Hyperspectral characterization of wildfire. Dr. Stefania Amici Istituto Nazionale di Geofisica e Vulcanologia The University of Manchester

Size: px

Start display at page:

Download "Hyperspectral characterization of wildfire. Dr. Stefania Amici Istituto Nazionale di Geofisica e Vulcanologia The University of Manchester"

Dulcie Edwards
5 years ago
Views:

1 Hyperspectral characterization of wildfire Dr. Stefania Amici Istituto Nazionale di Geofisica e Vulcanologia The University of Manchester

2 Introduction Introduction Importance of wildfires: Global scale Local scale Methodology for detection and burn severity What remote Sensing can do for open fire Spectral features of fires: Potassium emission CO 2 absorption Hyperspectral data for burn scar Remarks Acknowledgements

Mancheter Areas of interest: Remote Sensing Imaging

3 Intro: Dr. Stefania CV Amici B.Sc, Astrophysics (1997) University La Sapienza Rome. PhD Cal/val, University of Parma (2010) Researcher 70% Educational 30% Visiting researcher at 1 The University of Mancheter Areas of interest: Remote Sensing Imaging spectroscopy UAV-drone New ideas generation and project drafting dissemination stefania.amici@ingv.it

4 Remote sensing infrastructure at INGV

5 REAL-TIME ACQUISITION SYSTEMS GEO AND POLAR MULTIMISSION GROUND RECEIVING STATIONS in X/L BANDS 2017 AIST Tokyo

6 INGV MULTIMISSION RECEIVING STATIONS KONGSBERG MULTIMISSION SYSTEM L SEASPACE antenna L band receiving NOAA-AVHRR 2017 AIST Tokyo RED AVHRR COVERAGE YELLOW MODIS/METOP COVERAGE

7 INSTRUMENT CAL/VAL Strategy Field work Satellite acquisition New Instrument test Spectroradiomentry Surface temparature Roughness analysis Gases identification corner reflector ASTER LANDSAT SENTINEL 2 HYPERION AVHRR/MODIS Miniature Mass Spectrometer Drone 2017 AIST Tokyo

Permette la registrazione video radiometrica a 32 Hz alla risoluzione VGA di 640x480

8 In field instruments Laboratorio ottico e spettrometria: strumentazione Camera Termica Termocamera Optris PI640. Permette la registrazione video radiometrica a 32 Hz alla risoluzione VGA di 640x480 pixel, sensibilità termica (NETD) di 75 mk, campo spettrale da 7,5 a 13 micron e campi di misura da - 20 a 900 C. Peso: 320 grammi AIST Tokyo

Significance: Global and Local scale Economic Source http://www.earthzine.org/wp-content/uploads/2013/08/figure12.png Fires greatly affect Earth s atmospheric composition.

9 Significance: Global and Local scale Economic Source Fires greatly affect Earth s atmospheric composition. Fires affect the global climate through processes such as trace gas and aerosol production, and influence the terrestrial carbon dynamics. Giglio et al. (2010) indicates that ~3.4 % of the Earth s terrestrially vegetated area burns annually, with resulting large scale effects.

10 Significance: Global and Local scale Economic loss of infrastructure, natural and cultural resources, insurance Fire suppression high costs Scientific Transport gas to atmosphere climate change Social Loss of human lives, communities, Health (smoke, air quality ) Need for research Pre-fire Measurements that can be correlated to fire behaviour Active fire Localization, flame/ smouldering, evaluation of Parameters for modelling, Linking Energy to Emissions and Air Quality Post- fire Vegetation Mortality, Ecosystem Recovery, Land use change

fighters Evacuation orders are in place for

com/bigpicture/2009/09/wildfires_in_southern

11 Relevance of forest fire at local scale Station Fire (26 Aug Sept.2009) burned over 650 km 2, nearly 100 structures destroyed Loss of lives: 2 fire fighters Evacuation orders are in place for thousands in communities around the city, and resident Aug. 31, (AP Photo/Jon Vidar), _californ.html Proression map of Station Fire (Los Angeles County) by United State Forest Service

Nearly 50,000 Indonesians suffered respiratory diseases due to the haze. large amount of pollutants discharged to the atmosphere.

