**1. Introduction**

Lava flow emplacement is an important constructive geological process that contributes to reshaping natural landscapes [1–3]. To assess the hazards and long-term impacts posed by lava flows, it is vital to understand aspects such as the return period of effusive eruptions, to map the areas covered by eruptions in the past and to characterize the evolution of lava flow surfaces after emplacement [4,5]. In high eruption frequency areas, lava flows often overlap each other. If the overlapping lava flows erupt within a short time span and have similar chemical and surface characteristics, discrimination will be further complicated by their similar spectral signatures. Spectral reflectance plays an important role in visible and shortwave infrared (VIS-SWIR) remote sensing. Each material absorbs and reflects the incoming radiation in a characteristic way. In the 400–2500 nm range, minerals display absorption features due to the interaction of light with cations (Fe, Mg, Al) and anions (OH, CO3) [6]. Reflectance spectra provide information about the specific material and their composition. They are used for

different applications such as classification of remotely sensed data, identification of mineral features of rock, and environmental assessment [7,8]. The interest in reflectance spectra of volcanic rocks has increased recently as they can play an important role as planetary analogues. In fact, these spectra can be used to identify compounds by data acquired by ongoing solar system exploration missions [9,10].

Characterization of surface spectral reflectance by satellite remote sensing is constrained by the spectral range and resolution (i.e., number of spectral bands) as well as by the spatial resolution of the imagery. Whereas multispectral imagery can be acquired at very high spatial resolution (e.g., WorldView [11,12]); the spatial resolution of hyperspectral satellite data remains low (e.g., EO-1 Hyperion with a ground resolution of 30 m x 30 m); and spectral mixing is thus a major issue [13]. The spectral reflectance of lava of different compositions has also been documented using laboratory spectrometry with decimeter-size samples [14]. For accessible volcanic terrains, field spectrometry offers a useful alternative approach for characterizing the spectral reflectance of contrasted lava surfaces and for documenting its spatial variation at different spatial scales [5,14]. The great variety of morphologies observed in the 2014–2015 Holuhraun lava flows [1,15] encouraged a detailed study of their spectral characteristics, to obtain information about lava composition and detect possible differences in the spectra of the flow. In spectroscopy, the identification of the mineral constituents of major rock types is typically approached using spectral unmixing methods [5,16]. Usually, in the visible and near-infrared spectral range, mafic rocks are characterized by very low reflectance due to the presence of large amounts of dark mafic minerals [14]. The 2014–2015 lava flow at Holuhraun in NE Iceland offers an excellent diverse surface environment for investigating and characterizing lava deposits. Its intense volcanic activity [1,17–19], geomorphological complexity [20], and well-documented flank eruptions [1] perplex the remote sensing monitoring of the bulk volcanic edifice. However, the detailed field mapping of lithologies is frequently obstructed by difficulties in accessibility, the scale of lava flow fields, topography, while remote sensing has become increasingly important in mapping volcanic terrains and specifically in mapping lava flows. Mapping individual lava flows using satellite remote sensing is challenging for at least three reasons: vegetation cover, spatial overlapping, and spectral similarity [3,4]. Moreover, a high eruption frequency often leads to lava flows overlapping each other. If the overlapping lava flows are erupted within a short period and have similar chemical and surface characteristics, discrimination will be further complicated by their similar spectral signatures.

Hyperspectral remote sensing provides information on hundreds of distinct and contiguous channels of the electromagnetic spectrum, thus enabling the identification of multiple ground objects through their detailed spectral profiles. However, restrictions on the spatial resolution of hyperspectral data, the multiple scattering of the incident light between objects, and microscopic material mixing form the mixed pixel problem. Pixels are identified as mixed when they are composed of the spectral signatures of more than one ground object. Therefore, we adopted linear spectral mixture analysis (LSMA) techniques [8,21], which model the pixel spectra as a combination of pure components (endmembers) weighted by the fractions (abundances) that contribute to the total reflectance of the mixed pixel [22]. Ideally, each selected endmember from the hyperspectral image under study has the maximum possible abundance of a single physical material present and minimum abundance of the rest of the physical materials. Spectral unmixing typically consists of two main substages: (a) endmember extraction; and (b) abundance estimation [22]. In this paper, we focus on both endmember extraction and estimation of fractional abundances of the lava field products on 2014–2015 Holuhraun lava fields. For this purpose, an airborne hyperspectral image with an AisaFENIX sensor on board a NERC Airborne Research Facility (Natural Environment Research Council Airborne Research Facility) campaign was acquired at Holuhraun after the eruption and for the sub-pixel analysis we used the sequential maximum angle convex cone (SMACC) algorithm to identify the spectral image endmembers while the LSMA method was employed to retrieve the abundances. Our approach was narrowed to the eruptive fissure vent part since it is considered to have a more diverse surface. The resulting abundances from the LSMA method were both quantitatively and qualitatively compared with the spectral indices technique, aerial and field photographs, respectively. The objective was to

retrieve the main lava surface type contributing to the signal recorded by airborne hyperspectral at the very top surface of Holuhraun.

