Evaluation of geostationary EUMETSAT versus polar orbiting NASANOAA
Evaluation of geostationary (EUMETSAT) versus polar orbiting (NASA-NOAA) missions for detection/monitoring of thermal anomalies over southeastern Europe Julia Stoyanova 1, Christo Georgiev 1 Contribution: Wilfrid Schroeder 2, Erdem Erdi 3 1 National Institute of Meteorology and Hydrology, Bulgaria, 2 University of Meryland, USA, 3 TSMS, Turkey 5 th SALGEE Workshop, 18 - 20 September, Yerevan, Armenia ‘MSG Land Surface Applications: Heat waves, Drought Hazard and Fire Monitoring’
Observational studies have identified that the frequency of hot summer days and heat waves over Europe has increased during the past decades. Outline 1. Introduction 2. Fire products summary 3. Methodology and Data set for evaluation of fire algorithms behavior 4. Comparative analyses & Results 5. Concluding remarks
Introduction Landscape fires are frequent across much of Mediterranean Europe, and satellites are vital to assessing their terrestrial and atmospheric. Their dynamics at very high temporal resolutions can be monitored by using geostationary satellites. • EUMETSAT FIR products • Meteosat-10, Full scan mission, 0 , each 15 min • Meteosat-8, Full scan mission, 41. 5 , each 15 min • Meteosat-9, Rapid scan mission, 9. 5 , each 5 min - Europe • Fire Radiative Power (FRP) In addition to simple detection, give a potential to estimate a fire’s sub-pixel effective temperature based on biomass burning effects. • For SE Europe, operational satellite information distributed through EUMETCast is from Meteosat on geostationary orbit and from polar orbiters of NOAA-NASA. • Since February 2017 fire thermal anomalies product based on MODIS sensor is stopped, that is connected with replacement of Collection 5 (up to now available) with the new version MODIS Collection 6 (that requires great resources). • Since August 2017 in operational flow is distributed S-VIIRS 375 m (in NETcdf format).
Introduction This study follows and extend our previous work (Georgiev & Stoyanova, 2013) on application, validation and evaluation of fire detection algorithms for real fire situations with potential to be operationally used over southeastern Europe (eastern Mediterranean). This research is aimed to evaluate the behavior of Land Surface Analysis (LSA) Satellite Application Facilities (SAF) FRP-Pixel product two versions (operationally available and updated algorithm) against: • active fire data collected by the polar orbiting satellites (Moderate Resolution Imaging Spectoradiomer (MODIS) and Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (SNPP/VIIRS) Active Fire detection product (S-VIIRS ) and • actual forest fires. Comparison to those for the alternative active fire product derived from SEVIRI (MPEF FIR Fss and FIR Rss) is performed.
Fire Product Summary Geostationary active fire detection using MSG SEVIRI satellite data • Two Active Fire Monitoring (FIR) products (Joro et al. , 2008; EUM, 2007), operationally generated at Meteosat Products Extraction Facilities (MPEF) of EUMETSAT derived by using data from the two MSG missions: • MPER FIR Full Earth disc scanning generated data every 15 min, image rectification to subsatellite point location 0. 0° longitude. • MPER FIR Rapid scan, FIR Rss generates data at 5 -minute intervals, scan region from approximately 15° to 70° latitude, subsatellite point location longitude 9. 5°E. • MPER FIR Full Earth disc scanning generated data every 15 min, image rectification to subsatellite point location 41. 5° longitude. • The FRP-PIXEL product provides detailed information of pixels in which active fires have been detected (e. g. including the spatial location, thermal properties, atmospherically corrected FRP and uncertainty of pixels containing actively burning fires every 15 min (Roberts and Wooster (2008); Wooster et al. (2015)).
