Land cover and land use classification performance of machine learning algorithms in a boreal landscape using Sentinel-2 data

ABSTRACT

In recent years, the data science and remote sensing communities have started to align due to user-friendly programming tools, access to high-end consumer computing power, and the availability of free satellite data. In particular, publicly available data from the European Space Agency’s Sentinel missions have been used in various remote sensing applications. However, there is a lack of studies that utilize these data to assess the performance of machine learning algorithms in complex boreal landscapes. In this article, I compare the classification performance of four non-parametric algorithms: support vector machines (SVM), random forests (RF), extreme gradient boosting (Xgboost), and deep learning (DL). The study area chosen is a complex mixed-use landscape in south-central Sweden with eight land-cover and land-use (LCLU) classes. The satellite imagery used for the classification were multi-temporal scenes from Sentinel-2 covering spring, summer, autumn and winter conditions. Using stratified random sampling, each LCLU class was allocated 1477 samples, which were divided into training (70%) and evaluation (30%) subsets. Accuracy was assessed through metrics derived from an error matrix, but primarily overall accuracy was used in allocating algorithm hierarchy. A two-proportion Z-test was used to compare the proportions of correctly classified pixels of the algorithms and a McNemar’s chi-square test was used to compare class-wise predictions. The results show that the highest overall accuracy was produced by support vector machines (0.758 ± 0.017), closely followed by extreme gradient boosting (0.751 ± 0.017), random forests (0.739 ± 0.018), and finally deep learning (0.733 ± 0.0023). The Z-test comparison of classifiers showed that a third of algorithm pairings were statistically different. On a class-wise basis, McNemar’s test results showed that 62% of class-wise predictions were significant from one another at the 5% level or less. Variable importance metrics show that nearly half of the top twenty Sentinel-2 bands belonged to the red edge (25%) and shortwave infrared (23%) portions of the electromagnetic spectrum, and were dominated by scenes from spring (38%) and summer (40%). The results are discussed within the scope of recent studies involving machine learning and Sentinel-2 data and key knowledge gaps identified. The article concludes with recommendations for future research.     

基于GIS的岷江上游流域土地利用／覆被变化研究

以岷江上游流域为对象,选取3期9景TM/+ETM遥感影像,通过多步骤最大似然监督分类、变化检测,结合空间动态分析测算模型,分析近20年土地利用/覆被变化情况。结果表明:从整个流域分析,林地面积减少,主要转化为未利用地、建设用地和耕地;未利用地在前8年以每年3.7%、后8年以每年0.4%的速度增加;建设用地在1994—2002年以每年34%的速度增加,到2002—2010年增长速度减缓;耕地总面积减少54 431hm2;从县域分析,1994—2002年间,松潘和黑水县大面积林地转为未利用地;2002—2010年间,松潘县未利用地转为林地和建设用地,茂县和汶川县未利用地面积大幅增加。该研究结论不仅为国土资源管理部门优化土地利用结构提供依据,亦为当地政府实现生态资源可持续发展提供数据支撑。     

Hyperspectral imagery for disaggregation of land surface temperature with selected regression algorithms over different land use land cover scenes

Land surface temperature (LST), a key parameter in understanding thermal behavior of various terrestrial processes, changes rapidly and hence mapping and modeling its spatio-temporal evolution requires measurements at frequent intervals and finer resolutions. We designed a series of experiments for disaggregation of LST (DLST) derived from the Landsat ETM + thermal band using narrowband reflectance information derived from the EO1-Hyperion hyperspectral sensor and selected regression algorithms over three geographic locations with different climate and land use land cover (LULC) characteristics. The regression algorithms applied to this end were: partial least square regression (PLS), gradient boosting machine (GBM) and support vector machine (SVM). To understand the scale dependence of regression algorithms for predicting LST, we developed individual models (local models) at four spatial resolutions (480 m, 240 m, 120 m and 60 m) and tested the differences between these using RMSE derived from cross-validated samples. The sharpening capabilities of the models were assessed by predicting LST at finer resolutions using models developed at coarser spatial resolution. The results were also compared with LST produced by DisTrad sharpening model. It was found that scale dependence of the models is a function of the study area characteristics and regression algorithms. Considering the sharpening experiments, both GBM and SVM performed better than PLS which produced noisy LST at finer spatial resolutions. Based on the results, it can be concluded that GBM and SVM are more suitable algorithms for operational implementation of this application. These algorithms outperformed DisTrad model for heterogeneous landscapes with high variation in soil moisture content and photosynthetic activities. The variable importance measure derived from PLS and GBM provided insights about the characteristics of the relevant bands. The results indicate that wavelengths centered around 457, 671, 1488 and 2013–2083 nm are the most important in predicting LST. Nevertheless, further research is needed to improve the performance of regression algorithms when there is a large variability in LST and to examine the utility of narrowband vegetation indices to predict the LST. The benefits of this research may extend to applications such as monitoring urban heat island effect, volcanic activity and wildfire, estimating evapotranspiration and assessing drought severity.     

