Assessing RAT (Robust AVHRR Techniques) performances for volcanicash cloud detection and monitoring in near real-time:
The 2002 eruption of Mt. Etna (Italy)
C. Filizzola a, T. Lacava a, F. Marchese b, N. Pergola a,b,⁎, I. Scaffidi b, V. Tramutoli a,b
a Institute of Methodologies for Environmental Analysis - National Research Council, Z. I. 85050 Tito Scalo (Pz), Italyb Department of Engineering and Physics of Environment - University of Basilicata, Viale dell'Ateneo, 85100 Potenza, Italy
Received 25 May 2006; received in revised form 20 September 2006; accepted 23 September 2006
Keywords: Volcanic ash; Satellite monitoring; AVHRR; RAT
1. Introduction
Airborne volcanic ash annually causes significant damage andincreasing costs for companies in the aviation industry. Even atgreat distances from the eruption sites, ash clouds can causeserious damage to aircraft engines and, as a consequence, theymayrepresent a potential hazard for both crew and passengers (e.g.Casadevall, 1994). Consequently, surveillance systems of eruptiveash clouds have been created and the International AirwaysVolcano Watch (IAVW) has progressively been implemented. Asan integral part of the IAVW, Volcanic Ash Advisory Centers(VAACs) provide updated advisories on the distribution of ashclouds and on the prediction of their movements toMeteorologicalWatch Offices (MWO), which are responsible for forecasting and
officially warning the aviation companies. Owing to the highpotential hazard posed by volcanic ash, the optimal final goal is thecomplete avoidance of eruptive clouds (Campbell, 1994; Cantor,1998; Rossier, 2002; Salinas, 1999), which is only achievableprovided that a timely data collection, interpretation anddissemination, togetherwith an accurate integration of all availableinformation is routinely accomplished.
Theoretically, atmosphere should be considered volcanicash-free and there is no definition of a safe level of exposure(ICAO, 2004; FAA, 2001). This problem became veryimportant after the encounter of an NASADC-8 research aircraftwith a volcanic cloud from Hekla (Iceland) in 2000 (Grindle &Burcham, 2002, 2003; Pieri et al., 2002; Rose et al., 2003).
For the time being, several volcanoes have been studied andmonitored by means of specific ground-based systems (seismicnetworks, gas emission measurement apparatuses, SO2 plumecontent, electric, magnetic and electromagnetic devices, etc.) but
441C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
in many cases ground-based surveillance devices are inadequateto monitor rapid space-time dynamic phenomena, like eruptiveash cloud spread and dispersion in the atmosphere. Moreover,lots of volcanoes are situated in inaccessible areas, oftenunreachable by the traditional monitoring devices. On theother hand, satellites have been increasingly considered a newpossibility to monitor volcanic activity, because they are able toprovide global coverage and continuous observations at arelative low cost (Oppenheimer, 1998).
To accomplish an efficient remote sensing monitoringsystem, which may guarantee adequate safety for the aviation,the satellite methods should be prompt and reliable enough totimely detect the volcanic ash clouds with a low rate of falsealarms, discriminating ash from meteorological clouds inwhatever observational conditions (i.e. in daytime, in night-time, in different seasonal period and on whatever geographiclatitude, under variable atmospheric, environmental and illumi-nation conditions, etc.).
Until now, several methodologies and algorithms have beendeveloped and tested in order to improve our capability tounivocally detect and track volcanic ash from satellite observa-tions. In particular, starting from the pioneer work of Prata(1989a,b), significant efforts have been done, also exploiting theimproved spectral performances of new satellite sensors (e.g.Krotkov et al., 1999; Rose et al., 2000; Rose & Schneider, 1996;Schneider et al., 1995; Seftor et al., 1997; Wen & Rose, 1994).Innovative three-channel or four-channel detection schemes(using the mid-infrared channel and/or the visible band togetherwith the classical, spit-window IR thermal bands) have recentlybeen proposed, showing improved results and better reliabilityand sensitivity (Bonfiglio et al., 2005; Ellrod & Connel, 1999;Ellrod et al., 2003; Mosher, 1999; Pergola et al., 2004). Midinfrared information, in particular, may significantly improvedetection and discrimination of ash clouds, due to differentabsorption and reflectance of ash particles, water and ice. Also aSO2 signal may play a role in this band (Ellrod et al., 2003).
Unfortunately, no single method has been identified yetwhich is able to detect and track volcanic ash clouds in everycase and with full reliability. This has also been well documentedby Tupper et al. (2004). Most of the proposed methods are oftenaffected by residual problems, limiting their application in anoperational scenario (e.g. Prata et al., 2001; Pergola et al., 2001,2004; Rose et al., 2001; Simpson et al., 2000, 2001; Tupper et al.,2004). The nature as well as the properties of the volcanicclouds, generally different depending on volcanoes and eruptioncharacteristics, may prevent remote sensing techniques to beeffective enough to be successfully used in an operational way.
Generally, for proper aviation safety, an accurate, timely andunivocal detection is requested, especially in the first fewminutes after the eruption, together with an efficient continuoustracking of the ash cloud (Simpson et al., 2002). Both scientificand operational issues should be seriously considered andexamined, due to the complexity of the problem (Tupper et al.,2004). In particular, the requirements and procedures to meet anactual operational scenario and the potential impact of failuresin ash cloud detection on aviation safety, are well documentedand discussed in Hufford et al. (2000).
Recently, a contribution to the development of improvedsatellite techniques for ash cloud detection and monitoring hasbeen provided by Pergola et al. (2004), who suggested aninnovative IR detection method, based on a more generalapproach (the Robust AVHRR Techniques — RAT) whichexploits long-term time series of satellite records. In particular,the proposed approach, starting from a prior characterization ofthe signal, identifies ash contributions as signal anomaliescompared to the unperturbed conditions, well discriminatingthem from the random natural fluctuations of the signal by meansof a robust definition of “anomaly”. The information comingfrom themid-infrared (MIR) and thermal infrared (TIR) bands arejointly examined in order to derive local (both in space and timedomains) threshold values, dynamically derived at pixel level.
