Droughts are naturally (re-)occurring phenomena that can appear in different domains of the hydrological cycle and cause associated impacts (Tallaksen and Van Lanen, 2004; Stahl et al., 2016). Because of their multifaceted characteristics, droughts are often classified as different types (Wilhite and Glantz, 1985). Of these drought types, one is agricultural drought, which refers to the impact of lacking water availability on the health and growth of crops. These agricultural droughts can reduce yields and, thereby, cause large economic losses. A crucial first step for reducing the risk of (agricultural) drought impacts involves the effective monitoring and early warning of the drought hazard (UN/ISDR, 2009). Agricultural drought monitoring and early warning occurs at different scales, from plot-scale observations and simulations to regional-scale drought mapping. Regional-scale drought monitoring and early warning provides an overview of regions at drought risk, which raises awareness and helps decision-making. Accurately depicting areas affected by agricultural drought is complex, as its occurrence is influenced by a variety of factors, often including spatially heterogeneous climate and soil characteristics. A better understanding of how these climate and soil characteristics control (the development of) agricultural droughts is needed.

Droughts are often defined as a below-normal water availability, with the normal often depending on space and time (Tallaksen and Van Lanen, 2004). Such an anomaly-based definition allows the depiction of regions and episodes with below-normal water availability across the world, according to different hydro-meteorological variables. However, the identified events with below-normal water availability might not necessarily have the potential to cause drought-related impacts. The below-normal definition of drought forms the basis of many drought indices, which reflect whether a certain hydro-meteorological variable is anomalously low or high (e.g., Anderson et al., 2007; McKee et al., 1993; Samaniego et al., 2012; Vicente-Serrano et al., 2010). Soil moisture anomaly time series, or proxies of the latter, are often used for agricultural drought assessments (e.g., Sheffield et al., 2004; Andreadis et al., 2005; Samaniego et al., 2012). Different drought characteristics can be derived from these soil moisture anomaly time series, including drought magnitude, duration, and areal extent.

The data used for agricultural drought assessments stem from different sources. These data sources include direct soil moisture measurements, remote sensing observations, meteorological proxies, and hydrological or land surface model simulations (e.g., Berg and Sheffield, 2018). Soil moisture measurements provide the most realistic information about the soil moisture status at a certain depth but are point based and, thereby, limited in their spatial coverage. Remote sensing observations of soil moisture provide regional coverage but direct observations are only able to detect soil moisture changes in the upper soil layer, at least in the case of microwave remote sensing. On the other hand, remote sensing observations of heat fluxes and vegetation health can provide an estimate of the ratio between actual and potential evapotranspiration and, thereby, depict regions with soil moisture stress (e.g., Anderson et al., 2007). Meteorological proxies for agricultural drought include drought indices such as the Palmer Drought Severity Index (PDSI; Palmer, 1965) or standardized precipitation evapotranspiration index (SPEI; Vicente-Serrano et al., 2010). The strength of these meteorological proxies is their relative ease of computation and often low data requirements. However, meteorological proxies are often based on potential evapotranspiration and do not consider some other relevant terrestrial processes that influence soil moisture and agricultural drought, such as the reduction in evapotranspiration during soil moisture stress. Many of these terrestrial processes are included in physically based hydrological and land surface models. The physical basis of these models makes their use often preferable over the use of meteorological proxies for past and future agricultural drought assessments (e.g., Berg and Sheffield, 2018; Sheffield et al., 2012).

Various hydrological and land surface models have been used to assess past and future soil moisture drought events. An example is the Variable Infiltration Capacity (VIC) model, which has been applied to characterize major soil moisture drought episodes across different regions (e.g., US – Sheffield et al., 2004, and Andreadis et al., 2005; China: Wang et al., 2011; and globally – Sheffield and Wood, 2007). The latter analyses enabled the cataloguing of past soil moisture drought events according to a variety of characteristics, providing a benchmark for current and future drought events. Another example of a regionally applied model for the simulation of soil moisture (drought) is the mesoscale Hydrological Model (mHM; Samaniego et al., 2010). The output of the mHM has been used for both historic soil moisture drought assessments (Hanel et al., 2018) and for future soil moisture drought projections according to different climate change scenarios across Europe (as part of a model ensemble in Samaniego et al., 2018). The latter studies provide valuable insights about the severity of recent soil moisture drought events over Europe, e.g., 2003 and 2015, and also show that these recent events were not as rare when considered from a more long-term historical perspective, and that similar or worse events are more likely to occur under different climate change scenarios. The mHM is also run in near-real time, and its output is used by the German Drought Monitor (Zink et al., 2016).