Product Summary Polar orbiting active fire detection using Aqua/Terra and Suomi NPP satellite data • Moderate Resolution Imaging Spectroradiometer (MODIS) instruments on-board the Terra and Aqua satellites (relatively high spatial resolution of active fire observations (1 km at nadir), coupled with its better than daily availability from two platforms, ensure that the MODIS active fire product (e. g. Giglio et al. , 2003) is the standard against which geostationary active fire products are compared when performing product evaluations (e. g. Xu et al. , 2010; Schroeder et al. , 2014): • Collection 5 MODIS active fire detections (MOD 14 from Terra and MYD 14 from Aqua), (2012) • Collection 6 MOD 14 and MYD 14 (2016). • Visible Infrared Imaging Radiometer Suite (VIIRS) onboard Suomi-NPP (Suomi National Polar-orbiting Partnership; Csiszar et al. , 2014) • S-NPP VIIRS 375 m (2016) • S-NPP VIIRS 750 m (2016)
New Active Fire Products of NASA /update/replace now existing MODIS products with MODIS collection 6 & S-VIIRS/ https: //earthdata. nasa. gov/earth-observation-data/near-real-time/firms/viirs-i-band-active-fire-data Example: Fire spread, March 26 -31, 2013, Jul DOY 85 -90, Taim Ecological Reserve, southern Brazil 1 km Aqua/MODIS 750 m VIIRS 375 m VIIRS Near Real-Time (NRT) Suomi National Polar-orbiting Partnership Visible Infrared Imaging Radiometer Suite (SNPP/VIIRS) Active Fire detection product NEW
Methodology and Data set Evaluation Period The study is performed at two stages: 1. Evaluation of current versions of Fire Products Algorithms using Data Set covering information from 2016 for region of southeastern Europe (Bulgaria). 2. To evaluate the capabilities of updated algorithm of LSA SAF FRP (2016) comparing LSA SAF FRP current versions algorithm, reprocessed for 2012, with the operational Fire Products Algorithms for 2012. Geostationary satellite data from Meteosat SEVIRI (including the images, MPEF cloud mask (CLM) and fire products): • MPER FIR Full scan and Rapid scan, available through EUMETCast (2016, 2012) in NRT. • Meteosat FRP-Pixel operational product, through EUMETCast (2012 and 2016). • Meteosat reprocessed R-FRP-Pixel product (2012), available through LSA SAF archive. Polar orbiting satellite data • MODIS instruments , Terra and Aqua satellites, Collection 5 (2012) through EUMETCast • MODIS instruments , Terra and Aqua satellites Collection 6 (2016) through NASA archive • Suomi NPP VIIRS, 375 m and 750 m resolution, through NASA archive 2016. Each satellite fire product has different classes of output that are not taken into account in this study. All fire pixels (classifies by confidence levels, as active fires or potential active fires) are considered as Fire Pixels.
Methodology and Data set • Satellite Fire Products Evaluation of FRP-PIXEL products against active fire data collected by other satellite products. 1. To evaluate the consistency of Various Fire Products Algorithms 2. by using observation data of actual fires as a reference data set. • Data for active fire confirmed by ground observations Actual Forest Fires, State Forest Agency of Bulgaria, consisting of the following records: • Coordinates of the closest town/village • Start/End of fire • Total burned area (forested, grass) • Down/Crown affected area
Methodology and Data set Procedures for mapping various kind of data 1. All actual fire pixels detected by algorithms based on polar orbiting platforms are specially remapped to the geostationary projection, given each detected thermal anomaly MSG coordinates (line, column, Full Mission and Rapid Scan Missions). 2. Accordingly, the locations of Actual Forest Fire (reported by the coordinates of nearest village in the National Data Base of State Forest Agency of Bulgaria ) are remapped to SEVIRI’s imaging grid.
Methodology and Data set Reference Data and Matching Procedures 1. The detection capabilities of FRP-PIXEL product (2 versions) against active fire data collected by other satellite products are evaluated: The MODIS (as a reference) as well as active fire pixels / data collected by the two VIIRS algorithms and compare the results to those for two alternative active fire products derived from SEVIRI imagery. As a matched detection, we consider the presence of a FRP fire pixel in the following grid : • Within 3 x 3 pixel window centered on the active fire pixel under investigation. • At 1 h time difference between the SEVIRI detection and corresponding MODIS/VIIRS observations of the same fire. 2. We evaluate the consistency of the active fire products behaviour against actual forest fires as a reference. As matched detections, we consider the presence of fire pixels detected by satellite products in the following grid : • Within 7 7 SEVIRI pixel window centered on the pixel at which actual forest fire is located; • At a time interval from 1 h before the forest fire initiation to the end of the forest fire the corresponding MODIS, VIIRS and SEVIRI fire detections are taken into considerations.
Methodology and Data set Evaluation the SEVIRI FRP products (two versions) • SEVIRI FRP behavior is assessed by using MODIS active fire data from Terra and Aqua satellites. • SEVIRI FRP Omission errors are defined by the absence of a FRP SEVIRI active fire pixel detection in each MSG pixel of at least a MODIS fire detection at a matching time 1 hour in a 3 3 pixel window centered on the MODIS fire detections. • SEVIRI FRP Commission errors are defined by the absence of fire detections by MODIS, S-VIIRS 375 m and S-VIIRS 750 m at a matching time 6 hours within a 3 3 pixel window centered on the FRP SEVIRI Fire Pixel under investigation.
