EOS-MODIS 数据林火识别算法的验证和改进

EOS-MODIS 数据在森林火情监测中的应用研究日益受到世界各国的重视。为了获得适用于中国不同地区森林火情监测的成熟技术, 很有必要对现有MODIS 数据林火监测理论算法进行验证分析, 探讨其在中国不同地域和季节中使用时的通用性。为此, 利用中国境内9 起森林火灾事件对MODIS 数据火点识别的理论算法进行验证分析。结果显示9 起森林火灾有8 起被有效检测到, 1 起森林火情被遗漏。通过对9 起森林火点及其邻近像元的统计分析, 发现如下两个重要规则:利用火点亮温偏离统计均值3 倍标准差的关系来确定阈值, 可以避免火点的遗漏; 林火点在CH21和CH22 上的亮温值一般有CH21- CH22< 20 K, 而噪声点在两个波段上的差异却比较大。用以上规则改进的MODIS 林火热点识别算法可以检测出用来验证的全部9 起林火事件, 从而证明了改进算法的有效性和通用性。     

基于HJ-1卫星数据反演长江三角洲地区气溶胶光学厚度

探索利用我国HJ-1卫星CCD数据,运用深蓝算法开展长江三角洲地区气溶胶光学厚度反演的可行性,并将结果与其他气溶胶光学厚度产品进行比较。针对HJ-1A和HJ-1B数据,反演结果分别与MODIS气溶胶光学厚度产品以及AERONET地基观测数据进行对比验证。结果表明:深蓝算法得到A星、B星的反演结果与MODIS气溶胶产品呈显著相关,但在数值上普遍高于MODIS产品;反演结果与AERONET站点数据之间的误差介于0.008~0.364之间,研究时段内站点数据缺乏,未对误差进行严格的统计分析。基于深蓝算法的HJ-1卫星数据反演结果虽然在数值上与MODIS气溶胶光学厚度产品存在系统性偏差,但在空间上能够较好地反映长江三角洲地区大气气溶胶分布状况,且具有空间分辨率高的优势。     

京津冀冬季气溶胶光学厚度遥感反演

针对传统的暗像元算法难以满足植被稀疏陆表气溶胶遥感监测需求的问题,提出了冬季植被稀疏的京津冀地区气溶胶光学厚度的遥感反演方法。以2016—2018年连续3年1—2月的AQUA/MODIS L1B数据为数据源,采用暗像元算法与深蓝算法结合的方法对冬季京津冀地区的气溶胶光学厚度进行了遥感监测。使用AERONET数据对结果进行了验证,并与MODIS MYD04_L2暗像元-深蓝气溶胶产品进行了对比。结果表明,该算法在冬季京津冀地区的气溶胶监测效果远好于暗像元算法,并与MODIS气溶胶产品表现出了显著的相关性,且有效监测范围更大、空间分辨率更高。根据连续监测结果,分析了京津冀地区冬季气溶胶光学厚度空间分布特征及其影响因素。     

高分一号数据的气溶胶光学厚度反演和验证

针对高分辨率卫星遥感反演气溶胶光学厚度地表噪声难以分离的问题,利用国产"高分一号"(GF-1)的数据特点,提出了一种气溶胶光学厚度反演方法和处理流程。该方法分别基于暗像元和深蓝算法去除了浓密植被和城市亮目标地区的地表贡献,并应用于我国污染较为严重的京津冀、长三角、珠三角等示范区域。利用北京、杭州、香港AERONET地基观测数据,对GF-1反演得到的气溶胶光学厚度进行验证,结果表明:气溶胶高值均集中在三大区域工业排放大和人类活动密集的核心城市,年均光学厚度值在1左右。卫星和地基的相关性总体较好,三大区域的相关系数分别达到了0.71、0.55、0.54。受云识别、亮地表覆盖和气溶胶模式假设等影响,GF-1反演的气溶胶光学厚度存在一定程度的偏差。     

