Drought is a complex natural hazard that can affect all climates and produce devastating environmental, economic and social impacts, such as damage to wildlife habitat, diminishment of crop growth or mass migration. While these drought effects have been well documented [1
], there is not a single and precise definition of drought. It varies among climatic zones and is context-dependent. For this reason, drought assessments rely on the duration, severity and impact area. Droughts are usually classified into four types according to meteorological, agricultural, hydrological and socio-economic criteria [3
Among a wide variety of drought indices, most are climate-based, such as the Standardized Precipitation Index (SPI). The SPI is a meteorological index that compares the precipitation with its multiyear average [11
]. The Palmer Drought Severity Index (PDSI) is also a commonly used meteorological index, empirically derived from temperature and precipitation data. The PDSI evaluates the prolonged abnormally wet or dry periods using a two-layer soil model [12
]. The PDSI is an effective long-term index, but it does not consider different climate types [13
]. To improve this issue, a self-calibrating PDSI (sc-PDSI) was proposed [14
]. The Atmospheric Water Deficit (AWD) is also a meteorological index that uses the simple water balance between evapotranspiration (ET) and precipitation [15
To avoid the PDSI limitations [17
], the Crop Moisture Index (CMI) was developed. It is an agricultural index that is computed from ET deficits and moisture excess, based on the soil water holding capacity and a subset of calculations required to obtain the PDSI [18
]. As an advantage, the CMI responds faster than the PDSI to changing conditions [19
]. The two-layer soil model used by Palmer for the PDSI also computes, as an intermediate parameter, the moisture anomaly index (Z-index). The Palmer Z-index is a short-term agricultural index that measures the surface moisture anomaly of the current month without considering antecedent conditions [12
Similar to the PDSI, the Palmer Hydrological Drought Index (PHDI) has been developed. The PHDI uses the same two-layer soil model as the PDSI, but it applies a stricter criterion for determining the ends of drought or wet spells, which results in a longer-term index [12
]. In addition, the Soil Water Supply Index (SWSI) was developed as a more appropriate indicator of water availability than the PDSI, especially in regions where rainfall is not the main source of water. Thus, the SWSI is a hydrological index based on the probability distributions of snow runoff, precipitation, reservoir supply, and streamflow on the total surface water resources of the basin [21
In the literature, there are several drought indicators, i.e., variables or parameters used to describe drought conditions, such as precipitation, temperature, ET, soil moisture (SM) and streamflow [22
]. It can be assumed that SM is the governing variable for agricultural drought since this type of drought is considered to begin when the SM availability to plants drops to such a level that it adversely affects the crop yield [7
]. In addition, SM constrains soil evaporation and plant transpiration in several regions of the world, and it is a storage component for precipitation and radiation anomalies [23
]. The soil profile acts as a filter between incoming water and processes that remove water from the hydrological system [24
]. Therefore, the SM of the root-zone reflects antecedent conditions and recent precipitation and indicates potential and available water storage [25
]. A variety of SM-based agricultural drought indices have been developed over the last decade using SM simulations from models, in situ SM measurements or remotely sensed SM observations.