12 Fire Emissions and Air Quality Modis June 2013 Indonesia forest fires 400Km The major cause of the fires is clearing land for agriculture 51% on land managed by pulpwood, palm oil, and logging companies Haze crisis of June Nearly 50,000 Indonesians suffered respiratory diseases due to the haze. large amount of pollutants discharged to the atmosphere. high volumes of carbon released, contributing to climate change. e-emergency

13 EUROPE

and Informs the EC services and the European Parliament on forest fires in Europe

14 Relevance at local scale: Europe The European Forest Fire Information System (EFFIS) supports the services in charge of the protection of forests fires in the EU and neighbour countries, and Informs the EC services and the European Parliament on forest fires in Europe 2000->Operational >new products including maps of fire danger anomalies and maps of the fire danger index

15 Fire suppression in Italy 2016 fire season in Italy Total number of wildfire 4793 Total burned area: HA Canadair CL DPagina/319

Climate change and fire regime in north Europe Northern Europe countries experience an increasing number of vegetation fires when more frequent exceptional drought season occur.

It burned more than 150 square kilometers (60 square miles), killed one person, and forced thousands to evacuate their homes.

16 Climate change and fire regime in north Europe Northern Europe countries experience an increasing number of vegetation fires when more frequent exceptional drought season occur. Västmanland 31July -8 Aug In a week, the largest fire Sweden has seen in four decades. It burned more than 150 square kilometers (60 square miles), killed one person, and forced thousands to evacuate their homes. Hot and dry weather exacerbated the fire, stressing vegetation and priming it to burn. With a high-pressure system parked over Scandinavia, oppressive heat emerged across Sweden in July and August, bringing record or near-record temperatures to many towns and cities. In Sala, a city close to the fire, temperatures soared to 34.7 C on August Average temperatures are about (21 C) in early August. French and Italian airplane for suppression Source:

Climate change and fire regime in the UK The likelihood of

with projected warmer, drier spring and summer conditions

17 Climate change and fire regime in the UK The likelihood of wildfires occurring may increase between 10% and 50% by the 2080s with projected warmer, drier spring and summer conditions Knowledge for Wildfire KfWf ( MODIS data on 18 April False color reflectance at a resolution of 250m/pixel shows vegetation dominated area (in red) and the fire plume (white-greenish)

Relevance of open fire in UK Peat Moorlands 1) Wildlife wildfire affects unique habitats in Sites of Special Scientific Interest with impact on birdlife (e.g. Golden Plover and Dunlin).

18 Relevance of open fire in UK Peat Moorlands 1) Wildlife wildfire affects unique habitats in Sites of Special Scientific Interest with impact on birdlife (e.g. Golden Plover and Dunlin). 2)The equilibrium of the ecosystem is affected by wildfires removing vegetation (e.g. Sphagnum moss) which is an important species for carbon sequestration.. 3)More frequent severe open fires would lead to significant loss of biodiversity and ecosystem services such as carbon storage for ecosystems like peatland and heathland, which are particularly sensitive to fire. Contribute to Global Warming Organic Concentration of water in nearby streams & reservoirs causing discoloration to drinking water. Source

19 What Can Remote Sensing do for for open Fire? Fire Detection Active fire behaviour Post fire characterization Burn scar delineation Vegetation mortality Ecosystem recovery.

20 RS Detection methods Vegetation fires involve high temperatures, so thermal remote sensing is suitable to its identification and study. Actively burning fires emit IR so strongly, especially at MIR (3 5 μm) wavelengths that can be identify by Earth orbit Fixed threshold approach algorithms which provide hotspot counts and fire location maps (e.g. MODIS products Justice et al., 2002, Giglio et al., 2003, Dennisson et al., 2006.) Top-of-atmosphere spectral radiance simulated at four different target) using the MODTRAN 5 radiative transfer code. Simulations for a savannah surface at 300 K; the same surface but with a 1,000 K fire covering 0.5 % of the ground field-of-view (FOV), specularly reflected sunglint from a 300 K surface; and solar-heated (320 K) bare soil. The pixel containing the sub-pixel fire shows a signal highly elevated in the MIR (3 5 μm) spectral region compared to all other targets, equivalent to a brightness temperature of around 400 K ( Wooster et al )

Can Hyperspectral spectroscopy (VINIR (0.3-2.5μm ) contribute to active wildfire fire characterization? How?