In this paper, a further investigation of performances andlimitations of this technique is provided, analyzing resultsobtained during the recent Mt. Etna (Sicily — Italy) eruptionwhich occurred in October–November 2002. The proposedRAT protocol, with a double-step test application in succession,has been assessed and improvements and limitations of itsimplementation in an operational scenario were evaluated.
2. The RAT approach for volcanic ash cloud detection
The RAT approach is a very general satellite data analysisstrategy, first suggested by Tramutoli (1998). Its theoreticalbackground is widely discussed also in Tramutoli et al. (2000),Pergola et al. (2001), Di Bello et al. (2003), Bonfiglio et al.(2005) and Tramutoli et al. (2005), together with a tentativeinventory of its possible applications to environmental/naturalhazards and emergencies.
The RATapproach is exclusively based on satellite data, thenit is applicable at a global scale and fully exportable onwhichever satellite/sensor system. Its application to volcanicash cloud detection and tracking, by using AVHRR records,requires a computation of the following so-called ALICE(Absolutely Local Index of Change of Environment) indexes(Pergola et al., 2004):
�DT3�4ðx; y; tÞ ¼DT3�4ðx; y; tÞ−hDT3�4ðx; yÞi
rDT3�4ðx; yÞð1Þ
�DT4�5ðx; y; tÞ ¼DT4�5ðx; y; tÞ−hDT4�5ðx; yÞi
rDT4−5ðx; yÞð2Þ
where the quantities ΔT3–4(x,y,t)= [T3(x,y,t)−T4(x,y,t)] andΔT4–5(x,y,t)= [T4(x,y,t)−T5(x,y,t)] represent the brightness tem-perature differences in AVHRR channel 3 (MIR), 4 and 5 (TIR)while ⟨ΔTi–j(x,y)⟩ and σΔTi–j
(x,y) are, respectively, the relevantmean values and standard deviations of ΔTi–j, calculated over along-term series of homogeneous satellite records collected inunperturbed periods, as described in Pergola et al. (2004). Thejoint use of both indexes leads to a so-called “three-channel”RAT approach (as AVHRR channel 3, 4 and 5 are employed)whereas a simplified RAT version can be obtained by using thesole ⊗ΔT4–5
(x,y,t) index, within a “two-channel” configuration.Basically, the idea is to exploit the known spectral signature of
Table 2Night-time AVHRR passes
Date Time GMT (LT) Platform and orbit Panel (see Fig. 1)
442 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
ash in these bands but considering relative, rather than absolute,induced changes. In particular, the ash signature for ΔT4–5signal is a clear decrease because of the reverse absorption ofash compared to atmospheric water vapour (Prata, 1989a),whereas for ΔT3–4 an increase is expected, although with adifferent intensity depending on observational conditions (i.e.night/day). The fact is that, even though the signatures aresimilar, the intensities and magnitudes of these changes,induced by ash, might be very different, strongly dependingon geographic location, time of acquisition, season, etc. Theproposed ALICE indexes should take into account these local(both in space and time domain) factors, again by consideringrelative, rather than absolute, variations.
Therefore, the proposed indexes⊗ΔT4–5(x,y,t) and⊗ΔT3–4(x,y,t)should, respectively show negative and positive values incorrespondence to volcanic ash clouds regardless of, for example,atmospheric moisture conditions or underlying surface temper-ature. Higher (in modulus) are the ALICE values, moresignificant (from a statistical point of view) are the anomaliesand, consequently, higher should the relative ash content be.Obviously, such is the complexity of the problem that residualdifficulties may still survive, arising from the specific character-istics of the eruptive clouds and/or from very particularenvironmental and observational conditions, for instance.
In this paper, an extended evaluation of such a technique hasbeen carried out, analyzing the results obtained during the recentMt. Etna eruption, which occurred on 26 October 2002 andlasted for several weeks. The analysis has been performed bothfor daytime and night-time observations and the results achievedwith the joint application of both ALICE indexes (i.e. followingthe three-channel approach) have also been discussed forcomparison with those obtained applying the simplified RATversion (i.e. by using the two-channel configuration). Reliability(i.e. secure detection without confusion between meteorologicaland volcanic clouds) as well as sensitivity (i.e. capability toaccurately describe the ash clouds, characterizing them in termsof size, extent, position and direction) are discussed, in order toassess RAT performances and limitations on the whole.
3. Event description
On October 26, 2002, after a period of relatively weakactivity, at Mt. Etna (37.73N, 15.00E, Sicily, Italy) started oneof the most vigorous lateral eruptions, causing important ash fallthat paralyzed public life and airline traffic for several days.This eruption was one of the most explosive for this volcano inrecent times and, in this occasion, more than two-thirds of thetotal volume of erupted products were pyroclastic (Bertrandet al., 2003). In this case, explosive activity injected enormous
Table 1Characteristics of AVHRR datasets
Dataset Acquisition time Month Temporal span
d.1 23:00–02:00 GMT October 1997–2003d.2 11:00–14:00 GMT October 1997–2003d.3 23:00–02:00 GMT November 1997–2003d.4 11:00–14:00 GMT November 1997–2003
volumes of volcanic ash into the atmosphere which was carriedhundreds of kilometres away from the Sicilian coasts.
In particular, on 27th October, a thick black cloud rose about6 km (Carn et al., 2005) above the vents on the upper southernflank, while smaller ash columns rose from the vents situated onthe northeast Rift (Bertrand et al., 2003). The plume wasdirected southward, towards areas around the city of Cataniaand reached Mount Iblei's area, 100 km south (INGV, 2002).
In the following few days, a series of towns were stronglyaffected by ash fallout, in particular Catania, Nicolosi,Mascalucia, San Pietro Clarenza, Gravina, Pedara and Trecas-tagni and the ash reached the city of Siracusa, about 80 km awayfrom the eruptive center (INGV, 2002). The ash emissioncontinued copious for several days causing the shutdown of the“Fontanarossa” Catania's International Airport on 29, 30 and 31October. In less than three days, more than 2 kg/m2 of ash fellon the city of Catania (INGV, 2002).