Fusion of HJ1B and ALOS PALSAR data for land cover classification using machine learning methods

Image classification from remote sensing is becoming increasingly urgent for monitoring environmental changes. Exploring effective algorithms to increase classification accuracy is critical. This paper explores the use of multispectral HJ1B and ALOS (Advanced Land Observing Satellite) PALSAR L-band (Phased Array type L-band Synthetic Aperture Radar) for land cover classification using learning-based algorithms. Pixel-based and object-based image analysis approaches for classifying HJ1B data and the HJ1B and ALOS/PALSAR fused-images were compared using two machine learning algorithms, support vector machine (SVM) and random forest (RF), to test which algorithm can achieve the best classification accuracy in arid and semiarid regions. The overall accuracies of the pixel-based (Fused data: 79.0%; HJ1B data: 81.46%) and object-based classifications (Fused data: 80.0%; HJ1B data: 76.9%) were relatively close when using the SVM classifier. The pixel-based classification achieved a high overall accuracy (85.5%) using the RF algorithm for classifying the fused data, whereas the RF classifier using the object-based image analysis produced a lower overall accuracy (70.2%). The study demonstrates that the pixel-based classification utilized fewer variables and performed relatively better than the object-based classification using HJ1B imagery and the fused data. Generally, the integration of the HJ1B and ALOS/PALSAR imagery can improve the overall accuracy of 5.7% using the pixel-based image analysis and RF classifier.     

特征维数对支持向量机分类器性能影响的研究——以高光谱遥感影像为例

本文为验证SVM对高维特征的适应性和可靠性,针对不同特征提取方法与特征组合,以国产OMISⅡ传感器获得的北京昌平地区高光谱遥感据为例,对SVM分类器中特征维数对分类准确率的影响进行了试验,通过对主成分分析、最小噪声分离算法、相关系数分组后特征提取、导数光谱等的分析,表明SVM分类器的分类精度随着特征维数波动,其中主成分分析降维后提取的特征具有用于分类能够获得最高的准确率。通过与最大似然法和光谱角制图分类算法的比较,说明在同样的特征输入情况下SVM分类算法分类的准确率高于最大似然法和光谱角制图分类器。     

Monitoring of protected bands of Terkos drinking water reservoir of metropolitan Istanbul near the Black Sea coast using satellite data

In this study, remote sensing (RS) with computer-based geographic information systems (GIS) techniques are used as a tool for monitoring the water basin area and water quality in Istanbul's relatively less polluted and comparatively less destroyed catchment of the metropolis drinking water dam reservoir named Terkos. It is necessary to work with recent data to be able to identify the effects of urbanization on the water quality of the Terkos dam catchment area that supplies drinking water to the metropolis. RS is an important tool to monitor water quality and urban terrain. For this aim, a project has been initiated at the Technical University Remote Sensing Laboratory, under the Istanbul Water and Sewerage Administration (ISKI) sponsorship in Istanbul. The project uses SPOT-PAN, XS and IRS-1C/D PAN and satellite data of 1993 and 2000 for urban analysis and Landsat-TM and LISS-III satellite data of 1992 and 2000 for water quality. For calibration and validation, ground truth samples are collected from the experimental area. The RS data was converted into the UTM coordinate system and image enhancement and classification techniques are used. Raster data is converted to vector data to assess the study area for analyzing in GIS for the purpose of planning and decision-making on protected water basin zones. As a result of monitoring land use and water quality changes, recommendations are made for planning and management of the protected environment of the Terkos catchment protected area. Measuring land use change is a very important issue for controlling the future development of the basin, GIS techniques are performed and results are illustrated in established models on the four protected zones of Terkos water basin.     

Retrieval of land surface temperature from the Kalpana-1 VHRR data using a single-channel algorithm and its validation over western India

Indian geostationary satellite Kalpana-1 (K1) offers a potential to capture the diurnal cycle of land surface temperature (LST) through thermal infrared channel (10.5–12.5 μm) observations of the Very High Resolution Radiometer (VHRR) sensor. A study was carried out to retrieve LST by adapting a generalized single-channel (SC) algorithm (Jiménez-Muñoz and Sobrino, 2003) for the VHRR sensor over India. The basis of SC algorithm depends on the concept of Atmospheric Functions (AFs) that are dependent on transmissivity, upwelling and downwelling radiances of the atmosphere. In the present study AFs were computed for the VHRR sensor through the MODTRAN simulations based upon varying atmospheric and surface inputs. The AFs were fitted with the atmospheric columnar water vapour content and a set of coefficients was derived for LST retrieval. The K1-LST derived with the SC algorithm was validated with (a) in situ measurements at two sites located in western parts of India and (b) the MODIS LST products. Comparison of K1-LST with the in situ measurements demonstrated that SC algorithm was successful in capturing the prominent diurnal variations of 283–332 K in the LST at desert and agriculture experimental sites with a rmse of 1.6 K and 2.7 K, respectively. Inter comparison of K1-LST and MODIS LST showed a reasonable agreement between these two retrievals up to LST of 300 K, however a cold bias up to 7.9 K was observed in MODIS LST for higher LST values (310–330 K) over the hot desert region.     

条件随机场框架下基于随机森林的城市土地利用/覆盖遥感分类

针对"基于像素的条件随机场(conditional random fields,CRFs)模型能否在m级分辨率的多光谱遥感图像分类中表现良好"的问题,提出了集成图像的光谱、方向梯度直方图和多尺度多方向Texton纹理等多种线索的CRFs模型定义方法。利用上述特征,选择随机森林(random forests,RF)定义CRFs关联势函数;利用特征对比度加权的Potts函数定义CRFs交互势函数,并且建立了多标签的RF-CRFs模型;对该模型进行分项参数训练以及基于图割的α-膨胀算法推理;利用典型城区的Quick Bird多光谱图像进行模型的验证与精度评价。结果表明RF-CRFs模型的分类精度可达82.52%以上,比RF分类器的分类精度提高了3.35%。