Remote Sensing of Environment 113 (2009) 1778–1786

Contents lists available at ScienceDirect

Remote Sensing of Environment j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e

Remote sensing data types and techniques for lahar path detection: A case study at Mt Ruapehu, New Zealand Karen E. Joyce a,⁎, Sergey Samsonov a, Vern Manville a, Richard Jongens a, Alison Graettinger a, Shane J. Cronin b a b

GNS Science, PO Box 30368, Lower Hutt, New Zealand Soil and Earth Sciences Group, Institute of Natural Resources, Massey University, Private Bag 11 222, Palmerston North, New Zealand

a r t i c l e

i n f o

Article history: Received 4 February 2009 Received in revised form 6 April 2009 Accepted 11 April 2009 Keywords: ASTER SPOT5 ALOS-PALSAR LiDAR Crater Lake Mt Ruapehu Lahar

a b s t r a c t Mt Ruapehu is New Zealand's most active onshore volcano. In 2007, the volcano produced a large lahar following a break-out from the summit Crater Lake. Here, satellite and airborne remote sensing and image processing is used to extract the path of the lahar using ASTER and SPOT5 visible and near infra-red imagery, ALOS-PALSAR L-band synthetic aperture RADAR data, and airborne LiDAR. The results obtained from each of these datasets were compared to the lahar deposit manually digitized from aerial photography. SPOT5 imagery produced the most accurate map of the lahar deposit (77% correct), even though these data were acquired a year after the event. This is attributed to the spatial resolution of the data. The ALOS-PALSAR coherence mapping calculated from images acquired 2 months before and nine months after the lahar was not as accurate as that obtained using the optical imagery (43% correct), but this was still considered an important tool for acquiring data during cloudy periods. LiDAR topographic data, collected to constrain geomorphic changes caused by the lahar, was the least accurate in terms of mapping the lahar path (28% correct). No single technique was deemed to be the most accurate under all circumstances, and a combination of data types would produce the best results. By combining the satellite and LiDAR data, it was possible to accurately classify 92% of the lahar path. © 2009 Elsevier Inc. All rights reserved.

1. Introduction This paper investigates the ability of various remote sensing data types to detect a breakout lahar deposit from the summit Crater Lake of Mt Ruapehu, New Zealand on the 18th March 2007. Similar work performed around the world suggests that a remote sensing approach is a successful and cost-effective solution for mapping the source and extent of volcanic debris (Crowley et al., 2003; Harris et al., 2006; Kerle & van Wyk de Vries, 2001; Terunuma et al., 2005). However, commonly used techniques often have to be tuned to site-specific requirements, and little work has been done to compare some of the more common types of data available. For example, we found that optical data, successfully used for mapping of volcanoes since the 1980s, have limited applicability in New Zealand during the Austral winter because of dense cloud coverage. Alternatively, Synthetic Aperture Radar (SAR) data can be successfully used in cloudy environments but does not work well in densely vegetated regions or in areas covered by snow (Hanssen, 2001). Four data sets were tested in this study: (i) SAR imagery from the PALSAR sensor of the ALOS satellite; optical from (ii) SPOT5 and (iii) TERRA-ASTER

⁎ Corresponding author. Tel.: +64 4 570 4772; fax: +64 4 570 4600. E-mail address: [email protected] (K.E. Joyce). 0034-4257/$ – see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2009.04.001

satellites; and (iv) airborne LiDAR topographic data. If used together, SAR and optical data provides complete coverage and has no limitations from weather or daylight conditions. Although not tested here due to the lack of data, ASTER also acquires night-time thermal imagery that may be an additional option if the flow exhibits a higher temperature than its surroundings as has previously been demonstrated with pyroclastic flows (Carter et al., 2008). Synthetic Aperture Radar data from C-band ERS-1/2, RADARSAT-1, ENVISAT and L-band JERS-1 satellites have been successfully used for mapping of various ground changes caused by forest fires (Ranson et al., 2003), flooding (Geudtner et al., 1996; Oberstadler et al., 1997), and landslides (Rott & Nagler, 2006; Singhroy et al., 1998). Terunuma et al. (2005) used data from JERS-1 and ERS-1 satellites for mapping of lahar and pyroclastic flows at Mt. Unzen through the use of three products derived from SAR data: backscatter intensity, coherence, and differential interferometry. Some well known limitations of SAR data are temporal decorrelation and low spatial resolution (Hanssen, 2001). The resolution of the ALOS PALSAR sensor used in this study is 15 m, while the average resolution of most SAR sensors ranges between 10 and 30 m, and the highest resolution (3 m) is currently achieved by the RADARSAT-2 satellite in ultra-fine beam mode (Joyce et al., in press).The PALSAR data used here should provide results superior to those previously documented using lower spatial resolution SAR sensors.

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

The ability of differential SAR interferometry to measure deformations at sub-centimeter scales has been widely utilized in geophysics for mapping ground deformations of various types including, but not limited to, earthquakes (Jacobs et al., 2002; Wright et al., 2001), volcanic activity (Lundgren et al., 2003), and anthropogenic deformations due to mining and fluid extraction (Schmidt & Burgmann, 2003). However, differential interferometry is able to detect lahar-related changes only if significant deposition or removal of material has occurred. It is expected therefore that differential interferometry, coupled with relatively high resolution, should produce useful results with respect to lahar flow path detection. Progressing from historic use of the Landsat series (Francis, 1989), a variety of optical and infrared sensors are now available to provide data suitable for debris flow deposit mapping. Landsat and other sensors with similar spatial resolution such as SPOT and ASTER, are still commonly used as they provide a good compromise between spatial coverage and detail (Davila et al., 2007; Joyce et al., 2008; Kerle et al., 2003; Torres et al., 2004; Tralli et al., 2005). Hyperspectral imagery is less frequently available or utilized; although it has potential for compositional analysis of deposits (Crowley et al., 2003). Limited use of the very high resolution satellite sensors available, such as Quickbird or IKONOS, has been documented in the scientific literature (Huggel et al., 2006). This may be an indication of the high cost of these data, though it also lends credibility to the huge potential and utility of sensors with a lower spatial resolution between 10 and 30 m that provide coverage over larger regions. Lower spatial resolution satellites (e.g. AVHRR, MODIS) have been found to be inadequate for debris mapping (Kerle et al., 2003), but have proven useful for multiscale studies of volcanoes and are particularly beneficial for their high temporal resolution (Patrick et al., 2003). The use of additional geospatial information, such as pre-event images, is considered vital for accurate identification of volcanic deposits. Without the additional contextual information and knowledge of pre-conditions, damage assessment is not possible, and volcanic debris could be mistaken for other features in the image. It is possible to use the tonal and textural features of an optical image to detect deposits, or alternatively to create a DEM of the region from some satellite imagery (e.g. ASTER, SPOT), which can be useful in volumetric analysis of debris deposits where other methods such as airborne LiDAR is too expensive or ground surveys are impractical (Hubbard et al., 2007; Huggel et al., 2008; Lipovsky et al., 2008). The great advantage of satellite remote-sensing is that it gives a ‘before’ view: in general, most LiDAR data is acquired post-event so there is no equivalent baseline dataset.

