High-temporal resolution thermal volcano monitoring from space: a review of existing techniques Robert Wright Hawai’i Institute of Geophysics and Planetology

Lecture topics What do we want from a satellite thermal monitoring system? Underlying principles of hot-spot detection Some existing approaches to hot-spot detection Examples

Some requirements for a space-based thermal volcano monitoring system Be able to detect high-temperature bodies at the decimeter scale Depend on cost-free data Make repetitive, frequent observations (eruption intensity fluctuates on < hourly time scales) Minimise false positives Minimise transfer of actual image data Objective Communicate results ‘rapidly’ Any others you can think of……..?

Physical principles L l = c 1 l -5 exp(c 2 / l T)-1 L 4 ~T 4 L 12 ~ T 2 As the temperature of the emitting surface increases, the amount of radiance at all wavelengths increases and the wavelength of maximum emission shifts to shorter wavelengths Short-wave infrared radiance data are great for detecting and quantifying hot targets

300 K 4 m m, L l = 0.4 Wm -2 sr -1 m m 11 m m, L l = 9.5 Wm -2 sr -1 m m K (99.95%) 850 K 4 m m, L l = 1.3 Wm -2 sr -1 m m 11 m m, L l = 9.6 Wm -2 sr -1 m m -1 Sub-pixel-sized hot-spots High-temperature radiators are apparent at short wavelengths even if they are much smaller than the spatial resolution of the imaging system, which they often are…. 4  m data are very important: work-horse of low resolution thermal monitoring systems Image: Clive Oppeheiner 34 m

Sub-pixel-sized hot-spots High-temperature surfaces are easily distinguishable from surfaces at ambient temperatures when imaged at short and long wavelengths

ATSR – 1 km pixels Landsat TM – 30 m pixels ‘High’ versus ‘low’ spatial resolution data Many space-based resources available that acquire data in the important 4 and 12  m bandpasses High spatial resolution data can detect smaller, less intense thermal anomalies, but….. Their low temporal resolution, low duty cycle, data volume make them (largely) impractical as volcano monitoring tools (but possible OK as a volcano “surveying tool”; see work of Rick Wessels) Low spatial, high temporal resolution environmental/meteorological satellites are the best bet

Temporal resolution Temporal resolution very important for monitoring Data frequency varies depending on whether the satellite is in geostationary or low-Earth orbit GOES: geostationary: 7-30 minute repeat AVHRR/MODIS/AVHRR: LEO: hour repeat Frequency at which data are acquired can be improved by launching more satellites In the future……….highly elliptical orbits?

Sensors for hot-spot monitoring AVHRR:4 and 12 m m channels (1 km pixels) Temporal resolution = 6 hours, global coverage GOES:4 and 12 m m channels (4 km pixels) High temporal resolution = 7-30 mins, limited coverage, no coverage at high latitudes ATSR:1.6, 4 and 12 m m channels, 1 km pixels Temporal resolution = 3 days, global coverage MODIS:4 and 12 m m channels, 1 km pixels Temporal resolution = 24 hours, global coverage

Approaches for automatic detection of volcanic thermal unrest in low spatial resolution satellite data

MODIS band 22 (3.959  m) Brute force Acquire, enhance and manually inspect the images

‘Brute-force’ Not very practical for global/regional/small scale monitoring at high temporal resolution Humans introduce bias and are not to be trusted Need ‘non-interactive’ methods for identifying hot-spots

Simple thresholding of the 4  m radiance signal Pixels with a 4  m radiance > pre-determined threshold are classified as hot-spots Totally insensitive to variations in ambient background temperature (season, geography…) We need methods that account for variation of non-volcanic sources of scene radiance

The Spectral Comparison Method Calculate D T for each pixel Automatically accounts for variance in ambient background Flag pixel as a ‘hot-spot’ pixel if D T > chosen threshold Detects sub-pixel temperature ‘contrasts’, BUT…. Needs to include more checks to avoid returning ‘false positives’ caused by cloud edges, non-uniform surface emissivity, atmospheric transmissivity….) BT  = C2C2 ln[1+ C 1 /( 5L l )]

