Japan Geoscience Union Meeting 2024

Presentation information

[E] Poster

A (Atmospheric and Hydrospheric Sciences ) » A-CG Complex & General

[A-CG36] Satellite Earth Environment Observation

Mon. May 27, 2024 5:15 PM - 6:45 PM Poster Hall (Exhibition Hall 6, Makuhari Messe)

convener:Riko Oki(Japan Aerospace Exploration Agency), Yoshiaki HONDA(Center for Environmental Remote Sensing, Chiba University), Tsuneo Matsunaga(Center for Global Environmental Research and Satellite Observation Center, National Institute for Environmental Studies), Nobuhiro Takahashi(Institute for Space-Earth Environmental Research, Nagoya University)

5:15 PM - 6:45 PM

[ACG36-P07] Burned area detection method by using GCOM-C SGLI

*Kazuhisa Tanada1, Hiroshi Murakami1 (1.Japan Aerospace Exploration Agency)

Keywords:wildfire, satellite data, burned area, GCOM-C/SGLI, JAXA

The IPCC AR6 predicts that forest fires will increase in almost all tundra, boreal and some mountainous areas by the end of the century due to increased droughts and heat waves associated with global warming (IPCC, 2022). Fire models that can reproduce these fires are important for future climate projections, and need global monitoring of burnt area (BA) datasets for input and validation. However, the MODIS-equipped Terra and Aqua satellites, which have provided global monitoring of BA, are likely to cease operation soon and a follow-on data building system is needed.

In this study, a burnt area detection algorithm using GCOM-C/SGLI was investigated.

GCOM-C is a polar-orbiting satellite launched by JAXA in 2017, which observes the entire globe with high frequency (once every 2 or 3 days), multi-wavelength and 250 m spatial resolution. As fires occur suddenly all over the Earth, the SGLI's features were considered to be useful for burnt area detection.

In constructing the algorithm, we first prepared the burned area dataset (BARD) from the LANDSAT satellite with a resolution of 30 m as the correct data. Two algorithms were considered: a machine learning (DNN) method and a method using a classification tree.

For the DNN-based method, the teacher data was BARD and the input data was SGLI surface reflectance data, deviations in normalised burned ratio, deviations in NDVI, etc. The classification tree method was used to train the DNN. For the classification tree method, a decision tree was created to classify two categories, burnt and non-burnt areas, based on histograms of SGLI surface reflectance data, normalised burned ratio deviations and NDVI deviations in the two areas of burnt and non-burnt areas using BARD dataset.

The validation results and usefulness of these methods will be discussed in this presentation.