1779

Few methods of automatic detection of volcanic debris flow deposits using optical and/or infrared remote sensing have been reported in the literature, and it appears that the technique of favour is manual digitising. The Normalized Difference Vegetation Index (NDVI) is commonly used for its ability to enhance the difference between volcanic deposits and surrounding vegetated areas (Castro & Carranza, 2005; Harris et al., 2006; Kerle & Oppenheimer, 2002). The NDVI can also be combined with a threshold value to delineate the deposit. Other techniques that have been used to aid visual interpretation of changes due to volcanic activity include a multiband display incorporating different input dates (Calomarde, 1998; Castro & Carranza, 2005), principal components analysis (Davila et al., 2007), and image subtraction (Torres et al., 2004). The methods used in this study were executed with as little manual interpretation as possible, and only using contextual editing in the final stages of the processing. Therefore, with a calibrated time series of data it would be possible to reproduce the same method to provide similar and thus objective results. With this intention, four different datasets were evaluated (ALOS-PALSAR, ASTER, SPOT5, and LiDAR) for mapping a lahar deposit at Mt Ruapehu, New Zealand. The datasets and processing techniques can be used independently or combined for deposit mapping in other volcanic regions around the world. 2. Study site Mt Ruapehu is the largest and most active andesitic stratovolcano in the central North Island of New Zealand (Fig. 1). It lies at the southern end of the Taupo Volcanic Zone, an intra-arc rift developed in association with subduction of the Pacific Plate beneath the IndoAustralian Plate (Wilson et al., 1995). The 110 km3 composite cone rises to 2797 m above sea level and supports a number of small glaciers, permanent snow fields, and a summit Crater Lake. It is surrounded by a ring-plain constructed of distal pyroclastic fall, and lahar and fluvial deposits derived from the volcano (Donoghue, 1991; Hackett & Houghton, 1989). Historic activity at Ruapehu has consisted of very frequent, relatively small-to-medium phreatic and phreatomagmatic eruptions (e.g. 1895, 1969, 1975, 1988, 2007) (Gregg, 1960; Healy et al., 1978; Nairn et al., 1979) and more prolonged magmatic eruptions such as in 1945 (Beck, 1950; Oliver, 1945; Reed 1945) and 1995–96 (Bryan et al., 1996; Nakagawa et al., 1999). At least five vents have been active during the Holocene, but historic activity has been confined to the southern crater which is normally occupied by a hot, acidic Crater Lake with a (pre-1995) volume of c. 9 million m3 lying at an elevation of 2530 m (Christenson & Wood, 1993).

Fig. 1. Study site location (A) Mt Ruapehu location in New Zealand; (B) ASTER imagery before the lahar (9 February 2002); and (C) after the lahar flow (25 March 2007).

1780

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

Mt Ruapehu is one of the most lahar-prone volcanoes in the world, with more than 50 individual lahars recorded in the Whangaehu valley, the natural outlet to the summit Crater Lake, since historic observations began in 1861 AD (Graettinger, 2008). Lahars have been triggered in the Whangaehu and in a number of other catchments by a variety of mechanisms including: explosive eruptions that have displaced Crater Lake water over the outlet, or ejected it onto the snow-clad summit area of the volcano (Nairn et al., 1979); rainremobilization of thick tephra deposits on steep slopes (Hodgson & Manville, 1999); volumetric displacement over the outlet as a result of syn-eruptive changes in lake bathymetry (Cronin et al., 1997); and lake break-outs from Crater Lake following impoundment of excess water behind temporary barriers of tephra and/or ice emplaced over the outlet (Manville, 2004; O'Shea, 1954). Volcanic activity during 1995–96 emptied Crater Lake and deposited a c. 8 m thick blanket of tephra (ash, lapilli and blocks) over the hard lava ledges that comprised its former overflow channel, creating the potential for an overfilled lake to develop behind an unstable natural dam. Eleven years later, at 11:22 am on 18 March 2007, the refilling Crater Lake breached this barrier during a rainstorm event, releasing c. 1.3 million m3 of warm acidic water in less than 90 min as the 400 m diameter lake fell by 6 m. This outflow eroded and entrained snow, glacial ice, landslide debris, talus material, and older lahar deposits along the upper Whangaehu valley to transform to a non-cohesive debris flow by 7 km downstream. Ten kilometres downstream from the Crater Lake, the lahar debouched onto the Whangaehu fan where it braided into multiple distributary channels before collecting into a single incised meandering channel again and continued downstream, reaching the coast 215 km away at about 3 am the following morning. This was effectively a repeat of the 1953 lahar scenario (Manville, 2004; O'Shea, 1954) but, as it was foreseen, it provided an opportunity to prepare a multidisciplinary research scheme in advance of the break-out, which allowed for extensive scientific study (Manville & Cronin, 2007). In turn, the ability to accurately map the lahar deposit represents not only event documentation, but provides the basis of hazard zonation and validation of pre-existing maps. 3. Data and methods 3.1. Synthetic Aperture Radar Imagery Three Fine Beam Single (FBS) ALOS PALSAR images were used in this work. Two images were acquired, on 1 January and 16 February 2007, before the lahar event, while the next available image was acquired 9 months after the lahar on 4 January 2008. The raw data was processed to “Single Look Complex” format using the GAMMA MSP processor (Werner et al., 2000). All images were then co-registered to the master image acquired on 1 January 2007. A 3 × 6 pixel (range × azimuth) averaging window filter was performed in order to increase the signal-to-noise ratio, creating a final resolution of approximately 15 m. Three interferograms were created and the topographic phase was removed using the 90 m resolution Shuttle Radar Topographic Mission (SRTM) DEM. Higher resolution DEMs available for this region (for example, 10 m and 40 m DEMs from Land Information New Zealand) were not used here for interferometric processing because their accuracy had not yet been validated. In deformation studies it is impossible to separate true deformation signal from residual errors caused by the inaccuracies in DEM. Therefore it is beneficial to use interferometric pairs with the perpendicular component (Bp) of a spatial baseline (distance between satellite tracks) close to zero (Rosen et al., 2000). Unfortunately all three interferograms used in this study had large Bp values, which significantly decreased the accuracy of final differential interferograms. Two computed differential interferograms spanned the lahar (1 January 2007–4 January 2008, Bp = −1480 m

