自然资源遥感2026，Vol.38Issue(3)：93-101,9.DOI:10.6046/zrzyyg.2025145
地形校正与随机森林优化的火烧迹地提取方法
A method for information extraction of burned areas based on terrain correction and random forest feature optimization
摘要
Abstract
Forest fires tend to cause severe damage to the structures and functions of ecosystems.Burned areas that have not yet fully recovered from fires carry crucial post-fire information,holding significant implications for ecological monitoring and restoration.To address the challenge of insufficient accuracy in burned area identification under complex terrain conditions,this study proposed a high-precision information extraction method for burned areas that integrates sun-canopy-sensor(SCS)+C(SCS+C)terrain correction with random forest(RF)feature optimization.Based on the wide-field-view(WFV)images from the GF-1 satellite,a multi-dimensional feature set was constructed by integrating texture features and spectral indices.Through SCS+C terrain correction and RF feature optimization,the relative importance of feature subsets was assessed to identify the optimal feature combination for efficient information extraction of burned areas.The results indicate that SCS+C terrain correction significantly mitigated the shadow effect,avoiding misclassifying shadow areas as burned areas.The RF feature optimization retained feature subsets with higher contributions and eliminated redundant features,mitigating misclassification and omissions in the information extraction and effectively enhancing the model accuracy.The combination of terrain correction and RF feature optimization yielded extraction accuracy(including overall accuracy,user accuracy,and producer accuracy)of greater than 91%across multiple typical regions.In the information extraction of burned areas from 2019 to 2023 in Liaoning Province,the proposed method yielded overall classification accuracies of individual years ranging from 79.63%to 83.87%(average:81.90%),with Kappa coefficients varying between 0.503 9 and 0.806 2(average:0.707 3).These results further validate the universality,stability,and efficiency of the proposed method in the information extraction of burned areas.This study provides reliable technical support for post-fire ecological monitoring and restoration.关键词
火烧迹地/随机森林/地形校正/GF-1Key words
burned area/random forest(RF)/terrain correction/GF-1分类
信息技术与安全科学引用本文复制引用
刘英杰,武晋雯,孙龙彧,冯锐,纪瑞鹏,于文颖..地形校正与随机森林优化的火烧迹地提取方法[J].自然资源遥感,2026,38(3):93-101,9.基金项目
辽宁省气象局核心攻关项目"基于FY3-MERSI数据的增强随机森林火点识别算法研究"(编号:HXGGZ202401)、中国气象局气象能力提升联合研究专项重点项目"东北亚'高温-干旱-火灾'监测评估和风险预警技术研究"(编号:23NLTSZ006)、国家重点研发计划专项"中国东北区域陆—气跨圈层立体精细化协同观测国家重点研发计划专项"(编号:2022YFF0801301)和辽宁省农业气象灾害重点实验室联合项目"玉米关键发育期干旱寡照复合灾害对产量形成过程的影响评估"(编号:2024SYIAEKFZD08)共同资助. （编号:HXGGZ202401）