Some agricultural drought indices have been derived from the probability distributions fitted to a modeled SM. The Variable Infiltration Capacity (VIC) drought index was obtained from the modeled SM through the VIC land surface model [26
]. Later, the Soil Moisture Deficit Index (SMDI) was developed from SM anomalies. In this case, the SM was obtained from the Soil Water Assessment Tool (SWAT) hydrological model [27
]. The Normalized Soil Moisture (NSM) index was proposed using the SM simulated by the Tiled European Centre for Medium-Range Weather Forecasts (ECMWF) Scheme for Surface Exchanges over Land (TESSEL) model [28
]. More recently, the Standardized Soil moisture Index (SSI) and the Multivariate Standardized Drought Index (MDSI) were developed from the modeled SM based on a one-layer water-budget model [29
All of the previous agricultural drought indices were estimated from hydrological modeling instead of direct root-zone SM measurements. In this regard, the Water Deficit Index (WDI) was introduced as an appropriate way to express SM in terms relative to available water content [30
]. Later, the Soil Moisture Index (SMI) was developed from in situ SM data provided by the Automated Weather Data Network (AWDN) and estimated field capacity (FC) and wilting point (WP) water contents [31
]. Similarly, the Soil Water Deficit Index (SWDI), which is based on the WDI, was proposed using in situ SM data provided by the Soil Moisture Measurements Stations Network of the University of Salamanca (REMEDHUS), and the FC and WP were calculated from water retention curves [33
During the last decade, surface SM estimations from remote sensing observations have experienced great progress [34
]. Currently, there are two L-band missions in orbit that are specifically devoted to measuring global surface SM [37
]. The first mission was launched in November 2009 and is the Soil Moisture and Ocean Salinity (SMOS) satellite from the European Space Agency (ESA) [41
]. The second mission was launched in January 2015 and is the Soil Moisture Active Passive (SMAP) satellite from the National Aeronautics and Space Administration (NASA) [42
]. In addition, the ESA’s Climate Change Initiative (CCI) generated an SM database using data from passive and active microwave satellite sensors from 1978 to 2015 [43
]. These advances allow for the use of satellite surface SM observations for long-term drought monitoring [44
Preliminary research was performed to analyze the feasibility of using surface SMOS SM data to determine anomalous SM conditions [45
]. In this line, the SWDI was applied to SMOS SM data using different methods to estimate FC and WP water contents [46
]. The SMOS SM data have also been employed to derive the Soil Moisture Agricultural Drought Index (SMADI) together with the land surface temperature (LST) and the Normalized Difference Vegetation Index (NDVI) from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor [47
]. The historical CCI SM data were used to compute the Empirical Standardized Soil Moisture Index (ESSMI) [49
]. Alternatively, the Soil Wetness Deficit Index (SWetDI) was developed from the Soil Wetness Index (SWetI) anomaly [50
]. In this case, the SWetI played the role of the SM data instead of surface SM observations. The SWetI can be obtained from MODIS LST and NDVI using the LST−NDVI triangle concept to estimate the dry and wet edges [51
] following previous interpretations of the LST-NDVI space [52
]. The quantitative determination of these edges is key for an accurate estimation of ET or SM [54
]. In addition, the Centre Aval de Traitement des Données SMOS (CATDS) provides the SMOS Drought Index (SDI) at 25 km spatial resolution by including SMOS and MODIS observations into a hydrological model at the root-zone [56
In drought studies, and particularly in agricultural drought monitoring, the spatial resolution of the data is paramount for the efficient management of drought risks and impacts and water irrigation planning. However, both SMOS and SMAP SM radiometer observations have coarse spatial resolutions (~40 km) that are suitable for global-scale applications [34
] but are limited for local and regional drought assessments. During the last several years, there has been a growing interest in developing techniques to enhance the spatial resolution of SMOS and SMAP SM observations. For SMOS, pixel disaggregation approaches are usually based on the synergy of passive microwave and optical visible/infrared (VIS/IR) observations [57
]. Currently, the SMOS Barcelona Expert Centre (BEC) Level 4 (L4) SM maps at 1 km spatial resolution over the Iberian Peninsula are freely distributed (http://bec.icm.csic.es
). The downscaling algorithm is based in the LST-NDVI space and integrates SMOS, MODIS and Era Retrospective Analysis (ERA)-Interim data [59
]. An improved downscaling algorithm to obtain those SM maps at 1 km with no size or location limitations is being developed [65
The goal of this study was to obtain a comprehensive understanding of the spatiotemporal behavior of four innovative agricultural drought indices. To do so, the temporal and spatial performances of the indices were evaluated. This is of special interest in rainfed agro-ecosystems worldwide where the shortages of available water for plant growth directly affect agricultural activities, and these shortages can even induce social and economic conflicts, especially in developing countries. The SM-based indices that were analyzed are SWDI, SMADI, SMDI and SWetDI. All of them use SM (or a surrogate of it) as the main drought indicator. These two facts highlight the novelty of this study: first, the use of the SM variable, which is relatively new in drought studies; and, second, the agricultural drought perspective instead of the more common meteorological approach. The indices were computed over northwest Spain in the Castilla y León region on a weekly basis from June 2010 to December 2016 using the SMOS BEC L4 SM and/or MODIS reflectance and LST data at 1 km spatial resolution. In the case of SWDI, two different approaches to estimate the FC and WP were separately assessed. The temporal dynamics of the indices were compared to two well-known agricultural drought indices: the AWD and the CMI. The spatial distribution of each index and its intensity were assessed by comparing the estimated drought index maps under dry and wet SM conditions at the beginning of the growing season. This research could help enlighten the choice of the more appropriate drought index for developing an accurate agricultural drought monitoring system.