and 16 February 2007–4 January 2008, Bp = −2230 m), while the third (1 January 2007–16 February 2007, Bp = 750 m) did not. As a result of processing, three products were derived: backscatter intensity, differential interferograms and coherence images. Backscatter intensity was used because of its sensitivity to surface roughness and moisture content of the material reflecting the SAR signal. The differences in intensities before and after the lahar were estimated by subtracting one image from another and also by studying the absolute value of differences. Differential interferometry utilizes the phase information of SAR signals and is potentially capable of detecting sub-centimetre changes in line-of-sight distances. Usually it is used to map ground deformations that occur between consequent acquisitions, but changes in topography caused by removal or deposition of material will produce similar, easily detectable signals. Both filtered and unfiltered, and wrapped and unwrapped differential interferograms were studied. Coherence, which is the magnitude of cross-correlation between two co-registered SAR intensity images, quantifies the degree of similarity of two images (Hanssen, 2001). When two images are identical, coherence should be equal or close to one, while two completely different images should produce coherence close or equal to zero. It was anticipated that ground conditions affected by the lahar deposit would be easily detectable in coherence images as regions with lower than usual coherence. In order to eliminate signals produced by effects other than the lahar deposit, we subtracted from the coherence image calculated from pre-lahar SAR images from SAR images acquired both before and after the flow, e.g. DIF_COH = COH(1Jan2007– 16Feb2007) − COH(1Jan2007–4Jan2008). A threshold was then determined on the coherence pixel values to isolate the effects on coherence caused by the lahar.

3.2. Optical imagery Multi-temporal ASTER images with a 15 m ground resolution element (GRE) were acquired before (9 February 2002) and after (25 March 2007) the 18 March 2007 break-out lahar were used to distinguish the flowpath from its surroundings (Fig. 1). An additional SPOT5 scene (10 m GRE) was later obtained on 25 March 2008 to determine if the lahars deposit trace was still evident one year later. These sensors were primarily chosen for their availability and cost. Higher spatial resolution sensors (e.g. Quickbird, IKONOS) could also be used for additional cost. Lower spatial resolution satellites such as AVHRR have been found to be inadequate (Kerle et al., 2003). Unfortunately ASTER did not acquire short wave infrared (SWIR) or thermal data soon after the lahar, so it was not possible to test mapping methods using temperature anomaly techniques. Though outside the scope of this study, ASTER has the added advantage of providing stereo data that can be used to develop a DEM of the region, which can be useful in volumetric analysis of debris deposits (Hubbard et al., 2007; Huggel et al., 2008). All images were calibrated to at-sensor radiance using the coefficients provided with the data (USGS for ASTER; CNES for SPOT5). In the absence of access to atmospheric correction software and appropriate parameters, correction to at-surface reflectance was not performed. As an inland scene, it was also difficult to locate a suitably dark target (e.g. open ocean) that could be used for a dark pixel subtraction. De-striping was conducted on the ASTER imagery using a modified Lee filter implemented in ENVI software (ITTVIS). A DEM was used in conjunction with the provided rational polynomial coefficients to orthorectify the data to New Zealand Map Grid with a nearest neighbour resampling. An empirical line correction was used to radiometrically calibrate the 2002 ASTER image to the more recent 2007 image. This was originally completed in order to conduct image differencing between the two dates, a technique that was not eventually used in the final output.

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

A Normalized Difference Vegetation Index (NDVI) was calculated on the ‘after’ ASTER image, and a calculated threshold was used to restrict the area of analysis to unvegetated surfaces. An unsupervised classification of the resultant data was then performed using green, red and NDVI layers as input. Spectral angle mapper and principal components analysis techniques were also tested but proved less accurate than the unsupervised approach. Classes were iteratively sieved and clumped, before achieving the final classification. The same technique was used for the SPOT5 imagery (also with green, red and NDVI layers), though it was also noted that a simple threshold value using the green band was similarly effective. Although SWIR was available for the SPOT5 data, this was not used as it did not appear to show any contrast between the lahar deposit and its surrounds. This could be a combination of both the spatial and spectral characteristics of this band. 3.3. LiDAR Airborne LiDAR surveys have been increasingly used in the study of mass-flow events such as debris flows and rock/ice avalanches to extract geometric changes and estimate flow volumes (Baldo et al., 2009; Chen et al., 2006; Scheidl et al., 2008). As part of the scientific research plan designed to capture maximum data from the 18 March 2007 break-out lahar (Manville & Cronin, 2007), airborne LiDAR surveys (French, 2003) of the first 58 km of flow path were carried out approximately 12 months before the lahar and 3 weeks after in order to characterize and quantify geomorphic changes resulting from the flow. The acquired swaths represented a 500 m wide buffer centred on the active channel axis, widening to 4 km across the Whangaehu Fan. The surveys each covered a total area of 54 km2 with a mean data