基于NDTI指数的MODIS火情监测研究

在分析燃烧热点光谱辐射特征的基础上,提出基于MODIS红外辐射的NDTI指数火情监测模型,给出了该监测模型的详细算法,并通过几起较大森林火灾对该模型进行了分析和验证。这种双通道比值指数和NDVI指数一样具有衰减仪器噪声和抵消常规大气影响的优点。MWIR 4μm火点辐射通道和LWIR 11μm背景分量通道的组合突出了火点信息。研究表明,选择合适的NDTI门限,NDTI指数能够准确检测火点。MODIS NDTI火情指数模型可以推广到同类探测器的应用中,例如NOAA AVHRR、GOES VAS以及即将运行的NPP VIIRS和中国风云3号FY-3平台,特别是MODIS的下一代多光谱探测器NPOESS NPP VIIRS设计继承了MODIS的优势,为NDTI算法模型提供了广阔的应用前景。     

基于随机森林的高分辨率 PM2.5 遥感反演 ——以广东省为例

细颗粒物(PM2.5)监测是大气污染治理的重要手段,受限于地面观测点的数量,从遥感反演 PM2.5 是常规地面观测的有效补充,是当前的研究热点。通常遥感反演 PM2.5 的思路是先反演大气气溶胶光学厚度,然后基于统计关系由大气气溶胶光学厚度反演 PM2.5。该方法容易造成误差传递,从而 导致反演模型的不稳定。该文提出了一种基于随机森林算法(一种机器学习算法)的 PM2.5 遥感反演方法,直接建立中分辨率成像光谱仪(Moderate Resolution Imaging Spectroradiometer,MODIS)影像与地 面实测 PM2.5 的关系,可以避免传统反演 PM2.5 时先反演大气气溶胶光学厚度带来的误差,最终得到精度更高的 PM2.5 反演结果。该方法先用随机森林算法对 MODIS 影像和经过克里金插值后的地面监测站PM2.5 数据进行训练和测试;然后,根据测试的均方根误差从多个模型中选取最优(均方根误差最小)的模型;最后,将此模型用于整幅 MODIS 影像,得到整个区域的 PM2.5 反演结果。实验选取了广东省 四个季节多幅 MODIS 影像数据进行验证,并通过决定系数和均方根误差两个表现指标进行对比和分析,验证了所提算法的优越性。     

基于地球同步轨道卫星的输电线路山火监测方法研究与应用分析

为了满足输电线路山火易发地区的低漏检、高精度、大范围、高时效性火点近实时监测需求,本文以地球同步轨道卫星影像为基础,提出了一种基于MC-CNN的山火检测算法。通过结合大津算法(OTSU)和上下文算法来增加潜在火点,从而在一定程度上降低火点检测的漏检率;引入PCA算法对输入特征进行优化,构建多通道网络结构,并利用联合概率和PSO参数寻优算法获取不同通道火点识别权重,在加权平均的基础上最终判定火点;同时,采用固定高温热源和太阳耀斑对虚假火点进行去除,以降低误报率。为了验证所提算法的有效性,本文随机选取了2019年至2022年期间输电线路附近历史卫星监测山火案例,并利用已知火点样本对火点反演结果进行验证。计算结果显示,该算法的火点检测精度达到了89.4%。     

利用GF-4 PMS数据反演城市地区气溶胶光学厚度

高分四号是我国发射的第一颗地球同步轨道卫星,具有较高的空间分辨率及快速重复成像能力,在城市大气环境的高动态监测方面拥有较大的应用潜力。针对城市地区地表反射率高、地表类型复杂,传统的单一算法较难实现大气气溶胶光学厚度有效反演等问题,以北京市为研究区,使用了一种暗目标法和地表反射率数据库相结合的方法,分别对2017年5月25日和10月5日两个时相的气溶胶光学厚度进行了估算。结果表明,该方法能够有效实现城市暗目标区和亮目标区的气溶胶反演,且与MODIS标准陆地气溶胶产品及AERONET站点地面观测数值较为一致,相关系数超过了0.9。     

