Generally, drought is classified into meteorological drought, agricultural drought, hydrological drought, and socio-economic drought. Meteorological drought refers to the water deficit caused by an imbalance between precipitation and evaporation. Agricultural drought reflects the extent to which soil moisture is lower than the least requirement of plants by analyzing the characteristics of soil moisture and morphology of plants during growth. Hydrological drought occurs when river flow is lower than the normal value or when the water level of an aquifer decreases; and socio-economic drought is the phenomenon in which production and consumption are affected by the lack of water in both the natural system and human socio-economic system (Chen et al., 2009). Although the definitions of the four types of droughts are different, they are all water-deficit phenomena caused by the lack of precipitation, and they are all connected (Figure 1). When precipitation decreases, meteorological drought occurs first, followed by agricultural drought and hydrological drought, which gradually occur because of continuous water evaporation. Since agricultural drought is mainly concerned with water deficit in crops because of a reduction in water supply in the soil, loss of soil moisture caused by the decreases in precipitation is the earliest phenomenon. Because of transpiration, water in crops cannot meet the basic needs for physiological activities, and crop growth is suppressed, resulting in a reduction in crop yield or even failure. The influence of drought on different stages of plant growth is significantly different. Hydrological drought mainly causes a reduction in water resources in rivers and reservoirs and a decline in groundwater levels. Socio-economic drought is triggered when agricultural and hydrological droughts develop to a certain degree. Therefore, here is a simple understanding of the connection between all types of droughts: agricultural drought and hydrological drought refer to the influence of meteorological drought on agriculture and the hydrological system separately, and socio-economic drought refers to the influence of meteorological drought on the socio-economic system (Christopher et al., 2012; Zhang et al., 2014).

Drought events are normally characterized by drought indices because the phenomenon is very complicated and the time, development process, and scope of influence are difficult to observe directly (Dai, 2011; Heim et al., 2002; Solomon et al., 2007). Currently, there are hundreds of indices to characterize drought, and they can be divided into site-based and remote sensing-based indices (Heim et al., 2002). Site-based indices include the standardized precipitation index (SPI), Palmer drought severity index (PDSI), and crop moisture index (CMI). Remote sensing-based indices are divided into indices based on bare surface (including thermal inertia and microwave moisture inversion) and indices based on vegetation cover. Vegetation-based drought monitoring indices can be further classified into crop morphological indices (like condition vegetation index and normalized vegetation index), crop physiological indices (like canopy temperature and canopy moisture content), and crop comprehensive indices (like vegetation supply water index and temperature-vegetation- drought index) (Figure 2 and Table 1).

2.2.1 Site-based drought monitoring

Research on drought monitoring initiated in the US in the early 20th century, and most early indices only took precipitation into consideration (Gibbs, 1967; Henry, 1906; Kincer, 1919; Marcovitch, 1930; McGuire et al., 1957; McQuigg, 1954; Munger, 1916; Van Rooy, 1965) until Palmer raised the concept of the current climate suitable precipitation (Palmer, 1965) and proposed PDSI in 1965. This index became a milestone for drought monitoring and was used both in America and other parts of the world and by governments and scholars as a drought monitoring tool. Palmer took the water demand of crops into consideration and developed CMI, which was broadly applied for agricultural drought monitoring (Palmer, 1968). Scholars gradually understood the limitations of PDSI (Alley, 1984; Heddinghaus et al., 1991), and Wells (2004) reported self-calibrated PDSI to overcome the limitations. The biggest advantage of this improved index is that it decides different calibration parameters according to the local climate characteristics and therefore improves the ability of PDSI to regions. However, PDSI still had limitations because of the fixed time scale. In terms of agricultural drought monitoring, Shafer (1982) and Jackson (1988) proposed the surface water supply index (SWSI) and crop water stress index (CWSI), respectively, after comprehensive consideration of surface water supply and crop water demand, and they both worked well. In 1993, McKee (1993) found that the observed precipitation has a skewed distribution rather than a normal distribution and proposed SPI; SPI has become one of the most popular indices because the calculation is simple and can achieve multi-scale monitoring for different types of drought. However, since it only considers precipitation and neglects the influence of evaporation on drought, the method was also incomplete. To integrate the effects of precipitation and evaporation on drought, Vicente-Serrano (2010) established the standardized precipitation and evaporation index (SPEI), which had the multi-scale advantage of SPI and the advantage of considering evaporation of PDSI. SPEI has become one of the ideal drought monitoring tools. Then, Vicente-Serrano (2012) compared the performances of SPI, SPEI, and PDSI with respect to global drought monitoring and found that SPI and SPEI were better than PDSI for hydrological and agricultural drought monitoring and that SPEI was excellent for monitoring summer drought. A recent study established the standardized relative humidity index (SRHI) by applying relative humidity data, and this index can detect the onset start of a drought earlier than SPI and is considered an ideal index for drought warning (Farahmand et al., 2015).

