1 Introduction

Azraq Oasis, in Jordan, owes its name to the particularly light blue color of its water. The presence of pools of freshwater made Azraq a highly relevant spot in the Jordan Eastern desert: the first settlements in the area date back to the Middle Paleolithic era and are testified by the presence of an ancient Castle, a thirteenth-century Ayyubid dynasty fortress built on some Roman ruins (Ramsar Convention 1998; Cordova et al. 2013). Besides the human colonization, Azraq represents a strategically important area for local fauna and migratory birds, stopping on their way from East Asia, Africa and Europe.

The existence of this historically important site is now at high risk, since in the 1990s the Oasis went through a drastic drying up of the natural springs that were then releasing up to 16 million cubic meters (MCM) of water a year (Dottridge and Abu Jaber 1999; UN-ESCWA and BGR 2013).

The main reason for the depletion of the water level in the Oasis is connected with human abstraction from the groundwater basin for domestic and rural use, especially after the so-called green revolution carried out by the Jordanian Government in the 1980s (National Research Council 1999; Demilecamps 2010; IUCN 2010). The springs that supported the Oasis in fact were the natural discharge of the Azraq–Dhuleil Basin, which today is one of the biggest sources of potable water for the capital Amman (25 %, in a country where groundwater resources make up 57 % of total supply) (Mesnil and Habjoka 2012). The easily reachable water bed, in some points just a few meters under the soil level, allowed the exploitation of water resources by farmers that found in Azraq an easy and fast way of moneymaking, especially since at first the digging was not regulated or monitored. Subsequently and still currently the illegal digging is flourishing and not completely under control. Besides the rural use, which accounts for up to two-thirds of the total abstraction, the drawing of water for domestic purpose by the Water Authority of Jordan (WAJ) constitutes the second source of water exploitation (Demilecamps 2010). Figure 1 shows how the estimated yearly abstraction grew from few MCM in the 1980s up to the current 60 MCM (IUCN 2010), from a basin for which the calculated safe yield is between 20 and 25 MCM (Ramsar Convention 1987; Al-Kharabsheh 2000).

Fig. 1
figure 1

Source: based on data from the WAJ and Bajjali and Al-Hadidi (2005), El-Naqa et al. (2007), IUCN 2010, Mesnil and Habjoka (2012)

Springs flow and water abstraction from the Azraq–Dhuleil Basin since 1950, adapted from Dottridge and Abu Jaber (1999) and implemented.

Full size image

Besides the lowering of the water bed, the groundwater conditions are also experiencing a general deterioration, mainly alteration in the groundwater salinity as shown by recent studies (El-Naqa et al. 2007; Goode et al. 2013). The persistence of the present conditions will unlikely allow reversing the direction of the change, with a big loss in terms of desert biodiversity and cultural and esthetic values (Dottridge and Abu Jaber 1999; National Research Council 1999).

The situation at Azraq Oasis is a good example of how human behavior can pose a huge threat to the existence of an oasis, adding to the naturally imposed hazard of desertification. Oases’ characteristics in fact make them particularly vulnerable to desertification, a process defined as ‘land degradation in arid, semiarid and dry sub-humid areas resulting from various factors, including climatic variations and human activities’ (UNCCD 1994). Understanding the evolution of desertification processes has a huge relevance in a country like Jordan, with a water supply of 145 m3/person/year putting it high in the list of worldwide water scarcity countries (Edwards et al. 1999; Al Eisawi 2005; WAJ 2007; Al-Bakri et al. 2013; UNICEF 2014; WHO 2015), and poses a question on the sustainability of a society in a growing demand regime. The amount of water available per capita is in fact predicted to diminish down to 90 m3/person/year in 2025 (personal communication), and this prediction would probably worsen if the number of refugees arriving from neighboring Arab countries keeps on increasing (UNHCR 2015).

