|
| 基于Himawari-8静止卫星的野火过火迹地检测算法研究 |
| A Burned Area Detection Algorithm Based on Himawari-8 Geostationary Satellite Data |
| 投稿时间：2026-04-17 修订日期：2026-05-29 |
| 查看/发表评论 下载PDF阅读器 |
| DOI： |
| 基金项目:国家重点研发计划(2022YFC3003101)；国家自然科学基金(52321003) |
|
| 中文关键词: 过火迹地 对地静止卫星 Himawari-8 三步法 燃烧日期与置信度 |
| 英文关键词:Burned area Geostationary satellite Himawari-8 Three-step method Burn date and confidence |
| 摘要点击次数: 117 |
| 全文下载次数: 0 |
| 中文摘要: |
| 现有过火迹地产品主要依赖极轨卫星观测,受重访周期与云遮蔽制约,在时间分辨率与燃烧日期精度方面存在固有不足。对地静止轨道卫星的高频观测为突破这一局限提供了技术条件。本文以Himawari-8卫星先进葵花成像仪(AHI)数据为输入,以澳大利亚为研究区,提出了面向静止轨道观测特征的"三步法"过火迹地检测算法(AHIBAD)。算法将经典"两步法"(种子获取＋区域增长)的执行单元由月度合成影像下移至单个观测时刻,通过日内多时刻检测结果的叠加集成引入第三步——基于检出频次的像元级燃烧置信度构建,使单时刻空间缺口通过时间叠加得以弥补。以2020年9月至2021年8月共12个月AHI数据开展定量验证,结果表明AHIBAD月度过火面积时序与MCD64A1、FireCCI51季节性规律高度一致,月度Dice相似系数(DSC)均值分别为0.377和0.368；67.5%的重叠像元燃烧日期绝对偏差不超过1天；置信度指标与MCD64A1燃烧比例的Pearson相关系数为0.358(p<0.001)。以2019年澳大利亚黑色夏季为案例,AHIBAD在MCD64A1无有效记录期间检出约3 004 km2过火迹地,体现了静止轨道高频观测在极端火灾监测中的独特价值。 |
| 英文摘要: |
| Existing burned area (BA) products rely primarily on polar-orbiting satellites, which are inherently limited in temporal resolution and burn date accuracy due to revisit constraints and cloud obstruction. Geostationary satellites, providing continuous high-frequency observations over the same region, offer a technical pathway to overcome these limitations. This study uses data from the Advanced Himawari Imager (AHI) aboard the Himawari-8 geostationary satellite as the primary data source, with the Australian continent as the study region, to develop a burned area detection algorithm termed AHIBAD. The algorithm shifts the execution unit of the classical two-step approach (seed acquisition and region growing) from monthly composites down to individual observation epochs, and introduces a third step through intra-day multi-epoch accumulation: pixel-level burn confidence construction based on detection frequency, allowing spatial gaps caused by cloud obstruction at individual epochs to be progressively filled through temporal accumulation. Quantitative validation using 12 months of AHI data from September 2020 to August 2021 shows that AHIBAD monthly burned area time series are highly consistent with MCD64A1 and FireCCI51 in seasonal patterns, with mean monthly Dice similarity coefficients (DSC) of 0.377 and 0.368, respectively. 67.5% of overlapping pixels have an absolute burn date deviation of no more than one day. The Pearson correlation between the confidence index and the MCD64A1 burn fraction is 0.358 (p less than 0.001). Using the 2019 Australian Black Summer as a case study, AHIBAD detected approximately 3,004 km2 of burned area during periods when MCD64A1 contained no valid records, demonstrating the irreplaceable value of geostationary high-frequency observations for monitoring extreme wildfire events. |
| 关闭 |
|
|
|
