Radiative Transfer Modelling for the characterisation of natural burnt surfaces

AO/1-5526/07/NL/HE

Recommendations

P. LEWIS1, T. QUAIFE5, J. GOMEZ-DANS1,2, M. DISNEY1, M. WOOSTER2 , D. ROY3, B. PINTY4

1. NCEO/DEPT. GEOGRAPHY, UNIVERSITY COLLEGE LONDON, GOWER ST., LONDON WC1E 6BT, UK 2. NCEO/DEPT. GEOGRAPHY, KING'S COLLEGE LONDON, STRAND, LONDON WC2R 2LS, UK 3. GEOGRAPHIC INFORMATION SCIENCE CENTER OF EXCELLENCE, SOUTH DAKOTA STATE

UNIVERSITY, WECOTA HALL, BOX 506B, BROOKINGS, SD 57007-3510, USA 4. INSTITUTE FOR ENVIRONMENT AND SUSTAINABILITY (IES), EC JOINT RESEARCH CENTRE, VIA E.

FERMI 1, TP 440, 21020 ISPRA (VA), ITALY

5. NCEO/DEPT. GEOGRAPHY, EXETER UNIVERSITY,

Overview

EO technology overview (talk 1)– Wildfire detection and quantification– Brief summary of relevant results– ESA and related missions

Modelling fire impacts (talk 2)– Semi-analytical– 3D– Thermal– Linear modelling

EO technology overview: wildfire

Technologies:– Surface:

Optical (main focus here) Thermal (secondary focus) Microwave

– Atmosphere Not considered here

All technologies rely on spatial and temporal localisation of fire

Thermal: detection

Detect anomalous high T – Polar orbiting

Don’t view all fires– Orbital convergence

– Some methods night only (lower fire activity) Some methods rely on T saturation

– Geostationary Lower spatial, higher temporal resolution Low resolution at high latitudes

All impacted by cloud

Optical: detection

Sometimes feasible from single image: classification Mostly use time series Mostly use SWIR (also NIR)

– NBR/NDVI Compounded by BRDF effects

– Mostly considered noise, but can be treated Worse cloud/smoke problems that thermal

– Esp. if shorter wavelengths used Moderate resolution: global

– Polar orbiting (mainly), also geostationary Higher resolution:

– Low revisit used for specific study areas Or use longer time between

Active microwave: detection

Essentially classification mostly– Time series, generally

Issue of attribution of signal change to fire Complexity from moisture variations

– Not such an issue if materials dry? Complexity from dry materials Huge advantage in areas of high cloud cover

– Tropics

Fire impact

optical fire severity measures fire radiative energy

– integral over time of fire radiative power ‘direct’ measurements of pre-post biomass

– active microwave– lidar measurements– vegetation indices

area affected by fire from spectral unmixing atmospheric measurement of gasses and particles

released by fire

Fire detection and impact

Best strategy, combine information– Multiple moderate resolution optical– Constellations of higher resolution– Combined optical and thermal– Combine all sources

Need appropriate theoretical background, models, and algorithms

Contributions of this study

Model to estimate fcc– bottom-up approach to C release estimate

Needs fuel load

– Compare with FRE Fcc model generic to all optical sensors

– Should be able to combine information– Burn signal results suggest use as constraint– Measure is linear

Simple spatial scaling

The relevance of the algorithm to the exploitation of data from ESA and related sensors and missions

ENVISAT Earth Explorers Sentinels Meteorological missions Others

ENVISAT: MERIS

300m, VIS/NIR, many channels Issues:

– No SWIR sampling– Geolocation (?)– BRDF effects not too great

Main route to exploitation:– Detect fire from other sensors– Apply fcc algorithm

Noting issues wrt burn materials/dead vegetation confusion

– Would be aided by easy to use gridded surface reflectance product

ENVISAT: AATSR

Thermal– Detection (night time saturation or near saturation)

(WFA)– Can’t use for FRE

Optical– Relevant wavelengths– Only 4 wavebands

Issues with 3 parameter model Unless use burn signal constraint

– Dual view possibly interesting for fcc directional effects Same argument for MISR, CHRIS-PROBA

Earth Explorers

Earthcare: atmsophere (out of scope) – MSI worth considering?

Wind from ADM-Aeolus relevant (out of scope) SMOS soil moisture relevant (fire risk) (out of

scope)

PREMIER: atmosphere (out of scope) BIOMASS

BIOMASS

P band SAR Real potential for pre-post fire woody biomass

estimates– Would need to demonstrate acceptable

precision in change signal Unlikely viable for dry grass fires

– But distinguishing these of interest

Sentinels

Sentinel 1– C band SAR– Arguments above, re detection– but saturation at higher biomass

So biomass change issues if pre-fire biomass high

Sentinels 4,5– Atmosphere (out of scope)

Sentinel 2

13 bands across SW Varying spatial resolution 60m+ Satellite pair

– Increased viewing opportunity– 5 day (cloud free)

‘extended viewing capability’– BRDF issues would need treatment

Fcc product at 60m, or implement multi-scale for further localisation– Need detection algorithms– Very intersting platform to develop fcc-based detection

algorithm

Sentinel 3

OLCI and SLSTR similar to MERIS/AATSR SLSTR: dedicated low-dynamic range ‘fire’

channels – So FRP & day/night detections

Optical, similar to MERIS/AATSR uses and issues– BUT very interesting combination of

platforms (Sentinel-2,-3) for multi-scale strategy (possibly also then Sentinel-1/BIOMASS)

Meteo

MetOp– AVHRR instrument

Optical (includes SWIR)– BRDF effects (can be treated: MODIS algorithm prototyped with

AVHRR)– direct value if constrained burn signal used for fcc– Indirect, of value to moderate resolution constellation approach

thermal (fire saturation)

MSG/MTG– Operational FRE– Optical burned area/fcc difficult

Low signal/noise for any by largest fires BUT tracking fcc post fire would be of interest

– Need constraint to burn signal, but probably can get from active fire detections

Others

Medium resolution (sub 1m to 10s m)– VIS/NIR– SPOT Pléiades, DMC, SEOSAT-INGENIO, RapidEye, EROS

Revisit period for individual may not be high enough, but have constellation missions

Also CEOS LSI-concept – all as virtual constellation Probably most useful for localisation of information if fire known (e.g.

thermal)– Fcc could be applied if constrained burn signal– And atmospheric correction

EnMAP– Very relevant for characterisation of fcc– BRDF effects for off-nadir pointing– Don’t need full hyperspectral for this though

So maybe target more subtle information

Discussion

3 parameter fcc model– Best used with >> 3 bands– Or strong constraint to burn signal

2 single most interesting sensors– Sentinel-2 / EnMAP– Both aim at high repeat coverage high

resolution

Discussion

A major limitation to monitoring / characterisation is often viewing opportunity– So should develop multi-sensor concepts– Including multi-resolution– Issues:

different spectral sampling different spatial resolutions potentially different viewing and illumination angles

– Fcc approach can at least partially deal with all of these (BRDF modelling for latter)

– Longer term, may consider full DA system (e.g. ESA EOLDAS) But technology needs development

Key Recommendations

Fcc should be developed into operational algorithms to quantify fire impact

Method should be generic– needs testing ESA sensors and Sentinel prototypes

Investigate constraint for application to sensors with not >>3 wavebands

Issues in study in comparison of fcc-FRE– Need further investigation

Most application here to S. Africa– Wider application / testing

Develop method for multiple data streams– Including multiple resolutions