Several attempts have been proposed to choose valuable indicators of desertification, including soil erosion and sedimentation, perennial plant cover and biomass (Dawelbait and Morari 2011). In particular on a long-term temporal scale, remote sensing is a proper tool to monitor such type of change, using Land-Use–Land-Cover Change (LULCC) detection techniques, an effective device also for the investigation of desert environments (Mouat et al. 1997; Palmer and van Rooyen 1998; Lin et al. 2011). The changes to Azraq Oasis in recent history make it an exceptional case to test techniques for monitoring the desertification in drylands: by knowing a priori the change to terrestrial surface between two dates it is possible to look for indicators proving it, and thus check whether the technique is able to detect the change and to which extent. The technique chosen in this study is Change Vector Analysis (CVA) applied to the output of a Tasselled Cap (TC) transform, usually fit to perform LULCC detection in vegetated areas (Malila and Lafayette 1980). CVA gives as an output the direction of change between two dates for each pixel of an image, having the advantage, differently from canonical supervised or unsupervised classifications, of not needing a land-cover classification to be performed in advance. The use of this technique in desert areas bears some doubts, since the TC transform is intrinsically intended to be used in vegetated areas; the aim of this paper is therefore to resolve this issue, using for this purpose a new set of parameters for the TC transform, calculated for Landsat 8 OLI system specifically for desert conditions.

2 Case study

Azraq Oasis is located in Jordan, 80 km east of the capital Amman approximately at coordinates 32°N and 37°S. It lies at the bottom of a relatively shallow natural depression at around 500 m of altitude, which constitutes the northern part of Wadi Sirhan Depression, a huge hydrological system covering great part of central Northern Saudi Arabia desert (Ramsar Convention 1990a; UN-ESCWA and BGR 2013).

The groundwater system is rather complex and is part of a greater aquifer system that extends from southwestern Syria to Saudi Arabia, consisting mainly of three aquifers, one on top of the other. The upper one, called Basalt Aquifer, has been the most exploited one and is also the aquifer from which the Oasis’ springs originated. The superficial catchment area of this aquifer is a huge basin called the Azraq–Dhuleil Basin, that extends almost longitudinally, from the southeastern side of Jebel Al-Arab (or Jebel Al-Druze, 1800 m high) mountain in Syria to the northeastern desert of Jordan (Dottridge and Abu Jaber 1999; El-Naqa et al. 2007; Cordova et al. 2013). The Basin area is considered predominantly arid, with annual precipitation ranging from less than 50 mm in the southeastern depression to more than 300 mm in the northern part, in proximity of the most elevated point of the Basin (Al-Kharabsheh 2000; UN-ESCWA and BGR 2013).

Before drying up in the 1990s, the Oasis was made up of two main groups of springs surrounded by marshes: the Aura and Moustadhema springs surrounded by the Druze Marsh in the northern part, and the Souda and Qaisiyah Springs surrounded by the Shishan Marsh in the south (Ramsar Convention 1990a). The springs and the marshes created a seasonal cycle with the nearby Qa’ or Sebkha, the Arabic terms to refer to a mudflat, also called salt flat or playa lake (Alvarez Cobelas et al. 2005). The mudflat consists of a depressed area that is flooded during the winter season by the drainages from incoming streams (wadies) and then dries up due to the dry climate, creating temporary salty water ponds and swamps until the water evaporates (Cordova et al. 2013).

Figure 2 shows the position of Azraq in Jordan with the Azraq–Dhuleil Basin border and a Landsat look-like image of February 25, 2015, centered in the study area (in yellow), where the mudflat area and the RSCN Reserve are also shown.

Fig. 2
figure 2

Location of Azraq in the Middle East region and of Azraq–Dhuleil Basin in Jordan, with zoom on the study area (yellow border) showing the RSCN Reserve and the mudflat area (February 25, 2015, Landsat 8 look-like image)

Full size image

Object in the 1960s of several study campaigns, mainly by ornithologists and desert researchers (Mountfort 1965; Scates 1968; Nelson 1973; Green 1995), Azraq Oasis starts to gain attention far before the complete drying up of the sources in the 1990s: a first draft of the Natural Reserve was drawn as early as 1965. Later, in 1977 Jordan designated the Azraq wetland for the List of Wetlands of International Importance under the Ramsar Convention, an international agreement established in 1971. The Member States of the Convention commit to protect and preserve the wetlands that are enlisted in the agreement (the ‘Ramsar Sites’ across the world are more than 2000 for 160 Contracting Parties). Also thanks to the Ramsar Recommendations, which are periodically provided during the Ramsar Meetings, the Natural Reserve was finally instituted in 1978, under the jurisdiction of Royal Society for Conservation of Nature (RSCN) which got the mandate from the Ministry of Agriculture.