1781

density of 1–2 points/m2 on flat ground with vertical and horizontal accuracies of 0.15 m and 0.5 m respectively at 1 standard deviation. Independent check sites were surveyed in open ground outside the lahar flow path using differential GPS to verify the accuracy of the processed datasets. Height difference statistics between a TIN (triangulated irregular network) of the LiDAR ground points and the checkpoints show differences of +/−0.11 m at 1 standard deviation. Positional accuracy of the processed data was checked visually by overlaying the check site data over the LiDAR dataset. High resolution (c. 0.2 m GSD) orthoimagery of the flow path was acquired simultaneously with a horizontal accuracy of 0.5 m at 1 standard deviation. The ground return LiDAR data was reprocessed into a common format, projection, coordinate system and geoid. The natural neighbours interpolation method (ArcGIS) was used to produce a 0.5 m resolution DEM of the lahar channel, and the pre-event DEM was subtracted from the post-event DEM to produce a change raster for which cut and fill volume statistics were calculated. LiDAR backscatter intensity data was not acquired as part of the survey design. The LiDAR-derived topographic change map was reclassified to show areas of vertical change (positive and negative) greater than a threshold value. In this case, a value of ≥ ±0.5 m produced the best match when compared with the aerial photo derived flowpath. 3.4. Accuracy assessment A majority 5 × 5 filter was passed over all classifications to reduce extraneous classified pixels. All data were exported to a GIS (ArcMap 9.2) and converted to vectors. In the GIS environment, manual contextual editing was conducted, primarily to remove residual

Fig. 2. Comparison of (A) ASTER, (B) SPOT5, (C) ALOS-PALSAR coherence, and (D) LiDAR change map classifications with aerial photography results. ‘Accurate’ (green) refers to the area that matched between classifications, ‘commission’ (blue) is the area included in the classification but not in the aerial photography classification; and ‘omission’ (red) is the area excluded from the classification but included in the aerial photography. The ASTER green band has been used as a background.

1782

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

for verifying the different classification techniques. Some errors of omission and commission can be attributed to misregistration between the data types, but this is unavoidable due to differences in scale and resolution between each the datasets. It is difficult to compare the accuracy achieved here with other lahar studies because accuracy values are rarely reported. Presumably this is due to the lack of, and expense associated with acquiring adequate reference data, as we present here in the form of aerial photo analysis. The results of individual datasets and techniques are discussed in detail in the following sections; however the following summary comments can be made:

Fig. 3. Accuracy assessment as calculated based on ASTER, SPOT5, ALOS-PALSAR coherence, and LiDAR change maps.

spurious classified patches outside the expected path of the lahar. The 2002 ASTER image proved useful in assisting with this contextual editing. Manual editing did not involve reintroducing pixels that had been erroneously omitted in the classification process. The final vectorized layer for each classification was compared to the lahar path that was manually digitized from 18 to 21 cm spatial resolution digital aerial photography data captured three weeks after the flow. In areas of shadow, principally in the upper gorge, the flowpath was constrained by geomorphic changes mapped by comparing the pre- and post-lahar LiDAR data. Total area and spatial coverage was compared between all datasets. Note however, that only the area commonly imaged by all datasets was used for comparison despite the satellite data covering a larger area to the south and southwest of the lahar flow path. Where the lahar deposit derived from an individual technique corresponded to that obtained from the aerial photography, it was considered accurate. Areas that had been erroneously classified as lahar deposit are considered commission errors, whereas areas that were incorrectly excluded from the classification are considered omission errors. The total area and percent of area of accurate, committed and omitted pixels was reported for each technique and dataset. 4. Results and discussion A comparison between the mapping results achieved with ASTER, SPOT5, ALOS-PALSAR, and airborne LiDAR data is shown spatially in Fig. 2 and statistically in Fig. 3. The manually digitized aerial photography data were used as the basis of accuracy assessments

1. The narrow, upper reaches of the lahar deposit were most accurately mapped by the data acquired closest in time to the actual event (LiDAR and ASTER). 2. The ALOS-PALSAR coherence imagery had the highest commission error, though should not be discounted due to its ability to acquire data during periods of cloud cover. Also, it is presently not clear if an image acquired immediately after the lahar would produce a better result. 3. The LiDAR data was the least overall accurate in terms of flowpath mapping. It omitted large areas of the lahar path down the Whangaehu fan and committed large areas around Crater Lake that may be due to differences in snow coverage. 4. Although acquired a year after the event, the SPOT5 image data produced the most accurate results. This was considered to be primarily due to the higher spatial resolution compared with the ASTER imagery (10 m ground resolution element for SPOT5 compared to 15 m for ASTER). No single technique was the most accurate under all circumstances. When combining all classifications, 92% of the lahar deposit was accurately mapped with at least one dataset, though only 13% of the path was mapped accurately using all four datasets (Fig. 4). This indicates that in the absence of aerial photography, a combination of these data types tested herein would produce the most accurate result, though it would undoubtedly also be the most costly solution. 4.1. Synthetic Aperture Radar Imagery Coherence images produced the best results of all the SAR derived products. The lahar deposit was clearly visible as a region with lower than usual coherence on the E–SE slope of Mt Ruapehu (Fig. 5A). In order to exclude low-coherence regions not associated with the March 2007 lahar event we subtracted the coherence image calculated from two SAR images acquired before the event from the coherence image spanning the event (Fig. 5B). This differential image produced better

Fig. 4. Combined accuracy and number of datasets able to accurately map the lahar deposit when compared to manually digitized aerial photography. The color scale used on the map is the same as in the graphical legend.