During the last decade, the great importance of SM in agricultural drought monitoring led to the development of a variety of agricultural drought indices that mainly use this variable as drought indicator under different points of view. In this study, the performance of four innovative satellite SM-based agricultural drought indices was evaluated over a large agricultural area in northwest Spain. The indices were weekly estimated from June 2010 to December 2016 using the SMOS BEC L4 SM and/or the MODIS SR and LST data at 1 km spatial resolution. In the case of the SWDI, two approaches (SWDIRawls and SWDICYL) that were obtained from different PTFs were investigated separately. A temporal analysis was conducted by comparing these indices with two well-known agricultural drought indices: the AWD and the CMI. In addition, the spatial patterns of the estimated drought index maps were also compared under dry and wet SM conditions at the beginning of the growing season.
The SWDIRawls, SWDICYL and SMADI adequately reproduced the seasonality of the AWD and CMI. Conversely, this seasonality did not appear in the SMDI and SWetDI time series. On the one hand, the length of the available SM data, or its surrogate, was considered in all SM-based drought indices computed, except in the SWDI. This issue had a noticeable influence on the SMDI computation. On the other hand, the size of the study area influenced the SWetI estimation and, consequently, the SWetDI computation. Nevertheless, the SMDI trend was capable of capturing the very intense meteorological drought of 2012, which was not captured by the SWetDI trend.
In general, the SWDIRawls worked equally well as the SWDICYL. Therefore, the SWDI could be calculated in any region with available soil databases by applying suitable standards and adequately tested PTFs. However, both SWDI approaches were sensitive to the accuracy of the FC and WP estimations, which, in turn, were sensitive to the robustness of the soil database, especially when there was a very low SM content. By contrast, the SMADI was not directly affected by very low SM, because it includes additional LST and NDVI information.
In comparison with the AWD, both SWDI approaches obtained the highest correlation (R ≈ 0.68–0.80), followed by SMADI (R ≈ −0.60 to 0.83). At the same time, very low and non-significant correlations were found with the SMDI (R ≤ 0.3) and the SWetDI, respectively. Similar results were obtained when the drought indices were compared with the CMI (R ≈ 0.48–0.70 for both SWDI approaches, R ≈ −0.32 to −0.69 for SMADI, R ≈ 0.39 to 0.44 for SMDI and R ≤ 0.28 for SWetDI). Both SWDI approaches overestimated the drought weeks compared to the AWD and the CMI. The SMADI, the SMDI and the SWetDI underestimated the drought weeks compared to the AWD, whereas they overestimated the drought weeks compared to the CMI. In terms of similarity, the SWDI agreed with the AWD, and the SMADI, SMDI and SWetDI agreed with the CMI.
When analyzing the estimated drought index maps, a spatial pattern with two distinctive soil property areas was found in the SWDI maps, which displayed severe and extreme drought in fine-textured soils. The SMADI only estimated a few scattered pixels with mild drought under very low SM conditions, since its intensity was modulated by the temperature and vegetation conditions. No distinctive patterns were observed in the SMDI and the SWetDI maps.
The estimation of the soil water content from the optical VIS/IR observations through the LST-NDVI triangle does not always capture the same information as the passive microwave SM estimations. Then, the use of the SM variable itself is recommended. In addition, the absolute SM is preferred as agricultural drought indicator over the SM anomalies, since SM anomalies removed the inherent seasonality of the soil water content. Therefore, the SWDI and the SMADI would be good options for developing a feasible tool dedicated to water resource management and drought monitoring in many cropland regions worldwide.