Studies mentioned in the previous paragraph focus on characterizing past and future soil moisture drought events, whereas other studies aim to characterize its development. Drought is often referred to as a slowly developing phenomena that can take up to years to reach its full extent (Wilhite and Glantz, 1985). However, not all drought events are slowly developing phenomena, and soil moisture deficits can develop relatively quickly during dry weather conditions that favor high amounts of evapotranspiration (e.g., Hunt et al., 2009). These rapidly developing droughts, sometimes termed flash droughts, can severely impact agriculture (e.g., Svoboda et al., 2002; Otkin et al., 2018). Several case study flash drought events in the US have been described in Otkin et al. (2013, 2016). The latter studies show that precipitation deficits can be quickly followed by a reduction in evapotranspiration, which is indicative of low soil moisture levels, causing water stress for plants. Christian et al. (2019) aimed to make a regional assessment of past flash droughts and developed a framework of objective criteria to identify flash drought events from simulated soil moisture output. By applying this framework to soil moisture simulations over the US, they show that particular regions, such as the Great Plains, are more sensitive to flash drought occurrence.

Most of the above-described soil moisture drought assessments characterize drought as a below-normal anomaly according to different hydrometeorlogical variables, which is in line with the traditional definition of drought. However, from an agricultural drought impact perspective, it can make more sense to directly study the characteristics of (the development of) periods of lacking amounts of root zone soil moisture, i.e., soil moisture stress, which is in line with the soil moisture drought index proposed in Hunt et al. (2009). Following this reasoning, and being inspired by the methods used in previous soil moisture anomaly studies, we aim to study simulated soil moisture stress events across the agricultural regions of southwestern Germany. Our objectives are as follows:

characterize past soil moisture stress events,

Our first objective was to characterize the occurrence, development time, and persistence of simulated past soil moisture (SM) stress events. Results revealed a large temporal variability in the number of grid cells that reach SM stress in a certain year (Fig. 6). The most severe SM stress years were 2003 and 2018, during which up to 89 % of the agricultural grid cells reached SM stress. These percentages of grid cells were found to be (slightly) different, depending on the parameterization of Srootzone (Fig. S6), implying differences between, for example, shallow rooting crops with limited access to water and deeper rooting crops with a larger water availability. Nevertheless, the ordering of most severe drought years was not affected by the parameterization of the root zone, i.e., 2003 and 2018 were always characterized as most severe in terms of the number of grid cells that reached SM stress. Previous studies already showed that 2003 was an extreme drought year within and around the study region (e.g., Ionita et al., 2017). Results of this study imply that the recent 2018 event was comparable to 2003 in terms of the number of grid cells that reach SM stress. However, even during these most severe drought years, SM stress did not develop for some of the agricultural grid cells (unless a root zone with a fixed AWC of 100 mm was used), either because of (1) local variations in meteorological conditions (e.g., local rains storms) or (2) root zone soils having a large enough storage capacity that acted as a buffer during dry conditions. This illustrates that, even during the most extreme drought years, regional differences can occur. The factors that control these differences, i.e., the occurrence of local rainstorms and differences in soil characteristics, can be spatially heterogeneous. The latter implies that regional agricultural drought assessments and monitoring should occur at a relatively high spatial resolution to be able to capture these differences.

A large variability in the development time of simulated SM stress was found (Fig. 8b). SM stress could develop in less than 10 d, e.g., in shallow root zones with a low available water-holding capacity (AWC). This is faster than the minimum development time of 30 d used to identify rapid-onset (flash) droughts in, e.g., Christian et al. (2019). On the other hand, it could also take a lot longer (over 4 months) for SM stress to develop. This slower development matches better with the traditional description of drought being a slowly developing (creeping) phenomena (Wilhite and Glantz, 1985). The above-stated ranges in the development time were reduced when the starting point of SM stress development was set to a level lower than field capacity (Fig. S10), implying that it is important to keep track of partially depleted soil moisture stores that can be a precursor to more rapid development. The sensitivity analyses revealed that fixing Srootzone reduces the variability in development time; however, also showing the distinct differences between root zones with a relatively low and high storage capacity, which is indicative for differences between shallow and deep rooting crop species. Furthermore, the relative ordering of drought years in terms of their Sdevtime was often the same, besides few exceptions, which relates to specific differences in the configuration of meteorological dry spells and whether they caused SM stress under different Srootzone parameterizations. Overall, the large differences in development time suggest that different types of forecasting systems could be suitable for predicting the development of SM stress, medium-range weather forecasts for quickly developing events, and more long-term meteorological forecasts for slower developing episodes.