Methodology and Data set Comparison of per-fire FRP derived from SEVIRI and VIIRS-750 m observations 1. All fire pixels of VIIRS- 750 m are matched to the LSA SAF SEVIRI FRP fire pixel in the following grid : 3 x 3 pixel window centered on the SEVIRI FRP under. Algorithms investigation. Evaluation • of. Within the capabilities of all operationally available Firepixel Products • Atof southeastern 6 min time difference between SEVIRI detection and corresponding for the region Europe to detectthe forest fires (over Bulgaria) VIIRS-750 m observations of the same Omission errors are defined by the absence of anyfire. active fire pixel detection by 2. The area based results are derived from comparison themtotal FRP 750 measured by SEVIRI FRP / FIR Full Scan / FIR Rapid Scan, MODIS, S-VIIRSof 375 / VIIRS m in each all pixel VIIRS-750 detected fire pixels in a matching MSG with m coordinates of the village, where the. SEVIRI State detection Forest Agency of Bulgaria reported actual forest fire at 6 h time difference within 7 7 SEVIRI pixel window centered on the pixel of actual forest fire reported.
Evaluation the SEVIRI FRP products Intercomparison of 2016 and 2012 LSA SAF FRP-PIXEL versions FRP-PIXEL Algorithm Version Omission Errors (%) of MODIS detections Commission Errors (%) of MODIS detections 95. 8 95. 6 94 74 68 73 2012 Operational 2012 Reprocessed 2016 Operational Commission Errors (%) of detections by MODIS, S-VIIRS 375 m S-VIIRS 750 m 52 FRP product Evaluation • Better performance of FRP, last operational version as regards to the omission errors, as well as to the sensitivity of fires under the forest canopy. • Better performance of version 2016 Reprocessed than operational 2012 FRP version as regard to omission and commission errors. • Significant decreasing of commission errors when checking by more frequent polar orbiting observations.
Intercomparison and performance of fire products 2016 SEVIRI Algorithms 2016 MPEF FIR Full scan MPEF FIR Rapid scan LSA SAF FRP Full scan Omission Errors (%) of MODIS detections 86. 3 77. 8 94 Commission Errors (%) of MODIS detections 73. 6 78. 9 73. 2 Commission Errors (%) of detections by MODIS, S-VIIRS 375 m S-VIIRS 750 m 53. 9 62. 5 51. 7 • LSA SAF FRP-PIXEL product Active fire detection performance: • Best performance of MPEF FIR Rapid scan (5 min frequency observations) as regards to the omission errors, based on comparison to active fire pixels collected by MODIS. A critical point regarding omission error is the frequency of satellite observations. • Slightly better performance of LSA SAF FRP (15 min frequency observations) as regards to the commission errors, based on comparison to active fire pixels collected by MODIS and VIIRS • Significant decreasing of commission errors when checking by more frequent polar orbiting observations.
Evaluation for consistency of Fire Products Algorithms by using reference observation data for actual fires 2016 SEVIRI Algorithm and Version Omission Errors (%) of Actual Forest Fires (SFA) 2012 FRP Operational 2012 FRP Reprocessed 2016 FRP Operational 2016 FIR Full scan 2016 FIR RSS Among these Fire at Canopy (%of omitted) 95. 7 96. 3 96. 4 95. 5 92. 1 13. 96 13. 76 6. 4 6. 7 6. 1 • Consistency of fire product Active fire detection performance: • Best performance of MPEF FIR Rapid scan as regards to the omission errors, as well as to the sensitivity of fires under the forest canopy. A critical point regarding omission error is the frequency of satellite observations. • Better performance of version 2016 Reprocessed than operational 2012 FRP version.
