Introduction

Hyperspectral airborne images of the 1996 volcanic eruption of Mt. Etna, located in eastern Sicily, Italy (Fig. 1) were collected using the DAIS 7915 spectrometer (Digital Airborne Imaging Spectrometer) deployed during a multi-sensor campaign of Italian volcanic systems (Horne et al. 1997). We used these data to carry out an analysis of the thermal structures associated with volcanic activity.

Fig. 1
figure 1

Location map of Mount Etna: shaded relief draped over a three-dimensional image derived from a 10 m resolution digital elevation model (DEM) generated by Istituto Nazionale di Geofisica e Vulcanologia, Italy. A 3× exaggeration of the vertical scale was adopted. The square box indicates approximately the area imaged by DAIS (see Fig. 2)

Full size image

Crisp and Baloga (1990) describe the thermal flux for active lava flows as a function of the fractional area of two thermally distinct radiant surfaces: a spatially larger component corresponding to the cooler crust of the flow and a much smaller component related to fractures in the crust. These fractures exhibit a temperature (T h) assumed to be comparable to the temperature of the molten inner core and that is higher than the crust (T c). It has been observed that the temperature exposed in hot cracks is 100–200 °C cooler than the molten interior (Flynn et al. 1994; Calvari et al. 1994). This temperature variation for T h does not affect the results of the presented work since it uses a statistical approach. With this assumption, dual band techniques allow the calculation of T c and the fractional area of hot cracks (f h). This sub-pixel temperature retrieval technique (Dozier 1981; Matson and Dozier 1981; Wan and Dozier 1989) requires the availability of a sensor equipped with two bands within the short wave and/or thermal infrared.

This technique has been performed using different sensors characterized by low spatial resolution such as the 1 km2 data of the advanced very high resolution radiometer (AVHRR; Mouginis-Mark et al. 1994; Harris et al. 1995a, 1997; Harris 1996) and the along track scanning radiometer (ATSR; Wooster and Rothery 1997a,b). In addition it has been applied to Landsat thematic mapper (TM) data, an instrument with a fairly limited dynamic range of 8 bits (Rothery et al. 1988; Glaze et al. 1989; Pieri et al. 1990; Oppenheimer 1991; Flynn et al. 1994; Harris et al. 1998; Wright et al. 2001; Pieri and Buongiorno, 2001 personal communication). The high spatial resolution (30 m per pixel) of TM is important because it increases the number of samples (pixels) available over an active flow. However many pixels located over active lava flows are often saturated in AVHRR and TM data, reducing the number of ‘effective’ samples available for calculations. In fact, sensors characterized by an 8 bit dynamic range, such as Landsat TM, are badly affected by saturation of radiant pixels containing active vents or located in the middle of a lava flow (Pieri et al. 1990; Oppenheimer 1991; Rothery et al. 1992; Flynn et al. 2000, 2001). Data saturation makes statistical evaluations of the thermal structures that characterize an active flow complicated.

The high spatial resolution of DAIS data (2–20 m; Table 1), with its 15 bit dynamic range, provides a way to test and develop algorithms on a large number of unsaturated points. This makes DAIS-type data ideally suited to statistical investigations of radiant pixels. In addition, high spatial resolution data enhance the discrimination between the active part of lava flows and the much cooler ground surrounding the lava margins. This reasonably prevents the inclusion of an extra radiative contribution to pixel-integrated temperatures coming from a non-active lava background. Here, we illustrate our statistical approach using two DAIS images obtained on 16 and 18 July, 1996 and coinciding with a phase of summit activity at Mount Etna.

