Introduction

Land uses and land cover changes are one of the main driving forces of land degradation processes in the Earth system (Verburg et al. 2011). Understanding land use changes is indispensable to assess environmental changes (Amato et al. 2015; Chappell et al. 2018) and develop sustainable land plans (Bell et al. 2010; Debolini et al. 2015; Najafinasab et al. 2015; Li et al. 2017). Remote sensing can be defined as a tool that is able to assess these changes without physical contact. Remote sensing has been widely used because of its multidisciplinary applicability in geology (Siegal and Gillespie 1980; Immerzeel et al. 2014; Amri et al. 2017), forest evolution (Lim et al. 2003; Silva et al. 2017; Santos et al. 2017), natural hazard (Pourghasemi et al. 2012), hydrological processes (Schmugge et al. 2002; Etehadi Abari et al. 2017), geomorphology (Verstappen 1977; Moosavi et al. 2014; Kumar et al. 2018), vegetation vigor analysis (Jones and Vaughan 2010; Jabbari et al. 2015), land cover (Matinfar et al. 2013; Sabr et al. 2016), and land degradation (Zehtabian et al. 2014; Alanis et al. 2017; Ahmady-Birgani et al. 2017). Nowadays, remote sensing data are one of the most common resources to evaluate and monitor landscape evolutions and making reliable approaches for land cover changes (LCCs) (Schulz et al. 2010; Zewdie and Csaplovics 2015; Yuan et al. 2016; Kharazmi et al. 2016; Kharazmi et al. 2017b). Using multispectral imagery obtained from Landsat, it is possible to generate valuable data in non-well-studied regions (Shengbo and Rao 2008; Roy et al. 2014; Mandanici and Bitelli 2015).

The research of LCCs focuses on how surface landscapes evolve as a combination of different ecological terrains. Previous studies of LCCs have helped establish optimal distributions of human land uses (agricultural, urban, etc.) and aimed to protect the resources and stability of local populations (e.g., Luyssaert et al. 2014; Mirzaei et al. 2016; Sakieh and Salmanmahiny 2016). Driving factors that contribute to LCCs include human activities (Fu 2003; Nazari Samani et al. 2016a), political and economic decisions around urbanization and infrastructure (Yli-Pelkonen 2008; Briner et al. 2012; Bateman et al. 2013), and environmental changes (Shajaat Ali 2006; Szczypta et al. 2015). Land cover can be considered as the natural interface between ecological and anthropogenic activities (Rahdari et al. 2013; Yuan et al. 2016), affecting at the local and global scales (Luo et al. 2008; Issanova et al. 2015; Nazari Samani et al. 2016c).

The population growth rate of arid regions is roughly three times higher than the world average (Goudie 2013). Under the combined effects of climate change and population growth, semi-arid and arid regions have undergone severe stresses on local resources (Baron et al. 1998; Ranjeet et al. 2009). Many of these regions are difficult to access, making studies of these areas particularly problematic. Fortunately, during the last decade, the abundance of available remotely sensed data has helped to expand LCC monitoring efforts (Lu and Weng 2007; Rashki et al. 2014; Xiao et al. 2017). In particular, Landsat images, with records as far back as 1972, enhance large-scale researches that also are able to incorporate temporal changes (Wulder et al. 2012; Roy et al. 2014). The high frequency of monitored time and the large coverage areas make Landsat images ideal for studying LCCs (Vogelmann et al. 2016; Dawelbait et al. 2017). Also, remote sensing provides valuable information about landforms (Nazari Samani et al. 2016b).