Comparison of per-fire FRP (MW) derived from SEVIRI and VIIRS-750 m 2016 Two types of matching are performed 1. All fire pixels detected by VIIR- 750 are matched to the LSA SAF SEVIRI FRP fire pixel in the following grid : • Within the same SEVIRI FRP pixel under investigation. • At 6 min time difference between the SEVIRI detection and corresponding VIIRS -750 m observations of the same fire. The results show only 2 pixels, which meet this criteria for 2016: 2. All fire pixels of VIIR-750 m are matched to the LSA SAF SEVIRI FRP fire pixel in the following grid : • Within 3 x 3 pixel window centered on the SEVIRI FRP pixel under investigation. • At 6 min time difference between the SEVIRI detection and corresponding VIIRS-750 m observations of the same fire. Per-fire FRP from VIIRS-750 m (MW) Lat Lon Ful. Row Ful. Col Date time FRP (MW) VIIRS-75 Total FRP (MW) VIIRS-date VIIRS-time 42. 280 25. 340 3192 1220 01. 08. 2016 1200 243. 40 89. 92 01. 08. 2016 1206 42. 750 27. 310 3200 1181 11. 08. 2016 2300 91. 90 100. 86 11. 08. 2016 2254 850 R 2 = 0, 4041 750 650 550 450 350 250 150 250 350 450 550 Per-fire FRP from SEVIRI (MW) 650
Case study example: Satellite detection and monitoring of crown forest fire on 24 -28 August 2017 by all evaluated satellite algorithms
24 -29 August 2017, Large Canopy Forest Fires MPEF FIR Rss, 24 Aug 2017, 10: 15 -23: 30 UTC MPEF FIRdetected Full Scan , by Animation The fire was all operational 24 Aug 2017; 10: 15 UTC-25 Aug 2017 13: 30 UTC EUMETSAT MPEF FIR at the very beginning of its initiation. The fire behaviour was very well followed, including during the night on 25 – 26 August, before the beginning of active rescue operations. 25 Aug 2017, MPEF 02: 10 FIR –Rss, 26 Animation Aug 2017 03: 10 UTC
24 -29 August 2017, Large Canopy Forest Fires MODIS Collection 6 detections Animation, 24 -28/08/2017 Several hot spots are identified by MOD-6 algorithm.
24 -29 August 2017, Large Canopy Forest Fires VIIRS 325 m detections Animation, 24 -28/08/2017 Several hot spots are identified by VIIRS 325 m algorithm.
Satellites detections of active fire on 24 August 2017 (Large Forest Fire)
Satellite detection of active fire on 24 August 2017 (Large Forest Fire) Summary of active fire detection on 24 Aug 2017 • Early detection by all SEVIRI algorithms (FIR Rss, FIR Fss, FRP), 1015 UTC (earlier than the announcement for its ignition by ground observation) • FIR Rss: 1015 – 2315 UTC each 15 minutes detections • FIR Fss: 1015 – 2330 UTC each 5 minutes (159 time slots) detections • FRP: 1015 – 2315 UTC each 15 minutes detections. • Efficient night time detections by all SEVIRi algorithms of the canopy forest fire. • MODIS 6 and S-VIIRS 375 m also detect the fire
Satellite detection of active fire on 25 -28 August 2017 (Large Forest Fire)
Satellite detection of active fire on 25 -28 August 2017 (Large Forest Fire) Summary of active fire detection on 25 -28 Aug 2017 • 25 Aug: • FIR Rss: 1015 – 2315 UTC detections in all slots each 15 minutes • FIR Fss: 0210 – 1340 UTC each 5 minutes (37 time slots) • FRP: 0045 – 1430 UTC, more effective night time detection than FIR Fss • MODIS 6 and S-VIIRS 375 m with several hot spots identified • 26 Aug: Among the SEVIRI algorithms only FIR Rss has single detection at 0310 UTC • After active rescue operations SEVIRI doesn’t succeed to detect further fire development • 26, 27, 28 Aug: only MODIS 6 and S-VIIRS 375 m proceed to detect the fire (in addition
Satellite detection of active fire on 24 -28 August 2017 (Large Forest Fire) Summary of FRP active fire detections on 24 -28 Aug 2017 • All detected pixels 268 (72 time slots): • 17 detections with FRP equal/less 50 MW • 59 detections with FRP between 50 – 100 MW • 69 detections with FRP between 50 – 100 MW • All detections with FRP greater than 100 MW
Summary of satellite detection of active fire on 24 -28 August 2017 • SEVIRI algorithms are successful in detection of large forest canopy fire even during the night • FRP biomass burning is detected at minimum 31. 90 MW • Geostationary and polar orbiting fire detections complement each other in fire behavior monitoring and evaluation; • For developing an efficient fire warning system for the region of southeastern Europe, there is a need of a common framework taking the advantage of all algorithms for GEO and LEO.
Decoding archive S-VIIRS 750 m information Comparing datasets of Wilfrid Schroeder, Univ. Maryland - Erdem Erdi, TSMS 13 VIIRS 750 m registrations of active fire pixels, confirmed by actual forest fires .
PERSPECTIVES For an efficient fire detection, various available sources of information can be used in order to avoid as mush as possible the remote sensing constrains coming from sensor and satellite orbit geometry as well as cloud contamination. Further activities: • Validation studies of LSA SAF FRP product, regarding omission errors over South Eastern Europe, taking into account failure detections due to cloudiness. • For that purpose, a data set of SEVIRI cloud mask has been processed by TSMS, Turkey.
