Our analysis focuses on SESA (Fig. 1a) and, more specifically, on the region known as the core crop region bounded by 36–29∘ S and 65–59∘ W (dark brown rectangle in Fig. 1b–d), where most (about 80 %) of the Argentine production of wheat, corn, and soybean are found. The dark green color points out the regions in which each crop's production is more intense. Values of production for each crop are also shown in Fig. 1b–d. This region includes, almost entirely, the provinces of Córdoba and Santa Fe and part of the provinces of Entre Ríos, Buenos Aires, La Pampa, Santiago del Estero, and Corrientes (provinces are identified in Fig. 1a).

Wheat, corn, and soybean have different life cycles that last about 7 to 9 months (lower bars in Fig. 1b–d). Wheat is planted during late austral fall or early winter (May–June) and harvested during summer. It is most sensitive to water availability during its growth period in spring (October–November). Planting of corn and soybean occurs in austral spring (October–December), and both are harvested in the fall. Their most sensitive period takes place during the summer, specifically December–January for corn and January–February for soybeans. Therefore, a year's crop production could be largely impacted even if a dry period lasting just 1 month or even less occurs during the critical growth period. While these are the crops' traditional cycles, it has become possible to have double-cropping at specific locations, i.e., have two crops with different cycles in 1 year by making the second cycle shorter. Crop rotation – which also has the advantage of reducing the need for fertilizers – introduces the planting of corn or soybean right after the wheat harvest, and the second crop results in a smaller, but still profitable, production (Senigagliese, 2004).

This analysis of droughts focuses on precipitation (P), soil moisture (SM), and their derived standardized indices. Series of P and SM were turned into anomalies by removing their mean annual cycle. The monthly precipitation data cover 40 years, from January 1979 to December 2018, and were developed by the National Centers for Environmental Prediction (NCEP) Climate Prediction Center (CPC). It consists of in situ observations spatially interpolated to a regular 0.5∘×0.5∘ latitude–longitude grid cell (Chen et al., 2008). This product has been used as a benchmark for model evaluation in South America (Silva et al., 2011). In the absence of soil moisture observations, we employ products obtained from the Global Land Data Assimilation System (GLDAS; Rodell et al., 2004; Meng et al., 2012; Beaudoing and Rodell, 2019, 2020). GLDAS uses several land surface models to derive soil moisture from the surface water and energy balances forced by observations. The Noah model is considered here. It has four soil layers (0–10, 10–40, 40–100, and 100–200 cm) totaling 2 m depth (Rodell et al., 2004). The total soil moisture in a column is the sum of the content in the four layers. The soil moisture data set consists of monthly values at a spatial resolution of 0.25∘×0.25∘ over the same period of analysis as precipitation. Evaluation of GLDAS soil moisture products in the Humid Pampas was recently performed by Grings et al. (2015) and Spennemann et al. (2015, 2020). According to Grings et al. (2015), GLDAS is a good soil moisture benchmark in the Pampas region since it achieved the highest correlation (r>0.80) with in situ soil moisture measurements. Spennemann et al. (2015, 2020) also reported that GLDAS reproduces soil moisture observational patterns satisfactorily. They also found that GLDAS products can be used as soil monitoring indices in agricultural production management.

Several drought indices have been defined to characterize droughts. The World Meteorological Organization (WMO) recommends selecting a particular index, depending on the data available and ease of application (Byakatonda et al., 2018). It also recognizes the Standardized Precipitation Index (SPI) advantages for studying meteorological droughts (Hayes et al., 2011). SPI represents a standardized precipitation anomaly and stands among the most used indices to quantify and monitor droughts (Keyantash and Dracup, 2002; Mishra et al., 2009; Hayes et al., 2011). In addition to the SPI or any precipitation index, other environmental variables may need to be included, depending on the study region's characteristics and climate (e.g., Byakatonda et al., 2018). Soil moisture is particularly useful in agricultural areas, as they reflect the water content in the upper part of the soil where crops grow. Then, we used the SPI (McKee et al., 1993, 1995) for precipitation and the Standardized Soil Moisture Index (SSI; Hao and AghaKouchak, 2014; Hao et al., 2014) for soil moisture.