In arid territories, wetlands function as a part of the ecological landscape system (Mitsch and Gosselink 2000), which are essential because of their functions as a natural resource for million people (Dixon et al. 2016). Wetlands are characterized by a high water quality and are able to support high levels of biodiversity (Chen and Liu 2015). However, according to Patino and Estupinan-Suarez (2016), about half of the total worldwide wetland areas are being lost during the last century, causing a serious environmental problem for human and wildlife. The Ramsar Convention on wetlands in Iran ( 2014) reported that 24 international wetlands are located in Iran (1,486,438 ha) and some of them, due to human activities, have been highly polluted such as Anzali wetland (Vesali Naseh et al. 2012a; Vesali Naseh et al. 2012b; Hargalani et al. 2014), Hoor al-Azim wetland (Eydizadeh et al. 2014), and Hamoun wetland (Zolfaghari et al. 2016). Therefore, the protection of all of them should be vital for national and international organizations and policy makers. However, there is a lack of information about them since these drastic land degradation processes occurred. Previous studies of Iran’s wetlands have mainly focused on environmental pollution aspects, but less attention has been paid to land cover changes. It would allow us developing land use plans to manage the potential degraded areas and performing models to anticipate possible irreparable environmental losses.

Therefore, the main aim of this study was to assess land cover changes using Landsat images for a non-well-studied wetland area (as the Hamoun wetland) located in southeastern Iran. The population of the Hamoun wetland area is characterized by rapid expansion, and continued monitoring and prediction of LCCs will be essential to evaluate their future socioeconomic needs. Thus, two specific goals must be achieved: (1) to determine when the most significant landscape changes should be expected on a seasonal time scale and (2) to determine which season has the least year-to-year variability during 30 years (1987–2016).

Materials and methods

Study area

The Hamoun wetland is located in the Sistan Basin in southeast Iran and is divided into three regions: Saberi, Puzak and Hirmand (Fig. 1a) (Najafi and Vatanfada 2011; Mousavi-Sabet et al. 2014; Maleki et al. 2016). The Hamoun Lake is the third largest lake of Iran, and in 2016, it was designated as a biosphere reserve by UNESCO. At its maximum extent, the Hamoun Lake averages 4 m water depth and is fed by the Hirmand River which supports local agriculture and a growing population (Yousefi et al. 2017). Low precipitation amounts and a high evaporation have driven this area into multiyear droughts during the last 30 years (Table 1, Fig. 1b–d). This region is characterized by 120-day winds (Rezazadeh et al. 2013; Ekhtesasi and Gohari 2013; Alizadeh-Choobari et al. 2014; Rezaei and Ghofranfarid 2018), which blow from the north from June to September (Rashki et al. 2013), making the Sistan Basin one of the windiest regions over the world (Fig. 1e). During periods of high water level, these winds are able to lead to an increase of air humidity, and during periods of low water level, to the deposition of dust (Kaskaoutis et al. 2015), both of which stress the environment and the local population. Recent studies of the Sistan Basin have focused on these dusts (Alizadeh-Choobari et al. 2014; Behrooz et al. 2017), but further studies are very limited by the remoteness of the area.

Fig. 1
figure 1

a Study area. b Four springtime images showing the progressive drying and recovery of the Hamoun wetland. c Rainfall and temperature variations during the studied period (Zabol and Zahak synoptic climatic stations). d NDVI image of Sistan Basin (2016). e Distribution and sources of dusts for the Hamoun wetland (June 2012)

Full size image
Table 1 Mean annual meteorological conditions of the study area
Full size table

Landsat image selection

Landsat images of the Hamoun wetland taken from 1987 to 2016, freely available online from the USGS (https://earthexplorer.usgs.gov), were surveyed to quantify distributions and changes of land cover types (Fig. 2). Landsat images have 30 m/pixel resolution and are taken every 16 days (Pekel et al. 2016). For each of the 30 years studied, two clear and noise-free images were selected for each season and the best available images in each season were selected (Table 2). By convention, spring was defined from March to May, summer from June to August, and the fall season from September to November. Two images were necessary to ensure total coverage of the study area in each season. In winter season, due to the cloudy weather, assessments and interpretation of images could not be used. Thus, it was no possible to show significant changes in land cover from fall season to winter.