In China, scholars made considerable efforts to build drought monitoring indices continuously and have tried to integrate various meteorological indices to improve monitoring ability. Zhang (1998) proposed the comprehensive index (CI), which made weighted summation of the standardized precipitation index and relative humidity index, and this index is widely used for drought monitoring in meteorological departments of China. Wang (2007) proposed the K index, which is defined as the ratio of the relative variability in seasonal precipitation and the relative variability in evaporation; it is suitable for monitoring meteorological and agricultural droughts. In summary, the above review indicated that site-based drought monitoring indices have been developed over a long period, and it has become the main method for drought monitoring. In terms of data source, site-based indices mainly rely on the data records of meteorological stations. However, uncertainties still exist with respect to observed datasets, including uneven distribution in space and inconformity in time-series induced by site migration. Although a series of methods have been implemented to enhance observation station network density, such as addition of automatic weather stations, and develop data homogenization methods to correct abnormal sequences caused by non-climatic factors, there are few data records of new time series data and a shortage of stations in key ecological regions, especially in agricultural ecosystems.

2.2.2 Remote sensing-based drought monitoring

Since agricultural drought is closely connected to soil moisture and crop water deficit, remote sensing of water inversion in soil and vegetation is an effective way for large-scale agricultural monitoring. Data assimilation methods are generally used to estimate soil moisture (Kumar et al., 2014). Among these methods, thermal inertia models of different soil textures, established by Chen (1999), improved the accuracy of water inversion by introducing parameters of topography and wind field; however, the parameters are difficult to determine in practice. Then, Zhang (2001) integrated the thermal inertia model, heat balance model, and temperature difference model by using temperature differences of the soil and leaves facing the sun and away from the sun, developing a new method of soil moisture inversion on the basis of multi-angled remote sensing data. In addition, by applying the improved IEM model, Rajat Bindlish (2006) obtained an inversion result whose correlation with actual soil moisture was 0.95. Although microwave remote sensing can work continuously without being influenced by clouds, it is only capable of inverting the soil moisture of the surface (2-5 cm), while crop roots are usually 10-20 cm under the surface. Hence, water stress in crops cannot be examined accurately, and the result is highly uncertain (Chen et al., 2012). However, accurate estimation of soil moisture at different depths is very important, since it is a key parameter for agricultural drought monitoring. Therefore, despite the limitations of the applying remote sensing method to agricultural monitoring, further studies are still required; the microwave inversion results should be coupled with the terrestrial surface model, and field survey data should be collected to increase inversion accuracy and depth (AghaKouchak et al., 2015).