Table 1 Major characteristics of the DAIS 7915 sensor: IFOW Instantaneous Field of View, GIFOV Ground projected Instantaneous Field Of View, and detector types (Si Silicium, InSb Indium-Antimonide, MCT Mercury-Cadmium-Telluride), and FWHM for all bands—center wavelength and full width at half maximum value (FWHM) for all bands (spectral calibration results of April 1997)
Full size table

The 1996 Mt. Etna eruptive activity

The eruptive activity at Mount Etna during the 1996 DAIS campaign (Horne et al. 1997) was recorded on two dates, 16 and 18 July. This phase of the activity persisted from 1995 to 1999 and was one of the most complex and long lasting eruptive cycles at Etna of recent times (GVN 1997). Activity began at the Bocca Nuova crater during late July 1995, and was followed by a gradual reactivation of the North-East (NE) crater (Fig. 1). By July 1996, three of the four main summit craters of Mt. Etna (Fig. 1) were active: the North-East, La Voragine and Bocca Nuova (GVN 1996a). Figure 2 shows the locations of the vents and other morphologic features of the summit crater area of Mt. Etna, retrieved from an analysis of the 16 July 1996 (Fig. 2a) and of the 18 July 1996 DAIS images (Fig. 2b).

Fig. 2
figure 2

a Location of the vents and morphological features of the craters as revealed by band 36 (1.62 μm) of the 16 July 1996 DAIS image. Flight altitude is 4,000 m corresponding to a ground resolution of 2.3 square meters. b Location of the vents and morphological features of the craters detected by band 36 of the 18 July 1996 DAIS image. Flight altitude is 6,000 m corresponding to a ground resolution of 8.9 m2

Full size image

Activity at the NE crater between July 1995 and July 1998 can be divided into several distinct periods of eruptive behavior. Between mid-February and late-August 1996, persistent Strombolian activity occurred, punctuated by two paroxysmal episodes of fire-fountaining in June 1996 and culminating in four weeks of lava effusion. This intra-crater activity is recognizable in the DAIS images as a zone of highly radiant pixels within the crater. Radiance statistics for pixels associated with this activity are reported in Table 2. The initially weak activity at La Voragine was characterized by an open pit that had been degassing quietly for 2 years (GVN 1996b, 1997). This crater became active only after 1997 when it produced two of the most intense eruptive events of the 1995–1999 period (GVN 1998). The first episode began on June 1998 and lasted 4 months. Explosive activity, lava effusion and scoria emission characterized this eruptive phase. The second episode occurred on 4 September 1999, when a lava fountain rose hundreds of meters above La Voragine and a tephra-laden eruption column rose ∼2 km above the vent before being blown eastwards (GVN 1999). However, the only activity at La Voragine during July 1996 when the DAIS data were collected consisted of continuous gas emission. In the DAIS data there were 52 radiant pixels related to this activity, of which none were saturated (Table 2).

Table 2 Statistics for saturated pixels in the DAIS imagery obtained for each hot spot
Full size table

During the first two years of the 1996–1999 eruptive cycle, activity within the Bocca Nuova crater was moderate, resulting in a slow filling of the crater by intra-crater cones and lava flows (GVN 1996a). The eruption of lava within the crater was occasionally accompanied by gaseous emissions from fumaroles scattered around the intra-crater. Radiant pixels occur in two zones within the Bocca Nuova crater: (1) a circular zone, including some saturated pixels, in the center of the crater and (2) a zone scattered in the southern part of the crater (Fig. 2). Saturated pixels only occur in the central zone of 208 pixels (Table 2).

Data reduction

The DAIS instrument consists of four spectrometers known as VIS-NIR, SWIR-1, SWIR-2 and thermal, covering a spectral range of 0.4–12.6 μm for a total of 76 bands (see Table 1 for more details). To constrain lava temperatures, we consider spectrometers II (SWIR-1) and III (SWIR-2). These sensors are equipped with 37 bands in the 1.54–2.49 μm range. These are suitable for dual band style calculations (Rothery et al. 1988). Unfortunately, not all the bands could be used in the dual band equations because of strong noise contamination, especially in the SWIR-1 data across the 1.54–1.73 μm wavelength range

