Global hot spot monitoring with Landsat 8 and Sentinel-2. Soushi Kato Atsushi Oda Ryosuke Nakamura (AIST)

Transcription

1 Global hot spot monitoring with Landsat 8 and Sentinel-2 Soushi Kato Atsushi Oda Ryosuke Nakamura (AIST)

2 Motivation for Detecting Hot Spots Hotspot detection using satellite data To monitor wildfire and volcano 4 mm band of MODIS or VIIRS with ~1 km resolution AIST had a ground station of Landsat 8 until March 2015 Effective use of real-time and archival Landsat 8 data SWIR (2 mm) based method (Giglio et al., 2008) Limited to eastern Asian region Landsat 8 archive on Amazon Web Service Expand our application to global scale Few days of delay

3 Landsat 8 OLI (Operational Land Imager) 8 bands in visible shortwave infrared (30 m) + Panchromatic (15 m) TIRS (Thermal Infrared Sensor) 2 bands in thermal infrared (100 m)

4 Radiance (W/m 2 sr mm) Bands Sensitive to Hot Target NIR SWIR TIR B5 B6 B7 B10 B11 900K Planck curves K K 500K 300K l (mm)

5 SWIR 2.33mm Detection Method by Giglio et al. (2008) Fire detection by ASTER reflectance Obvious fire Candidate fire Candidate fires are further evaluated by comparing with the surroundings to identify fire. Schroeder et al. (2016) applied the same method to Landsat 8 data. NIR 0.81mm

6 Band 7: SWIR2 Detection Algorithm Modified to Landsat 8 Band 7: SWIR2 Day Reflectance thresholds L max Night Radiance thresholds L max Hotspot Cloud Hotspot Reflected radiance Min. radiance Light Min. radiance Band 5: NIR Band 5: NIR

7 Example of Detection Natural color B7 (SWIR) B10 (TIR) Nshinoshima Volcano Mar. 30, 2014 Detected Hot Spots 1 km (erupted during Nov., 2013 Nov. 2015)

8 WebGIS System to Visualize Results

9 Detected Example: Wildfire Hubei, China Jan. 23, 2014 L8 detected VIIRS MODIS

10 Detected Example: Open Burning Jewish Autonomous Region, Russia Twice a year May 3, 2015 L8 detected VIIRS MODIS 0.5 km 2 Octob er 26, km 2 2km

11 Detected Example: Volcano Satsuma-Iojima, Japan in 2014 erupted in Jun Jan. 1 Mar. 22 Apr. 23 Jul. 5 Not detected Jul. 21 Jul cloud free images in 46 scenes during 2014 (= 13%)

12 To Solve limited Observation Opportunity Sentinel-2A Comparable waveband and spatial resolution with Landsat series Revisit cycle of 10 days (16 days: Landsat 8) Sentinel-2B (the same design) will be launched in NASA

13 Apply the Same Method to Sentinel-2 Data Sentinel-2A Jan. 4, 2016 Landsat 8 Mar. 17, 2016 Tokyo Tokyo 30 km Background: true color Too many detected results in comparison with Landsat 8

14 False Positives Sentinel-2 Jan. 4, 2016 Roof and solar panel Greenhouse Specular reflection resulted in increased SWIR radiance.

15 Distinguish Hotspot from Specular Reflection Reflectance Examples of reflectance spectra Greenhouse Roof Steelplant Openburning Wavelength (mm) Additional empirical criteria R 2.2mm R 0.87mm R 1.6mm R 0.87mm > r min

16 Result Sentinel-2 Jan. 4, 2016 Accepted Discarded 15% of the results satisfied the criteria.

17 Other Seasons May 13, 2016 Aug. 31, 2016 Accepted Discarded % of the results were discarded.

18 Inspection of Cause Land use was identified using Google Map/street view Land use Jan. 4 May 15 Aug. 31 Accepted result Steel plant Chimney stack Farmland Forest Roof Discarded result Roof Solar panel Cloud Chimney stack Farmland 1 0 0

19 Validation of Hotspot Detection Prescribed fire Jan. 24, 2015 Joso, Japan Overpass time 21:30 JST f 1.3m Spectroradiometer Thermocouple

20 At the Overpass Time Spectral radiant intensity (W/sr mm) FOV of spectrometer Landsat 8 OLI Detected f 20cm NIR: B5 SWIR1: B6 SWIR2: B7 Overlayed on Open street map Spectrometer radiance Landsat K blackbody K m Wavelength (mm)

21 Time Series Comparison Difference in measurement positions produced overestimated temperature from OLI data.

22 Hotspot Detection and Temperature Retrieval Jan. 24, :30 JST 232 results in total Only 6% were obviously false detection due to light.

23 Further Study: Integration with the Other Data Hotspot detection using TIR data ASTER by GSJ Spatial resolution: 90m 2000~ Criteria: TOA BT > 330K CIRC by JAXA Spatial resolution: 200m (ALOS2) 130m (CALET on ISS) May 2014~ (ALOS2) Aug. 2015~ (CALET) Criteria: spatial anomaly Level 2 product compiles detected results in text file Integrating these with our result will increase detection opportunity.

24 Further Study: Hyperspectral Data Hyperspectral sensors with spatial resolution of 30 60m will be launched in few years. EnMap, DLR PRISMA, ASI HyspIRI, NASA HISUI, METI HISUI sensor ISS HISUI will be onboard ISS. Hyperspectral data will be useful to detect hotspots and retrieve their temperature from spectra.

25 Summary Hotspot detection using SWIR data Finer spatial resolution than MODIS or VIIRS fire product Automated detection system was developed. Results are provided by web service. Integration of data from multiple satellite will potentially increase data acquisition frequency. Landsat 8, Sentinel-2A, ASTER, CIRC, Temperature can be retrieved from nighttime NIR-SWIR spectra. Hyperspectral data will provide suitable data to temperature retrieval.

26 Thank you for your attention.

27 False Detection: Rooftop Residential area No fire occurred Niigata, Japan Oct. 29 Specular reflection? TOA reflectance l (mm) 0.79 Roof 26.5 = roof pitch 6/12 Industrial area Shangxi, China Google map Apr. 27 Jun.30 Oct. 18 Roof slope 11.6?