With respect to crop water deficit, CWSI was developed by analyzing the empirical relationship between air vapor and temperature differences of the canopy and air (Idso et al., 1981). Later, Moran (1994) developed the water deficit index (WDI) on the basis of the two-layer model of the energy balance model, while Gao (1996) proposed the normalized difference of water index (NDWI). To eliminate the influence of both Normalized Difference Vegetation Index (NDVI) spatial variation and other parameters of geography and ecosystem, Kogan (1995) proposed the vegetation condition index (VCI) for drought monitoring, and then Wang (2003) developed the vegetation temperature condition index (VTCI) in 2003. On basis of this, Kogan proposed the vegetation health index (VHI) by linear integration of TCI and VCI (Boken et al., 2004; Kogan et al., 2013), which was proven to be effective in reflecting the drought situation of crops (Mu et al., 2007). In 2004, Haboudane proposed the vegetation supply water index (VSWI), which is a relatively simple synthesis index for vegetation and temperature (Haboudane et al., 2004). Previous studies have shown that VSWI is appropriate for regions with high vegetation coverage, and it is widely applied in practice (Liu et al., 1998). Sandholt (2002) proposed the temperature vegetation drought index (TVDI) to estimate soil moisture on the basis of the relationship between land surface temperature (LST) and vegetation index (VI), which is an important method to reflect agricultural drought conditions through soil moisture monitoring. A hypothesis of TVDI is that NDVI is negatively correlated with LST (Karnieli et al., 2010). However, NDVI is negatively correlated with LST when water is the limitation factor of vegetation growth, whereas NDVI is positively correlated with LST when energy is the limiting factor of vegetation growth. Moreover, TVDI can well explain regions with drought episodes, but failure in performance of agricultural monitoring and earlier warning.

2.2.3 Comprehensive drought monitoring on the basis of both meteorological and remote sensing data

With the introduction of multi-source data, considerable efforts for drought monitoring were made to integrate meteorological and remote sensing data. By a review of recent studies on comprehensive drought monitoring indices, Hao (2015) pointed out that the US drought monitoring (USDM) model is a relatively successful model. However, application of USDM at a regional scale has many uncertainties because of its limited spatial resolution. Based on the classification and regression model, Brown (2008) proposed the vegetation drought response index (VegDRI) by combining meteorological drought indices (SPI and PDSI), the vegetation index, and DEM. The index can provide near real country-scale drought information and has become a model for comprehensive drought monitoring indices. Wu established an integrated drought monitoring model for different growth stage of crops in China (Wu et al., 2013; Wu et al., 2015). Du (2013) developed a synthesis drought index by using TRMM precipitation data, LST data, and VI data in Shandong Province, and it has achieved good results. By linear combination, Rhee (2010) proposed a drought monitoring index suitable for arid and humid regions by using a linear combination of LST, NDVI, and TRMM datasets, whereas Zhang (2013) established a meteorological drought index based on satellite-derived precipitation, AMSER-E soil moisture, and NDVI data. In addition, Mu (2013) developed a satellite-based near real drought severity index (DSI) by using land surface evapotranspiration and VI data, which could successfully detect the drought episode in Europe in 2003 and the Amazon drought in 2005 and 2010. Moreover, estimation by DSI is highly correlated with the results of PDSI at a site scale. By comparing different drought indices, Hao and AghaKouchak (2014) proposed the multi-variable drought index (MSDI), which combined precipitation and soil moisture data and is considered a valid drought monitoring index. In addition, AghaKouchak (2015) modified MSDI through introduced ensemble runoff prediction and carried out drought monitoring and early warning in East Africa. Deepthi (2015) proposed the multi-variable drought index (MDI), which comprehensively considers precipitation, runoff, evapotranspiration, and soil moisture and can simultaneously monitor meteorological, agricultural, and hydrological droughts. In addition, scholars recently attempted to assimilate historical data and real-time data to establish a data drive for near real drought monitoring indices (Shah and Mishra, 2015). On the basis of the aforementioned analysis, scholars have made considerable efforts to establish comprehensive drought monitoring indices and obtained valuable results. It should be noted that although several comprehensive drought indices have been proposed, the research is still in the infant stage. Moreover, multi-spatiotemporal scale drought monitoring indices are still required, and comprehensive drought monitoring indices have limitations. For example, although the linear combination method integrated multiple factors, it is difficult to explain the physical implication, whereas the copula function can obtain probability, facilitating risk analysis; however, it only considers the statistical characteristic of data and lacks a description of the physical process.