The dual band system equation adapted for the DAIS bands can be summarized as follows (e.g. Rothery et al. 1988):

$$\begin{array}{*{20}c} {R{\alpha } = f_{\rm h} (R_{\rm h} {\alpha}) + (1 - f_{\rm h} )R_{\rm c} {\alpha }}\\ {R{\beta } = f_{\rm h} (R_{\rm h} {\beta }) + (1 - f_{\rm h} )\, R_{\rm c} {\beta}}\end{array}$$
(1)

where R α and R β are respectively the total radiance in two suitable bands (α and β) and R hα and R hβ are the radiance values calculated for the same bands using the Planck equation with 1,080 °C as the hot temperature (T h) for Etna lavas (GVN 1996a, 1999; Calvari et al. 1994; Archambault and Tanguy 1976; Gauthier 1973). The simultaneous solution of the equations now allows for the calculation of the fractional area of the hottest component   f h, as well as the cooler crust temperature T c. In applying Eq. (1), our aim was to obtain T c and   f h for every pixel across the thermal anomaly.

Previous results achieved from Landsat TM imagery

The bands used in Eq. (1) were selected to allow a comparison between our new results and the previous results achieved at Etna using Landsat TM data (Lombardo et al. 2004). The DAIS bands used for the dual band calculations were thus selected so that they were as close as possible to the corresponding SWIR bands of the Landsat TM. These are TM bands 5 and 7, corresponding to central wavelengths of 1.65 and 2.22 μm, respectively. The DAIS instrument provides two bands (band 37 and 54, Table 1) that fall within the same wavelength region as TM bands 5 and 7. Unfortunately DAIS band 37 (1.650 μm) is one of the bands affected by noise, having an extremely poor signal to noise ratio. For this reason band 36 (1.619 μm) was preferred in the solution of the dual-band equations.

The statistics derived from our Landsat TM results (Lombardo et al. 2004) show a recurring trend when we plot f h vs. T c for distinct eruptive events. A logarithmic curve provides the best fit for this distribution of points (Lombardo et al. 2004). This trend shows a systematic increase in f h as T c increases. Such a positive relationship is consistent with a lava-flow surface structure where larger areas of the exposed hot core often occur when the crust temperature is higher (i.e., more active flows).

Statistics for saturated pixels in our Landsat data are reported in Table 3. The percentage of unsaturated pixels available to perform the dual band system equation, and thus the calculation of T c and   f h, ranges from 59 to 29 percent. The total number of non-saturated pixels varies from 41 to 1,428; the upper value corresponding to data collected when a large lava field was active within Valle del Bove during 1992. In comparison, the average number of non-saturated pixels available for all the other eruptions is about 220 (Table 3). We can compare these statistics with a value of 95% for the percentage of unsaturated pixels available from the 16 July 1996 DAIS image. This corresponds to a total number of 2,933 unsaturated pixels. The lower spatial resolution (30 by 30 m) and limited dynamic range (8 bits) of Landsat TM data accounts for this difference.

Table 3 Statistics for saturated pixels in the time series of Landsat TM images analyzed by Lombardo et al. (2004)
Full size table

Data analysis

We analyzed the scatter plots of derived crust temperature T c vs. the hot crack fractional area f h for each of the intra-crater hot zones identified in Fig. 2. We found that each distribution revealed a similar trend and matched the expected thermal structure fingerprint for active lava. Figure 3a displays a logarithmic trend linking T c and f h for the unsaturated pixels of the 16 July 1996 lava flow that erupted onto the floor of the NE crater. This distribution thus shows a trend that can be summarized as follows. Between T c of 0 and approximately 160 °C, increases in T c occur without a significant increment in the fractional area of the hot crack. With higher values of T c (>160 °C), f h increases exponentially so that small variations in T c correspond to a proportionally larger increment in the fractional area. Figure 3b shows the same data plotted on a log scale, confirming the logarithmic relationship between T c and f h. This plot reveals that the derived data lie along a line with a very narrow dispersion range. This trend is expressed by:

$$ T_{\rm c}=61\,\ln (f_{\rm h}) + 652$$
(2)
Fig. 3
figure 3figure 3

a Arithmetic and b logarithmic plot of T c vs. f h for the North-East crater hot-spot obtained from the 16 July 1996 DAIS image. Dashed lines in logarithmic plot identify typical scatter about the best-fit trend line (gray)

Full size image

with an R 2 of 0.67.

