Introduction

Remote sensing methods are utilized in numerous industries, including agriculture, environment, urban planning, meteorology, and disaster management. Additionally, these methods are commonly implemented within the realm of earth sciences such as mining, geology, and mineralogy. For conducting mineralogical studies in environmentally sensitive areas, including special conservation areas, protection of the area is crucial. Remote sensing techniques are commonly utilized in such studies to avoid the possibility of any damage to the area.

Lake Salda, a medium-sized crater lake in southwestern Türkiye, which came to prominence with the Jezero Crater on Mars, was considered the most similar region on Earth to look for evidence of life on Mars (Horgan et al. 2020). The first goal of the Perseverance rover, which explores Jezero Crater, is to identify past environments that can support microbial life and to look for signs of possible past microbial life. In this regard, Lake Salda, whose shoreline is formed by microbialites, has been the focus of research (Garczynski et al. 2019).

Russell et al. (1999) report that hydromagnesite stromatolites are formed in and around Kocaadalar Burnu peninsula, Salda Lake provide excellent opportunities for study of carbonate morphologies in preparation for an exobiological exploration of Mars. Kazancı et al. (2004) examined the limnological characteristics of Lake Salda, which is a deep, soda and highly alkaline lake.

Wang and Zheng (2010) used the Support Vector Machine (SVM) classification method in their study using EO-1 Hyperion hyperspectral data from BeiYa region in northwestern YunNan Province, China. Because of their study, they found that alteration zones can be identified very effectively with Hyperion data and that the mineralogical and lithological information generated from these data is largely consistent with the geological map and previous research results.

Salaj et al. (2012) used Hyperion data to map mineral abundance in Udaipur region of Rajasthan, India. They used Blend Adjustment Matched Filtering and Spectral Angle Mapper (SAM) methods in their study, using FLAASH software and the United States Geological Survey (USGS) mineral library. They concluded that although the spatial distribution of minerals is influenced by topographic effects and vegetation cover, Hyperion data are useful for determining mineral abundance and mapping geologic features.

Pour and Hashim (2014) used Landsat ETM+ and Hyperion satellite data to perform mineral mapping in the study area and surrounding terrain in the Bau gold mining area in the state of Sarawak, East Malaysia, Borneo Island. Because of the study, they have demonstrated the importance and advantages of using ETM+ and Hyperion remote sensing data together to identify potential areas in tropical/subtropical regions. They also demonstrated that indicators of structurally controlled gold mineralization, including iron oxides, clay minerals, faults, and fractures in the Bau gold mining area, can be detected using satellite remote sensing data.

Ducart et al. (2016) conducted a study using remote sensing methods in Serra Norte iron deposits in Caracas, Brazil. They used the EO-1/Hyperion satellite data as satellite images, and used the Spectral Angle Mapper (SAM) classification method. As a result of the study, they obtained good results in mapping high-grade iron ore, hematite–goethite, and clay minerals of regolith origin, and stated that EO-1/Hyperion imagery is an excellent tool for rapid mineral mapping in open pit areas, as well as for mapping waste and waste disposal facilities.

Oskouei and Babakan (2016) used Hyperion satellite data and a standardized hyperspectral processing method developed by Analytical Imaging and Geophysics (AIG) to map the mineral distribution in the study area in the Lahroud region of Ardabil province, northwestern Iran. As a result of the study, they found that the mineral distribution maps they produced were in high agreement with the geological map of the region. They also stated that it can be successfully used to reveal the mineralization trend in the region.

Balcı et al. (2018) investigated the geomicrobiology of Lake Salda and stated that stromatolites forms under a complex biological, chemical and physical biogeochemical processes and also indicate that bacteria serve as a nucleation site and control kinetic factors such as precipitation rate by lowering activation energy rather than thermodynamic factors (e.g. alkalinity, pH). Garczynski et al. (2019) reported that Lake Salda is good compositional analogs for a Jezero paleolake.

Kah et al. (2020) investigated the mineralogy of authigenic and detrital sediments of Salda Lake to understand the nature of carbonate accumulation in the Jezero crater. As a result of this study, it is reported that hydromagnesite is rare in deltas and is associated with paleo-shorelines, and the formation of darker-toned Fe/Mg-smectite muds in Salda Lake may be due to groundwater or stream input, but further studies are needed to confirm the formation mechanisms.

Varol et al. (2020) investigated the geological, hydrological, hydrogeological and hydrochemical characteristics of water resources in the south of the Salda Lake basin. The results of this investigation showed that groundwater and surface waters are in the Mg-Ca-HCO3 and Ca-Mg-HCO3 facies.