Fig. 2
figure 2

Flowchart of the methodology used for this research

Full size image
Table 2 Landsat images used in different seasons during the 30-year period (row 157, paths 38 and 39)
Full size table

Moreover, for some years, Landsat data were not available, and after May 2003, a fault in the instrument caused a 22% loss of coverage in Landsat ETM+ images (Howard and Lacasse 2004; El Fellah et al. 2017). Fortunately, this defect affected < 10% of the Hamoun study area and, for these gaps, land cover type was interpolated from the nearest neighbor pixels. In total, 154 Landsat images were processed for this research.

Landsat image treatments

An atmospheric correction and a conversion of the brightness values from radiance to reflectance (Vahtmäe and Kutser 2013; Feyisa et al. 2014; Bernardo et al. 2017) were performed using FLAASH (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes) module in ENVI v. 5.1 (Othman et al. 2018). Atmospheric correction is not strictly necessary for image interpretation (Wu 2004), but the results obtained by Okin and Gu (2015) and El Fellah et al. (2017) showed that these algorithms allow improving the datasets.

The boundary of the study area encloses a total area of 5.95 × 105 ha, which supposes the total extension of the Hamoun Lake at its highest water level. That interpretation can be noted from Google Earth images in Fig. 1a. Classification of the land cover was performed using the ERDAS Imagine maximum likelihood algorithm (Abd El-Kawy et al. 2011; Jaleta et al. 2016; Kharazmi et al. 2017b). According to Bolstad and Lillesand (1991), Gevana et al. (2015), and Rebelo et al. (2017), that approach, tying a training algorithm to ground-truth control data, is the most statistically robust in assessing remotely sensed images. In countries such as Iran where a limited ground control data occurs, this algorithm is ideal for that application. Unsupervised classification was conducted on two images from 30 June 2014 and 19 July 2015. Based on reflectance profiles, the region was mapped into five major classes of land cover: (1) surface water, (2) bare land, (3) dry bed, (4) swamp, and (5) grasslands. Then, 60 control points were randomly selected within the mapped area and were compared with the ground-truth observations. Those comparisons confirmed that the algorithm predicted the land cover type to within acceptable levels of accuracy (Fig. 3). The highest error was in distinguishing “bare land” and “dry bed” regions. Both of them were merged and named “bare land” to be referred to areas devoid of vegetation and surface water. The ERDAS algorithm was trained to identify land cover types using a combination of reflectance profiles, visual interpolation (Dawelbait et al. 2017), and Google Earth images (Karkon Varnosfaderani et al. 2017) and to map the study area into water, swamp, grassland regions, and bare lands (Table 3).

Fig. 3
figure 3

Verification of model results. a Landsat images of a representative part of the studied area (July 2015). b Unsupervised land cover classification of a (the location of the ground control points (GCPs) in a, b is marked with a cross). c Example of representative grassland area close to ground control point 1 (GCP1). d Reflectance profiles of the five types of land cover originally defined by the algorithm

Full size image
Table 3 Land cover classification types
Full size table

To quantify LCCs, the pixel-based change detection (PBCD) was used for determining “from-to” changes in data that has already been classified into zones (post-classification comparison) (Chen et al. 2012; Zewdie and Csaplovics 2015; Kharazmi et al. 2016; Demissie et al. 2017). Gain was defined as the percentage of a land cover class that is added from t1 to t2 and loss as the percentage of elimination of that land cover. Land cover classes that did not change among times were called persistent. Additionally, swap was calculated to account for shifts in the land cover where the percentage cover was conserved. Swap is the simultaneous loss and gain of area for a given land cover type, but in different regions within the study area. For a land use class (m), the amount of swap (Sm) was calculated as

$$ {S}_{\mathrm{m}}=2\times \min\ \left({a}_{\mathrm{m}}-{b}_{\mathrm{m}},{c}_{\mathrm{m}}-{b}_{\mathrm{m}}\right) $$

where am is the total column proportion of the land cover class, bm represents the pixel persistence of the land cover class (Padonou et al. 2017), and cm means the total row proportion of the land cover class. Additionally, the total change of a land cover type was either the sum of the net change and the swap or the sum of the gains and losses.

