日本地球惑星科学連合2023年大会

講演情報

[E] 口頭発表

セッション記号 M (領域外・複数領域) » M-AG 応用地球科学

[M-AG33] Satellite Land Physical Processes Monitoring at Medium/High/Very High Resolution

2023年5月24日(水) 09:00 〜 10:15 103 (幕張メッセ国際会議場)

コンビーナ:Jean-Claude Roger(University of Maryland College Park)、Shinichi Sobue(Japan Aerospace Exploration Agency)、Eric Vermote(NASA Goddard Space Flight Center)、Ferran Gascon(European Space Agency)、Chairperson:Jean-Claude Roger(University of Maryland College Park)、Eric Vermote(NASA Goddard Space Flight Center)

09:00 〜 09:15

[MAG33-01] NASA Harmonized Landsat Sentinel-2 (HLS) 30 m burned area product generation, Comparison with the NASA 500 m burned area product, and Validation with 3 m PlanetScope burned area data

★Invited Papers

*David Roy1、Haiyan Huang1、Luigi Boschetti 2、Louis Giglio 3 (1.Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI 48823, USA、2.College of Natural Resources, University of Idaho, Moscow, ID 83844, USA 、3.Department of Geographical Sciences, University of Maryland, College Park, MD, 20742, USA)

キーワード:Fire, Satellite, Burned area

The free-availability of Landsat-8 and Sentinel-2 data provides the opportunity for systematic generation of medium spatial resolution land products. Under NASA funding we developed and published an algorithm to map 30-m burned areas using Landsat and Sentinel-2 nadir BRDF-adjusted surface reflectance time series. The different sensor data are combined through a random forest change regression, trained with synthetic data built from laboratory and field spectra and using a spectral model of fire effects on reflectance. The random forest regression is applied independently at each gridded 30-m pixel location on a temporally rolling basis considering three months of sensor data to map the approximate day of burning in the central month. Temporal consistency checks are used to reduce commission errors due to non-fire related spectral changes, and a region growing algorithm is used to reduce omission errors due to temporally sparse observations. Recently we refined and ported the algorithm for application to NASA Harmonized Landsat Sentinel-2 (HLS) data and generated annual Africa 30-m burned area products. The Africa 30-m burned area product and statistical comparison with the contemporaneous NASA MODIS 500-m product are first presented. Temporally, the 30-m product reports the day of burning on average two days later than the 500-m burned area product due to the lower revisit frequency of the HLS observations. Spatially, the 30-m product captures more detail than the 500-m product, with systematically higher monthly burned area estimates, and seasonal differences related to changing burn spatial characteristics and combustion completeness. A comprehensive validation by comparison of the Africa 30-m product with 3-m burned areas mapped from two-date PlanetScope image pairs classified using a recently published deep learning algorithm is presented. A total of 1,316 PlanetScope image pairs were considered across Africa and provided 33% and 50% 30-m burned area omission and commission errors relative to the 3-m maps and an overall 95% classification accuracy which is better than the accuracy reported when we validated the global NASA MODIS 500-m burned area product relative to 30-m Landsat burned area interpretations.