Bouzidi et al. (2022) conducted a study at the Tamera open pit mine in the Nefza mining district of northwestern Tunisia, which shows strong potential for heavy iron mineralization. They examined iron oxide mineralization using EO-1 Hyperion hyperspectral data and compared laboratory analyzes of 56 samples with satellite data. As a result of their study, they demonstrated the presence of a mineralogical composition consisting primarily of iron oxide minerals (hematite and goethite) and minerals such as clays (kaolinite) in all samples.

Kaya et al. (2023) investigated trace and Rare Earth Element, geochemistry of recently formed stromatolites at Lake Salda. One of the results of this study indicated that hydromagnesite composition of the Salda Lake stromatolites were precipitated from the waters influenced by Mg-rich meteoric waters fed from the serpentinite rocks around the Lake.

In this study, the mineral distribution in Lake Salda was studied using the Hyperion sensor data on the EO-1 satellite capable of hyperspectral sensing, the HMI tool in the THOR module within the ENVI software, the USGS spectral library, and the ACE algorithm.

Materials and methods

Study area

Lake Salda is a somewhat salty karst lake in the Yeşilova district of Burdur, Türkiye, 4 km from the district seat. It is flanked by forest-covered hills, stony terrain, and tiny alluvial plains. The Lake Salda basin is located between 37° 30′ 33″ and 37°35′10″ northern latitudes and 29° 38′ 28″ and 29° 43′ 34″ east longitudes in the geographical coordinate system. In the Lakes Region, it has a closed basin structure with no outflow. It is Türkiye’s third deepest lake, with a surface size of around 44 km2 and a depth of up to 184 m. In this study, an area of approximately 57 hectares in size in and around the peninsula of Kocaadalar Burnu, located on the southwest shore of Lake Salda, was examined. The altitude of the study area is 1145 m above sea level, it has a flat topography and no vegetation. (Fig. 1).

Fig. 1
figure 1

Study area

Full size image

Lake Salda was designated and declared a Special Environmental Protection Area by the President’s Decision dated 14.03.2019 and numbered 824, which was published in the Official Gazette on 15.03.2019 and numbered 30,715.

Lake Salda is a tectonically-formed lake that began to form at the end of the Neogene period and took on its present shape as a result of subsequent tectonic movements (Akkuş 1987). The accumulation of surrounding waters in an anticline area and a segment where the fold axes decline led to the initial formation of the lake (Altınlı 1955). The structure is situated in the “Taurus Tectonic Union” of Anatolia’s tectonic units. It was formed during the Alpine Orogenesis and is occasionally found among serpentinised ophiolites. Limestones are exclusively present on the Kale hill north of Kaledibi village, southeast of the lake. The lake’s surrounding is made up of Peridotides, which at times are found alongside limestones from the Mesozoic era (Erentöz 1974). The geological map of the study area made by the Institute of Mineral Research and Exploration (MTA) is given in Fig. 2.

Fig. 2
figure 2

Study area geological map (MTA 2023)

Full size image

Materials

The Earth Observing Mission-1 (EO-1) satellite, launched by the United States on a Delta II rocket from Vandenberg Air Force Base on November 21, 2000, was the first spacecraft of the American New Millennium Program (NMP) (NASA 2023). The EO-1 satellite has three instruments: the hyperspectral Hyperion instrument, which we used in this study, the multispectral Advanced Land Imager (ALI) instrument, and the LEISA Atmospheric Corrector (LAC) instrument. The Hyperion instrument has a spatial resolution of 30 m and can detect in 220 spectral bands from 0.4 to 2.5 µm. The ALI instrument was a multispectral sensor with the same spatial resolution as Hyperion and can detect in 10 spectral bands from 0.4 to 2.4 µm. The LAC instrument covers the 0.89–1.6 µm wavelength IR band using a wedged etalon filter and three arrays of In-Ga detectors, each array providing 250 m spatial resolution (Barry 2001).

There are 220 unique spectral channels captured with a total spectral range of 357–2576 nm. The Level 1 radiometric product has a total of 242 bands, but only 198 are calibrated. There are only 196 unique channels due to overlap between the VNIR and SWIR focal planes. Calibrated channels are 8–57 for VNIR and 77–224 for the SWIR. The lack of calibration of all 242 channels was mostly due to the low sensitivity of the detectors. In some channels, the uncalibrated bands are set to zero (Beck 2003). The characteristics of the Hyperion data used in the study are shown in Table 1, and the footprint of the satellite is shown in Fig. 3.