The persistence of SM stress (total days and maximum duration) varied strongly between years and grid cells (Fig. 8c, d). The results of this study showed that the total days and maximum duration of SM stress was generally highest in 2018, making this event more severe than earlier (recent) benchmark events, such as 2003. The long nature of the drought of 2018 was also found in a recent study for Switzerland, the country directly south of our study region, in Brunner et al. (2019). The ordering of most extreme drought years according to duration was often found to be independent of the parameterization of Srootzone or whether SM stress or drought was analyzed (Fig. S9). On the other hand, distinct differences in duration were found, especially between SM stress and drought, i.e., SM was generally much longer in an anomalously low state compared to the time when it was in a state of SM stress. This can be partially explained by the fact that SM can be anomalously low without being severely depleted, especially towards the end of the year, after a severe drought year, when SM stores are not completely filled to field capacity again (as would normally be the case). We also found that the annual maximum duration and total time of SM stress never exceeded 6 months, and most of the root zones reached field capacity again each year before the start of the new growing season. Thus, SM stress was never a multi-year phenomenon for the considered agricultural grid cells. SM droughts, on the other hand, can last longer and could more easily persist into the next year.

Our second objective was to investigate the dominant controls on the probability, development time, and persistence of SM stress. Both probability and development time were most strongly related to the AWC of the root zone and less to the climatological setting (Fig. 7; Table 1). SM stress was generally more likely to develop, and it evolved faster and earlier in the year in shallow root zones with a lower AWC. These findings are in line with results for the 2012 flash drought in the US presented by Otkin et al. (2016), where anomalous soil moisture conditions generally first appeared in the topsoil layer (lower AWC) and only later in the entire soil layer (higher AWC). Results also confirm that AWC of the root zone is an important factor for determining the vulnerability to agricultural drought, as was also stated in, e.g., Wilhelmi and Wilhite (2002). Here, it is important to state that AWC is not only a soil parameter but also encompasses differences between, for example, a shallow or deep rooting crop, as was exemplified by the differences between the two root zone parameterizations with a fixed AWC found in the sensitivity analyses (Figs. S6, S8). Finally, these results imply that agricultural drought assessments purely based on meteorological proxy indicators should be interpreted with care as most meteorological proxy indicators do not consider differences in root zone soil characteristics.

The persistence of SM stress was only weakly correlated with the AWC of the root zone and more strongly with the climatological setting (Table 1), especially when considering a parameterization of Srootzone with a fixed AWC (Fig. S9). The reason for the overall weaker correlations with the AWC might be related to the different mechanisms that govern the persistence of SM stress in different types of root zones. In root zones with a low AWC, SM stress can develop rather quickly. However, the total deficit that can build up is limited, and only a small rainfall event is enough to alleviate SM stress conditions. In root zones with a high AWC, larger SM deficits can potentially develop. However, this development takes longer, and the SM stress threshold is only exceeded towards the end of the growing season, after which further development is limited because of lacking evapotranspiration. The most persistent SM stress events might, therefore, occur for root zones with an intermediate AWC. In these root zones, SM stress can develop reasonably fast but can also build up a large enough deficit that can endure some smaller rainfall events. This is different for the duration of SM drought (anomaly), which is positively correlated with the AWC of the root zone, i.e., SM droughts tend to last longer for root zones with a higher storage (Fig. S9). A reason for this is that SM (anomaly) time series derived from root zones with a larger AWC often exhibit a much more gradual behavior, whereas SM (anomaly) time series derived from root zones with a smaller AWC are often flashier. Another reason for this is that it can take much longer for root zones with a larger AWC to reach a level of field capacity towards the end of the year (normal conditions) after a prolonged meteorological dry spell.

The third objective of this study was to portray the meteorological anomalies during (the development of) simulated SM stress. During the development, especially the precipitation needed to be anomalously low, particularly during the more rapidly developing events (Fig. 9a), suggesting that lacking precipitation was the most important prerequisite for SM stress to develop. However, air temperature and evapotranspiration were also often higher than normal during the development of SM stress, implying an enhancing (compound) effect of these variables (see also Manning et al., 2018), especially during rapid onset events. During the annual maximum duration SM stress events, precipitation was often below normal as well, especially for the longer events (Fig. 9b). However, precipitation anomalies during the maximum duration events were not as extreme as during the development, possibly because SM only needed to remain in a steady state condition of SM stress, rather than having to decline from field capacity to a level of SM stress. Temperature and simulated evapotranspiration show contrasting anomalies during the annual maximum duration SM stress events, with temperature generally being above normal and simulated evapotranspiration generally being below normal, particularly during the longer events. The reason for these contrasting anomalies might be related to a different energy partitioning of heat fluxes during SM stress (described in, e.g., Seneviratne et al., 2010). During SM stress, simulated evapotranspiration was anomalously low because of the water stress for vegetation that causes plants to limit their evapotranspiration assumed in the model. The incoming solar radiation that is normally consumed by evapotranspiration (latent heat flux) is now used to warm up the soil and lower atmosphere (sensible heat flux), possibly explaining the above-normal temperatures during SM stress (Miralles et al., 2014). This energy partitioning during SM stress, and the resulting contrasting temperature and evapotranspiration anomalies, highlights that agricultural drought assessments derived from meteorological proxy indicators based on potential evapotranspiration should be interpreted with care.

