The implementation of drought management plans contributes to reduce the wide range of adverse impacts caused by water shortage. A crucial element of the development of drought management plans is the selection of appropriate indicators and their associated thresholds to detect drought events and monitor the evolution. Drought indicators should be able to detect emerging drought processes that will lead to impacts with sufficient anticipation to allow measures to be undertaken effectively. However, in the selection of appropriate drought indicators, the connection to the final impacts is often disregarded. This paper explores the utility of remotely sensed data sets to detect early stages of drought at the river basin scale and determine how much time can be gained to inform operational land and water management practices. Six different remote sensing data sets with different spectral origins and measurement frequencies are considered, complemented by a group of classical in situ hydrologic indicators. Their predictive power to detect past drought events is tested in the Ebro Basin. Qualitative (binary information based on media records) and quantitative (crop yields) data of drought events and impacts spanning a period of 12 years are used as a benchmark in the analysis. Results show that early signs of drought impacts can be detected up to 6 months before impacts are reported in newspapers, with the best correlation–anticipation relationships for the standard precipitation index (SPI), the normalised difference vegetation index (NDVI) and evapotranspiration (ET). Soil moisture (SM) and land surface temperature (LST) offer also good anticipation but with weaker correlations, while gross primary production (GPP) presents moderate positive correlations only for some of the rain-fed areas. Although classical hydrological information from water levels and water flows provided better anticipation than remote sensing indicators in most of the areas, correlations were found to be weaker. The indicators show a consistent behaviour with respect to the different levels of crop yield in rain-fed areas among the analysed years, with SPI, NDVI and ET providing again the stronger correlations. Overall, the results confirm remote sensing products' ability to anticipate reported drought impacts and therefore appear as a useful source of information to support drought management decisions.
Drought is defined as a temporary water shortage in part caused by
anomalous climatic conditions but strongly influenced by
socioeconomic factors
The implementation of drought management plans by governing agencies can contribute to reducing the negative effects of drought by guiding decision-makers in taking appropriate mitigation actions. However, the effectiveness and cost efficiency of these actions rely on the selection of suitable indicators to monitor drought conditions and to detect events at an early stage, gaining valuable time for mitigation measures to be implemented effectively and impacts to be mitigated. Examples of actions that can be taken include retention of water; reallocation of available water resources; curtailment of current allocations; recommendations to plant less water-demanding or drought-resistant crops; or prohibition of certain water uses (e.g. watering gardens or washing cars).
Indicator systems consist of drought indices with associated
thresholds that allow classifying the event in categories of drought
severity. A classical example is the division of river flow into
several categories. When the value of the indicator crosses one of the
thresholds, managers should decide whether to activate the
corresponding responses defined in the drought management plan for
that situation. Indicators and associated thresholds should be
problem, context and user-specific
Measurements from in situ networks and from remote sensing are
complementary sources that can be used to build the system of
indicators for early detection and monitoring of drought
conditions. In situ data are generally collected at specific points
only. The advantage is the high temporal frequency of observations
and the availability of longer-term records. Remote sensing
techniques, on the other hand, offer cost-effective and spatially
continuous information over extended regions. Satellites allow drought
events to be categorised over a certain area, rather than at point
locations
Since mitigating impacts is the purpose of drought indicators included
in drought management strategies, impact data are especially suitable
as a benchmark in this case. Drought impacts, however, are difficult
to evaluate and are rarely monitored
Despite their important role in mitigation of drought impacts, the
selection and use of indicators and thresholds for decision-making
often suffers from a lack of scientific justification: only a few
studies have analysed the choice of drought indicators in relation to
drought management in practice
The Ebro Basin, with an extent of 85 600
The period 2000–2012, selected for the analysis, encompasses a wide
range of different conditions: the hydrological year 2004–2005 was
characterised as one of the most intense droughts of the record in the
Iberian Peninsula
The Confederación Hidrográfica del Ebro (CHE) is the organisation responsible for the management, regulation and conservation of water in the Ebro Basin. The basin is divided into 18 management units, each of which has a board constituted of representatives of the different water users as well as of the basin authority to coordinate the use of the hydraulic infrastructures and water resources in their area.