#### **2. The 2014–2015 Eruption at Holuhraun**

The eruption took place in the tectonic fissure swarm between the Bárðarbunga-Veiðivötn and the Askja volcanic systems (Figure 1a). It lasted about six months (31 August 2014 to 27 February 2015) and produced a bulk volume ~1.44 km3 of basaltic lava [1]. Lava effusion rates during the eruption period range from 320 to 10 m3/s. Averaged values are ∼250, 100, and 50 m3/s during the initial (August–September 2014), intermediate (October–December 2014) and final phase (December 2014 to February 2015), respectively [1,17] (Figure 1b). The lava was emplaced on the sandur plains (glacial outwash sediment plains) north of the Vatnajökull/Dyngjujökull glacier, partially covering the previous two Holuhraun lava flow fields south of the Askja caldera [1]. The area is gently sloping (average inclination <0.5%; i.e., ∼0.3◦) to the east-northeast. The shallow gradient resulted in low topographic forcing of the flow and, therefore, rather slow lava flow advance. During its emplacement history, the lava field was initially dominated by channels and horizontal expansion. Then it transitioned to grow in volume primarily by inflation, tube-fed flow (i.e., transport of lava through roofed over partially or filled channels) and vertical stacking of lava-lobes. The 2014–2015 effusive eruption products originate from intense activity in the vent, in which high oxidation occurs in this area. The main lava channel shows significant inflation (5–10 m). Lava advancement rates were generally low ∼0.0167 m/s during the initial eruption phase [1] and dropped to ∼0.0017 m/s during the middle of November 2014 [23]. The six-month-long effusive eruption features diverse surface structures and morphologies. The 2014–2015 lava flow at Holuhraun in NE Iceland offers an excellent diverse surface environment to investigate and characterize lava deposits.

**Figure 1.** Bárðarbunga volcano and the Holuhraun lava flow field. (**a**) geological setting by the Icelandic Meteorological Office (after modification) [24], (**b**) coverage of the three main phases after Pedersen et al [1].

#### **3. Spectral Unmixing on Lava**

Various spectroscopy studies [2,5,7,14,25] over the volcanic area have examined the mineralogical composition of the extensive lava fields. Usually, in the visible (VIS) and near-infrared (NIR) spectral range, mafic rocks are characterized by very low reflectance due to the presence of large amounts of dark mafic minerals [14]. Spectral indices provide the first efficient way to emphasize subtle spectral variations at the surface [26]. More elaborate methods have been developed to discriminate and quantify mixtures of mafic minerals. They have been used to derive composition maps of mafic minerals [27–29]. However, some lava flows can have a similar chemical/mineralogical composition but dissimilar spectral behaviour due to the different grain size, surface texture, and presence of weathering [13,14]. The main components of igneous rocks do not display any peculiar spectral features in the visible and near infrared spectral range. In the case of basalts, the only spectral feature commonly found is an absorption peak, due to iron, located around 1000 nm [26]. However, in the case of hydrothermal alteration, hydroxyl bearing minerals show distinctive absorption features in the 2000–2500 nm spectral region [30]. Because of the heterogeneity of the lava surface, mixed pixels are very common which is illustrated in Figure 2a,b.

**Figure 2.** Illustration of (**a**) the mixed pixel in the lava surface caused by the presence of small, sub pixel targets within the area; (**b**) variability of lava surfaces in Holuhraun lava field which include the oxidizing surface, sulfate mineral, and lava.

Spectral Mixing Analysis (SMA) has been specifically developed to account for mixtures [10]. Analysis of the data sample can simply be performed on these abundance fractions rather than the sample itself. This method is well-suited for spectroscopic analysis because most of the spectral shapes are due to different materials. The signal detected by a sensor at a single pixel is frequently a combination of numerous disparate signals. Unmixing techniques were applied to the volcano of Nyamuragira for discriminating lava flows of different ages by Li et al. [5]. The most recent study by Daskalopoulou et al. [16], used unmixing techniques to segregate lava flows and related products from the historical Mt. Etna. Nonetheless, there are no findings concerning lava flow delineation through unmixing in Iceland.

#### **4. Data Acquisitions and Methods**

#### *4.1. Airborne Hyperspectral Data Acquisitions*

Airborne hyperspectral data were acquired on 4 September 2015 between 16.56 and 17.58 (local time) with an AisaFENIX sensor (Specim, Spectral Imaging Ltd, http://www.specim.fi) [31] on board a NERC Airborne Research Facility (Natural Environment Research Council Airborne Research Facility http://www.bas.ac.uk/nerc-arf) aircraft [32]. Pushbroom VNIR and SWIR sensor, are two separate detectors with common fore-optics. The hyperspectral data contain 622 channels with spectral range from ~400 nm to 2500 nm (break at ~970 nm). The pixel size of this data is explained in Section 4.2.2. In total, eight flights were acquired at the Holuhraun lava flow during this period with an average altitude of 2.4 km (Figure 3a). The data are delivered as level 1b ENVI BIL format files which means that radiometric calibration algorithms have been applied and navigation information has been synced to the image data (Figure 3b). In this study, we subset the data to focus on the area around the eruptive fissures vent (Figure 3c) which is thought to have a diverse surface and has field photographs. Very high-resolution aerial photographs of the lava field (0.5 m spatial resolution) from Loftmyndir ehf (http://www.loftmyndir.is/) [33] were used for comparison and validation of the unmixing results.

**Figure 3.** (**a**) Map showing line acquisition of FENIX hyperspectral image in the Holuhraun lava field; (**b**) Image mosaic from eight FENIX lines collected during the campaign (red box shows the image subset location); (**c**) Image subset of the focusing study area in the eruptive fissure vent of Holuhraun.

#### *4.2. Spectral Unmixing and Abundance Retrieval*

The processing workflow towards unmixing and generating abundance consists of four steps: (1) Atmospheric correction to retrieve surface reflectance; (2) Data masking, geocorrection, reprojection, and resampling; (3) An endmember selection algorithm was adopted to select the endmembers; then a linear spectral mixing analysis method was employed to retrieve the abundance (Figure 4).