To better understand the development stages of studies on agricultural drought monitoring, we searched for papers in the ISI Web of Science by using the terms “agricultural drought” and “drought monitor” on June 26, 2015. Statistics showed that both the number of papers and citations of them increased exponentially (Figure 3), and we also observed an increasing trend when we searched for “agricultural drought” and “drought monitor” in CNKI. The results showed that research on agricultural drought monitoring has become the focus of scholars worldwide, and the trend is significant after 2000.

Since drought is associated with various disciplines, scholars in different areas, such as geography, ecology, meteorology, and disaster have performed systematic studies on drought. To better analyze the studies, we searched the ISI Web of Science with the term “Drought” and obtained a total of 52376 papers (period: 1990-2014, search time: 6/26/2015).Through a literature review, we found large differences in the number of papers of different research orientations. The top three were botany, agronomy, and ecological environment, and they accounted for 25.97%, 22.68%, and 20.48%, respectively, of the total number of papers; and water resources and meteorology accounted for 7.98% and 7.72%, respectively; and geography only accounted for 6.82%.This distribution indicates that vegetation, agriculture, and ecological environment are most heavily and directly impacted by drought, and they have attracted worldwide attention (Figure 4).

We observed four main characteristics of the studies on agricultural drought monitoring on the basis of the paper review and citation. First, the distribution of research orientation for drought is diverse. Majority of the papers are about the influence of drought on the ecosystem, like on vegetation growth, production, and the fluctuation in carbon storage. In agriculture, previous studies mainly focused on assessing the impact of drought on crop growth and production. The IPCC report pointed out that the production of rice, maize, and wheat in Asia has decreased in the past decades because of increasing drought, mainly reflecting climate warming, frequent ENSO, and an increase in dry days (IPCC, 2007). Geographical and disaster studies focused on the spatial and temporal distribution, mechanism, and reduction of disasters. While papers on disaster spatiotemporal distribution are comparatively abundant and there are also profound studies on the cause of drought with respect to an analysis of the climate, there are few studies on drought monitoring, which is significant for disaster reduction (Du et al., 2013; Mishra and Desai, 2005; Mishra and Singh, 2009; Mishra et al., 2009). Second, the method suggested a transformation of agricultural drought monitoring from traditional single meteorological monitoring indices to meteorology and remote sensing-integrated monitoring indices, and new data mining methods, including classification and regression tree, linear weighting, copula, and Bayes network, were introduced. For example, Brown (2008) integrated VI, meteorological drought index, and auxiliary data to develop the vegetation drought response index, which is widely used in the US for drought assessment. Rhee (2010) and Zhang (2013) discussed the performance of drought monitoring from different linear combinations of precipitation, vegetation, and LST in meteorological and agricultural droughts. Anderson (2012) proposed a drought monitoring method by combining multiple soil moisture datasets through the triple collocation analysis method, while Hao (2013) established multi-variable drought monitoring indices by using precipitation and soil moisture data. Third, with respect to the data collection method, traditional ground-based observation cannot reflect spatial characteristics because of its uneven distribution. A combination of gauged data and remote sensing data has become popular in recent studies, especially for remote sensing-based drought monitoring, because of the increase in different types of sensors and enhancement in spatial and temporal resolution. However, relatively short data records have limitations. Thus, we should highlight the construction of long-term surface parameter products by assimilation of multi-source datasets. In addition, development of satellite-based drought monitoring models are needed to enhance our ability to forecast drought, for example, the microwave remote sensed drought model (Kongoli et al., 2012; Rott et al., 2010). Fourth, in terms of drought monitoring products, several global-scale products were widely used, including the PDSI product developed by Dai (2004), SPEI products with a 0.5°×0.5° resolution developed by Vicente-Serrano (2010), and remote sensed global terrestrial drought severity index products developed by Mu (2013). Recently, Hao et al. (2014) released the global integrated drought monitoring and prediction system (GIDMaPS), including the near real-time drought monitoring model and seasonal prediction model, and it can provide global-scale meteorological drought and agricultural drought products. Together, the aforementioned drought products provide data support for global drought monitoring and assessment.