The logarithmic trend observed in our DAIS data (Fig. 3a) is comparable to the plot derived from the Landsat imagery. However, the relationship defined using Landsat TM data is incomplete. This is due to the inferior dynamic range of the TM sensor, which produces saturation at lower temperatures and causes the high-temperature limb of the relationship to be clipped (Fig. 4). Figure 4 shows a plot of T c vs. f h obtained using the 30 September 1992 Landsat image in the 0–250°C temperature range. It is easy to recognize the same trend shown in Fig. 3a.

Fig. 4
figure 4

Plot of T c vs. f h derived from the 30 September 1992 Landsat TM image (red dots) and the 16 July 1996 DAIS data (blue dots). We use these scatter distributions to compare the effect of the dynamic range on T c and f h solutions. We observe that saturation in TM data (A curve) occurs at a lower level than DAIS data (B curve)

Full size image

A comparison of these two plots (Fig. 4) is instructive in understanding how the dynamic range affects the dual-band technique. The most relevant detail is that the 15-bit DAIS data allows solution for a wider range of T c. Indeed, the T c upper limit for Landsat TM data is one half (about 250 °C) of the DAIS T c upper limit of 450 °C. It follows that it is impossible to entirely determine the full relationship between T c and f h, using Landsat TM data alone.

Figure 5a and b map the derived crust temperatures and fractional areas respectively, obtained for the 16 July 1996 DAIS image. The spatial distribution of T c in the NE crater (Fig. 5a) shows a concentric arrangement. Cooler values are located along the lava boundaries while the hottest pixels are prevalent at the center of the anomaly. These results relate to a ground-based situation in which a lava pond was present (GVN 1996a). This is not to be confused with an active lava lake which forms directly on top of the magma column (Swanson et al. 1979). Effusive activity recorded at the NE crater consisted in a ponding of lava that was trapped in the topographic low (GVN 1996a). Degassing from the vent beneath the pond will cause disruption of the surface crust at the feature center and would thus explain the spatial distribution of T c and f h observed by us for this case. Also, the fractional area occupied by the cracks increases from the periphery to the center of the ponds with highest values of f h occurring at feature center (Fig. 5b).

Fig. 5
figure 5figure 5figure 5

a Crust temperature T c and b fractional area f h spatial distributions obtained using the dual band technique with the 16 July 1996 DAIS image. c Location of two spatial profiles taken across the North-East crater

Full size image
Fig. 6
figure 6

Crust temperature T c and fractional area f h profiles (horizontal and vertical) across the NE crater hot spot on the 16 July image (Fig. 5)

Full size image
Table 4. Maximum, minimum and mean T c and f h values derived from DAIS data for each hot spot
Full size table

Figure 6 shows spatial profiles of T c and f h across the NE crater lava pond (see Fig. 5c for profile locations). These exhibit a rapid increase of T c when moving from the edge to the feature center, which is not matched by a proportional increment in f h, which tends to vary at a slower rate. Approaching the middle of the lava body, however, f h begins to increase rapidly so that T c and f h reach their maxima at the same central location.

Similar results are obtained from the 18 July 1996 DIAS image. Compared to the 16 July image, the percentage of unsaturated pixels is higher (91%) but the total number of radiant pixels is lower (1,726). This is related to the lower spatial resolution of 8.9 m due to the higher flight altitude of 6,000 m adopted for the 18 July 1996 acquisition. The 16 July flight altitude was 4,000 m, corresponding to a ground resolution of 2.3 m. The 18 July image is useful because it allows for the comparison between results achieved with different spatial resolution data. In this regard, the 18 July results for the NE crater (Fig. 7) shows a similar distribution of T c vs. f h to that obtained on 16 July. Once more, the relationship is logarithmic, with a best fit of:

$$ T_{\rm c}=59\,\ln (f_{\rm h}) + 667 \quad {\rm and} \quad R^2=0.76.$$
(3)
Fig. 7
figure 7