Results and discussion

The accuracy of the mapping procedure was calculated by ERDAS Imagine based on the field study results of 2014 and 2015 (see above), visual interpretation using a standard RGB composite display of some images in the time series (e.g., Karkon Varnosfaderani et al. 2017; Dawelbait et al. 2017), Google Earth images, and recent literature on this area. Based on Yuan et al. (2016), the reliability of estimates of LCCs can be approximated by the image classification accuracy. The overall accuracy of our classification was between 81.3 and 97.6%, and the kappa index (Foody 2002) was between 0.83 and 0.94 (Fig. 4). An overall accuracy of 70% or more (Chuang et al. 2011) was considered a successful classification and mapping of the study area (Table 4).

Fig. 4
figure 4

Overall accuracy and kappa coefficient for classification of land cover types per season

Full size image
Table 4 Overall accuracy (%) and kappa coefficient in other arid and semi-arid regions
Full size table

Over the 30 years of the research period, the surface water area was highly predictable from spring to summer (R2 = 0.94) and from summer to fall (R2 = 0.91) (Tables 5, 6, and 7). From summer to fall, surface water extent decreased, causing a corresponding increase in the extent of grasslands (R2 = 0.36) and bare land (R2 = 0.90).

Table 5 Correlation coefficient (R2) of land types among each paired season
Full size table
Table 6 Standard regression analysis (F value)
Full size table
Table 7 Standard regression analysis (T value)
Full size table

Annual summertime changes in vegetation, dependent on spring humidity, resulted in a low correlation in the extent of grassland between fall and spring (R2 = 0.18) (Fig. 5). Bare land was highly predictable from summer to fall (R2 = 0.90), showing a slight decrease going into the later season. Bare land was the lowest during spring and the highest during the fall, relating to the counter-balancing changes in surface water and grasslands (Fig. 5).

Fig. 5
figure 5

Land cover comparisons among seasons. a Surface water. b Swamp. c Grasslands. d Bare land

Full size image

From 1988 to 1998, surface water dominated the study area, comprising more area than any other land category for all seasons in most of this decade (Fig. 6). During this period, the Hamoun Lake reached at its peak and the results of this study demonstrated its subsequent decline. From 2000 to 2004, the Hamoun Lake became nearly completely dry, constituting the lowest water level of the lake during the 30-year study period (Fig. 6). Correspondingly, the extent of bare land, including dry exposed lakebed, increased dramatically during this period. The expansion of bare land accelerated after 2000, and during two periods (2000–2004 and 2011–2016), the bare land class covered more than 90% of the region, reaching maximum values of 99% in 2004 (Figs. 6 and 11). On a seasonal scale, the greatest extent of bare land was in fall (Fig. 6c).

Fig. 6
figure 6

Distributions of land cover types from 1987 to 2016. a Spring. b Summer. c Fall. d Mean annual land cover with a single combined class for swamp and surface water classes

Full size image

Variations between the north and the southern areas included higher springtime water levels in the north (Saberi, Puzak) relative to the south after 2000, and higher relative water levels in the south from 1991 to 1993 (Figs. 7, 8, and 9). This variation suggests that spring floods from the Farahrood and Harut rivers have long been an important control on the water supply of region. After 2004, pulses of river discharge into the Hamoun Lake led to an increase in water level that was not maintained from year to year. Grassland was the most extensive in 2006 and 2013 during the spring, covering 22 and 23% of the total area of region, respectively. The 85% loss in grassland shortly after these periods is attributed to human intervention (e.g., overgrazing of livestock, Fig. 10). Repeated water discharge into the Hamoun wetland was recorded from 2005 to 2010 (Figs. 6, 7, 8, and 9), although during summer, an increase in temperatures and dry periods avoided water level increases.