中国东南地区及近海海域气溶胶反演遥感研究

大气气溶胶在很多生物地球化学循环中具有重要作用,但是由于它的来源广泛并且具有很大的时空变化性,难以在全球范围内精确、实时确定气溶胶的性质、组成及时空分布,因而对大气气溶胶的研究依赖于监测手段的发展。地基试验能获取点源的大气气溶胶光学厚度(AOD)的地面测量数据,得到的气溶胶光学厚度用于卫星数据的预处理以及气溶胶光学厚度反演的精度验证。而经过地基校验后的卫星遥感数据,可以反映大范围内实时动态的气溶胶信息。利用MODIS资料和地基探测的太阳光度计资料,对中国东南地区及近海海域的大气气溶胶光学特性进行了分析,讨论了适用于中国东南地区的大气气溶胶模型;利用连续的太阳光度计数据对MODIS资料的反演结果进行校验,结果表明：改进气溶胶模型和采用连续波段太阳光度计探测数据,可以提高MODIS AOD的校验结果。     

基于MODIS数据的城市气溶胶光学厚度反演方法

遥感的方法为大面积获取气溶胶光学厚度提供了手段。目前使用MODIS(中分辨率成像光谱仪)数据反演气溶胶光学厚度常用的是暗像素方法以及对比法。由于城市地区符合暗像素标准的像素较少,而且城市地区的气溶胶成分复杂,难以确定其具体成分,所以暗像素方法会产生较大误差。根据MODIS第7、3、1波段对气溶胶的不同响应,以三者的反射率为自变量,结合地面实地观测数据,给出了计算北京地区气溶胶光学厚度的计算公式,误差分析表明该方法更适合城市地区。     

An improved algorithm for small and cool fire detection using MODIS data: A preliminary study in the southeastern United States

Traditional fire detection algorithms mainly rely on hot spot detection using thermal infrared (TIR) channels with fixed or contextual thresholds. Three solar reflectance channels (0.65 μm, 0.86 μm, and 2.1 μm) were recently adopted into the MODIS version 4 contextual algorithm to improve the active fire detection. In the southeastern United States, where most fires are small and relatively cool, the MODIS version 4 contextual algorithm can be adjusted and improved for more accurate regional fire detection. Based on the MODIS version 4 contextual algorithm and a smoke detection algorithm, an improved algorithm using four TIR channels and seven solar reflectance channels is described. This approach is presented with fire events in the southeastern United States. The study reveals that the T₂₂ of most small, cool fires undetected by the MODIS version 4 contextual algorithm is lower than 310 K. The improved algorithm is more sensitive to small, cool fires in the southeast especially for fires detected at large scan angles.     

Enhanced contextual forest fire detection with prediction interval analysis of surface temperature using vegetation amount

Early detection of small forest fires is important for forest management because it can prevent fires from spreading and causing severe environmental and economic damage. This study proposes a new method that uses a negative relationship between vegetation amount and land surface temperature to determine a temperature threshold for detecting small forest fires. The proposed method analyses the differences between brightness temperature in remote-sensing data and that estimated from a regression model of brightness temperature and vegetation amount measured by the normalized difference vegetation index. The upper prediction interval of estimated brightness temperature based on a statistical test of the differences was used for the temperature threshold. This method was compared with the Moderate Resolution Imaging Spectroradiometer contextual algorithm using two accuracy measures: precision and recall. The results showed that the proposed method improved the recall accuracy, and its precision accuracy was similar to that of the contextual algorithm. This indicates that the proposed method detected more small forest fires with a similar false detection rate as that of traditional methods.     

Enhancement of a fire-detection algorithm by eliminating solar contamination effects and atmospheric path radiance: application to MODIS data

Traditional fire-detection algorithms with either fixed or contextual thresholds mainly rely on the temperature contrast between a fire pixel and its surrounding pixels in the mid-infrared (MIR) and thermal-infrared (TIR) bands. Solar contamination and thermal atmospheric path radiance in the MIR band can weaken the contrast between the high- and low-temperature objects and undermine the capability of detecting fires during daytime. The degree of solar contamination in the MIR band depends on variable surface albedo, solar zenith angle and atmospheric conditions. A method is proposed to eliminate the solar radiation and thermal path radiance received by the MODerate Resolution Imaging Spectroradiometer (MODIS) sensor in the MIR band. The modified MIR brightness temperature is incorporated into the existing fire-detection algorithm (referred to as ‘MOD14’) after re-tuning the daytime thresholds. The performance of the revised algorithm (referred to as ‘MOD-MOD’) was tested using coincident data from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) onboard the Terra satellite and visual inspection of large quantities of MODIS imageries. Moderate improvements are achieved in the detection rate while retaining low commission errors. Improvement of the detection by MOD-MOD depends on land-cover type. The majority of the false detections occur over deforested area.     