In 1980, a special cabinet committee created to establish a plan of action, assessed the safe yield at 20 MCM; in the meantime the Amman Water and Sewage Authority (AWSA) set up a governmental wellfield for domestic use abstraction, and during the 1980s the pumping from the Oasis kept on growing, despite the calls from Ramsar Recommendations to reduce it at least by 50 % (Ramsar Convention 1987, 1990b, personal communication). Eventually the sources dried up completely in 1992 and this led to a Global Environment Facility (GEF) funding of a three-year project aimed to restore and manage Azraq Wetlands Reserve. This resulted in a variety of management activities during 1994–1999, including the opening of a Visitors Center (active from 2000) and a guided natural trail through the Reserve, together with the return of a supply of water to the Shishan Marshes from June 1994, following lack of water for 2 years. In 1998 the artificial water supply started to be pumped bypassing the former springs (Souda and Quasiya), where a large amount of water was being lost through infiltration, and transported directly to the central marsh (Ramsar Convention 1999).

3 Materials and methods

3.1 Landsat data

For the present research five images centered on Azraq Oasis were chosen, namely from path/row 173/38 of the Landsat database available at http://glovis.usgs.gov/. Cloudless images taken in summer season dates were chosen, this in order to avoid the climatic seasonal fluctuation and to have the most similar conditions on the terrestrial surface. Given the desert and arid conditions, we can assume that no big climatic variation occurs in summer in the analyzed area and that the atmospheric correction is negligible. The earliest available image respecting the above-described criterion, and considering also the availability of TC transform for that type of sensor, is from the summer of 1984. Images taken around the 30th of August have been chosen over a span of 30 years, in order to detect the long-term change but also the shorter-scale variation. Tables 1 and 2 give the list of selected dates for the analysis and the characteristics of the involved Landsat satellites. Later in this article the images will be denoted just by the year (i.e., 1984 for August 30, 1984).

Table 1 Selected dates for Landsat 5 and Landsat 8 satellites for the multitemporal change detection
Full size table
Table 2 Technical data of selected spectral bands used in the TC transform for the Landsat 5 satellite TM (Thematic Mapper) and Landsat 8 satellite OLI (Operational Land Imager) sensors, with resolution of 30 m
Full size table

3.2 Tasselled Cap (TC) transform

The Tasselled Cap (TC) transform was introduced by a work of Kauth and Thomas (sometimes it is also called KT transform) in 1976 (Kauth and Thomas 1976) and consists of a linear transformation of the pixels’ values of a satellite image of terrestrial surface, in order to convert the originally highly correlated bands (in the Infrared and Visible region, see Table 2) to a new set of uncorrelated axes. The new axes, also called TC features, bear a physical meaning in terms of the characteristics of the surface: Brightness, Greenness and Wetness. As a linear transform, the TC consists of a set of parameters (frequently referred to as coefficients) that weigh the original spectral bands’ values and sum up to give as an output a new value for that pixel in the new set of axes, therefore literally transforming the coordinates reference system passing from the satellite bands space to a new TC features space. Unlike other transformations, Principal Component Analysis (PCA) as one example, TC transform is not image dependent; therefore once a set of parameters for the transformation is given, it can be used to convert any image with the same characteristics (type of satellite, data and region) (Ivits et al. 2008; Yarbrough et al. 2012; Zanchetta et al. 2015). In particular when dealing with desert conditions, special attention should be paid when applying the existing coefficients, noticing whether the original studies were carried out specifically for vegetated areas.

The coefficients present in the literature for Landsat 8 were calculated for vegetated conditions (Baig et al. 2014; Liu et al. 2015); therefore a new set of parameters have been calculated for desert conditions, using Top Of Atmosphere Reflectance (TOAR) images (Table 3). The method chosen to calculate the coefficients uses the Gram–Schmidt orthogonalization as extensively illustrated by Jackson (1983), as well as the original study by Kauth and Thomas and in other sources (Chang 1992; Ivits et al. 2008). Another common way of calculating the coefficients is through PCA followed by a rotation, to match some preferential directions given by the position on the bands space of selected categories of the surface cover (Crist and Cicone 1984; Huang et al. 2002; Yarbrough et al. 2012). In the present paper the first method was chosen following the considerations made by Zanchetta et al. (2015), source from which are also taken the coefficients used in the present study for Landsat 5 images. The Gram–Schmidt orthogonalization is performed on selected pixels that belong to three categories of surface cover (indicative of soil, vegetated and wet surfaces) and on a range of images from different seasons and selected geographic locations in the Middle East (Jordan and Egypt), in order to overcome the dependency of the TC calculation on temporal conditions and location. These criteria allow the use of the new Landsat 8 coefficients to perform a LULCC detection against products of older Landsat systems.

