water Resour. Res., 47, W00J12, https://doi.org/10.1029/2010WR010283.
Wilhite, D. A., 2000: Drought as a natural hazard: Concepts and definitions. Drought: A Global Assessment, Vol. I, edited by D. A. Wilhite, Routledge, 3–18.
World Meteorological Organization, 2012: Standardized precipitation index user guide. WMO-1090, 16 pp., https://library.wmo.int/doc_num.php?explnum_id=7768.
Wu, H., M. D. Svoboda, M. J. Hayes, D. A. Wilhite, and F. Wen, 2007: Appropriate application of the standardized precipitation index in arid locations and dry seasons. Int. J. Climatol., 27, 65–79, https://doi.org/10.1002/joc.1371.
Xia, Y., and Coauthors, 2012: Continental-scale water and energy flux analysis and validation for the North American Land Data Assimilation System project phase 2 (NLDAS-2): 1. Intercomparison and application of model products. J. Geophys. Res., 117, D03109, https://doi.org/10.1029/2011JD016048.
Xia, Y., M. B. Ek, D. Mocko, C. D. Peters-Lidard, J. Sheffield, J. Dong, and E. F. Wood, 2014: Uncertainties, correlations, and optimal blends of drought indices from the NLDAS multiple land surface model ensemble. J. Hydrometeor., 15, 1636–1650, https://doi.org/10.1175/JHM-D-13-058.1.
Yuan, X., M. Zhang, L. Wang, and T. Zhou, 2017: Understanding and seasonal forecasting of hydrological drought in the Anthropocene. Hydrol. Earth Syst. Sci., 21, 5477–5492, https://doi.org/10.5194/hess-21-5477-2017.
Zargar, A., R. Sadiq, B. Naser, and F. I. Khan, 2011: A review of drought indices. Environ. Rev., 19, 333–349, https://doi.org/10.1139/a11-013.
Zhai, J., B. Su, V. Krysanova, T. Vetter, C. Gao, and T. Jiang, 2010: Spatial variation and trends in PDSI and SPI indices and their relation to streamflow in 10 large regions of China. J. Climate, 23, 649–663, https://doi.org/10.1175/2009JCLI2968.1.