Assessing the utility of the TET-1 hotspot detection and characterization algorithm for determining wildfire size and temperature

The detection and characterization of wildfires using satellite remote-sensing systems has improved considerably over the last 30 years, with daily, global coverage maintained by a number of satellite systems. The recent deployment of new satellite systems, such as Technologie-Erprobungs-Träger-1 (TET-1/Technology-Experiments-Carrier-1), with a higher spatial resolution than the current satellite sensing systems employed for global fire detection, presents an opportunity to investigate the utility and accuracy of the TET-1 detection and characterization algorithm in detecting fires over a range of areas and temperatures. In this study, simulated fire landscapes were generated based on varying fire area (1–100,000 m²) and fire temperature (450–1200 K) and passed through the TET-1 algorithm. The results indicate the TET-1 algorithm to estimate fire area to ±12% and fire temperature to ±3%, which implies that under the test conditions, the products generated by TET-1 have a low systematic error.     

Evaluation of global fire detection algorithms using simulated AVHRR infrared data

This paper provides a comparison of selected algorithms that have been proposed for global active fire monitoring using data from the NOAA Advanced Very High Resolution Radiometer (AVHRR). A simple theoretical model was used to generate scenes of AVHRR infrared channel 3 and channel 4 data containing fires of various sizes and temperatures in a wide range of terrestrial biomes and climates. Three active fire detection algorithms were applied to the simulated AVHRR images and their performance was characterized in terms of probability of fire detection and false alarm as functions of fire temperature and area, solar and viewing geometry, visibility, season and biome. Additional comparisons were made using AVHRR imagery. Results indicate that while each algorithm has a comparable probability of detecting large (1000m2) fires in most biomes, substantial differences exist in their ability to detect small fires, their tolerance of smoke and neighbouring fires, the number of false alarms, and their overall suitability for global application. An improved automatic algorithm is finally presented. It offers enhanced active fire detection with comparable or reduced false alarm rates in most biomes.     

基于改进YOLOv3的火灾检测与识别

现阶段火灾频发,需要自动进行火灾的检测与识别,虽然存在温度、烟雾传感器等火灾检测手段,但是检测实时性得不到保证.为了解决这一问题,提出了基于改进YOLOv3的火灾检测与识别的方法.首先构建一个多场景大规模火灾目标检测数据库,对火焰和烟雾区域进行类别和位置的标注,并针对YOLOv3小目标识别性能不足的问题进行了改进.结合深度网络的特征提取能力,将火灾检测与识别形式化为多分类识别和坐标回归问题,得到了不同场景下火焰和烟雾两种特征的检测识别模型.实验表明,本文提出的改进YOLOv3算法对不同拍摄角度、不同光照条件下的火焰和烟雾检测都能得到理想的效果,同时在检测速度上也满足了实时检测的需求.     

A self-adaptive algorithm based on AVHRR multitemporal data analysis for small active fire detection

The present study proposes and improved self-adaptive algorithm (ISAA) for the detection of active fires using only channel 3 data of the Advanced Very High Resolution Radiometer (AVHRR). ISAA is specifically devised for the detection of small fires. The fire detection procedure is mainly based on the multitemporal approach (TN-ALT) devised by Cuomo et al . (2001a) and makes use of statistical analyses of real fires from different regions of the Italian peninsula. Such analyses allow the characterization of these fires as well as the computation of dynamic threshold values, which are variable in time and space and calibrated on local environmental conditions. ISAA was developed using an initial data sample of 1000 fires that occurred in 1996, and then in order to achieve a highly satisfactory performance in fire detection, the statistical analyses are updated yearly, so that a wider data sample can be considered for subsequent years. The evaluation tests made use of multitemporal satellite data (from 1997 to 1999) and ground observations provided by the Italian Forestry Service. The results obtained in different regions of North and South Italy demonstrated that ISAA detected about 80% of fires (with a low rate of false alarms at 15%) and showed a high fire discrimination capability both in the worst and good light conditions. The most recent contextual methods of fire detection were applied to significant test cases and compared with the results obtained from ISAA. This comparison showed that ISAA was able to find an increased number of fires as well as to reduce false alarms in all different light conditions.     