On 2 November, the direction of the ash cloud changed,moving in a Northern direction, towards Tyrrhenian sea andthen it was suddenly pushed to an Eastern direction, reachingthe southern and central regions of continental Italy, asconfirmed also by independent direct measurements fromLIDAR systems (Pappalardo et al., 2004).
This Etna eruption lasted several weeks and the eruptivecolumn was continuously present above the volcano (Scolloet al., 2005). Particularly during 5 and 6 November, ash fell onNNE slopes while on 7 November the axis dispersion plumechanged and wreaked ash near Zafferana and Milo regions.After November the 14th, the release of ash was significantlyreduced and the products fell mainly in areas near the craters(INGV, 2002).
Afterwards, the eruption continued in a more or less irregularmanner until 28 January 2003 (Bertrand et al., 2003).
4. Technique implementation
All AVHRR images, selected according to homogeneitycriteria as quoted in Pergola et al. (2004), have been stratifiedaccording to the time and the period of acquisition andorganized in four datasets, reported in Table 1. In particular,datasets d.1 and d.2 included all passes acquired in the month ofOctober, respectively around 24:00 and 12:00 GMT (01 A.M.and 01 P.M. Local Time, respectively) and datasets d.3 and d.4collected all passes received in November, within the sametemporal ranges. In order to build the background referencefields, the images covering a time span of six years (i.e. from
443C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
1997 to 2003) were analysed, excluding all passes of 2002because they were perturbed by the event under study.
Before further processing, all passes have been calibratedfollowing standard calibration procedures (Kidwell, 1991;Lauritson et al., 1979) and navigated employing an automaticand precise (i.e. one AVHRR pixel of residual error) navigationprocedure proposed and validated by Pergola and Tramutoli(2000, 2003).
A specific Region of Interest (ROI) has been chosen and, foreach pass, a subsetting procedure has been applied. The resultingscenes have been reprojected in a common geographic grid
Fig. 1. Results of the three-channel RAT configuration. Panels identify night-time pintensities. In particular, yellow pixels represent areas where ⊗ΔT4–5
(x,y,t)b−1 and ⊗(x,y,t)b−2 and ⊗ΔT3–4
(x,y,t)N2. (For interpretation of the references to colour in thi
(Lambert Azimuthal Equal Area) so as to have a final set of co-located images, stratified along the time dimension. Further-more, for each homogeneous dataset (d.1, d.2, d.3, d.4), theabove defined reference fields (i.e. mean ⟨ΔT(x,y)⟩ and standarddeviation σΔT (x,y)) have been calculated for all pixels within theROI, for a final number of eight reference images.
Later on, in order to locate ash clouds, ALICE indices⊗ΔT4–5
(x,y,t) and ⊗ΔT3–4(x,y,t) as reported in Eqs. (1) and (2), have
been computed for each image of the sequence acquired duringthe considered eruptive event, following prescriptions describedin Pergola et al. (2004).
asses listed in Table 2 and different colours describe anomalies with different
ΔT3–4(x,y,t)N2 whereas red zones indicates the more restrictive condition ⊗ΔT4–5
s figure legend, the reader is referred to the web version of this article.)
444 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
5. Results
5.1. Night-time cases
All the available, cloud-free AVHRR images acquired withinthe first days of the eruption occurrence (i.e. from 27 October to7 November 2002) have been analysed by means of the RATapproach and the results obtained have been discussed in orderto assess its performances and limitations. The techniqueassessment has been performed both in terms of reliabledetection (i.e. capability to well discriminate ash frommeteorological clouds) and of efficient sensitivity (i.e. ability
Fig. 2. Results of Spilt-Window Temperature Difference (SWTD) method. Panels idΔT4–5=T4−T5b0) are marked in red. (For interpretation of the references to colour
to accurately describe ash cloud properties such as size, shape,structure, etc.).
In Table 2 all the nocturnal passes analysed with theproposed approach are reported.
In Fig. 1 the results obtained by the joint use of both ALICEindices in Eqs. (1) and (2), (i.e. by following the more completethree-channel RAT configuration) according to the prescriptionsquoted in Pergola et al. (2004), are shown together with ageographical map of the investigated ROI. Panels from a to grefer to passes listed in Table 2 and different colours identifyanomalies with a different intensity. Looking at this figure itappears clear as the proposed approach was able to correctly
entify night-time passes listed in Table 2. Detected ash clouds (i.e. pixels wherein this figure legend, the reader is referred to the web version of this article.)
445C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
detect, in all scenes, the position of the ash cloud, withoutproducing any confusion with meteorological clouds, which arealso present in the images. In particular, in this RATconfiguration the detection reliability is such that ‘false alarms’(i.e. signals coming from no-ash contributions) are reduced tovery few, sporadic pixels, permitting a nearly univocal locationof the ash dispersion area. Moreover, the approach seems able toidentify with sufficient accuracy the ash size and shape and tofollow its direction well.
Fig. 3. Detailed comparison of detection accuracy and sensitivity of the SWTD methoare the ones used in Figs. 2 and 1, respectively. Dates of passes are put on the top olegend, the reader is referred to the web version of this article.)
In order to better assess the results achieved, the latter havealso been compared with the ones obtained by the application ofthe standard SWTD (Split-Window Temperature Difference)technique, one of the most well-established and largely usedsatellite methods for ash cloud detection and tracking (Prata,1989a,b). Comparisons have been analysed under the assump-tion that the ash position was known over all the consideredsatellite scenes. SWTD results (i.e. ΔT4–5=T4−T5b0) areshown in Fig. 2, for the same night-time passes listed in Table 2.
d (on the left side) with the proposed RAT approach (on the right). Colour codesf each image pairs. (For interpretation of the references to colour in this figure
Fig. 3 (continued).