SPI and SSI were computed, following the approaches of Hao and AghaKouchak (2014) and Farahmand and AghaKouchak (2015), which allows us to obtain nonparametric standardized indices for precipitation and soil moisture. A growing body of research attests that a nonparametric approach is better than a parametric one for studies of droughts. Unlike parametric approaches, nonparametric methods do not rely on any theoretical distribution. Parametric and nonparametric (empirical) probability density functions tend to have differences in the tails, where the parametric distribution may not be a good fit (Farahmand and AghaKouchak, 2015). A comparison of parametric and nonparametric estimates of SPI (Soláková et al., 2014) found that differences can be significant in terms of drought severity and not as much in terms of duration. According to Mallenahalli (2020), the nonparametric SPI can better categorize the drought classes, representing the extent of dryness and normality conditions better than parametric approaches. For these reasons, we adopted a nonparametric methodology that uses an empirical function (Gringorten, 1963; Farahmand and AghaKouchak, 2015). This method circumvents the use of theoretical functions, avoids issues with zero precipitation values, and is suitable in precipitation and soil moisture studies. Moreover, it is also an opportunity to provide a different approach to the index construction that has not been tested yet in the region.

The SPI and SSI were calculated following a nonexceedance empirical probability function for extreme events (Gringorten, 1963). 1pxi=i-0.44n+0.12. Equation (1) represents the associated probability of nonexceedance for the ith element of the series, where x is either P or SM, i is the rank of nonzero values of the sample, and n is the size of the sample. This probability is then transformed into standardized indices (SIs), applying the inverse of the standard normal distribution function (∅) to the results of p(xi) (Farahmand and AghaKouchak, 2015) as follows: 2SI=∅-1pxi. This approach is applied to precipitation and soil moisture to create the corresponding indices of SPI and SSI.

Here, SPI and SSI are defined for two different timescales, 3 and 6 months, to facilitate the monitoring of meteorological and agricultural droughts. SPI3 (values for SPI at the 3-month scale) reflect wet or dry conditions for short and medium time ranges and estimate the climate conditions at critical stages of the crops' growth. SPI6 (values for SPI at the 6-month scale) provides information between seasons and can be a reference point for the start of the anomalous behavior of flows and reservoir levels, which usually have larger timescales than precipitation itself. As defined by soil moisture content, the less variable SSI index can identify and monitor seasonal agricultural droughts more directly (Hao et al., 2014). While SPI is widely used for drought monitoring and prediction, SSI produces a reliable representation of drought persistence (Farahmand and AghaKouchak, 2015). Additionally, we determined SPI1 (values for SPI at 1-month scale) as an index with higher variability for comparison in Figs. 2 and 4.

The time series of wheat, maize, and soybean yields cover the seasons from 1979–1980 to 2018–2019 for the provinces of Santa Fe and Córdoba, covering most of the core crop region (see Fig. 1d). Data are available from the Ministry of Agriculture, Livestock, and Fisheries (MAGyP, 2020).

A drought is a sustained period of below-normal water availability (Tallaksen and Van Lanen, 2004; Van Loon, 2015). Droughts are identified as being meteorological droughts when there is a precipitation deficit over a period of time. A continued precipitation deficit can lead to a scarcity of soil moisture that does not meet the plants' water demand. In this case, the drought is called an agricultural drought. This study focuses on meteorological and agricultural droughts and their impacts on crop yields within the region of interest.

For this analysis, droughts are defined as those periods in which SPI or SSI depart from the mean by at least minus 1 standard deviation. Drought events below that threshold range from moderate to extreme droughts (McKee et al., 1995). Weaker or milder droughts were estimated using a threshold of one-half the standard deviation. Droughts persist as long as they continue to exceed the threshold. We also examined different periods, starting at 1 month and continuing to more prolonged periods. We included the 1-month results in Fig. 6 for completeness, but most of our analysis and conclusions are based on longer periods.

Low-frequency variability modes in the drought indices were identified using a singular spectrum analysis (SSA) approach (Ghil et al., 2002; Wilks, 2006). SSA decomposes the time series in temporal–empirical orthogonal functions (T–EOFs) and temporal–principal components (T-PCs) and facilitates the interpretation of processes related to interannual modes of climate variability and the cases of drought. The SSA was used to identify the nonlinear trends and interannual quasi-oscillatory modes. Following Von Storch and Navarra (1995), we choose a window length (W) of 120 months, as it does not exceed one-third of the length of the whole period and resolves quasi-periods in the interannual band (1 year < T < 10 years).

Dry events were analyzed by studying their frequency, duration, severity, and areal extent. Drought frequency (F) indicates the percentage of droughts during the time of analysis, with respect to the total possible cases, in scales of months or the critical month periods for crop growth. Therefore, the frequency analysis is performed at monthly steps for the whole period and for each crop's critical period. The frequency distribution of drought events also depends on the duration (D); that is, the length in time that an index remains below the threshold until it reaches the threshold again. The drought magnitude is defined as the average deficit of an index during the duration of the event. The drought severity (S) is equivalent to the accumulated water deficit on the event (Dracup et al., 1980), and it is defined as the magnitude times the duration, i.e., S=D×M (see Yevjevich, 1967; Keyantash and Dracup, 2002). The properties of frequency, duration, and severity of droughts are unique to the thresholds that define them. The analysis is completed with the examination of the droughts' areal extent (A).