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

1783

Fig. 5. (A) Coherence image calculated for 1 January 2007/4 January 2008 pair. The lahar flow path is observed as low coherence (in black) on E–SE slope of Mt Ruapehu but can not be easily distinguished because of a large amount of noise (low-coherence regions) and low contrast with the surrounds. (B) Difference between coherence images calculated for 1 January 2007/16 February 2007 pair and 1 January 2007/4 January 2008 pair produces better results with the lahar deposit clearly visible in white.

mapping results with the useful lahar deposit signal shown in white. The threshold applied to the ALOS-PALSAR coherence image was 43% accurate, with 46% committed (Fig. 3). Analysis of SAR backscatter intensities (Fig. 6A) was not particularly successful. It was possible to distinguish some parts of the lahar deposit, but mostly because of topographic features associated with it. These topographic features were observed on all intensity images acquired before and after lahar. The differential image calculated between pre- and post-lahar intensities as well as an absolute value of the differences did not reveal any noticeable lahar deposit signal. This result is somewhat expected since the post-event image was acquired nine months after the lahar, giving ample time for the flowpath and lahar deposits to dewater and dry out. It is, however, unexpected that change in roughness caused by the lahar did not produce any noticeable signal. Changes in the surface elevation of lahar deposits as a result of post-depositional dewatering and compaction are

minimal in coarse-grained framework-supported deposits such as those of the 18 March 2007 lahar. Analysis of a differential interferogram (Fig. 6B) spanning before and after the event was partially successful. Both filtered and unfiltered interferograms were unwrapped and some topographic changes observed. However, it was not possible to distinguish useful signals from the surrounding noise caused by inaccuracies in the original DEM used to remove the topographic phase from the interferogram and tropospheric noise. It is anticipated that highresolution and high-accuracy DEMs would produce better results. 4.2. Optical With the exception of a few minor areas of cloud cover, the classification of optical data proved to be fast, accurate, and the most cost-effective method for mapping the lahar deposit. The path of the

Fig. 6. (A) ALOS PALSAR intensity image 4 January 2007. (B) Differential wrapped interferogram calculated for 1 January 2007/4 January 2008 pair. The wrapped phase changes from −π to π and a complete color cycle corresponds to approximately 12 cm of line-of-sight deformations. Lahar deposit related features cannot be clearly distinguished.

1784

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

lahar was clearly apparent as a highly reflective mudflow when comparing the before and after ASTER imagery (Fig. 1), especially as the after image was acquired only 7 days after the lahar and before rain had washed away the blue-grey muddy highwater and inundation marks. When observing the imagery it was also easy to see several other spectrally similar features (e.g. fallow fields, turbid water) that could be confused for the lahar deposit if contextual editing was not applied. Of the area classified as lahar deposit using ASTER, 74% of this corresponded to the manually derived aerial photography based lahar path. The omitted areas are most apparent in the regions of narrow lahar flow near the source and downstream. A commission error of 23% is partially due to mixed pixels (e.g. in the eastern portion of the image) resulting in an overestimation of the classified area. The benefits of ASTER for lahar deposit mapping have previously been noted (Davila et al., 2007), however a quantitative assessment of classification accuracies has been unavailable. A change detection routine between the two ASTER images was also attempted; however, it produced many more false alarms than the classification. In the absence of atmospheric measurements at the time of image capture and the availability of robust atmospheric correction software, it was difficult to match the radiometry of these two scenes that were acquired five years apart. Automated change detection would be more effective had the scenes been closer in acquisition dates. It was still possible to see the majority of the path of the lahar deposit a year after the event in the SPOT5 data. In some places, the SPOT5 imagery even produced a more accurate classification of the lahar trace than the ASTER image acquired much sooner after the event. This is due to the increased spatial resolution of the SPOT5 scene, as the areas of increased accuracy were those areas where the trail was too narrow to be mapped with the ASTER scene. The SPOT5 classification was less accurate than the ASTER near the top of the lahar path in the western portion of the scene. Here, the ASTER classification was able to map the narrow trace of the lahar deposit high on the mountain; whereas the contrast with surrounding rock was not apparent one year later when the SPOT5 image was acquired (Fig. 2). Despite the spatial differences between the ASTER and SPOT5 image classifications, it was noted that the increase in resolution with the SPOT5 data resulted in an overall increase in accuracy. The SPOT5 classification had the lowest commission and omission errors (Fig. 3). The SPOT5 image classification was 77% accurate, with 17% area committed to the lahar deposit that was not digitized with the aerial photography. SPOT5 has an increased spatial resolution of 10 × 10 m compared with its predecessor, SPOT4 (20 × 20 m), that was also considered adequate for lahar deposit delineation (Kerle et al., 2003). In their work, Kerle et al. (2003) stated that SPOT4 was inadequate for determining the nature of the flow (debris avalanche or lahar) though the advanced spatial resolution of SPOT5, in particular with the panchromatic band (5 × 5 m), may hold potential for this application. Also of note is the considerably greater area that was able to be captured within the 60 km wide swaths of the ASTER and SPOT5 scenes. In this case, the path of the lahar down the Whangaehu River was captured, though was not surveyed with aerial photography. At $80 USD per scene of ASTER data compared with the required $4000 USD (excluding mobilization costs) for the aerial photography, the ASTER imagery presents a significant cost saving. The SPOT5 image data used was provided free of charge for this research project through the New Zealand Ministry for Environment, however it would normally be a more costly alternative to ASTER (approximately $3200 USD at the time of writing). The slightly higher spatial resolution and corresponding increased accuracy may be considered a worthwhile trade off. ALOS AVNIR-2 imagery is another option, although the tasking schedule for acquisition is somewhat more restrictive than for either SPOT or ASTER.