T c vs. f h plot for the North-East crater hot-spot on the 18 July 1996 DAIS image and logarithmic interpolation

Full size image

The T c vs. f h distribution obtained for the 18 July hot spot within La Voragine, however, is entirely different. Below T c of 350 °C, the distribution of points falls along a nearly vertical line parallel to the T c axis (Fig. 8). This means that over this temperature range, f h remains constant with increasing T c. For T c greater than 350 °C, however, increasing T c are associated with decreasing f h. This trend is entirely different from the logarithmic distribution encountered for the NE crater (Fig. 7). This different distribution obtained for La Voragine is consistent with ground-based observations (GVN 1996a) that reveal that a degassing vent, rather than an active lava body, was active in La Voragine during July 1996. We can thus infer that our data relate to active degassing vent(s) inside this crater when the image was taken. The typical range of temperatures given for the degassing feature in Fig. 8 is 50–400 °C. This range is comparable with the T c range obtained for the lava feature given in NE crater (Fig. 3). Table 4 reports maximum, minimum and mean T c and f h obtained for each hot spot. These data show that there is no apparent difference in derived T c between degassing and lava features. Thus, the features cannot be separated on the basis of temperature alone. However, when the features are examined in terms of a scatter plot of T c vs. f h, the two features are apparent from their different distribution. This result indicates that we can use the scatter plot of T c vs. f h as a method to discriminate between different types of volcanic activity in remotely sensed thermal date.

Fig. 8
figure 8

T c vs. f h distribution and observed trend (dashed line) for La Voragine crater hot spot in the 16 July 1996 DAIS image

Full size image

We examined the 18 July hot spot within Bocca Nuova (BN) in terms of these criteria. The results indicate a mixed situation comprising degassing and lava flow activity. The plot of the dual band parameters (Fig. 9) is clearly split into two separate distributions: a linear distribution parallel to the T c-axis (expanded in Fig. 10a) and a second distribution that approaches the logarithmic relationship found for our lava flow (Fig. 10b). In effect, the analysis of the pure cases at the NE Crater and La Voragine allow these features now to be distinguished at the BN with some confidence; where the first trend indicates the presence of a degassing vent and the second trend indicates an active lava flow.

Fig. 9
figure 9

Plot of T c vs. f h for the Bocca Nuova crater hot spot in the 16 July 1996 DAIS image. Note the two apparent trends

Full size image
Fig. 10
figure 10figure 10

Plot of T c vs. f h for the inferred a degassing and b lava components of the Bocca Nuova hot-spot

Full size image

Conclusions and future work

Our analysis of the parameters calculated using the dual band technique indicates that unique relationships in plots of flow crust temperature, T c, vs. the fractional area of hot cracks, f h, allow us to distinguish two different hot volcanic features (lava flows and degassing vents). Finding this relationship was enabled by the use of a high spectral and spatial resolution spectrometer data which has a wide dynamic range of 15 bits, and therefore pixels are rarely saturated. These data attributes allow for a complete analysis of a large number of unsaturated pixels and yield results that are statistically significant. When dual-band solutions are derived for pixels corresponding to an active lava flow or lava pond and assuming a suitable value for the hottest temperature T h, a logarithmic relationship is obtained from the distribution of T c vs. f h. This result is in agreement with the solutions achieved using Landsat TM data for an active lava flow. In contrast, the relationship obtained from pixels containing active degassing vents is entirely different. A sub-vertical trend, parallel to the T c axis, reveals the presence of a degassing activity.

The next step will be to apply this method to a larger selection of data in order to verify this technique and improve its accuracy. The availability of a large number of images of different active sites in Italy (Mt. Etna, Vulcano, ‘Campi Flegrei’) characterized by gas emissions will allow for the further testing and refinement of this approach. Then this technique will be suitable to be applied on those sites where field measurements are difficult such as Erebus or Nyragongo. For extra-terrestrial volcanic activity such as the one observed on Jupiter's satellite IO, the presented methodology could be used to discriminate surface thermal features which maybe correlated to lava flows, lava lakes and high temperature degassing vents or fracture.