Contextual algorithms Combine spectral AND spatial analysis Each pixel in image treated as a “potential” hot-spot and its multi-spectral characteristics compared against adjacent non-hot-spot pixels. Thresholds are  less empirical and more scene dependent A potential hot-spot is reclassified as an actual hot-spot if: T 4 > T 4b + n  T 4 b AND  T >  T b + n   T b Detection does not rely on radiance threshold but does rely on  threshold Neighbourhood operation – computationally intensive

Dealing with daytime data ‘Cold’ but ‘reflective’ surfaces can generate false positives (e.g. snow, sand) Use the “mean” approach or the “per-pixel” approach ‘Raw’ 4  m daytime data Corrected 4  m daytime data The Earth emits AND reflects at 4  m Need to isolate the portion of the signal thermally emitted by the target Spectral/contextual algorithms account for extra emitted energy What about the reflected energy? L 4corr = L 4 – × L 1.6

Dealing with daytime data ‘Sun-glint’ – specular reflection anomaly that can produces ‘false positives’ Identify ‘potential’ sun-glint pixels on the basis of sun-sensor geometry and exclude them  g < n º, where cos  g = cos  v cos  s  sin  v sin  s cos 

Nighttime short, short-wave infrared data Wooster and Rothery (1997a,b); Wooster et al., (1997) Night-time 1.6  m data acquired by the Along-Track Scanning Radiometer Only detects material at magmatic temperatures: makes thresholding very simple Hopeless during the day due to contamination by reflected sunlight Wooster and Rothery, 1997

A multi-temporal approach Pergola et al. (2004) use a multi-temporal approach at Etna and Stromboli  T 4 (x,y,t) = T 4 (x,y,t) – T 4ref (x,y)  T 4 (x,y) ‘Stack’ co-registered images of an area of interest Characterise the thermal ‘behaviour’ of each pixel over an extended period of time Hot-spots identified when a pixel begins to behave (thermally) ‘differently’ than it has in the past Great potential for detecting low temperature events

Some case studies describing applications of the data

What kind of activity can we detect? Ability to detect the thermal emission associated with volcanic activity depends on: The temperature of the lava/process The area it covers Its longevity Basaltic lava flows Basaltic lava lakes Lava domes Strombolian activity Phreatic activity Phreatomagmatic activity Fumarolic activity Easier Harder Block lava flows

Cycles of dome growth at Popocatepetl Dome growth resumed at Popocatepetl in 1996 Satellite remote sensing only method useful for routine observations of the crater interior GOES images summit crater once every 15 minutes High temporal but low spatial resolution: what can we learn?

Dome growth at Popo 10 × 10 kernal centred at Popo’s summit Record the peak radiance from the group (P r ) and the mean of the remainder (B r ) In the absence of any time-independent forcing mechanism, P r and B r should be well correlated Wright et al., 2002

Dome growth at Popo However, a volcanic radiance source, radiance from which is time-independent will cause P r and B r to de-couple Use adjacent inactive volcano to normalise for environmental effects Easy to identify volcanic activity in GOES radiance time-series Wright et al., 2002

Elevated GOES radiance coincides temporally with periods of heightened explosivity of the dome Periods of heightened explosivity follow substantial decreases in SO 2 flux Restricted degassing = overpressure = explosions Dome growth at Popo Wright et al., 2002

Cyclic activity described in terms of generation of overpressure within the dome due to degassing induced decreases in permeability and compaction Satellite measurements of radiance corroborate physical model Dome growth at Lascar Wooster and Rothery, 1997

Estimating lava eruption rates Very easy to detect lava flows Spectral radiance from the flow surface is related to the area of lava at a given temperature within the field of view In other words….the higher the eruption rate the greater area lava will be able to spread before it cools to a given temperature, and the higher the corresponding at-satellite radiance will be Pieri and Baloga (1986) Harris et al. (1997) Wright et al. (2001) Harris et al, 1997

Conclusions Principles of satellite detection of volcanic hot-spots are well established and much work continues to be done in both the volcanological and wildfire communities Many different “flavours” of hot-spot detection algorithms Trade-off between detecting low intensity anomalies and false positives An approach tailored to your volcano of interest is probably the best solution