A drought management plan for the basin was developed in 2007 to guide
drought management actions
The north-east of the basin, where the larger irrigation districts of
the Ebro Basin are located, was selected to evaluate the set of
drought indicators against the qualitative text reports
(Fig.
The north-eastern part of the Ebro Basin with selected agricultural land cover information.
The analysis focuses on medium-resolution global remote sensing
products that are related to land surface hydrological and vegetation
growth processes. Six commonly used remote sensing parameters were
investigated: precipitation (
The Climate Hazards Group InfraRed
Precipitation with Station data (CHIRPS) is a gridded precipitation
data set based on satellite and station data, designed with the main
objective to support agricultural drought monitoring. It is a daily,
quasi-global product, with a resolution of 0.05 The MODIS (Moderate
Resolution Imaging Spectroradiometer) product MOD11A2 offers day and
night LST data sets, available at 1 GPP describes the
daily gross carbon flux as a result of the photosynthetic process and
is thus suitable to detect the effects of drought on biomass
production. The MODIS GPP product (MOD17) applies a light-use
efficiency model based on MODIS FPAR (fraction of photosynthetically
active radiation) data, meteorological data and biome-specific
parameters. The product also includes net photosynthesis (PsnNet),
which corresponds to the GPP minus the maintenance respiration for
leaves and roots. It is available at 1 The soil moisture product considered is
taken from the Soil Moisture Climate Change Initiative (CCI)
project, which is part of the ESA Programme on
Global Monitoring of Essential Climate Variables (ECV)
There are currently three global
data sets of actual ET in the public domain. These are the MODIS ET
product (MOD16;
Selected remote sensing products.
In order to have one common time interval, precipitation data in
In situ data of reservoir levels and inflow and river flow from the
basin measurement network were used to calculate the status index
(
The indicators selected by CHE for each of the management areas were used for the analysis presented here. These are the values of reservoir volume for the regulated areas (122, 123, 131, 132, 141, 151), inflow into the corresponding reservoir(s) for the upstream areas (120, 140, 150) and runoff at a selected station for management area 130.
Two different tests were carried out using drought impact data sets as a benchmark to assess the ability of remote-sensing-based indicators to provide early drought detection information during the period 2000–2012. The short length of the remote sensing data series available was one of the reasons to base the definition of drought we use to build the reference not on a frequency analysis, in which drought is defined as an extreme event with respect to the historical series, but on the occurrence of drought impacts. The other reason is that managers need to identify the conditions that may lead to drought impacts in order to take mitigation actions. In the first test, text-based records of drought occurrence and impacts collected from a review of local news (i.e. qualitative information) were used to reconstruct the onset and evolution of drought conditions during the period of analysis and as a benchmark for the comparison of the remote sensing data sets. Newspaper records were selected as a data source because they allowed a systematic collection of impact occurrence data of all affected sectors with a monthly time step for the whole period of analysis. In the second test, the use of crop yield statistics (i.e. quantitative information) is considered as a benchmark of drought impact on agriculture. The correlation of remote sensing data, especially SPI and NDVI, to agriculture yield data has been widely researched and applied (see Bachmair et al., 2016, for a review). This second type of impact data was included to provide a comparison of the results obtained in the correlation to text-based impact data and results obtained with the most commonly used type of impact data, and discuss the advantages and limitations of one with respect to the other.
Text-based data sets were collected from a review of regional news. “El periódico de Aragón”, the second largest newspaper in average daily circulation in the Aragón region was selected for the review because it has an online record going back to September 2001. All news items containing the word “drought” were reviewed and relevant records of drought events and impacts referring to the area of study were tabulated. For each entry, the location, period, description and, in the case of reported impacts, the affected sector were noted. The affected sectors were labelled as “rain-fed agriculture”, “irrigated agriculture”, “livestock”, “water quality”, “fire”, “water supply”, “energy” and “others”. The records of drought occurrence are classified according to the source of the information, making a distinction between non-official sources such as journalists and water users, labelled “mention of drought occurrence” in Fig. 2, and official sources labelled as “drought acknowledged by the authorities”, “ongoing mitigation measures” and “periods retrospectively defined as anomalously dry”. This last type corresponds mainly to news about the publication or communication of analysis performed by the scientific community or the water managers describing an ongoing or past drought.