4.3. LiDAR The LiDAR topographic dataset was not acquired for the primary purpose of delineating the flowpath, but rather for mapping and quantifying patterns of erosion and aggradation associated with the lahar. In this respect, it obtained unique and highly accurate results that would have been impossible to achieve using any other method. Only 28% of the area classified using a ± 0.5 m topographic change threshold was considered correct when compared with the flowpath digitized from the simultaneously acquired orthoimages with a commission error of 36%, which was lower only than the ALOSPALSAR classification. However, LiDAR analysis was the only data type and technique (with the exception of the orthophotos) able to accurately map the narrow path of the lahar as it emerged from Crater Lake, presumably due to its very high (sub-metre) spatial resolution compared with the satellite data tested here (Fig. 2). Unfortunately, the flow path classification was also surrounded by large commission errors, so it would be difficult to solely use the LiDAR data set for discrimination in this region. Backscatter intensity data, if collected as part of the LiDAR mission may have been a more accurate method of determining the flowpath because of the likely high reflectivity of the fresh muddy deposits, which showed up clearly on optical imagery. Intensity data were not purchased at the time in an attempt to minimise the cost of the mission. In retrospect we consider this a potential oversight given the possible utility of those data for providing insight into surface grain size distribution and water content, amongst other things. The commission error for LiDAR mapping was 36%. Considerable areas of commission error are apparently due to seasonal topographic changes on the upper mountain related to melting or accumulation of snow and ice, and vegetation growth at lower elevations. A 200 m × 300 m lahar triggered landslide just downstream of the Crater Lake was also included in the commission error. None of the other datasets appeared to have detected this landslide. Some commission errors, particularly along steep slopes or cliff edges, are a derivative of the way in which the LiDAR data is generated, with each survey essentially sampling a different set of ground points. Large areas of omission occurred on the ring-plain, where the lahar braided into multiple shallow and broad channels. These areas, although inundated, did not experience significant topographic change under the consequent low energy flow conditions. Lower threshold values, while producing more complete delineation of the ring-plain flowpaths, introduced excessive commission errors because the vertical topographic changes detected approached the resolution of the airborne LiDAR dataset. Conversely, higher threshold values omitted more of the ring-plain flow path where topographic changes resulting from the lahar were more subdued. Near the downstream limit of the LiDAR data, substantial areas of commission errors arose from changes in vegetation (forestry and pasture) over the year between the surveys; non-vegetated areas such as forestry roads showed errors within the vertical accuracy limit of the dataset. 5. Conclusions The preceding work confirms that a variety of remote sensing datasets and processing methods can be successfully used for mapping lahar and debris flow deposits in the New Zealand environment. Manual interpretation of aerial photography was considered the most accurate mapping method, so each other data type and processing technique was compared to this reference. Each of the data sets tested performed better in detecting some regions of the lahar deposit than others. Optical data from SPOT5 satellite is preferable for mapping because of its higher spatial resolution than corresponding data from the Terra-ASTER satellite. The acquisition of optical satellite data represents a significant cost saving when compared to airborne data and the accuracy level achievable is

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

between 70% and 80%. The accuracy of mapping using SPOT5 data is expected to exceed this level if acquired closer in time to the event, though it was still useful a year later. Airborne LiDAR topographic data was less effective for delineating the lahar flow path as it could only detect geomorphic changes caused by the flow where these exceeded 2× the vertical resolution of the method (i.e. N30 cm). Where the lahar was shallow and or slow, minimal erosion or deposition occurred and the only indicator of inundation areas was an ephemeral muddy film. This color contrast between areas within and outside the active channel proved to be useful for classifying the lahar path on aerial photographs captured shortly after the event. Despite the low comparative accuracy of Synthetic Aperture Radar for delineating the lahar path, it should not be excluded as a potential option for mapping lahar deposits, given that can be acquired under any weather or light conditions. However, the user needs to be aware of the limited achievable mapping accuracy level. Coherence images calculated from two SAR images acquired before and after the lahar successfully mapped ground changes as regions with lower than average coherence, and were enhanced by subtracting a coherence image calculated from SAR data acquired before the event. However, this technique can be applied only in regions with a high degree of initial coherence. Backscatter intensity information was not particularly successful for mapping of ground changes in this work. However, we expect that if a post-event image was acquired soon after the event it would have produced better results due to the sensitivity of SAR signals to soil moisture contents. Some changes associated with the 18 March 2007 lahar event were observed on differential interferograms but it would not have been possible to differentiate these changes from noise if other sources of data (air- and space-borne optical) were not available. In order to utilize SAR interferometry it is necessary to have a high resolution DEM for removing topography noise, or to use SAR pairs with spatial baselines close to zero. Based on these results it is recommended that SAR data be acquired regularly with small spatial baselines. It is clear that a combination of data types and processing techniques will provide the optimal solution for lahar deposit mapping. Where finances permit and environmental conditions are appropriate, manual interpretation of aerial photography will undoubtedly provide the most accurate solution. However, satellite data provide a more cost-effective option, and the increasing availability of very high spatial resolution data will soon come close to providing similar results to that achieved with aerial photography.

Acknowledgments The research contained herein was supported by the New Zealand Foundation for Research Science and Technology under contracts C05X0403 and C05X0006. This manuscript includes material © CNES 2007, Distribution SPOT Image S.A., France, all rights reserved, that was provided through the New Zealand Ministry for the Environment, Land Use and Carbon Accounting System project. The ALOS PALSAR data has been used in this work with the permission of JAXA and METI and the Commonwealth of Australia (Geoscience Australia) (“the Commonwealth”). JAXA, METI and the Commonwealth have not evaluated the data as altered and incorporated within this work, and therefore give no warranty regarding its accuracy, completeness, currency or suitability for any particular purpose. ASTER imagery was provided courtesy NASA/JPL-Caltech, 2002/07. LiDAR data and associated aerial photography was acquired by New Zealand Aerial Mapping and Fugro Spatial Solutions Party Ltd. under contract from GNS Science: co-funding was from Massey University, New Zealand. Fiona Links and Kathryn Lyttle assisted with LiDAR and aerial photography processing and interpretation. We thank the anonymous reviewers for constructive comments and advice on our initial manuscript.