An analysis of the relation between drought occurrence and annual crop yields of wheat, corn, and soybean is performed for Santa Fe and Córdoba. First, crop yield data were detrended to remove the increasing yields resulting from technological and genetic improvements. The detrended series can be better related to drought characteristics. Then, we examined the Pearson correlation coefficients between the annual detrended crop yields and the drought indices for the critical crop months (October–November, ON, for wheat; December–January, DJ, for corn; January–February, JF, for soybean).

Due to the importance for regional economies, it is always of interest to stress the droughts' negative impact on crops. Changes in crop yield, defined as crop production per unit area, expressed as kilograms per hectare (kg ha-1), reflect not only the effects of climate variability but also nonclimatic factors like technological and biotechnological advances (including seed quality, different use of fertilizers, and sawing or harvesting dates), usually in the form of a positive nonlinear trend. This can be seen in Fig. 8, which presents the 1979–2018 area-averaged time series of corn, wheat, and soybean yields for the provinces of Santa Fe and Córdoba. The wheat and soybean trends show a significant change around the mid-1990s, whereas, for corn, a change occurred earlier in the late 1980s. On average, wheat and soybean yields increased from 1000 to 3000 kg ha-1 (Fig. 8a and c), while corn yield increased from 3000 to almost 8000 kg ha-1 (Fig. 8b). As stated, at least most of the increases may be due to advances in the production process. These trends should be removed when examining the crop yield variability and its relation to droughts. Crop yield time series were fitted with a cubic polynomial trend (see dotted lines in Fig. 8a–c). Then, the trends were subtracted from the original series, leaving the shorter-term variability (see Fig. 8d–f). Detrended time series of one or more crop yields exhibit the largest negative anomalies concurrently with the most severe droughts identified by SPI6 and soil moisture anomalies (Fig. 4c and d) recorded in 1988–1989, 1995–1996, 2008–2009, and 2017–2018 (Fig. 8d–f). Not all crops are affected equally by drought, as slight differences in the onset of the drought and the crops' critical growth periods may affect them differently.

Table 3 presents the correlation coefficients between detrended crop yields and SPI/SSI during the crops' critical growth periods. The results suggest a direct relation between summer crops (corn and soybean) and deficits in precipitation and soil moisture during both crops' critical growth periods. Of the three crops, wheat yields have the lowest correlations with the indices. Table 3 also shows that the shorter-scale indicators (SPI3 and SSI3) achieve a better correlation with crop yields than the longer-scale indicators (SPI6 and SSI6), making them a good descriptor of crop yield losses.

Figure 8d–f shows that large negative anomalies of detrended corn and soybean yields (Fig. 8e and f) are consistent with the lowest values of SPI3 during the drought events in 1988–1989, 1995–1996, and 2017–2018 (Fig. 7). Severity, derived from SSI3 and SSI6, reached extreme negative values around -8 during the crops' critical growth period. Similarly, a considerable reduction in wheat production in 2009 is related to large (negative) drought severity values, particularly for the Santa Fe province (see the blue line in Fig. 8d). Although there are some differences in the anomaly values, the detrended series of corn and soybean yield are in phase and exhibit a close resemblance (Fig. 8e and f). Increases and decreases in production take place nearly simultaneously, unlike the behavior of wheat (Fig. 8d). Both crops present significant yield losses approximately in the years of major droughts (see Fig. 7). This could be related to the 1-month overlap during both sensitive periods, as the two crops have January as a common month during their critical growth in the summer.

The detrended time series (Fig. 8d–f) show declines in production due to major drought events. The losses in production may reach up to 1500 kg ha-1 for corn and between 500 and 1000 kg ha-1 for wheat and soybean. Correlations between SPI3 and the different crop yields (Table 3) suggest that corn and soybean are more sensitive to water availability. Figure 8e and f show that SPI3 values and crop production have a better representation with a detrended series of corn and soybean yields. Notably, a good fit is observed for the most severe drought events in 1988–1989, 1995–1996, and 2017–2018. Conversely, from 1998 to 2007, no severe events occurred (see Fig. 7a and b). For those years, crop production ranges from neutral to positive values, with a maximum for corn in Córdoba. However, in this period without severe droughts, a drop in the Córdoba' soybean yield occurred in 2004 that is correlated with a moderate drought represented by an SPI close to -1.2 during the soybean's critical period. This could imply that SPI3 may also be used as a drought indicator during moderate drought events. We have not discussed the compound effect of water scarcity and heat waves, which may intensify crop yield losses. Llano and Vargas (2016) suggested that the compound event of precipitation and maximum temperature in the corn critical growth period have the greatest influences on crop production in central and eastern Argentina.