Table 1 Hyperion instrument specifications (Beck 2003)
Full size table
Fig. 3
figure 3

Hyperion image footprint

Full size image

Hyperion data are processed on three different levels: L1R, L1Gst and L1T. At the L1R level, radiometric corrections were applied to the images, while at the L1Gst level, in addition to the L1R level, a 90-m digital elevation model (DEM) was used for topographic accuracy and systematic geometric corrections were applied from spacecraft ephemeris data. At the L1T level, radiometric and systematic geometric corrections including ground control points were applied in addition to the L1Gst level for geodetic accuracy (Simon and Beckman 2006; USGS 2021). Since only the L1R and L1Gst levels of the study area are available in this study, the L1Gst data, which have increased topographic accuracy through the use of DEM data, were used in addition to the L1R level data. Hyperion data were acquired from EarthExplorer, an online platform that facilitates online search, browsing, metadata exportation, and data downloading for earth science data archived by the USGS. As a result of the archive search, it was determined that there was only one image in the study area. The name and date of this image are listed in Table 2 (USGS 2023).

Table 2 The name of the Hyperion image and date, used in the study
Full size table

Methods

The Environment for Visualizing Images (ENVI) software was used to perform radiometric and atmospheric corrections on the Hyperion data, which can detect at 357 and 2576 nm wavelengths and 30 m resolution in 220 bands used in the study. ENVI is a software with advanced image processing capabilities that allows meaningful information to be extracted from a wide range of data. In hyperspectral satellites, some bands can cause significant interference, particularly at wavelengths associated with atmospheric absorption and water vapor. These bands must be removed before analysis. In EO-1 Hyperion data, bands 1–7, 58–76, and 221–242 are automatically set to 0 by the data provider (Barry 2001). Also, as noted by Datt et al. (2003), bands 121–126 and 167–180 were removed in the preprocessing step due to the strong noise generation corresponding to strong water vapor absorption.

After removing the bad bands, the radiometric corrections necessary to use the image were calculated using the ENVI software, and the Quick Atmospheric Correction (QUAC) module of the same software was used for the atmospheric corrections calculated according to Canty (2014). QUAC is a scene-based empirical method that converts radiance values to apparent surface reflectance. Scene-based means that the atmospheric correction parameters are derived strictly from the pixel spectra within the scene and not from ancillary data (Bernstein et al. 2005, 2012). Mineral analysis of radiometrically and atmospherically corrected images was performed using the Hyperspectral Material Identification (HMI) tool in the Tactical Hyperspectral Operations Resource (THOR) module of the ENVI software.

In the THOR HMI analysis, the United States Geological Survey (USGS) spectral library was used and the best matching minerals were determined using the Adaptive Coherence Estimator (ACE) algorithm. Hyperspectral Material Identification (HMI) tool in the Tactical Hyperspectral Operations Resource (THOR) module of the ENVI software generally focuses on qualitative mineral identification. It helps identify and classify materials based on their spectral signatures.

The USGS spectral library contains a collection of spectra collected with laboratory, field, and airborne spectrometers (Clark et al. 2007). The wavelengths covered by the instruments range from ultraviolet to far infrared (0.2–200.0 μm) (Kokaly et al. 2017). The spectra in the library were measured using four different types of spectrometers, and data from the BeckmanTM 5270 spectrometer in the 0.2–3.0 m spectral range were used in this study.

The ACE technique identifies whether a pixel spectrum could contain a known target signature. This is particularly useful when the conditions of the background are uncertain and varied. It is not affected by the scaling of input spectra, and all the end members in the scene do not need to be known (Manolakis et al. 2003; Kraut et al. 2005). Its performance is reliable in the detection of sub-pixel targets (Wolfe and Black 2018). The ACE algorithm is used for the mineral that is dense in the pixel. ACE values range from − 1 to 1, with values close to 1 indicating the best match (Nv5 2023). This study was conducted according to the flow chart shown in Fig. 4.

Fig. 4
figure 4

Flowchart of the study

Full size image

The examined points were created using the ArcMap program (ESRI 2011). For analysis, 634 points were constructed based on the midpoint of each pixel of Hyperion data with a spatial resolution of 30 m (Fig. 5).

Fig. 5
figure 5

Map showing the location of pixels

Full size image

Results and discussion

ACE values for each pixel were calculated separately for all 428 minerals in the “Minerals” category of the USGS spectral library. In addition, hydromagnesite, which is not in the “Minerals” category of the USGS spectral library, but in the “Soils and Mixtures” category, was also analyzed. Hydromagnesite has been reported to occur frequently in this region, which is the reason for this analysis.