1785

References Baldo, M., Bicocchi, C., Chiocchini, U., Giordan, D., & Lollino, G. (2009). LiDAR monitoring of mass wasting processes: The Radicofani landslide, Province of Siena, Central Italy. Geomorphology, 105, 193−201. Beck, A. C. (1950). Volcanic activity at Mt. Ruapehu from August to December, 1945. New Zealand Journal of Science and Technology, B31, 1−13. Bryan, C., Christenson, B. W., Hodgson, K. A., Houghton, B. F., Johnston, D. M., JuradoChichay, Z., et al. (1996). Volcanic eruption at a New Zealand ski resort prompts reevaluation of hazards. EOS, Trans. AGU, 77, 189−191. Calomarde, R. (1998). An application of remote sensing and geographic information system (GIS): A case study in Sto. Tomas-Marela River, Mt. Pinatubo, Zambales, Philippines. Asian conference on remote sensing. Carter, A. J., Girina, O., Ramsey, M. S., & Demyanchuk, Y. V. (2008). ASTER and field observations of the 24 December 2006 eruption of Bezymianny Volcano, Russia. Remote Sensing of Environment, 112, 2569−2577. Castro, O. T., & Carranza, E. J. M. (2005). Remote sensing of temporal variations in spatial distributions of lahar and pyroclastic-flow deposits, West Mount Pinatubo, Philippines. In S. Erasmi, B. Cyffka, & M. Kappas (Eds.), Remote sensing and GIS for environmental studies Gottinger Geographische Abhandlungen, Gottingen, Vol. 113 (pp. 223−229). Chen, R. F., Chang, K. J., Angelier, J., Chan, Y. C., Deffontaines, B., Lee, C. T., et al. (2006). Topographical changes revealed by high-resolution airborne LiDAR data: The 1999 Tsaoling landslide induced by the Chi-Chi earthquake. Engineering Geology, 88,160−172. Christenson, B. W., & Wood, C. P. (1993). Evolution of a vent-hosted hydrothermal system beneath Ruapehu Crater Lake, New Zealand. Bulletin of Volcanology, 55, 547−565. Cronin, S. J., Neall, V. E., Lecointre, J. A., & Palmer, A. S. (1997). Changes in Whangaehu River lahar characteristics during the 1995 eruption sequence, Ruapehu volcano, New Zealand. Journal of Volcanology and Geothermal Research, 76, 47−61. Crowley, J. K., Hubbard, B. E., & Mars, J. C. (2003). Analysis of potential debris flow source areas on Mount Shasta, California, by using airborne and satellite remote sensing data. Remote Sensing of Environment, 87, 345−358. Davila, N., Capra, L., Gavilanes-Ruiz, J. C., Varley, N., Norini, G., & Gomez Vazquez, A. (2007). Recent lahars at Volcán de Colima (Mexico): Drainage variation and spectral classification. Journal of Volcanology and Geothermal Research, 165, 127−141. Donoghue, S. L. (1991). Late Quaternary volcanic stratigraphy of the southeastern sector of the Mount Ruapehu ring plain, New Zealand. Palmerston North: Massey University. Francis, P. W. (1989). Remote sensing of volcanoes. Advances in Space Research, 9, 89−92. French, J. R. (2003). Airborne LiDAR in support of geomorphological and hydraulic modelling. Earth Surface Processes and Landforms, 28, 321−335. Geudtner, D., Winter, R., & Vachon, P. (1996). Flood monitoring using ERS-1 SAR interferometry coherence maps. International geoscience and remote sensing symposium, Vol. 2 (pp. 966−968). Graettinger, A. H. (2008). Depositional record of historic lahars in the Whangaehu Gorge, Mt. Ruapehu Department of Earth and Ocean Sciences (pp. 176). Hamilton: Waikato University. Gregg, D. (1960). The geology of the Tongariro Subdivision (pp. 152). Hackett, W. R., & Houghton, B. F. (1989). A facies model for a Quaternary andesitic composite volcano: Ruapehu, New Zealand. Bulletin of Volcanology, 51, 51−68. Hanssen, R. F. (2001). Radar interferometry: Data interpretation and error analysis. Kluwer Academic Publishers. Harris, A. J. L., Vallance, J. W., Kimberly, P., Rose, W. I., Matias, O., Bunzendahl, E., et al. (2006). Downstream aggradation owing to lava dome extrusion and rainfall runoff at Volcan Santiaguito, Guatemala. In W. I. Rose, G. J. S. Bluth, M. J. Carr, J. W. Ewert, L. C. Patino, & J.W. Vallance (Eds.), Volcanic hazards in Central AmericaGeological Society of America Special Paper, Vol. 412 (pp. 85−104). Healy, J., Lloyd, E., Rishworth, D., Wood, C., Glover, R., & Dibble, R. (1978). The eruption of Ruapehu, New Zealand on 22 June 1969. New Zealand Department of Scientific and Industrial Research Bulletin, 224. Hodgson, K. A., & Manville, V. (1999). Sedimentology and flow behaviour of a raintriggered lahar, Mangatoetoenui Stream, Ruapehu volcano, New Zealand. Geological Society of America Bulletin, 111, 743−754. Hubbard, B. E., Sheridan, M. F., Carrasco-Núñez, G., Díaz-Castellón, R., & Rodríguez, S. R. (2007). Comparative lahar hazard mapping at Volcan Citlaltépetl, Mexico using SRTM, ASTER and DTED-1 digital topographic data. Journal of Volcanology and Geothermal Research, 160, 99−124. Huggel, C., Kaab, A., & Salzmann, N. (2006). Evaluation of Quickbird and Ikonos imagery for assessment of high-mountain hazards. EARSeL eProceedings, 5, 51−62. Huggel, C., Schneider, D., Miranda, P. J., Delgado Granados, H., & Kaab, A. (2008). Evaluation of ASTER and SRTM DEM data for lahar modeling: A case study on lahars from Popocatépetl Volcano, Mexico. Journal of Volcanology and Geothermal Research, 170, 99−110. Jacobs, A., Sandwell, D., Fialko, Y., & Sichoix, L. (2002). The 1997 (M 7.1) Hector Mine, California, earthquake: Near-field postseismic deformation from ERS interferometry. BSSA Special Issue on Hector Mine Earthquake, 92, 1433−1442. Joyce, K. E., Belliss, S., Samsonov, S., McNeill, S., & Glassey, P. J. (in press). A review of the status of satellite remote sensing image processing techniques for mapping natural hazards and disasters. Progress in Physical Geography. Joyce, K. E., Samsonov, S., Jongens, R., Lee, J. M., & Glassey, P. J. (2008). Using remote sensing for mapping the effects of natural hazards in New Zealand. IEEE International Geoscience and Remote Sensing Symposium paper II-1251. Kerle, N., Froger, J. L., Oppenheimer, C., & De Vries, B. V. (2003). Remote sensing of the 1998 mudflow at Casita volcano, Nicaragua. International Journal of Remote Sensing, 24, 4791−4816.