Fig. 7
figure 7

Land cover changes in the study area during spring

Full size image
Fig. 8
figure 8

Land cover changes in the study area during summer

Full size image
Fig. 9
figure 9

Land cover changes in the study area during fall

Full size image
Fig. 10
figure 10

Examples of grassland in the Hamoun wetland. Black arrows indicate anthropogenic activities associated with nomads livestock over grazing

Full size image

In 2005 after a significant flood, grassland experienced the most gain (23%) and, then during the subsequent years, a lost higher than 80% of its total extension (Fig. 12). From 2007 to 2011, spring floods were additionally correlated with an increase in grassland extension, although grassland cover is typically the dominant ecological control of dry land ecosystems (e.g., Dubovyk et al. 2016; Gessner et al. 2013). Our findings demonstrated that very little fluctuation occurred in total grassland areas in comparison to variations in swamp and surface water land types.

To smooth short-time-scale variations in LCCs caused by seasonal changes, Fig. 6d shows the annual average values for each land cover type. Furthermore, the swamp and surface water land types were combined to highlight changes in the presence and absence of vegetation relative to water-logged landscapes (Fig. 11). Our results showed that the grasslands exhibited a very little fluctuation relative to the highly fluctuating one conformed by the water swamp class in the Hamoun wetland area, but in other wetlands, vegetation cover is entitled as a main factor for LCCs (Munyati 2000; Tareq et al. 2004; Sosnowski et al. 2016). We considered that grasslands were not the best indicators for analyzing dynamics of Hamoun wetland, because no significantly changes were observed during a long-term period (Fig. 6d). The combined water class (watery land with vegetation) showed a general decreasing trend that is expected to continue, with high fluctuations into the coming decade. LCCs occurred at different rates during the 30 years covered in this study, and changes in water level had the most significant effect on the health of this ecosystem. Many studies based on periodic long-term monitoring, carried out in the Hamoun wetland (Vekerdy et al. 2006; Shafie et al. 2011; Sharifikia 2013; Rashki et al. 2013; Kharazmi et al. 2017a; Maleki et al. 2018), are not comparable with our study because they do not allow obtaining an objective forecast of the real ecological situation. Our results showed that long-term monitoring assessments based on seasonal changes can provide more logical and precise results of ecosystem dynamics too.

Fig. 11
figure 11

Inter-annual LCCs during different seasons. a Surface water. b Swamp class. c Grasslands class. d Bare land

Full size image

The fall season was least affected by highly variable controls on the water level (e.g., spring flooding or summer heat waves) and was determined by this study to be the best representative season for continued long-term studies of the Hamoun wetland. Data from the fall season were the most representative of the annual means, and therefore, monitoring of this season can be considered as the most effective proxy of inter-annual variability (as it was seen in snapshots of time). The increasing trend in the extent of bare land could be the result of a positive feedback between overgrazing, exposed bare land, wind erosion, and plant-choking dust deposition in semi-arid and arid lands as confirmed by several authors (Pulido-Fernández et al. 2013; Zezin et al. 2015; Borrelli et al. 2016). In this way, several authors also claimed that bare lands enhance land degradation processes related to soil erosion (Berendse et al. 2015; Ahmed et al. 2016), loss of biodiversity, and impoverishment of nutrients (Borrelli et al. 2014; Nearing et al. 2017). According to that, policy makers and rural populations have to develop effective solutions such as revegetation strategies and the use of amendments, geotextiles, or mulches (Bienes et al. 2016; Hueso-Gonzalez et al. 2016). Naturally, these strategies have to be conducted in order to take into account the mentioned driving factors such as water competence, long dry periods, and land use changes such as grazing or abandonment (Lasanta et al. 2000; Lasanta et al. 2015).

Using the fall season as a representative proxy of the inter-annual variability, the highest gains (26%) in watery land (with vegetation) occurred in 1989 and 1995. Conversely, the most significant losses occurred in 1999 when the extent of surface water dropped by 31% (Fig. 12). Despite this variability, the persistence values of the water land class showed that water as a resource was available from 1987 to 1999, which demonstrated that ecosystem dynamics were healthier during this period. After 2000, water levels were extremely lower and then showed fluctuations unable to recover to the levels of the previous decade. The water land class exhibited high levels of swap in 1989 (20% of the class value). Additionally, grassland swap was high in 1999 (10% of the class value). Significant changes in water lands and grasslands show both net change and total change from 2005 to 2008 (Fig. 12).