446 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
Looking at panels in Fig. 2, it should be noted as the SWTDmethod does not suffer, in this case, from a false alarm problem.In all considered days, in fact, meteorological clouds over thescene were not misinterpreted as ash contributions. On the otherhand, some limitations of such a traditional method can berecognized and should be mentioned. It is well known that theSWTD technique shows possible failures if volcanic clouds areoptically opaque and/or in case of particles with a size greaterthan 5 μm (e.g. Krotkov et al., 1999; Rose et al., 2001; Schneideret al., 1995; Simpson et al., 2000; Wen & Rose, 1994),circumstances that typically occur within the first hours after aneruptive event. A recent study carried out by using AIRS(Atmospheric Infrared Sounder, orbiting on EOS/Aqua) spectra,demonstrated that ash particle size, during the first days of theEtna eruption, was in the 10 μm range (Carn et al., 2005),possibly higher over the ROI of this work. Although this was, asthe authors themselves say, a rough estimate, such a particle sizeis also consistent with results of direct measurements on samplesof previous Etna emissions (Allard, 1999). This could explain thepoor results achieved by SWTD during the first days of theeruptions (e.g. panels a, b and c, in Fig. 2, for instance): during allthese days, in fact, the SWTDdetection sensitivity is dramaticallylimited, with only few pixels (when detected) identified as ash.Especially on October 28 and 29 (panels b and c), i.e. the secondand the third day after the eruption onset, the SWTD approachseems to show a strong omission problem, making it practicallyunable to identify any volcanic cloud on both scenes. On the otherhand, in the following days, the SWTD technique seems tobehave similarly to RAT (compare for instance panels e and g ofFigs. 1 and 2), as the actual ash contributions correctly identifiedare more or less the same. An exception is represented by theimage of November the 2nd (panels f in Figs. 1 and 2) where,again, the SWTDmethod shows a reduced sensitivity, limiting itscapability in tracking the abrupt change of direction of the ashcloud which occurred that day. The RATapproach is, in this case,more sensitive and, consequently, it is able to better map the rightposition and extent of the cloud, correctly identifying itsmovement toward northern latitudes.
A detailed documentation of the above mentioned improvedRAT sensitivity, is reported in Fig. 3, where an accurate
comparison with the SWTD results (on the left side) is shownfor some of the previous cited days: in all cases the bettersensitivity of the RAT approach is clear, allowing for the samerate of false alarms, a more accurate description andcharacterization of the ash cloud extent, shape and direction.
Therefore, the ‘three channel’ RAT configuration (i.e. a jointscheme with the combined use of both ALICE indexes) mayguarantee high detection reliability and an improved (comparedto the standard SWTD method) sensitivity, useful to betterdescribe ash cloud characteristics.
Finally, some residual limitation of this index must bementioned, regarding an occasional inability to perform areliable and effective detection of ash cloud when it is locatedover land locations (see for instance Fig. 1 panels a and b).Although such a limitation is common to the standard SWTDmethod (see Fig. 2), it may represent a residual problem whichdeserves further investigations.
In order to better understand the RAT potential andcapabilities, the simplified two-channel version has also beenassessed, evaluating the results obtained by using the soleALICE index defined in Eq. (2) (i.e. ⊗ΔT4–5
(x,y,t)). Thisindicator, in fact, disregarding the information provided bythe MIR band, is more strictly related to the traditional split-window brightness temperature difference (SWTD) method.These results are reported in Fig. 4 similarly for all the nocturnalpasses listed in Table 2. As before, different colours identifyanomalies with different intensity. From the analysis of thefigure, the following considerations can be drawn:
i) Looking first at anomalies with the higher level ofintensity (i.e. the red areas), the RATapproach, also in thissimplified configuration, seems able to clearly distin-guish, with sufficient accuracy and reliability, the ashclouds from meteorological clouds. The latter, in fact,although clearly contaminating part of the scenes (see forinstance the central part of panel a, as well as the westernzones in both panels b and c) are practically neverenhanced by the proposed approach during the first threedays (i.e. panels a, b and c). In the following days, somefew, sporadic pixels located over meteorological clouds
Fig. 4. Results of the two-channel, simplified RAT configuration. Panels identify night-time passes listed in Table 2 and different colours describe anomalies withdifferent intensities. In particular, for all panels, yellow pixels represent areas where⊗ΔT4–5
(x,y,t)b−1 whereas red zones indicate more intense anomalies satisfying themore restrictive condition⊗ΔT4–5
(x,y,t)b−2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
447C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
are erroneously detected as ash (see for instance left sidein panel d or bottom right corner in panel g) suggesting aweak degradation of performances of this simplified RATscheme as far as the ash cloud disperses into theatmosphere. On the other hand, even looking only atthe more intense anomalies, an improved capability tocharacterize the volcanic cloud, in terms of shape, extent,direction, size, etc., is achieved with this simplifiedconfiguration.
ii) Looking at the low level intensity of anomalies (i.e. theyellow zones), more diffuse non-ash contributions areerroneously detected on several scenes, mostly after the
first three days. Even if such erroneous signals are oftenquite randomly distributed over the scene, withoutfollowing some preferred direction and, then, lookingvery different from the ash patterns (which, in fact, stillremain clearly distinguishable in all panels), this resultmay represent a limitation, discouraging its application, inthis configuration, in an operational scenario. On theother hand, the ash cloud is very well described in termsof size, shape, extent, etc. by looking at these areas,suggesting an employment of such a simplified config-uration of RAT scheme only if an ash cloud accuratedescription, rather than a reliable detection, is requested.
448 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
Therefore, a compromising choice favouring reliabilityrather than sensitivity or vice versa should be accom-plished if the simplified RAT configuration is used.Therefore, if a reliable detection and discrimination frommeteorological clouds is desired, the more intenseanomalies have to be analysed, whereas low levelsshould be used when a detailed description andcharacterization of volcanic clouds in terms of size,extent, shape, direction, is requested.
In conclusion, comparing the results obtained with the twoRAT configurations, we can summarize that detection reliabilitysignificantly improves if the two ALICE indices are jointly
Fig. 5. Results of the two-channel, simplified RATconfiguration. Panels show the firsof the figure maps different ALICE index (i.e.⊗ΔT4–5
(x,y,t)) values, for all the satellitethe references to colour in this figure legend, the reader is referred to the web versio
used, making the technique more suitable for a prompt anddefinite detection, also in an unsupervised implementation asrequired in an operational context (e.g. Hufford et al., 2000).Moreover, the recourse to a joint usage of the proposedindicators, also provides satisfying results in terms of ash clouddescription and characterization, without limiting the techniquesensitivity which still remains good enough to map the ashcloud well in terms of its size, extent and direction.