Soil moisture, a key parameter that integrates the responses of climate, soil, and vegetation to water balance and the influence of water balance on vegetation dynamics, plays a very important role in the terrestrial water cycle. Agricultural drought reflects the extent to which soil moisture is lower than crop demand water, resulting in crop wilting and even failure. Since the tolerance to soil water deficit is different among crops and regions, time lags to precipitation shortage and soil moisture deficit also differ, and the difference varies with the growing stage as well. With an increase in water shortage, evaporation decreases and the surface temperature increases, forming a positive feedback process. The precipitation model that neglects the abovementioned process cannot explain agricultural drought monitoring with respect to crop water balance. Currently, the comprehensive response process of crops to precipitation and temperature is still unclear, and we should further understand the crop drought mechanism. We should also make better use of the time lag between water deficiency and crop drought stress for early warning of drought.

In addition, agronomic parameters are the basic parameters for describing crop growth, and they are, therefore, an important aspect of research on the agricultural drought mechanism. The crop reflectance spectrum is significantly influenced by agronomic parameters, and the red edge, which is closely related to chlorophyll content, is one of the most important parameters. When crops are under water stress, the parameters values will change and further result in the displacement of the red edge. Therefore, analyzing variations in the crop reflectance spectrum by hyperspectral remote sensing provides a new angle for research on the agricultural drought mechanism. Since monitoring models can hardly detect the start and end time of agricultural drought, studies on the mechanism may be a key to solving the problem. Upon understanding the agricultural drought mechanism, we can predict the onset of drought in time and establish an agricultural drought remote sensing model on the basis of crop physiological and ecological characteristics, facilitating prompt measures for reducing drought-related losses and influences.

Agricultural drought is caused by an abnormal decrease in precipitation, and it is influenced comprehensively by land surface temperature, evapotranspiration, soil properties, and physiological or ecological characteristics of the crop. We should pay more attention to factors closely related to agricultural drought in the future, understand the interaction mechanism, screen the key influencing factors to cause variations in drought, and create agricultural drought monitoring models based on multiple geological factors. In practice, since satellite remote sensing can acquire precise spatial and temporal information on the land surface and at a large scale, it is widely used in agricultural drought monitoring. It is considered an ideal data source because it provides not only information on the land surface environment but also information on crop growth, like vegetation index, land surface temperature, and precipitation. By using remote sensing as a data source, we can integrate soil characteristics and topography features and add both environment and crop information in the agricultural monitoring model to achieve the integration of multiple geological factors for drought monitoring.

Currently, some countries have already developed agricultural drought monitoring systems at country or regional scales, and satellite data-driven comprehensive models are being established (Brown et al., 2008). However, integration of remote sensing data with meteorological data obtained by stations has not been achieved. It should be noted that even though remote sensing data can be an effective supplement for meteorological data, because they cover large regions at a high frequency, they cannot fully replace the latter. The reason is that although land surface observation stations are few and unevenly distributed, the data are precise with a long time series and the most important data source for remote sensing data validation. Therefore, it is important to build a bridge between remote sensing monitoring models and meteorological observation models, integrating the advantages of both model types. To achieve this goal, technologies and methods still need improvement. For example, we should build a more complete ground observation network, develop a perfect coupling method, and determine a uniform influencing factor framework.