21 Can Hyperspectral spectroscopy (VINIR ( μm ) contribute to active wildfire fire characterization? How? Hyperspectral remote sensing data typically cover the visible (VIS, nm), near infrared (NIR, nm) and shortwave infrared (SWIR, nm) regions of the spectrum. In addition to capturing emitted radiance, hyperspectral data possess narrow bands that may be appropriate for creating effective fire detection indices Lab. scale experiment (Y) Airborne experiments (Y) UAVs? Satellite hyperspectral sensors: EO1- Hyperion (Y) New missions such as Enmap, Prisma, HSUI?

22 Spectral features for active fire detection: K emission and CO 2 absorption

23 Potassium emission Fuel biomass is largely composed of carbon (~45%), hydrogen (5.5%), oxygen (41%), and nitrogen (3.5%), and the molecular combustion products are dominantly CO 2, H 2 O, CO, CH 4, and various nitrogenous compounds (Levine, 1991) In addition trace elements: K: up to 7% Na: 0.1% P: up to 1% Ca: up to 5% When ionized alkalis can make transitions resulting in very strong emission lines.

24 How the potassium line technique works When vegetation burns at high temperatures associated with flaming combustion, trace elements like K are mobilised. This produces a sudden increase in reflectance at 766.5nm and nm, which very narrow band (hyperspectral) sensors detect as a sharp emission peak or line. NIR Oxygen absorption lines K Lines The strength of the peak decreases as temperature falls. Advantages: being specific to flaming combustion a K emission approach theoretically allows for the separation of smouldering from flaming areas of vegetation and active fire detection. Potassium emission line simulation Transmission was calculated as viewing the Earth from100 km elevation at nadir and assuming a US 1976 Standard Atmosphere and a 23 km rural aerosol. Simulation was conducted at 0.1 cm 1 wavenumber resolution using the high spectral resolution mode of MODTRAN 5.2 (Berk et al., 2008)

25 Laboratory scale Sensor Sptial resolut ion Centra l band Hyperion 30m 772nm How different sensors see the K emission? HyspIRI 60m 770nm PRISMA 30m 770nm Spectrometer PRISMA HyspIRI EO1-Hyperion Flaming Mixed Mixed Smouldering Data courtesy Prof. M. Wooster

26 Mediterranean Land cover AKBD metric applied to airborne data to simulate a range of spatial resolution. - Amici et al. 2011

27 Potassium Emission -Space True colour composite EO-1 Satellite + A +C + B SWIR colour composite A & B within fire area C outside of fire area 2007 Escondito Wildfire (lat N, lon W) Amici et al. 2011

Heather dominated vegetation ( land cover) Can flaming and smouldering combustion be distinguished in heather? The k-line technique works in Italy and California.

28 Heather dominated vegetation ( land cover) Can flaming and smouldering combustion be distinguished in heather? The k-line technique works in Italy and California. It was tested during experimental burns in heather moorland in Northumberland during an airborne campaign funded by NERC-ARSF funded on March Figure: hyperspectral imageries by Eagle airborne sensor (400nm-1000nm) are used for K study (data courtesy Prof. M. Wooster). Open flame (1.A, visible bands composite colour) results in a very strong K emission peak and weak Na and P emissions. Flame location 2A is compared against LEIKA super resolution camera. Apparently smouldering phases (2.A and 3A) results in a weak but distinctive K peak (2.B and 3B) which results in distinct signal in 2C and 3C and recognised as flame in 2D and 3D data processing and analysis Dr. Stefania Amici

29 AKBD Spatial resampling Land plot size 80mx40m data courtesy Prof. Martin Wooster data process S. Amici

30 CDAI (Carbon Dioxide Absorption Index) Dennison 2009 Saturation effect Detection map by threshold 1m/px e 30mpx. CDAI detection 1m/px 30m/px Poor SNR effect data courtesy Professor Martin Wooster data processing and CDAI adaptation to PRISMA and analysis Dr. Stefania Amici.

CDAI map by using AVIRIS Wallow Fire -Giugno 2011 AVIRIS 14.7m /px nel range spettrale 365.9nm -2496.

