基于GNSS-R数据和机器学习算法探测森林火灾研究  

Research on Forest Fire Detection Based on GNSS-R Data and Machine Learning Algorithm

作　　者：杨金虎 YANG Jinhu(Chongqing Research Institute of China Coal Science and Industry Group Co.,Ltd.,Chongqing 400039,China;College of Resources and Security,Chongqing University,Chongqing 400044,China)

机构地区：[1]中煤科工集团重庆研究院有限公司,重庆400039 [2]重庆大学资源与安全学院,重庆400044

出　　处：《地理空间信息》2025年第4期71-74,共4页Geospatial Information

基　　金：中煤科工集团重庆研究院有限公司自立重点项目(2023ZDYF15);山西焦煤能源集团股份有限公司西山分公司自立项目(23140109202306G);中国煤炭科工集团有限公司科技创新创业资金专项国际合作项目(2020-2-GJHZ003)。

摘　　要：基于巴西东南部地区的全球导航卫星系统反射技术(GNSS-R)和累计火灾燃烧面积数据,利用卷积神经网络(CNN)和随机森林(RF) 2种算法,提出基于GNSS-R技术的火灾探测新方法。研究结果表明,CNN方法预测火灾面积和真实火灾面积的符合度显著高于RF方法,显示出较少的异常值。2种方法在探测火灾面积的相对误差方面均小于6%,其中CNN方法在精度上比RF方法提高了3.3%,在拟合度上提高了17.4%。误差分析表明,2种方法的探测误差均遵循正态分布,说明模型无显著系统误差,与RF方法相比,CNN方法探测火灾的误差较低。在空间分布分析中,CNN方法探测到的火灾面积与实际情况更为吻合,而RF方法倾向于对实际火灾面积进行高估。本研究为利用GNSS-R技术进行火灾探测提供了新视角,对于火灾预测与防控具有重要的实践意义。Based on GNSS-R and accumulated fire burn area data,we used convolutional neural network(CNN) and random forest(RF) algorithm to detect fire in the southeastern region of Brazil.The results show that the CNN method significantly outperforms the RF method in terms of congruence with actual fire areas,exhibiting fewer anomalies.Relative error in detecting fire areas is less than 6% for both methods,with the CNN method demonstrating 3.3% increase in accuracy and 17.4% enhancement in model fit over the RF algorithm.Error analysis indicates that the detection errors for both methods adhere to a normal distribution,suggesting an absence of significant systemic errors in the models.Spatial distribution analysis reveals that the fire areas detected by the CNN method align more closely with the actual conditions,in contrast to the RF algorithm which tends to overestimate the actual fire areas.This research offers a novel perspective in utilizing GNSS-R technology for fire detection,holding significant implications for future fire prediction and prevention strategies.

关 键 词：全球导航卫星系统反射技术 火灾 卷积神经网络 随机森林 相对误差 

分 类 号：P228[天文地球—大地测量学与测量工程]