The limit between indicators and impacts is not always clear. For example, low flow or reservoir levels are considered an impact of meteorological drought in some analyses, while these serve as indicators of hydrological drought in others. Here, we limit the definition of drought impacts as the effects of drought on people, economy and/or the environment.
Crop yield data of winter cereals both for irrigated and rain-fed cropping systems were obtained for the five selected districts in Huesca (H, S, L, M and B). Winter cereals are the cereal crops that are planted in the autumn, and they are the crops that cover the largest surface area. Their importance for the region results in better data availability than for other crops, and for this reason this type of crops was selected for the analysis. Only winter cereal crops with larger cultivated areas were considered: two- and six-row barley (irrigated and rain-fed), wheat (irrigated and rain-fed) and rice (irrigated). The two- and six-row barley types refer to the number of fertile spikelets in the spike.
The correlation between each of the remote sensing parameters and both
the timeline that aggregates all types of drought event records and
the timeline that aggregates all types of drought impacts
(Fig.
The sample cross-correlation function (CCF),
Here,
The reference drought periods used for the correlation provide
a binary record, indicating the occurrence or non-occurrence of
drought events in each month, without quantifying their intensity. To
obtain insight into the severity of the events, the use of annual
crop yield data was explored. The correlation of each annual crop
yield value to the monthly values of the remote sensing time series
from the start of the hydrological year in September to the end of the
following calendar year was analysed. This was done to detect the key
months in which the occurrence of drought conditions led to impacts
on the (annual) crop yield. The comparison was performed for three
rain-fed areas and three irrigated areas. These were selected to
correspond with the management units so that a relation could be
established with the areas of influence of the reservoirs
(Fig.
The timelines of drought events and impacts derived from the review of
local news are illustrated in Fig.
The first coloured row (yellow) in the figure represents the months in which drought was taking place according to the records found in the newspapers. The first line of the second block (red) reflects the occurrence of drought impacts described in the newspaper, while in the following rows these impacts are disaggregated by the affected sector.
Timeline of drought events (upper part) and impacts (lower part) for the north-eastern Ebro Basin during the period 2001 to 2012 based on a comprehensive local newspaper review.
Based on the records gathered from the newspaper records, the following descriptions of the hydrological years affected by drought episodes were constructed.
In 2002, after a dry winter, the availability of water in the
reservoirs was low. A first reference to drought in the press appeared
in February 2002. At the start of the spring, which is the beginning
of the irrigation season, water curtailments were reported for the
Bardenas irrigation system. In the beginning of April, agricultural
associations reported losses of 20 % of rain-fed cereal crops in
Aragón and at the end of the month the impact of drought in the area
was acknowledged by the ministry as well as by the local
government. In July, the flow of the Ebro in Zaragoza was half of the
minimum 30
The hydrological year 2004–2005 was depicted as the driest on record. The combination of cold and dry conditions during the first part of 2005 produced significant losses in the agriculture and livestock sectors. First impacts were reported in February 2005 (lack of pastures' production after 5 months without rain). From then until September 2006, the newspaper reflected a succession of impacts in different sectors, including all crop types, pastures, forests, livestock production, water supply to the population, wildlife, economy, recreational activities, hydroelectricity, water quality, employment and politics. The drought was already acknowledged by the authorities in March 2005, and the first mitigation measures were announced shortly after. This was that the regional government increased to 50 % the area of land, rain-fed or irrigated, that could be set aside to remain fallow. In June, aid measures were approved by royal decree.