Our regional assessment of SM stress is subject to inaccuracies, challenges, and assumptions, which is common for these kinds of analyses. A source of the inaccuracies relates to the modeling of SM. Previous studies showed that the physically based TRAIN model was able to provide a good temporal representation of soil moisture over agricultural fields (e.g., Stork and Menzel, 2016). However, it is important to bear in mind that the studied results are regional model simulations for specific soil and land use parameterizations that can be different from the heterogeneous real world. An evaluation of the simulated hydrological fluxes with observed streamflow suggests that TRAIN provides a reasonable estimation of the water balance and its variability (Figs. 5 and S2–S5). However, there are other models, model structures, and model parameterizations to simulate soil moisture, implying a dependency between the used model (parameterization) and the results (shown in, e.g., Samaniego et al., 2018; Zink et al., 2017). The latter studies use ensembles of different models or different model parameterizations to consider model- or parameter-related uncertainties, which is beyond the scope of the current study.

Another source of inaccuracies stems from the data used to set up and force the model. A challenge was the interpolation of several different meteorological variables over a rather complex terrain which is prone to biases, especially for variables such as wind speed. Another challenge was the spatially accurate representation of the root zone soil, both in terms of the interpolation of heterogeneous soil and land use characteristics and in the parameterization of the rooting depth. The interpolation of soil and land use characteristics was based on the majority class within a 1 km grid cell. However, each grid cell can still exhibit a large variability in soil and land use characteristics, implying that the simulated SM dynamics might not be representative for the entire grid cell. The parameterization of the rooting depth of each grid cell was further based on soil characteristics, which is a procedure that is often used to parameterize regional models. However, roots do not necessarily utilize the water in the entire soil column, and rooting depth depends on other factors, such as the type of crop. For example, a soil might have a maximum rooting depth of 1 m; however, if a shallow rooting crop species is grown in this soil, roots may not have access to all water. A sensitivity analyses revealed that derived results change depending on the used parameterization. Differences in (controls on) simulated SM stress characteristics are small between a soil-based root zone and a soil-based root zone constrained to 1 m depth, implying that the latter depth constraint does not have a great impact on simulated SM stress characteristics. The differences were larger when the volume of the AWC of the root zone was constrained to a fixed value, i.e., mimicking shallow and deeper rooting crop species with lower and higher water availabilities, respectively. An option not considered was a climate-based parameterization of the root zone, which works with the hypothesis that the (catchment average) size of the AWC of Srootzone (dynamically) develops to deal with meteorological droughts of certain return periods (e.g., 10 years). Various studies show improved model performances for a selection of catchments when defining the root zones in such a way as opposed to a soil-based definition (e.g., de Boer-Euser, et al., 2016). The reason why we did not apply this parameterization in our study is that (1) we focus on annual agricultural crops that are harvested every year and, thus, do not have the opportunity to gradually adapt their root zones over time, and (2) such analyses require a study with a different scope. In the end, an accurate spatiotemporal representation of the root zone, considering the influence of soil-, climate- and crop-specific characteristics (as well as their interactions), remains an important challenge. With the sensitivity analyses, we cover four possible scenarios, but different assumptions might apply, depending on the scope of the study.

The soil-based parametrization of the root zone, the variability in soil, and land use characteristics within a single grid cell and possible biases in interpolated meteorological variables mean that results might not always be accurate for a specific grid cell or for a single agricultural field located within this grid cell. However, by analyzing a large sample of grid cells and by including a sensitivity analyses to the parameterization of the root zone, we cover a large number of combinations of root zone characteristics and climatological settings that occur within the study region (Fig. 1). Lessons learned from these large samples, for example, about the relationship between SM stress characteristics and soil or climate properties (e.g., Fig. 7; Table 1), provide insights that might be relevant for smaller (local) scales within the study region; however, this is only the case when the modeling assumptions, for example, behind the parameterizations of Srootzone, apply. Here, the most suitable assumption can vary, depending on the studied crop, for example, whether the crop being studied makes full use of all plant-available water in the root zone soil or whether the crop being studied is a shallow rooting crop that only uses of part of it.

