气候变化领域集成服务门户(CCPortal): Spectral harmonization and red edge prediction of Landsat-8 to Sentinel-2 using land cover optimized multivariate regressors

CCPortal

DOI	10.1016/j.rse.2020.111723
	Spectral harmonization and red edge prediction of Landsat-8 to Sentinel-2 using land cover optimized multivariate regressors
	Scheffler D.; Frantz D.; Segl K.
发表日期	2020
ISSN	00344257
卷号	241
英文摘要	Multi-sensor remote sensing applications consistently gain importance, boosted by a growing number of freely available earth observation data, increasing computing capacity, and increasingly complex algorithms that need as temporally dense data as possible. Using data provided by different sensors can greatly improve the temporal resolution of time series, fill data gaps and thus improve the quality of land cover monitoring applications. However, multi-sensor approaches are often adversely affected by different spectral characteristics of the sensing instruments, leading to inconsistencies in downstream products. Spectral harmonization, i.e., the transformation of one sensor into the spectral domain of another sensor, may reduce these inconsistencies. It simplifies workflows, increases the reliability of subsequently derived multi-sensor products and may also enable the generation of new products that are not possible with the initial spectral definition. In this paper, we compare the effect of multivariate spectral harmonization techniques on the inter-sensor reflectance consistency and derived products such as spectral indices or land cover classifications. We simulated surface reflectance data of Landsat-8 and Sentinel-2A from airborne hyperspectral data to eliminate any sources of error originating from unequal acquisition geometries, illumination or atmospheric state. We evaluate different methods based on linear, quadratic and random forest regression as well as linear interpolation, and predict not only matching but also unilaterally missing bands (red edge). We additionally consider material-dependent spectral characteristics in the harmonization process by using separate transformation functions for spectral clusters of the input dataset. Our results suggest that spectral harmonization is useful to improve multi-sensor consistency of remote sensing data and subsequently derived products, especially if multiple transformation functions are incorporated. There is a strong dependency between harmonization performance and the similarity of source and target sensor's spectral characteristics. For spectrally transforming Landsat-8 to Sentinel-2A, we achieved the lowest radiometric inter-sensor deviations with 50 spectral clusters and linear regression. Based on simulated data, deviations are below 1.7% reflectance within the red edge spectral region and below 0.3% reflectance for the remaining bands (RMSE). Regarding spectral indices, our results show a reduction of inter-sensor deviation (vegetation pixels only) to 38% of the initial error for NDVI (Normalized Difference Vegetation Index) and to 43% for EVI (Enhanced Vegetation Index). Furthermore, we computed the REIP (Red Edge Inflection Point) with an accuracy of 3.1 nm from Sentinel-2 adapted Landsat-8 data. An exemplary multispectral classification use case revealed an increasing inter-sensor consistency of classification results from 92.3% to 97.3% mean error. Applied to time series of real Landsat-8 and Sentinel-2 data, we observed similar trends, albeit intermingled with non-sensor-induced inconsistencies. © 2020 Elsevier Inc.
英文关键词	Analysis ready data; Machine learning; Satellite image harmonization; Spectral harmonization; Time series analysis
语种	英语
scopus关键词	Classification (of information); Decision trees; Errors; Learning systems; Random forests; Reflection; Remote sensing; Time series analysis; Vegetation; Airborne hyperspectral data; Analysis ready data; Multispectral classification; Normalized difference vegetation index; Red edge inflection points; Remote sensing applications; Satellite images; Spectral harmonization; Metadata; forest ecosystem; land cover; Landsat; machine learning; multiple regression; multivariate analysis; NDVI; optimization; prediction; satellite data; satellite imagery; Sentinel; surface reflectance; time series analysis
来源期刊	Remote Sensing of Environment
文献类型	期刊论文
条目标识符	http://gcip.llas.ac.cn/handle/2XKMVOVA/179373
作者单位	Helmholtz Center Potsdam GFZ German Research Center for Geosciences, Section Remote Sensing, Telegrafenberg, Potsdam, 14473, Germany; Geography Department, Humboldt-Universität zu Berlin, Unter den Linden 6, Berlin, 10099, Germany
推荐引用方式 GB/T 7714	Scheffler D.,Frantz D.,Segl K.. Spectral harmonization and red edge prediction of Landsat-8 to Sentinel-2 using land cover optimized multivariate regressors[J],2020,241.
APA	Scheffler D.,Frantz D.,&Segl K..(2020).Spectral harmonization and red edge prediction of Landsat-8 to Sentinel-2 using land cover optimized multivariate regressors.Remote Sensing of Environment,241.
MLA	Scheffler D.,et al."Spectral harmonization and red edge prediction of Landsat-8 to Sentinel-2 using land cover optimized multivariate regressors".Remote Sensing of Environment 241(2020).