Fig. 12
figure 12

Annual LCCs from 1987 to 2016

Full size image

More than 25% of the total jobs in this region involve the care of livestock which pasture in the Hamoun area (Dahmardeh et al. 2009). Therefore, over half of the inhabitants of the Sistan Basin will be directly affected by changes to the Hamoun wetland. We confirmed that the desertification process is leading to a decrease in vegetation cover, particularly annual vegetation and grassland on which the livestock feed. In Fig. 10, it can be observed how grazing resources were limited to small areas of the wetland, where enough moisture existed to sustain grasslands. Overgrazing in these limited areas exposed wetlands to wind erosion (Kairis et al. 2015) and led to the spread of drought-tolerant plants and invasive species in rapidly dewatering regions (Flory and D’Antonio 2015).

Furthermore, changes in land cover distribution showed three distinct periods in the recent history of the Hamoun wetland, each of which can be attributed to a change in the available water budget (Figs. 6d, 7, 8, and 9). Firstly, a wet system period (1988–1998) is in its prime, balancing ~ 40% surface water with vegetation and comparable bare land exposures. Fed continually by plentiful rivers, the Hamoun wetland maintained a significant amount of water over this decade, sustaining the small population with a steady water supply. Secondly, a dry system period (1999–2004) is when Hamoun wetland experienced a significant drought which included the almost complete desiccation of the lake and the absence of sustained swamplands. The evaporation increased, and the water supply from upstream watersheds decreased, being likely a result of harsher climatic conditions during this time. This period represents the most severe hydrologic conditions in the Hamoun wetland. Finally, thirdly, a recovering system period (2005–2016) is when spring floods from the rivers are feeding the Hamoun Lake episodically, increasing the water level. However, during the summertime, climatic conditions instigated high evaporation rates again and these levels were not maintained from year to year. An associated increase in vegetation areas showed a sustaining influence on the fluctuating water level, suggesting that the system is in recovery from the drought period, but it is not yet able to reestablish the wet system by its self.

Causes of the drought are hypothesized to be a variety or combination of factors including a lack of water supply into the lake, increasing population, increasing agricultural and grazing areas, a decrease of water storage in the four main reservoirs (“Chah Nimeh” holding 1530 Mm3 of water; Vekerdy et al. 2006), and a new construction of two dams in Afghanistan (“Kajaki” and “Kmal Khan”; Whitney 2006). Therefore, we claim that this study of LCCs demonstrates the fragility of the Hamoun wetland ecosystem and the unstable feedbacks among winds, dust mobilization, and high evaporation rates that stressed and drove the ecological system to its current recovering state.

Conclusions

Characterizing landscape changes and evolutions in the Hamoun wetland demonstrated the importance of the application solutions for local water budgets, and the necessity to quantify these dynamics and impacts on the ecosystem at the local and regional scales. We observed that the Hamoun wetland showed several fluctuations over short-time scales and that its seasonal variability at long-term trends can provide insights into the future evolution of the landscape. Water loss over the last 30 years has pushed the region from a period of fertility and ecological health to a period of drought and desertification and, finally, to a recovering ecological regime. This recovering state reflects both the expansion of the grassland in response to an increase of water input and the instability of that grassland for overgrazing. This study further determined that (1) fall was the best season for long-term monitoring, being the most representative of the inter-annual variability; (2) water levels were the highest during spring, a result attributed to springtime flooding and fall-winter precipitation; (3) conversely, water levels were the lowest in fall after high evaporation rates during hot and dry summer seasons; (4) vegetation, in both swamp and grassland terrains, showed the lowest expansion from 2000 to 2005 when the drought stressed and reduced the extent of water-loving plant life; and (5) wet-environment land classes (surface water and swamp) exhibited the most inter-annual variability, and the responses of grassland extent to that change in water supply were universally less extreme. We conclude that this study case provides further evidence of ecological transformation in the Middle East and desertification processes associated with both global warming and anthropogenic landscape forcing.