Table 3 TC coefficients for Landsat 8 TOAR data for desert conditions
Full size table

3.3 Change Vector Analysis (CVA)

Change Vector Analysis (CVA) was first implemented by Malila and Lafayette (1980) for detecting forest changes and then widely adopted for LULCC detection by several authors also for desert areas (Lorena et al. 2002; Bayarjargal et al. 2006), sometimes further developed and variated (Lambin and Strahlers 1994; Chen et al. 2003; Flores and Yool 2007; Dawelbait and Morari 2011; Dubovyk et al. 2013; Singh and Talwar 2014). The technique uses two (or more) spectral variables, like spectral bands or surface features or spectral indices, to produce in output a map of the magnitude and a map of the direction of the change between two dates. For each date, the pixel’s values in the two bands are plotted on the bands space, generating a vector of the change in time. The vector’s length and angle give, respectively, the magnitude and direction of the change. Once a threshold is chosen, the significant change is given as those pixels that exceed that value in the map of the direction.

As spectral variables, some biophysical indicators connected with surface characteristics are chosen. In this way, the combined variation can have a meaning in terms of surface change. In general (considering the two bands space) an indicator of the soil reflectance and an indicator of the vegetation vigor are used, like Albedo and NDVI (Normalized Difference Vegetation Index) (Karnieli et al. 2014) or the Tasselled Cap features Brightness and Greenness (Malila and Lafayette 1980).

3.4 CVA interpretation

The determination of the threshold and the interpretation of the direction of the change are critical points when applying the CVA technique and have been faced in several ways by different authors. The threshold in particular can be set with the use of empirical values, or through interactive trial-and-error procedures, or semi-automated approaches (He et al. 2013). For this work, the threshold given by the average plus the standard deviation of the magnitude values was chosen; therefore it is not a fixed value and it changes for each pair of images taken into consideration. This choice respects a statistical requirement and does not require human supervision, but in the same time it bears some uncertainty connected with different statistical distributions of the values of magnitude when used for multitemporal analysis, meaning when, like in our case, the change is investigated in more than one time interval. To address this issue, the CVA was carried out on a vast area surrounding the Oasis (891 km2), and not only the actual area of interest around the Shishan marsh (see Fig. 2, yellow border).

Once the angle of the change vector is mathematically determined, the meaning of the CVA direction is not straightforward. Generally a simplified interpretation of the quadrants of the bands space is given: the first quadrant in fact represents a positive change for both variables, the third quadrant a negative change for both, and so on. In line with the interpretation given by other authors for desert conditions (Lorena et al. 2002; Karnieli et al. 2014), we defined the four change directions as follows. The first quadrant, which is characterized by an increase in both Greenness and Brightness, indicates moisture reduction and drying up of salty surfaces (Karnieli et al. 2014); the second quadrant, with increase in Brightness, is indicative of chlorophyll increase and of regrowth and regeneration of vegetated features in general; the third quadrant, where both spectral features have a decrease, indicates burning or water and in general a change toward higher moisture land; the fourth quadrant, with increasing Brightness, is strongly related to great losses of vegetation biomass and to bare soil expansion (Fig. 3).

Fig. 3
figure 3

CVA explanation and physical meaning of the quadrants

Full size image

4 Discussion and results

Before performing CVA on the selected dates, a visual interpretation of images was carried out in order to visualize the expected change between the pre-drying up conditions and the present conditions. As a reference for the pre-event situation, a map from 1979 was digitized and georeferenced (see Fig. 4). The map is available from the 1990s RAMSAR Report, and it contains also some updates from an expedition organized by the RAMSAR committee in March 1990 (Ramsar Convention 1990a). The annotations describe a general deterioration of the Oasis: stressed or tainted vegetation in the central area of the marsh (on dunes and water) and low water level in the pools, with the Ingilesi pool almost dry and the ditches totally dry. As a reference for the present conditions, missing ground truth data, a comparison with updated base maps available online (OpenStreetMap, Google) was taken in consideration (Fig. 5). Further help in interpreting the intermediate dates comes from aerial pictures available from the APAAME online catalog (Aerial Photographic Archive for Archaeology in the Middle East) that dates from 1998 to present.

Fig. 4
figure 4

(Source: Ramsar Convention 1990a)

Major habitats/vegetation communities in Azraq Wetland Reserve, digitization of a 1979 map.

Full size image
Fig. 5
figure 5

Google Earth image of the Azraq Reserve in 2015, 1979 RAMSAR fence (in black) is shown

Full size image

The visual comparison over the entire time span shows indeed a deterioration of the general status of the Oasis area, while rural areas on the north east part of the plateau seem to flourish. These impressions were confirmed during a field visit in June 2014.