A contextual algorithm for AVHRR fire detection

A contextual algorithm for fire detection with NOAA-AVHRR-LAC data was developed. Unlike ‘traditional’ fire detection algorithms (e.g., multichannel thresholds), the decision to record a fire is made by comparing a fire pixel with the pixels in its immediate neighbourhood. The algorithm is self-adaptive and therefore very consistent over large areas as well as through seasons. The algorithm appears to operate successfully in most areas of the world. This Letter presents the contextual approach and describes the algorithm.     

激光雷达数据辅助的FY-3D影像阈值自适应云检测方法

国产风云系列卫星可为全球范围内大气、陆地和海洋的遥感监测提供重要数据支撑,由于光学卫星影像不可避免受到云覆盖的影响,通过云检测获取准确的云掩膜是风云系列卫星影像精细处理与应用的关键。现有的云检测方法大多采用简单高效的阈值法,然而由于传感器光谱响应以及不同场景云覆盖下垫面的辐射差异,在缺少大量真实云覆盖标记情况下,现有方法往往难以确定最优的检测阈值。鉴于此,提出了一种阈值自适应的云检测方法（TACD）,顾及传感器波段特性以及云覆盖下垫面差异,设置不同场景下的多通道阈值测试,包括反射率及反射率组合测试、亮度温度测试、亮度温度差异值测试、卷云测试等,联合具有高精度云层信息的激光雷达数据构建全球范围的云检测样本集,实现基于样本集真实云标记的迭代阈值优化,最终基于最优的阈值进行云检测。以风云三号（FY-3D）MERSI-II影像为例,联合CALIOP云层数据构建全球范围的云检测数据集,并将所提出的TACD方法云检测结果与官方云掩膜产品进行对比,结果表明该方法较官方云检测算法精度有明显提高,其中平均交并比从80.35%提升至84.09%,召回率可达92.67%,具有业务化应用的潜力。     

Quantifying burn severity in a heterogeneous landscape with a relative version of the delta Normalized Burn Ratio (dNBR)

Multi-temporal change detection is commonly used in the detection of changes to ecosystems. Differencing single band indices derived from multispectral pre- and post-fire images is one of the most frequently used change detection algorithms. In this paper we examine a commonly used index used in mapping fire effects due to wildland fire. Subtracting a post-fire from a pre-fire image derived index produces a measure of absolute change which then can be used to estimate total carbon release, biomass loss, smoke production, etc. Measuring absolute change however, may be inappropriate when assessing ecological impacts. In a pixel with a sparse tree canopy for example, differencing a vegetation index will measure a small change due stand-replacing fire. Similarly, differencing will produce a large change value in a pixel experiencing stand-replacing fire that had a dense pre-fire tree canopy. If all stand-replacing fire is defined as severe fire, then thresholding an absolute change image derived through image differencing to produce a categorical classification of burn severity can result in misclassification of low vegetated pixels. Misclassification of low vegetated pixels also happens when classifying severity in different vegetation types within the same fire perimeter with one set of thresholds. Comparisons of classifications derived from thresholds of dNBR and relative dNBR data for individual fires may result in similar classification accuracies. However, classifications of relative dNBR data can produce higher accuracies on average for the high burn severity category than dNBR classifications derived from a universal set of thresholds applied across multiple fires. This is important when mapping historic fires where precise field based severity data may not be available to aid in classification. Implementation of a relative index will also allow a more direct comparison of severity between fires across space and time which is important for landscape level analysis. In this paper we present a relative version of dNBR based upon field data from 14 fires in the Sierra Nevada mountain range of California, USA. The methods presented may have application to other types of disturbance events.