Finally, once a reliable identification of the ash cloud hasbeen obtained, the recourse to lower ALICE levels as well as theuse of the simplified two-channel configuration (also at a lowerlevel of anomaly intensity), may surely improve the sensitivityof the technique in characterising ash cloud properties such as
t four passes (panels from a to d) shown in Figs. 1 and 2. Colour bar at the bottomscenes. A black triangle indicates the position of Mt. Etna. (For interpretation ofn of this article.)
Table 3Daytime AVHRR passes
Date Time GMT (LT) Platform and orbit Panel (see Fig. 5)
449C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
size, shape, direction, etc. Such results are in accordance withour previous achievements, obtained in particular analysing theEtna eruption which occurred in July 2001 (Pergola et al.,2004), confirming the significance of the information providedby the mid-infrared band for the development of a prompt andreliable detection tool and the usefulness of the intrinsictuneability of the proposed approach for a better ash cloudcharacterization.
5.2. Night-time results: discussion
To summarize the results achieved by the analysis ofnocturnal passes, some general considerations can be drawn.The more complete three-channel RAT configuration, obtainedby jointly using the two ALICE indexes, as already demon-strated in Pergola et al. (2004), seems able to identify ash cloudwith an improved level of efficiency, strongly reducing the ash/no-ash discrimination problem and, at the same time, savingdetection sensitivity. Therefore, the information provided by themid-infrared channel, as documented also by other authors (e.g.Ellrod et al., 2003, Mosher, 1999) seems to be of valuableimportance for a reliable and efficacious ash cloud detection, tobe possibly implemented in an operational scenario. On theother hand, the ⊗ΔT4–5
(x,y,t) ALICE index, when applied alone,might be able to provide quite good results, also in terms ofdetection reliability, provided that higher level of anomalyintensities are considered. Besides, a better ash cloud size andshape description can surely be achieved with this simplifiedscheme, which may also provide information about the ashcloud internal structure. Therefore, the intrinsic tuneability ofthe proposed ALICE indicator might allow a definition ofspecific procedures and protocols that, for instance, can use themore robust and reliable joint scheme (and/or the simplifiedconfiguration but looking at very high intensity anomalies) for apreliminary and univocal identification of the ash plume.Furthermore, once all the possible no-ash sources areeliminated, the ‘two-channel’ simplified RAT version mightbe used to improve the mapping capabilities, better describingthe size, shape, extent and direction of the detected plume. Withregards to this, one example is shown in Fig. 5, where the single⊗ΔT4–5
(x,y,t) ALICE index computed for some passes (i.e.panels from a to d in Figs. 1 and 2) is depicted in false coloursand reported only over the area previously detected (by meansof the joint scheme) as affected (with the highest probability) bythe ash cloud. The ash size and extent is better described, asexpected, when also lower values of the index are considered(see blue and green tones in figure). Besides, some consider-ation about the ash column internal structure (i.e. relative ashcontent, particle size, column density, etc.) could be inferredperhaps by analysing such a map with the help of volcanolo-gists' expertise and knowledge. For example, results achievedfor October 30 (Fig. 5c) seem to be correlated quite well,although in a qualitative way and for a limited portion of theplume, with independent estimations of column ash densityobtained for the same day by means of AIRS observations (Carnet al., 2005). This result is particularly important if anotherretrieval obtained by Carn et al. (2005) is considered: the mean
ash particle diameter. The latter, in fact, is for the same day andfor the same portion of the plume evaluated by AIRSmeasurements in the 10 μm size range, then in a range makingthe SWTD method ineffective. This probably explains the poorresults obtained by the traditional SWTD technique in trackingthe ash cloud on that day (see Fig. 3) and gives to the proposedapproach an additional valence: the RAT technique appears notto be limited by these intrinsic plume characteristics and then itmight be regarded as a complementary method to be used whenenvironmental/observational conditions as well as specificplume properties disfavour the application of traditionalmethods.
In conclusion, the recourse to such a RAT protocol, usingboth the detection schemes in succession, seems to be the betterway to perform a reliable detection together with a detaileddescription of eruptive ash clouds. Such a scheme has actuallybeen investigated, confirming its usefulness, in a pre-opera-tional way during this eruptive event when, by developing anautomatic procedure and processing chain, we experimented forthe first time a real-time provision of ash satellite products to theItalian National Department of Civil Protection, who had tomanage the Catania's International Airport operations. DPCreported a very positive feedback from this experiment; in fact,the timely delivery of satellite products as well as their frequentupdating, allowed Civil Protection to have eruptive cloud undera continuous control for more than two months and gave tothem the opportunity to validate and check performances oftheir operational plume dispersion model.
5.3. Daytime cases
A similar analysis has been carried out for AVHRR imagesacquired in daytime, in order to assess RAT results in differentobservational conditions. In Table 3 all the diurnal passesacquired after the eruption onset and analysed following RATapproach are reported, listed by date, acquisition time and orbitnumber. Acquisition dates in this case span from October 27until November 7, including several days with double passes.Similarly to Table 2, the last column in Table 3 associatessatellite passes with panels shown in the following figures.