The CVA image resulting from the comparison between 1984 and 2013 is given in Fig. 6, where Landsat look-like images are also shown. The highest change occurs in the ‘bare soil expansion’ category, which accounts for more than 80 % of the total changed pixels (see Table 4). A bare soil expansion was detected in and around the Reserve area, while ‘generation/increase in chlorophyll’ and ‘moisture increase’ were detected in the eastern side and around the mudflat area, with clear identification of agricultural fields (Fig. 6).

Fig. 6
figure 6

Landsat look-like image for August 30, 1984 (a) and August 30, 2013 (b) and results of the CVA between the two images for the study area: directions (c) and magnitude (d) images before applying the threshold, final change detection analysis (e); RSCN Reserve border (in red) and the 1979 RAMSAR fence (in black) are shown

Full size image
Table 4 CVA results for the studied time periods in km2 and % on the total study area (for the meaning of the categories refer to Fig. 3)
Full size table

The long-term trend is replicated in the shorter-term periods, where the highest change occurs in the ‘bare soil expansion’ category with a maximum in the 2003–2013 period (9.39 %). The intermediate periods show in general a more varied picture, with change occurring homogeneously in all the categories. The whole area study includes also the north east side of the image, where a high rural activity developed in the period taken into account; therefore a closer view on the Oasis area, centered on the RSCN Reserve, for shorter temporal scales is also analyzed (Fig. 7).

Fig. 7
figure 7

Look-like Landsat images (a 1984, b 1990, c 1998, d 2003, e 2013) and CVA results images (f 1984–1990, g 1990–1998, h 1998–2003, i 2003–2013, j 1984–2013) for the Reserve area from 1984 to 2013 (for the dates specification see Table 1)

Full size image

The centered CVA image for 1984–2013 (Fig. 7j) detects a high change toward ‘bare soil expansion’ in the Reserve area, and this is affirmed by the base maps. The CVA image for 1984–1990 (Fig. 7f) detects ‘bare soil expansion’ and ‘moisture reduction’ all along the Reserve area and the mudflat. This trend continues in the 1990–1998 CVA image (Fig. 7g), where the drying up of the major pools adjacent to the mudflat is evident. This trend is partly halted in the following CVA image for 1998–2003 (Fig. 7h) likely as a result of the restoration project completed in 1998 through the GEF fund, and in fact a replenishment of green areas is detected in the central part of the marsh. The following CVA image for 2003–2013 (Fig. 7i) shows again a change toward ‘bare soil expansion,’ and this is attributable to the fire that broke out in October 2010, after which the Reserve was temporarily closed and could open again just in April 2011: the fire’s consequences on the vegetation were still evident during the field visit in June 2014.

A closer analysis of the CVA results on the area of the pools surrounding the former Shishan springs was also considered (not shown here). The results show a main change toward drier conditions between 1990 and 1998, as expected, and a partial recovery between 1998 and 2003. Finally between 2003 and 2013 the Oasis achieves its present conditions, with the former springs’ pools left empty and a partial recovery of the marsh and of the stream that flows toward the former Burgees pool (Fig. 5).

5 Conclusion

An analysis was performed of the capability of the Tasselled Cap (TC) transform combined with Change Vector Analysis (CVA) to detect surface changes in arid areas, using as a case study the wetland of the Azraq Oasis, Jordan. The analysis was carried out on a 30-year interval (1984–2013) on Landsat satellites images. For the purpose of the research, a new set of TC coefficients was calculated, using several Landsat 8 images of selected dryland locations in the Middle East, and is here presented for the first time.

CVA gives as an output a map of the magnitude and a map of the direction of the change between two dates, having the advantage of not needing a previous land-use/land-cover classification. There are two difficulties though that are intrinsic to the use of the CVA technique: the first is deciding what the meaningful change is, namely choosing a threshold for the change detection, and the second is the absence of an objective validation method. The first issue was addressed using a statistical criterion for the selection of the threshold (average of the magnitude values plus the standard deviation), allowing for an automatic selection of the value for each couple of dates. The second one was faced with the support of data and material collected in a field visit in June 2014, together with several sources available in the literature.

The results show that the CVA applied to desert-adapted TC transform features is able to detect the expected changes in the studied area and can therefore be considered a valuable technique for change detection studies in areas subject to desertification and deterioration processes in drylands.