ACE levels between 0.9 and 1.0 are very acceptable, good between 0.75 and 0.9, poor-appropriate between 0.5 and 0.75, and inappropriate for values less than 0.5 (Nv5 2023). Minerals with ACE values greater than 0.75 are given in Table 3. There were 23 minerals with ACE values greater than 0.75, 202 minerals with values between 0.5 and 0.75, and 203 minerals with values less than 0.5.

Table 3 Results of ACE analysis
Full size table

Figure 6 shows the wavelength plot in the USGS spectral library of six minerals that have a distribution greater than 5% in the study area, as well as the wavelength plot of the Hyperion pixel that has the largest distribution in the ACE analysis. Since the hydromagnesite (Mg5(CO3)4(OH)2·4H2O) was detected in all pixels, it is not included here.

Fig. 6
figure 6

The wavelength graphs of the six minerals that have a distribution greater than 5% in the study area

Full size image

Trona mineral with the chemical formula Na3H(CO3)2 2H2O (Chukanov 2014) is shown in Fig. 6a, ulexite mineral with the chemical formula NaCaB5O9 8H2O (Korbel and Novák 2001) is shown in Fig. 6b, tincalconite mineral with the chemical formula Na2(B4O7) 5H2O (Kokaly et al. 2017) is shown in Fig. 6c, vermiculite mineral with the chemical formula (Mg,Fe+2,Al)3((Al,Si)4O10)(OH)2 4H2O (Kokaly et al. 2017) is shown in Fig. 6d, colemanite mineral with the chemical formula Ca2B6O11 5H2O (Kokaly et al. 2017) is shown in Fig. 6e, rivadavite mineral with the chemical Formula Na6MgB24O40 22H2O (Kokaly et al. 2017) is shown in Fig. 6f.

The band depths of the minerals detected as a result of ACE analysis and whose wavelength graphs are given in Fig. 6 were also analyzed to determine at which wavelengths they were absorbed. Band depth (absorption) is one of the most widely used methods in the parameterisation of spectral data (Clark and Roush 1984). For this purpose, the wavelength of the pixel where the minerals are most intense was removed and shown as “Continuum Removed” (Fig. 7). The Continuum Removed method normalises the wavelength by comparing it to a continuum curve, thus defining a common baseline for measuring absorption feature depth and position. A continuum curve is a mathematical function created by fitting a convex hull to the spectrum. The resulting reflection curve gives a better picture of band centres and depths (Wolfe and Black 2018). As a result of the analyses, the wavelengths at which the minerals are absorbed in the most intense pixel and the graph showing the wavelengths absorbed by the minerals taken from the USGS spectral library are shown in Fig. 7.

Fig. 7
figure 7

The Continuum Removed graph in the pixel with the highest ACE values of the six minerals

Full size image

Analysis reveals that the examined area consists of hydromagnesite along with trona and borates such as ulexite, tincalconite and colemanite. Hydromagnesite mineral (Mg5(CO3)4(OH)2∙4H2O) is detected in all of the study area as a result of ACE analysis. Trona (Na3H(CO3)2 2H2O) detected in 35% of the study area as a result of ACE analysis. The mineral was detected with higher values in the northern part of the study area, starting at a distance of approximately 100 m from the shore of Lake Salda and inland. In the eastern part of the study area, it was detected inland at a distance of approximately 250 m from the shore (Fig. 8a). Trona deposits, which have a limited distribution in the Earth’s crust, can be found in unusual concentrations in some places, usually together with borate and other salt deposits. Their origin is related to alkaline volcanism, hot spring activity, closed basins, and arid climate, and Quaternary sodium carbonate minerals (trona minerals) are found in salars, playa lakes such as Lake Van, and salt lakes (Helvacı 2019a).

Fig. 8
figure 8

Distribution maps of the six minerals

Full size image

According to ACE analysis, ulexite (NaCaB5O9 8H2O) mineral is found in 26.2% of the study area, 50 m away from the beach and in the central parts of the study area. (Fig. 8b). Ulexite, an evaporite mineral commonly found with colemanite in sedimentary rocks in areas where borax deposits occur (Korbel and Novák 2001). Balcı et al. (2018) reported that Na2O and Boron (B) was detected in 10 different samples taken from the Salda Lake sediments, hydromagnesite terrace and Kocaadalar Burnu peninsula. Typically, in playa and salt marsh deposits in arid regions, and in bedded sedimentary deposits formed from them, boron is supplied by surrounding hot springs (Anthony et al. 2003). Although many minerals contain boric oxide, the three most commercially important minerals in the world are borax, ulexite, and colemanite, which are produced in a limited number of countries. Türkiye has the largest reserves of borax, ulexite, and colemanite in the world (Helvacı, 2019b).