Because of the 3A/B AVHRR channels switching (Goodrumet al., 2000), the daytime analysis has been carried out only in
Fig. 6. Results of the two-channel, simplified RAT configuration. Panels identify daytime passes listed in Table 3 and different colours describe anomalies withdifferent intensities. In particular, for all panels, yellow pixels represent areas where⊗ΔT4–5
(x,y,t)b−1 whereas red zones indicate more intense anomalies satisfying themore restrictive condition⊗ΔT4–5
(x,y,t)b−2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
450 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
Fig. 7. Results of Spilt-Window Temperature Difference (SWTD) method. Panels identify daytime passes listed in Table 3. Detected ash clouds (i.e. pixels whereΔT4–5=T4−T5b0) are marked in red. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
451C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
452 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
the simplified RAT configuration, being the mid-infraredinformation not available on NOAA-16 in daylight. In Fig. 6the RAT results are shown, using the same colour code as Fig. 1.In this case the proposed approach, although in its simplifiedconfiguration, seems to perform better and longer than thenight-time condition. In fact, for more than six days, fromOctober 27 to November 2 (from panel a to panel i with a soleexception on panel h), the RAT reliability is good enough,regardless of the intensity of ALICE index, to assure quite anunivocal identification of the ash cloud, without any confusionwith no-ash contributions possibly due to meteorologicalclouds. Also the sensitivity of the approach seems to fulfillthe requirements for an accurate description of ash extent andfor a complete identification of its position, direction andevolution. With regards to this aspect, the result achieved onNovember 2 (panel i) deserves to be mentioned as, again, thetechnique was able to identify and describe the unexpecteddirection change of the volcanic cloud observed that day. On theother hand, starting from November 4 and for the next few days(i.e. from panel l to panel n), a degradation in performances,although limited to lower ALICE values (i.e. yellow areas), ismanifested. Mostly at these lower intensities, some meteoro-logical clouds, though far from the volcano, are erroneouslyinterpreted as ash, reducing reliability of RAT in thisconfiguration. This more or less confirms the results previouslyobtained (in night-time), even if with a late occurrence time. Onthe other hand, looking at higher ALICE values (i.e. the redzones), such a limitation seems to be completely overcome. Infact, provided that the higher (in modulus) ALICE indices areconsidered (i.e. the signals more significantly deviating fromthe expected value), most of the false identifications disappear.The sensitivity is, of course, similarly reduced by this way andsuch a drawback still remains to be accurately evaluated. Inorder to do that, again a comparison with the standard SWTDmethod has been carried out.
Therefore, the standard SWTDmethod has been investigatedand results are shown in Fig. 7. Once more, such a traditionalmethod does not produce confusion between meteorologicaland volcanic clouds but, similarly to the night-time case, suffersfrom a sensitivity (i.e. omission) problem, especially as far asthe first days of the eruption are considered (see for instancepanels a, b). In the following days SWTD performs well, apartfrom panel i (i.e November the 2nd), where, again, the directionchange of the plume, documented also by other independentobservations (e.g. Pappalardo et al., 2004) is only partlyrecognized and the actual extent of ash cloud is not properlydetected and identified.
Therefore, to summarize, in daytime performances of theproposed approach are adequate to monitor and describe ashcloud evolution with sufficient reliability and sensitivity forabout one week, also performing better than the traditionalSWTD method which, instead, suffers from a sensitivityproblem, especially (but not only) in the first few days afterthe eruption onset. In the following days, a degradation inperformances of the proposed ALICE index is observed,although limited to lower anomaly intensities. Starting fromNovember 4 (i.e. panel l), in fact, some no-ash contributions,
mainly due to meteorological clouds, are erroneously detected.However, such a performance degradation seems to be almostnegligible when more intense anomalies are considered.Anyway, this limitation may be mainly ascribed to the lack ofthe AVHRR 3B band and, consequently, a further improvementis to be expected, as shown before, by the introduction of themid-infrared signal. Fortunately, at the time of writing NOAAlaunched and made another ‘afternoon’ polar platform opera-tional (i.e. NOAA-18), where a modified version of AVHRRradiometer is operational, with the traditional 3.7 μmMIR bandswitched back to working all time long, also in daytime. Thiscircumstance will surely contribute to improve RAT capabilitiesunder whatever environmental and observational conditions.
6. Conclusions
AVHRR records have been analysed during the recent Etna(Sicily, Italy) eruption which occurred in October 2002 in orderto assess performances and limitations of a previously proposedrobust scheme (the RAT approach) to detect and track ashclouds with sufficient reliability and sensitivity. The RATapproach has been assessed both in its simple (i.e. two-channels) and advanced (i.e. three channels) configurations andunder different observational conditions (e.g. night/day passes).
The discrimination reliability (between ash/no ash) achievedin this case is dependent on the selected configuration and/or onthe level of anomaly intensity: simple RAT configuration seemsto provide good reliability only if higher level of anomalyintensity is considered, both in night-time and in daytime. Onthe other hand, the three-channel RAT scheme performs betterand longer, regardless of the different intensity of anomalies.
Concerning detection sensitivity, both the configurationschemes benefit from the intrinsic tuneability of RAT, providinggood capability to describe, for all the considered days andinvariantly for day/night observations, the actual size, shape,extent and direction of the ash plume with sufficient accuracy, aswell as to infer some qualitative eruptive cloud characteristics,such as the ash column density. Therefore, specific proceduresand protocols have been experimented first in an operationalway and can be better investigated in order to achieve the bestfrom such an approach. In particular, provided that a first step iscarried out by using the advanced scheme configuration in orderto optimize the ash/no ash discrimination (i.e. excluding as muchas possible no-ash contributions), a further analysis can beapplied in succession, zooming over the areas where theprevious detection scheme suggested the highest probability tofind ash clouds, exploiting the tuneability of the two-channelALICE index to better describe in detail ash cloud propertiesand, possibly, to infer considerations about internal structure andcharacteristics of the plume column. In particular, preliminarypromising (qualitative) correlations between the ash plume“structure” observed by RAT and independent observations andretrievals, obtained by AIRS data, have briefly been discussed.These RAT performances, especially achieved in the early stageof the eruption, are particularly valuable and candidate RATapproach as, at least, a complementary method of the well-established SWTD technique. In fact, SWTD methodology
453C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
suffers from some well-known limitations which may oftenoccur within the first moments of ash cloud life, exactly when,instead, the proposed approach has shown better responses.Limitations of the proposed approach regards a sort ofoccasional inability (mainly in night-time) to detect ash whenit is located over land (but the standard SWTD technique suffersfrom a similar problem) and a weak degradation with time. Thelatter drawback is, however, mostly due to the lack, in daytimepasses, of the MIR band on the AVHRR sensor configurationonboard NOAA-16, confirming the importance of this infraredinformation for this kind of application. Fortunately, such aresidual limitation should be completely overcome because ofthe new configuration of AVHRR/3 sensor, aboard the new‘afternoon’ NOAA polar platform (i.e. NOAA-18), which doesnot perform the 3A/B channel switching in daytime anymore.Such a circumstance should contribute to improve RATcapabilities in whatever environmental and observationalconditions, suggesting this method may be profitably used toincrease our ash cloud satellite monitoring capability.