Reservoir levels increased during the first half of the hydrological year 2005–2006, but the system failed to recover completely from drought before levels started decreasing again in April 2006, and at the beginning of the summer levels were lower than the previous year. After a hot summer, storage started to recover again, and in December 2006 the government considered the drought to have ended. Intense rains starting in February 2007 were followed by a period of precipitation deficit from May to February 2008. A few problems of water supply to certain villages were reported in August 2007 and flows were below the minimum required to warrant water quality in October. Impacts on agriculture and hydroelectricity started to be reported again in October. Abundant rains during spring 2008 constituted a first step towards the end of the drought episode.
The hydrological year 2010–2011 was characterised by lower-than-average precipitation and high temperatures. In February 2011, the newspaper showed the first reference to an emerging drought and its impact on the sprouting of winter cereal. This drought especially affected the Bardenas irrigation district. The Riegos del Alto Aragón and Canal de Aragón y Cataluña districts were also affected. All the systems managed to reach the end of the irrigation season, but with restrictions of more than 60 % on water quotas. Grapes and olives were the most damaged crops, but in general the food production in the area was defined as satisfactory at the end of the season. The following hydrological year (2011–2012) started with low reserves and a dry winter and spring, with the exception of November, which was a particularly wet month. In particular, the middle sector of Huesca revealed drought-affected areas. Extensive livestock farming, fodder and cereal production were the most impacted sectors. The risk of fire was reported to be high, even during the winter, which translated in a higher number of fires.
The information on drought occurrence and impacts obtained in the
previous step was used as a benchmark data set to assess the ability
of the remote-sensing-based data sets to provide early
detection. Figures
Cross correlation of drought indicators and drought events at
multiple time lags. The numbers in the
Cross correlation of remote sensing data sets against the timeline of reported drought impacts (all types).
Correlation between remote sensing drought indicators and crop yield data. Rain-fed areas and crops are marked with 0 and irrigated areas and crops with 1. The crops are irrigated and rain-fed wheat (W1 and W0), irrigated rice (R1), irrigated maize (M1), irrigated and rain-fed six-row barley (6B1 and 6B0) and irrigated and rain-fed two-row barley (2B1 and 2B0).
Correlation between drought indices (SPI and state index) and crop yield. Rain-fed areas and crops are marked with 0 and irrigated areas and crops with 1. The crops are irrigated and rain-fed wheat (W1 and W0), irrigated rice (R1), irrigated maize (M1), irrigated and rain-fed six-row barley (6B1 and 6B0) and irrigated and rain-fed two-row barley (2B1 and 2B0).
Figures
The time plots obtained for each of the parameters present no trends
or discontinuities, and the values in the autocorrelation plots show
that the autocorrelation diminishes quickly with increasing lag. An
exception are the series of the reservoir indices. In that case, for
some of the series, it is not clear from the plot if the series is
stationary. For one of them (management unit 122), it clearly is
not. This management unit corresponds to a reservoir (Rialb) that
started to be filled in the year 2000, and therefore the levels cannot
be considered stationary for the period of study. Most of the
autocorrelation plots for the reservoir level series present a small
peak of autocorrelation at a lag of 12 months, and one of them
(management unit 132) presents autocorrelation values declining more
slowly (significant values until lag 20). In the
The results of the correlation analysis between the remote sensing
data time series and the annual crop yield for the main irrigated and
rain-fed cereal crop types in the selected districts in Huesca are
represented in Figs.
NDVI and ET present some of the strongest positive correlations, especially between the remote sensing measurement during the spring (MAM) and the yield of rain-fed crops. LST shows also strong correlations with rain-fed crops in March and at the beginning of the season in September (S). The pattern is less clear for irrigated crops, probably because their water supply is less dependent on the rainfall. The strongest correlations in this case appear for rice crops with ET and NDVI, mainly at the start of the year.