An assumption that was made in this study relates to the definition of SM stress. We characterized periods of SM stress (absolute) rather than SM drought (anomaly). We used one fixed threshold of 30 % of the AWC to define SM stress. This threshold is in line with the indicative threshold for potential SM stress used by the German Weather Service (DWD, 2018). However, it should be noted that this threshold, and the relationship between the degree of SM stress and the amount of available water left in the root zone, varies depending on, for example, crop species, climatological conditions, and soil type (Allen et al., 1998). Notwithstanding these assumptions, we believe that, from an agricultural drought impact perspective, the used definition of SM stress is more closely related to actual water stress experienced by plants than an anomaly-based definition. This is especially so because SM anomalies can be significantly different from SM stress, and below-normal anomalies often correspond to situations with sufficient soil moisture (Fig. 10). SM stress often still relates to an anomalously low state that develops and persists during periods with below-normal precipitation (Fig. 9). However, SM stress also incorporates temporal variability, with an increased occurrence during the growing season and a limited occurrence during the non-growing season, whereas SM drought occurs equally distributed over the year (Fig. 10). Furthermore, the rareness of SM stress is affected by the plant-available water-holding capacity of the root zone soil and the climatological setting as revealed by, for example, the ranges in Fig. 10 or the difference between Fig. 10c and d, whereas this is not the case for SM drought. On the other hand, it should be noted that derived SM stress characteristics are more sensitive to modeling assumptions and uncertainties. SM stress characteristics derived from simulations using different parameterizations of the root zone reveal more variation (Fig. 10) but, therefore, also a higher degree of disagreement on whether SM stress was reached (Fig. S11a). Soil moisture anomalies show a higher degree of agreement, i.e., results are more robust and much less sensitive to the (uncertainties in) parameterization of the root zone (Fig. S11b). Overall, the definition of SM stress used in this study might be applicable in other regions or for other research purposes, for example, those that aim to investigate changes in agricultural drought vulnerability under climate change.

Meteorological droughts cause soil moisture levels to decline. Diminished root zone soil moisture can largely affect agricultural productivity, as crops might experience soil moisture stress. In this study, we investigated the characteristics of simulated past soil moisture stress events across the agricultural regions of southwestern Germany and their relationship with soil and climate variables. The total agricultural area that reached soil moisture stress conditions was found to vary strongly among the years and was highest in 2003 and 2018. In terms of the development time, 2003 was not much different from 2018. In both years, development time varied from as little as 10 d to over 4 months. What made 2018 distinctively different from 2003 was the generally longer total time and maximum duration of simulated soil moisture stress, highlighting the extraordinary severity of the most recent event studied.

Both the occurrence and development time of soil moisture stress were found to be strongly related to the available water-holding capacity of the root zone and not so much to the climatological setting. That is, when we assume roots can make use of all available water in the root zone column by being either constrained or not constrained at a depth of 1 m. When we assume root zones of fixed sizes, the influence of the climatological setting increases, yet the difference between a shallower rooting crop (lower AWC) and a deeper rooting crop (higher AWC) remains. Thus, the above findings stress the importance of considering differences in root zone storage characteristics for agricultural drought assessments and monitoring and early warning, independent of whether these differences in storage are related to the difference in soil or crop species. Nonetheless, a major challenge remains with respect to the accurate spatiotemporal characterization of the root zone soil that considers (the interactions between) soil, climatological, meteorological, and crop-specific factors.

Results of this study further imply that below-normal precipitation was the most important reason for soil moisture stress to develop. However, the often above-normal anomalies of temperature and, especially, simulated evapotranspiration during development suggest an augmenting effect of these variables. During soil moisture stress, temperature anomalies were found to often be above normal, which contradicted with the often below-normal simulated evapotranspiration anomalies. These contrasting anomalies of temperature and evapotranspiration imply that agricultural drought assessments derived from meteorological proxies based on potential evapotranspiration should be interpreted with care. The same is the case for agricultural assessments based on soil moisture anomalies, as below normal anomalies were found to not necessarily correspond to a situation of soil moisture stress, especially for periods outside the growing season. In addition, the sensitivity analyses revealed that SM drought characteristics, and controls on these characteristics, can differ significantly from (controls on) SM stress characteristics. Overall, the approach presented in this study of directly characterizing simulated soil moisture stress events for agricultural drought assessments might, in some cases, be a suitable alternative to approaches based on meteorological proxies or soil moisture anomalies.