1786

K.E. Joyce et al. / Remote Sensing of Environment 113 (2009) 1778–1786

Kerle, N., & Oppenheimer, C. (2002). Satellite remote sensing as a tool in lahar disaster management. Disasters, 2, 140−160. Kerle, N., & van Wyk de Vries, B. (2001). The 1998 debris avalanche at Casita volcano, Nicaragua — Investigation of structural deformation as the cause of slope instability using remote sensing. Journal of Volcanology and Geothermal Research, 105, 49−63. Lipovsky, P., Evans, S., Clague, J., Hopkinson, C., Couture, R., Bobrowsky, P., et al. (2008). The July 2007 rock and ice avalanches at Mount Steele, St. Elias Mountains, Yukon, Canada. Landslides, 5, 445−455. Lundgren, P., Berardino, P., Coltelli, M., Fornaro, G., Lanari, R., Puglisi, G., et al. (2003). Coupled magma chamber inflation and sector collapse slip observed with synthetic aperture radar interferometry on Mt. Etna volcano. Journal of Geophysical Research, 108(B5), 2247. doi:10.1029/2001JB000657. Manville, V. (2004). Palaeohydraulic analysis of the 1953 Tangiwai lahar: New Zealand's worst volcanic disaster. Acta Vulcanologica, XVI, 137−152. Manville, V., & Cronin, S. J. (2007). Break-out lahar from New Zealand's Crater Lake. EOS Transactions AGU, 88, 441−442. Nairn, I., Wood, C., & Hewson, C. (1979). Phreatic eruptions of Ruapehu: April 1975. New Zealand Journal of Geology and Geophysics, 22, 155−173. Nakagawa, M., Wada, K., Thordarson, T., Wood, C. P., & Gamble, J. A. (1999). Petrological investigations of the 1995 and 1996 eruptions of Ruapehu volcano, New Zealand: Formation of discrete and small magma pockets and their intermittent discharge. Bulletin of Volcanology, 61, 15−31. Oberstadler, R., Honsch, H., & Hutch, D. (1997). Assessment of the mapping capabilities of ERS-1 SAR data for flood mapping: A case study in Germany. Hydrological processes, 11, 1415−1425. Oliver, R. L. (1945). Further activity of Mount Ruapehu, May–July, 1945. New Zealand Journal of Science and Technology, B26, 24−33. O'Shea, B. E. (1954). Ruapehu and the Tangiwai disaster. New Zealand Journal of Science and Technology, B36, 174−189. Patrick, M. R., Dehn, J., Papp, K. R., Lu, Z., Dean, K., Moxey, L., et al. (2003). The 1997 eruption of Okmok Volcano, Alaska: A synthesis of remotely sensed imagery. Journal of Volcanology and Geothermal Research, 127, 87−105. Ranson, K., Kovacs, K., Sun, G., & Kharuk, V. (2003). Disturbance recognition in the boreal forest using radar and Landsat-7. Canadian Journal of Remote Sensing, 29, 271−285.

Reed, J. J. (1945). Activity at Ruapehu, March–April, 1945. New Zealand Journal of Science and Technology, B26, 17−23. Rosen, P., Hensley, P., Joughin, I., Li, F., Madsen, S., Rodriguez, E., et al. (2000). Synthetic aperture radar interferometry. Proceedings of the IEEE, 88, 333−382. Rott, H., & Nagler, T. (2006). The contribution of radar interferometry to the assessment of landslide hazards. Advance in Space Research, 37, 710−719. Scheidl, C., Rickenmann, D., & Chiari, M. (2008). The use of airborne LiDAR data for the analysis of debris flow events in Switzerland. Natural Hazards and Earth System Sciences, 8, 1113−1127. Schmidt, D., & Burgmann, R. (2003). Time-dependent land uplift and subsidence in the Santa Clara valley, California, from a large interferometric synthetic aperture radar data set. Journal of Geophysical Research, 108, 2416. doi:10.1029/2002JB002267. Singhroy, V., Mattar, K. E., & Gray, A. L. (1998). Landslide characterisation in Canada using interferometric SAR and combined SAR and TM images. Advances in Space Research, 21, 465−476. Terunuma, T., Nishida, K., Amada, T., Mizuyama, T., Sato, I., & Urai, M. (2005). Detection of traces of pyroclastic flows and lahars with satellite synthetic aperture radars. International Journal of Remote Sensing, 26, 1927−1942. Torres, R., Mouginis-Mark, P., Self, S., Garbeil, H., Kallianpur, K., & Quiambao, R. (2004). Monitoring the evolution of the Pasig–Potrero alluvial fan, Pinatubo Volcano, using a decade of remote sensing data. Journal of Volcanology and Geothermal Research, 138, 371−392. Tralli, D. M., Blom, R. G., Zlotnicki, V., Donnellan, A., & Evans, D. L. (2005). Satellite remote sensing of earthquake, volcano, flood, landslide and coastal inundation hazards. ISPRS Journal of Photogrammetry and Remote Sensing, 59, 185−198. Werner, C., Wegmüller, U., Strozzi, T., & Wiesmann, A. (2000). Gamma SAR and interferometric processing software. ERS — ENVISAT Symposium. Wilson, C. J. N., Houghton, B. F., McWilliams, M. O., Lanphere, M. A., Weaver, S. D., & Briggs, R. M. (1995). Volcanic and structural evolution of Taupo Volcanic Zone, New Zealand: A review. Journal of Volcanology and Geothermal Research, 68, 1−28. Wright, T., Fielding, E., & Parson, B. (2001). Triggered slip: Observations of the 17 August 1999 Izmit (Turkey) earthquake using radar interferometry. Geophysical Research Letters, 28, 1079−1082.