31 CDAI map by using AVIRIS Wallow Fire -Giugno 2011 AVIRIS 14.7m /px nel range spettrale 365.9nm nm a) visible bands RGB combination for Wallow fire b) RGB faulse colors c) CDAI map Data courtesy Dr. S. Veraverbeke data processing Dr. S. Amici

32 Post fire effect

33 Post fire: Burn scar From unburned to burned Surfaces: Decrease of reflectance in Visible NIR and increase in SWIR. Several remote sensing Methods Uses this changes. Normalized Difference Vegetation Index Normalized Burn Ratio differenced Normalized Burn Ratio Available satellite data: Multispectral, Landsat, ASTER, hyperspectral EO1, spatial res. 30m/px temporal res. 20, Modis ( 2 data per day, but 250m/) Data quality depends on the weather condition. SAR is a solution. However, validation is needed.

34 Peak District National Park (PDNP) case study - PDNP is located between Manchester and Sheffield - The PDNP is one of the most degraded moorland landscapes in the UK. - The Dark Peak is dominated by heather moorland and brown peats with an underlying geology of gritstones and shales. Map credits Google maps

Edale (25 May 2008) - The Edale wildfire

with areas of back burn especially over the

- The CEH LCM2007 classifies the area as mainly

35 Edale (25 May 2008) - The Edale wildfire occurred on 26/05/08 at Grindsbrook Clough and is a small burn scar at 0.10 km 2. It is reported that the fire burnt for 3 days with areas of back burn especially over the southern plateau edge. - The CEH LCM2007 classifies the area as mainly Bog with some Heather Grassland to the southern tip of the burn scar. Landsat 7 ETM Pre- 22 May 2008 Pre- 22 May 2008 Post- 27 Sept May September 2008 Landsat 7 ETM processing S. Amici Crown Copyright Ordnance Survey. An EDINA Digimap/Jisc supplied service. Photo courtesy J. McMorrow

36 Optical Data Validation ESA 2008 USGS/EROS 2008 SAR data processing and analysis implemented by Dr.Gail Millin Chalabi Landsat 7 data processing and analysis implemented by Dr. Stefania Amici

37 Hyperspectral data for Edale validation Amici, S et al. Living Planet 2016

38 REMARKS Imaging spectroscopy offer alternative method for fire detection in peatland when based on high spectral resolution NIR measures. K-emission signal from fires is unique, works day/night. May provide info. related to fire emission source strength. Spaceborne measurement is possible, but good performance requires higher spectral resolution, Band positioning and SNR than existin gones New generation of Hyperspectral sensors such as HyspIRI, PRISMA, EnMap might be able to contribute to fire detection and flame location. CO2 indexes can be of interest if concentration could be retrieved Post fire analysis and burn severity SAR products validation Low weight hyperspectral sensor have been developed and in the next future can be used on Unmanned Aerial Vehicle.

39 Acknowledgements The study visit and associated project are part of a study funded by Italian Space Agency as part of a work package entitled Spectral characterization of open fire, ASI-AGI ( ). Professor John Dold (University of Manchester and FireLab Ltd.) for organising the experimental burns at Debdon, Northumberland, March Professor Martin Wooster, Kings College London, Debdon airborne data. Prof. Mark Danson & Julia McMorrow Airborne Research Survey Facility 01/07/08 aerial photography, Edale data. Gail Millin Chalabi, SAR data processing and analysis Airborne Research and Survey Facility, Natural Environment Research Council (NERC) for SPECIM Eagle and Hawk images and aerial photographs. Landsat 7 ETM+ from EarthExplorer ERS-2, ASAR & ALOS PALSAR data as part of Category, 1 Project 2999 PI Dr. Kamie Kitmitto The AIST for invitation to contribute at the workshop

40 Thanks for your attention

NEAR REAL TIME DETECTION OF BURNED SCAR AREA USING LANDSAT-8 IMAGERIES

NEAR REAL TIME DETECTION OF BURNED SCAR AREA USING LANDSAT-8 IMAGERIES Danang Surya Candra Kustiyo - Indonesian National Institute of Aeronautics and Space (LAPAN) - Outline INTRODUCTION: Background, Problems,