References
Allard, P. (1999).Mitigation of volcanic risk by remote sensing: Airborne plumevalidation, final report. MVRRS research contract ENV4-CT96-288Bruxelles: Eur. Union Fourth Framework Programme, 55 pp.
Bertrand, C., Clerbaux, N., Ipe, A., & Gonzalez, L. (2003). Estimation of the2002 Mount Etna eruption cloud radiative forcing from Meteosat-7 data.Remote Sensing of Environment, 87, 257−272.
Bonfiglio, A., Macchiato, M., Pergola, N., Pietrapertosa, C., & Tramutoli, V.(2005). AVHRR automated detection of volcanic clouds. InternationalJournal of Remote Sensing, 26(1), 9−27.
Campbell, E. E. (1994). Recommended flight-crew procedures if volcanic ash isencountered. In T. J. Casadevall (Ed.), Volcanic ash and aviation safety:Proceedings of the First International Symposiumon volcanic ash and aviationsafety, Seattle, WA. U.S. Geological Survey Bulletin, Vol. 2047 (pp. 151157).
Cantor, R. (1998). Complete avoidance of volcanic ash is only procedure thatguarantees flight safety. ICAO Magazine, 53, 18−19 (26).
Carn, S., Strow, A., de Souza-Machado, L. L., Edmonds, S., & Hannon, Y.(2005). Quantifying tropospheric volcanic emissions with AIRS. The 2002eruption of Mt. Etna (Italy). Geophysical Research Letters, 32, L02301,doi:10.1029/2004GL021034.
Casadevall, T. J. (1994). The 1989–1990 eruption of Redoubt Volcano, Alaska;impacts on aircraft operations. Journal of Volcanology and GeothermalResearch, 62, 301−316.
Di Bello, G., Filizzola, C., Lacava, T., Marchese, F., Pergola, N., Pietrapertosa,C., et al. (2003). Robust satellite techniques for volcanic and seismic hazardsmonitoring. Annals of Geophysics, 47(1), 49−64.
Ellrod, G. P., & Connel, B. H. (1999). Improvements in volcanic ash detectionusing GOES multispectral image data. 8th Conference on Aviation, Range,and Aerospace Meteorology (pp. 326−329). American MeteorologicalSociety Preprint volume.
Ellrod, G. P., Connel, B. H., & Hillger, D. W. (2003). Improved detection ofairborne volcanic ash using multi-spectral infrared satellite data. Journal ofGeophysical Research, 108(D12), 4356, doi:10.1029/2002JD002802.
FAA (U.S.A.) Aviation Weather Directorate (2001). User Needs AnalysisDocument: Volcanic Activity. Version 1.0, 5 Sept 2001.
Grindle, T. J., & Burcham, F. W. (2002). Even minor volcanic ash encounterscan cause major damage to aircraft. ICAO Journal, 57(2), 1−14 (29).
Grindle, T. J., & Burcham, F. W. (2003). Engine Damage to a NASA DC-8-72airplane from a high-altitude encounter with a diffuse volcanic ash cloud.Technical Memorandum NASA/TM-2003-212030. 22 pp.
Goodrum, G., Kidwell, B. K., &Winston, W. (2000). NOAA KLMUser's Guide.U.S. Department of Commerce National Oceanic and Atmospheric
Administration National Environmental Satellite, Data, and InformationService. Available on http://www2.ncdc.noaa.gov/docs/klm/index.htm
Hufford, G. L., Salinas, L. J., Simpson, J. J., Barske, E. G., & Pieri, D. C. (2000).Operational implications or airborne volcanic ash. Bulletin of the AmericanMeteorological Society, 81(4), 745−755.
ICAO (2004). International Civil Aviation Organisation International Hand-book on the International Airways Volcano Watch (IAVW)—Operationalprocedures and contact list. ICAO Doc 9766-AN/968, Second Edition-2004Available at: http://www.icao.int/icao/en/download.htm#Docs
INGV-Sezione di Catania staff (2002): http://www.ct.ingv.itKidwell, K. (1991). NOAA polar orbiter data user's guide, NCDC/SDSD.
Washington, DC: National Climatic Data Center.Krotkov, N. A., Torres, O., Seftor, C., Krueger, A. J., Kostinski, A., Rose, W. I.,
et al. (1999). Comparison of TOMS and AVHRR volcanic ash retrievalsfrom the August 1992 eruption of Mt. Spurr. Geophysical Research Letters,26(4), 455−458.
Lauritson, L., Nelson, G. J., & Porto, F. W. (1979). Data extraction andcalibration of TIROSN/NOAA radiometers. NOAA Technical MemorandumNESS, Vol. 107. Washington, D.C., U.S.A.: U.S. Department of Commerce.
Mosher, F. R. (1999). Four channel volcanic ash detection algorithm, AviationWeather Center Staff publication. Available on http://aviationweather.noaa.gov/publications/mosher/paper3.html
Oppenheimer, C. (1998). Volcanological applications of meteorologicalsatellites. International Journal of Remote Sensing, 19(15), 2829−2864.
Pappalardo, G., Amodeo, A., Mona, L., Pandolfi, M., Pergola, N., & Cuomo, V.(2004). Raman lidar observations of aerosol emitted during the 2002 Etnaeruption. Geophysical Research Letters, 31, LXXXXX, doi:10.1029/2003GL019073.
Pergola, N., Pietrapertosa, C., Lacava, T., & Tramutoli, V. (2001). Robustsatellite techniques for volcanic eruptions monitoring. Annals of Geophys-ics, 44(2), 167−177.
Pergola, N., & Tramutoli, V. (2000). SANA: Sub-pixel automatic navigation ofAVHRR imagery. International Journal of Remote Sensing, 21(12),2519−2524.