Acknowledgments This study is funded by EUMETSAT in the frame of SALGEE Project 2016 -2017. TSMS has developed software for processing and visualisation of archive S-VIIRS 750 m data (in the frame of SALGEE Project). LSA SAF provided data for FRP-PIXEL product to fill the gaps for the test period. LSA SAF FRP-PIXEL new algorithm product for the test period was kindly provided by the CDOP-3 archive. State Forest Agency of Bulgaria has provided information for actual forest fires and their characteristics (used as a reference in this study).
References Csiszar, I. , Schroeder, W. , Giglio, L. , Ellicott, E. , Vadrevu, K. P. , Justice, C. O. , and Wind, B. (2014) Active fires from Suomi NPP Visible Infrared Imaging Radiometer Suite : Product status and fire evaluation results, J. Geophys. Res. -Atmos. , 119, 803– 816. EUMETSAT: Active Fire Monitoring with MSG, Algorithm Theoretical Basis Document, EUM/MET/REP/07/0170, Darmstadt, Germany, 2007. Freeborn, P. H. , Wooster, M. J. , Hao, W. M. , Ryan, C. A. , Nordgren, B. L. , Baker, S. P. , and Ichoku, C. (2008) Relationships between energy release, fuel mass loss, and trace gas and aerosol emissions during laboratory biomass fires, J. Geophys. Res. , 113, D 01301, doi: 10. 1029/2007 JD 008679. Georgiev, C. G. and Stoyanova, J. S. (2013) Parallel use of SEVIRI LSA SAF FRP and MPEF FIR products for fire detection and monitoring, 2013 EUMETSAT Meteorological Satellite Conference, 19 th American Meteorological Society, AMS Satellite Meteorology, Oceanography and Climatology Conference, 16– 20 September 2013, Vienna, Austria, ISSN 1011 -3932. Giglio, L. , Descloitres, J. , Justice, C. O. , and Kaufman, Y. J. (2003) An enhanced contextual fire detection algorithm for MODIS, Remote Sens. Environ. 87. 2 -3. 273 -282. Jenkins, B. M. , Baxter, L. L. , Miles Jr. , T. R. , and Miles, T. R. (1998) Combustion properties of biomass, Fuel Processing Technol. , 54, 17– 46. Joro, S. , Samain, O. , Yildirim, A. , van de Berg, L. , and Lutz, H. J. (2008) Towards an improved active fire monitoring product for MSG satellites. www. eumetsat. int/cs/idcplg? Idc. Service=GET_FILE&d. Doc. Name=pdf_conf_p_s 8_47_joro_v&allow. Inter rupt=1&no. Save. As=1&Revision. Selection. Method=Latest. Released. Schroeder, W. , Oliva, P. , Giglio, L. , and Csiszar, I. A. (2014) The New VIIRS 375 m active fire detection data . product: Algorithm description and initial assessment, Rem. Sens. Environ. , 143, 85– 96 Roberts, G. and Wooster, M. J. (2008) Fire Detection and Fire Characterization over Africa using Meteosat SEVIRI, IEEE T. Geosci. Remote Sens. , 48, 1200– 1219.
References Wooster, M. J. , Zhukov, B. , and Oertel, D. (2003) Fire radiative energy for quantitative study of biomass burning: derivation from the BIRD experimental satellite and comparison to MODIS fire products, Rem. Sens. Environ. , 86, 83– 107. Wooster, M. J. , Roberts, G. , Perry, G. L. W. , and Kaufman, Y. J. (2005) Retrieval of biomass combustion rates and totals from fire radiative power observations: FRP derivation and calibration relationships between biomass consumption and fire radiative energy release, J. Geophys. Res. , 110, D 24311, doi: 10. 1029/2005 JD 006318. Wooster, M. J. , Roberts, G. , Freeborn, P. H. , Xu, W. , Govaerts, Y. , Beeby, R. , He, J. , Lattanzio, A. , Fisher, D. , and Mullen, R. (2015) LSA SAF Meteosat FRP products – Part 1: Algorithms, product contents, and analysis, Atmos. Chem. Phys. , 15, 13217– 13239, doi: 10. 5194/acp-15 -13217 -2015. Xu, W. , Wooster, M. J. , Roberts, G. , and Freeborn, P. (2010) New GOES imager algorithms for cloud and active fire detection and fire radiative power assessment across North, South and Central America, Rem. Sens. Environ. , 114, 1876– 1895.
- Slides: 32