The concept of scale, both temporal and spatial scales, is frequently mentioned. For agricultural drought monitoring, there are intrinsic differences among the results at different spatial and temporal scales. For example, the monitoring result of months may be different from the result of years, and at a different spatial scale, the result may also vary with the input data. Previous studies have tried to create agricultural drought monitoring models at different temporal and spatial scales and achieved some models with good validation results at a regional scale (Du et al., 2013; Rhee et al., 2010). However, most studies were limited to specific regions and time scales, and the scale of the models should be further expanded. Therefore, how to integrate the existing agricultural drought monitoring models and develop new multi-scale agricultural drought monitoring models will be a key question for establishing a drought monitoring platform with multi-spatiotemporal scales in the future.

Agricultural drought monitoring provides information on crop growth and production worldwide, and it facilitates crop production prediction as well. However, since there are differences in the developing stage, technology level, and disaster reduction ability among different countries, the request for agricultural drought monitoring information also varies. Therefore, building an agricultural drought monitoring platform at multi-spatiotemporal scales not only meets the demand of different countries but is also significant for reducing the influence of drought and improving the ability to cope it worldwide. More specifically, (1) at temporal scales, monitoring models should be able to manage early warning at 10 days, a month, a season, a year, and a decade in regular situations, whereas during emergencies, the models should be able to monitor near real-time early warning. (2) At spatial scales, the model should satisfy the needs of the world, different continents, different countries, and different regions. How to achieve a monitoring platform at multi-spatiotemporal scales is a key issue for future research on agricultural drought monitoring.

Qualitative description accounts for the most in the results of existing agricultural drought monitoring model, and the lack of quantitative monitoring and warning assessment limits the association between monitoring results and practical loss assessment. Although technological methods for drought monitoring are developing fast, attention to methods for qualitative assessment of the results is still lacking. With the rapid development of numerical simulations and further understanding of the drought mechanism of crops, the crop growth model has made great progress with respect to quantitative assessment of agro-meteorological hazards, and it is considered the core step to push agricultural drought monitoring into the stage for quantitative assessment. The advantage of the crop growth model in agro-meteorological hazard assessment is that the mechanism is clear and it can reflect the active relationship between the growing process; crop production; and temperature, precipitation, and soil moisture at every growth stage. Therefore, we should use functions of the crop growth model when creating agricultural drought monitoring models in the future.

In terms of technology, the spatial crop vulnerable model can be drawn from station-scale vulnerable models through the scale upscaling algorithm, which is a promising method for coupling an agricultural drought monitoring model at a large scale and the crop growth model; it will become the core and critical point of agricultural drought monitoring. On the basis of spatial information, especially the advantage of remote sensing technology, and by improving the integration of indices for crop physiology, morphology, and soil moisture, we can achieve the coupling of an agricultural drought monitoring model at a large scale and the crop growth model, and create a remote sensing-based agricultural drought monitoring platform with a clear physical mechanism, multiple factors, and progress and at multiple scales. This is an ideal way for realizing the goal of reducing agricultural hazard loss to the largest extent.

Remote sensing is a very important data source for agricultural drought monitoring, and with launching of satellites for different uses, abundant remote sensing data are acquired by scholars to analyze land surface processes. Compared with ground observation data, the biggest challenge for remote sensing data is the short time sequence, which can barely present the variation in drought at a long time scale. The diversified temporal and spatial scales of different data sources also limit the comprehensive use of the data. Although we have developed transforming technology for multi-scale remote sensing data, comprehensive applications have not been completely achieved, especially with respect to applications of microwave remote sensing data for the influence of drought on vegetation (Andela et al., 2013). In future, we should enhance the use of remote sensing in drought monitoring by exploring more land surface parameters and increasing the applications of remote sensing data (Rodell, 2012).Uncertainties of remote sensing data are a key problem with respect to their applications, like the difference in uniformity caused by the change in sensors. Therefore, to develop data assimilation technology, upgrading the comprehensive applications and quantitative estimation of the uncertainties of remote sensing data are important. Finally, successful launches of new types of satellites such as the Soil Moisture Active Passive is offering new opportunities for agricultural drought monitoring, hence increasing the importance of remote sensing data in agricultural drought monitoring (Figure 6).