The distribution of tincalconite (Na2(B4O7) 5H2O) detected in 19.6% of the study area is shown in Fig. 8c. It is a hydrous sodium borate mineral closely related to borax and is a secondary mineral formed as a dehydration product of borax (Anthony et al. 2003). Tincalconite is typically found as a fine-grained white powder. It is formed when borax is exposed to air and decomposes (Helvaci et al. 2017), as well as naturally occurring, such as in Searles Lake, California. It has been found in California, USA; Salta, Argentina; Eskisehir, Türkiye; Tuscany, Italy; and the Kerch Peninsula, Ukraine (Anthony et al. 2003).

The distribution of vermiculite ((Mg, Fe+2,Al)3((Al,Si)4O10)(OH)2 4H2O) detected in 9.8% of the study area is presented in Fig. 8d. In vermiculite minerals, which are expandable phyllosilicates, the Fe oxidation state is directly affected by both external redox conditions and microbial activities. These reactions are important for microbial respiration and, more generally, in soils where water saturation conditions, microbial activity, and organic matter concentrations modify the Fe oxidation state in phyllosilicates (Cuadros et al. 2019). An alteration product of biotite or phlogopite by weathering or hydrothermal action. Formed at the contact between felsic and mafic or ultramafic rocks, such as pyroxenites and dunites, and in carbonatites and metamorphosed limestones (Anthony et al. 2003).

The distribution of colemanite (Ca2B6O11 5H2O) detected in 7.1% of the study area is shown in Fig. 8e. A common constituent in borate deposits formed in arid alkaline lacustrine environments, deficient in sodium and carbonate, typically under warm conditions (Anthony et al. 2003). Borax, ulexite, and colemanite, which are the most commercially dominant and most important borate minerals worldwide, are produced in a limited number of countries, essentially by Türkiye and the United States. These minerals are largely a constituent of economic nonmarine evaporates formed concentrated in continental Tertiary deposits of western Anatolia (Türkiye) and the American continents (Helvacı et al. 2017).

Rivadavite, a hydrated sodium and magnesium borate, was detected in 5.0% of the study area and its distribution is shown in Fig. 8f. The mineral, which is found in sedimentary borate deposits and deposited from borate-rich hot springs, has been identified in the Salta region of Argentina, the California region of the USA, and the Balıkesir region of Türkiye (Anthony et al. 2003).

The Karakova Stream and its tributaries, descending from the hills up to 2050 m high around Salda Lake, carry eroded materials with high energy and form the Karakova fan delta in the study area. Magnesium and sodium concentrations in wells and cold water springs in the ultramafic surroundings of the lake average 60 ppm and 5 ppm, respectively. Temperature and pH vary between 12–15 °C and 8–10, respectively (Zedef 1994; Braithwaite and Zedef 1996). Water samples collected from the lake surface and up to a depth of 80 m always contained 300 ppm Mg and 200 ppm Na at pH 9.1. After evaporation and precipitation of alkaline earth carbonate minerals, they transform into high pH magnesium carbonate brines that precipitate hydromagnesite (Mg5(CO3)4(OH)2∙4H2O) (Russell, 1999). High pH sodium carbonate brines can also cause trona (Na2CO3 NaHCO3 2H2O) to precipitate. Helvacı (1998) reported the presence of magnesite (MgCO3) in the Beypazarı/Türkiye trona field.

Conclusion

Understanding the historical process of Mars, which used to have water and is now arid, is important for predicting the future of our planet. However, although these studies on Mars are carried out with the help of expensive satellites, landers, and rovers, they remain weak in comparison with the studies to be carried out on Earth. Instead of the Jezero crater on Mars, it is essential to make detailed analyzes in Lake Salda, which is similar to it in our world, not only to understand the history of our Earth, but also to understand the history of Mars.

Collecting representative samples in a large area of approximately 57 hectares in Salda Lake brings with it many difficulties in terms of time, cost and manpower, and in a protected area, such activities can even harm the area in terms of research. Therefore, remote sensing applications are particularly useful especially in protected areas such as Lake Salda limits the direct geological exploration. It is concluded that EO-1 Hyperion satellite observations will help to better understand the mineralogical characteristics of the Salda Lake surroundings. In this study, mineral analysis was performed using remote sensing methods without damaging such a specially protected sensitive area.

Analysis reveals that the examined area consists of hydromagnesite along with trona and borates such as ulexite, tincalconite and colemanite. Hydromagnesite is the most abundant minerals in the study area. As a result of mineral analysis using the Hyperion satellite with hyperspectral detection, 35% trona, 26.2% ulexite, 19.6% tincalconite, 9.8% vermiculite, 7.1% colemanite and 5.0% rivadavit minerals were detected on the surface of the 57-hectare study area.