Despite irrigated crops directly depending on reservoir supply, only
rice shows significant positive correlations with the index based on
reservoir levels for the two irrigated areas tested (HM, corresponding
to management unit 141 and LB, corresponding to management
unit 131). The reason can be that rice is especially drought
sensitive, since it has shallow roots and consequently a low depth of
readily available soil water, which is the fraction of total available
soil water that crops can obtain from the root zone without
experiencing water stress. This fraction is 0.2 for rice
Rain-fed two-row barley (2B0) in March stands out as the crop with the
stronger overall correlation with the different indicators. 2B0 is one
of the major crops in the area, with a maximum cultivated surface for
the period 2000–2012 of 170 914
Three years stand out in Fig.
Correlation of the remote sensing drought indicators for the month of March to annual rain-fed two-row barley yield in Monegros-Bajo Cinca districts (MB). For LST, NDVI, SM, ET, GPP and PsnNet, monthly means were used.
Crop yields were very similar in 2005 and 2008 (1662.3 and
1800.1
There is a second group in the middle sector of the plots that
includes the rest of the years for which drought impacts on rain-fed
agriculture were reported in the analysed media. This includes 2011
(3551
The results of this second test present a consistent behaviour of the indicators with respect to the different levels of crop yield among the analysed years in rain-fed areas. As in the previous test, NDVI, ET and SPI stand out for having stronger correlations. Most indicators present similar March values for the years of severe drought, clearly differentiated from the behaviour of years of moderate drought and years of no drought. The only exception is LST, in which a year where drought was not reported and yields were normal, such as 2009, has similar LST values in March to the years of severe drought. This indicates that LST may not be a good indicator of drought on its own but can still be useful in combination with other indicators.
The review of text-based records allowed a detailed reconstruction of
the drought events during the period studied. The cross correlation of
the timelines of drought events derived from this review to the
indices derived from remote sensing data revealed the potential of the
latter to provide early detection of drought events. However, this
binary information has the limitation that it does not allow to
objectively quantify the severity of the events. For example, in the
case of rain-fed agriculture, the information on impacts collected from
the newspaper does not allow for differentiation between those years
in which production was extremely low as a consequence of drought
conditions and those years in which production was only partially
affected by drought. Other studies have suggested a link between
impact severity and the number of records reporting it
A few additional aspects concerning reliability were noticed while
processing the records from the press: The information
on drought occurrence reported in the newspaper may not be
accurate. For example, impacts due to other causes may be attributed to
drought, or other phenomena such as normal summer shortages may be
described as drought. This issue was the reason to classify the
records of drought occurrence according to the source of the
information to make a distinction between official sources such as
mandated authorities, managers and scientists, and non-official
sources such as journalists or water users. This second type of source
is the one that is most susceptible to accuracy issues. Particularly
for the case of the mandated authorities, there are clear procedures
with which drought is officially acknowledged, which are defined in
the drought management plan. In the records reviewed, only the mention
of drought conditions recorded in 2003 is not backed up by the mention of
drought from official sources during the same period and may
therefore be regarded as a misuse of the word. Thus, we consider
accuracy issues to have little impact on results. Reporting of drought occurrence in the newspaper is not systematic, and
therefore some impacts may be missing. In Fig. 2, some unlikely
situations can be identified. For example, there are impacts on livestock in May
and July 2006 but not in June. Records referring to specific types of
impacts are more likely to have gaps. However, when all types are
aggregated, part of the gaps in each of the disaggregated data sets
will likely be filled with records from the other data
sets. Drought events affecting only a small area within
the region covered by the newspaper may not be reported. The results
of the test with crop yield data show values for the hydrological year
2006–2007 for which no drought impacts were identified in the
reviewed regional newspaper that are similar to three other
hydrological years for which drought impacts were recorded. Local
press for the specific area of the test (Alto Aragón), however,
reported a lack of rain from October to March, aggravated by high
temperatures, in Monegros and Bajo Cinca that had an impact on rain-fed
cereals and pastures. Public or political interest or
concern about drought (or even scarcity of other relevant news) can
motivate overstatement of drought impacts. These do not have an
influence in our analysis since we are only considering binary data of
occurrence or non-occurrence, but this issue could have a significant
impact on the reliability if the records were used to estimate the
severity of the event.