Pergola, N., & Tramutoli, V. (2003). Two years of operational use of SANA(Sub-pixel Automatic Navigation of AVHRR) scheme: Accuracy assess-ment and validation. Remote Sensing of Environment, 82(2), 190−203.
Pergola, N., Tramutoli, V., Marchese, F., Scaffidi, I., & Lacava, T. (2004).Improving volcanic ash cloud detection by a robust satellite technique.Remote Sensing of Environment, 90, 1−22.
Pieri, D., Ma, C., Simpson, J. J., Hufford, G., Grindle, T., & Grove, C. (2002).Analyses of in-situ airborne volcanic ash from the February 2000 eruption ofHekla Volcano, Iceland. Geophysical Research Letters, 29, 19-1−19-4.
Prata, A. J. (1989a). Observations of volcanic ash clouds in the 10–12 μmwindow using AVHRR/2 data. International Journal of Remote Sensing, 10,751−761.
Prata, A. J. (1989b). Infrared radiative transfer calculations for volcanic ashclouds. Geophysical Research Letters, 16, 1293−1296.
Prata, A. J., Bluth, G. J. S., Rose, W. I., Schneider, D. J., & Tupper, A. C. (2001).Comments on “Failures in detecting volcanic ash from a satellite-basedtechnique”. Remote Sensing of Environment, 78, 341−346.
Rose, W. I., Bluth, G. J. S., & Ernst, G. G. J. (2000). Integrating retrievals ofvolcanic cloud characteristics from satellite remote sensors: A summary. Phi-losophical Transactions of the Royal Society of London. A, 358, 1585−1606.
Rose, W. I., Bluth, G. J. S., Schneider, D. J., Ernst, G. G. J., Riley, C. M., &McGimsey, R. G. (2001). Observations of 1992 Crater Peak/Spurr volcanicclouds in their first few days of atmospheric residence. Journal of Geology,109, 677−694.
Rose, W. I., Gu, Y., Watson, I. M., Yu, T., Bluth, G. J. S., Prata, A. J., et al.(2003). The February–March 2000 eruption of Hekla, Iceland from asatellite perspective. In A. Robock & C. Oppenheimer (Eds.), Vol-canism and Earth's atmosphereAGU Geophysical Monograph, Vol. 139(pp. 107132).
Rose, W. I., & Schneider, D. J. (1996). Satellite images offer aircraft protectionfrom volcanic ash clouds. EOS, Transactions—American GeophysicalUnion, 77, 529−530.
Rossier, R. N. (2002, February). Volcanic ash: Avoid at all cost. Business andCommercial Aviation, 70−71.
454 C. Filizzola et al. / Remote Sensing of Environment 107 (2007) 440–454
Salinas, L. (1999). Volcanic ash clouds pose a real threat to aircraft safety. EightConference on Aviation, Range, and Aerospace Meteorology, Dallas, Texas(pp. 253−256). American Meteorological Preprints.
Schneider, D. J., Rose, W. I., & Kelley, L. (1995). Tracking of 1992 eruptionclouds from Crater Peak of Mount Spurr Volcano, Alaska, using AVHRR.U.S. Geological Survey Bulletin, 2139, 27−36.
Scollo, S., Coltelli, M., Prodi, F., Folegani, M., & Natali, S. (2005). Terminalsettling velocità measurements of volcanic ash durino the 2002–2003 Etnaeruption by an X-band microwave rain gauge disdrometer. GeophysicalResearch Letters, 32, L10302, doi:10.1029/2004GL022100.
Seftor, C. J., Hsu, N. C., Herman, J. R., Bhartia, P. K., Torres, O., Rose, W. J.,et al. (1997). Detection of volcanic ash clouds from Nimbus 7/total ozonemapping spectrometer. Journal of Geophysical Research, 102(D14),16749−16759.
Simpson, J. J., Hufford, G., Pieri, D., & Berg, J. S. (2000). Failures in detectingvolcanic ash from a satellite-based technique. Remote Sensing ofEnvironment, 72, 191−217.
Simpson, J. J., Hufford, G., Pieri, D., & Berg, J. (2001). Response to“Comments of Failures in detecting volcanic ash from a satellite-basedtechnique”. Remote Sensing of Environment, 78, 347−357.
Simpson, J. J., Hufford, G., Pieri, D., Servranckx, R., Berg, J., & Bauer, C.(2002). The February 2001 eruption of Mount Cleveland, Alaska: Casestudy of an aviation hazard. Weather and Forecasting, 17, 691−704.
Tramutoli, V. (1998). Robust AVHRR Techniques (RAT) for environmentalmonitoring: Theory and applications. In G. Cecchi & E. Zilioli (Eds.), Earthsurface remote sensing. II. Proceedings of SPIE, Vol. 3496 (pp. 101113).
Tramutoli, V., Cuomo, V., Filizzola, C., Pergola, N., & Pietrapertosa, C. (2005).Assessing the potential of thermal infrared satellite surveys for monitoringseismically active areas. The case of Kocaeli (Ýzmit) earthquake, August17th, 1999. Remote Sensing of Environment, 96, 409−426.
Tramutoli, V., Cuomo, V., Lanorte, V., Pergola, N., Pietrapertosa, C.,Ricciardelli, E., et al. (2000). Self-adaptive algorithms for environmentalmonitoring by SEVIRI and GERB: A preliminary study. Proceedings of TheEUMETSAT Meteorological Satellite data User's Conference, Bologna,Italy, 29 May–2 June (pp. 79−87). Meteorological Organisation forExploitation of Meteorological Satellites.
Tupper, A., Carn, S., Davey, J., Kamada, Y., Potts, R., Prata, F., et al. (2004). Anevaluation of volcanic cloud detection techniques during recent significanteruption in the western ‘Ring of Fire’. Remote Sensing of Environment, 91,27−46.
Wen, S., & Rose, W. I. (1994). Retrieval of sizes and total masses of particles involcanic clouds using AVHRR bands 4 and 5. Journal of GeophysicalResearch, 99, 5421−5431.