The length of the period of analysis does not have an influence in the identification of drought events based on impact records. However, having a longer series, and therefore potentially a larger number of drought events, would provide more robust results in the correlation analysis. Ideally the results should be updated as the period of record of remote sensing data grows.
The drought events identified by the textual search for a sector of
the Ebro Basin correspond with events observed at a larger scale. For
example,
Crop yield data, on the other hand, allowed for a more objective
identification of the drought events that had higher impact on
agriculture, though the yield data do have the disadvantage that may
only be reported on an annual basis. March was the month that
presented higher correlations. This is in agreement with the results
obtained by
Crop yield data can also be a useful reference to identify thresholds of drought severity classes. These thresholds could be derived based on the differences observed between the groups of years with severe, moderate and no drought conditions, although a longer data series than was used in this study is recommended to provide a more robust estimate of threshold values.
There are several factors that play a role in the severity of the
impacts due to drought conditions, including coping capacities and
water management (e.g. drought may not lead to impacts in irrigated
areas). Variations in these factors can alter the relationship between
the indicators and the impact. It should also be noted when using
drought impacts as a benchmark of drought occurrence, the absence
of certain types of impacts as a result of sound drought management
does not imply that there is no drought
Early information on emerging droughts benefits mitigation strategies
by increasing the time available for managers and affected communities
to take action. The requirements for drought early warning range from
a few weeks to several months
The weaker correlations obtained for SM data in the first test may be
due to the coarser spatial resolution of the data set. Higher-resolution
soil moisture products
The trade-off between the anticipation of the information and its reliability is also illustrated by the results. The lower reliability associated with earlier information detection of conditions that may lead to drought implies that often the situation may not evolve into a drought event. However, that information is still highly valuable as it allows the stakeholders to get ready to undertake mitigation actions if necessary.
The remote sensing products tested can enhance early warning capacity
and therefore contribute to the shift from reactive to proactive
management recommended by the European Commission
The aim of this research was to test the ability of remotely sensed data sets to detect early stages of drought at the river basin scale, with particular attention to their capacity to anticipate drought impacts and gain time to inform operational land and water management. Media records from a regional newspaper proved to be a helpful source of information that allowed a detailed reconstruction of drought events and impacts. The analysis using these data as a benchmark revealed the potential of the tested medium-resolution remote sensing products to anticipate reported drought impacts on irrigated and rain-fed areas at basin scale up to 6 months. The best correlation–anticipation relationships were obtained for SPI, NDVI and ET. SM and LST also showed potential to anticipate drought but with weaker correlations. GPP and PsnNet from MOD17 presented weak or no correlation for most of the areas, with only some of the rain-fed areas having moderate positive correlations. The index based on in situ data currently used in the basin also provides early detection, and with the exception of two of the management units, the anticipation of drought impacts is better than that provided by the remote sensing indicators. However, the correlation of the indices based on SPI, NDVI and ET to anticipate drought impacts was found to be stronger. The use of quantitative impact data of crop yields as a benchmark showed a consistent behaviour of the remote sensing indicators with respect to the different levels of crop yield in rain-fed areas among the analysed years. SPI, NDVI and ET stand out for having stronger correlations, reinforcing the findings of the first analysis. In both analyses, drought on irrigated land showed less clear correlation patterns than drought in rain-fed areas.
Altogether, the results confirm remote sensing products' ability to anticipate reported drought impacts and therefore provide a useful source of information to support drought management decisions at the basin scale. However, further analysis of managers' information requirements and response options is required to better assess the usefulness of these types of products in informing specific operational drought management decisions.
The remote sensing data used in this research are
openly available. The sources are mentioned in Sect. 2.2.1. In situ
data from the basin measurement network can be downloaded from the
Ebro Basin authority site (
The authors declare that they have no conflict of interest.
This research received funding from the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement no. 603608, “Global Earth Observation for integrated water resource assessment”: eartH2Observe. This work is a contribution to the Hymex Drought and Water Resources Science Team. The authors would like to thank the reviewers for their thoughtful comments. Edited by: Hannah Cloke Reviewed by: Azin Wright and one anonymous referee