Inconsistency among Landsat Sensors in Land Surface Mapping: A Comprehensive Investigation Based on Simulation

Feng Chen,Qiancong Fan,Lin Liu,Yuejun Song,Yuansheng Zhang,Zhenshi Yi,Chenxing Wang

doi:10.3390/rs13071383

Abstract

Comprehensive investigations on the between-sensor comparability among Landsat sensors have been relatively limited compared with the increasing use of multi-temporal Landsat records in time series analyses. More seriously, the sensor-related difference has not always been considered in applications. Accordingly, comparisons were conducted among all Landsat sensors available currently, including Multispectral Scanner (MSS), Thematic Mappers (TM), Enhanced Thematic Mappers (ETM+), and Operational Land Imager (OLI)) in land cover mapping, based on a collection of synthesized, multispectral data. Compared to TM, OLI showed obvious between-sensor differences in channel reflectance, especially over the near infrared (NIR) and shortwave infrared (SWIR) channels, and presented positive bias in vegetation spectral indices. OLI did not always outperform TM and ETM+ in classification, which related to the methods used. Furthermore, the channels over SWIR of TM and its successors contributed largely to enhancement of inter-class separability and to improvement of classification. Currently, the inclusion of MSS data is confronted with significant challenges regarding the consistency of surface mapping. Considering the inconsistency among the Landsat sensors, it is applicable to generate a consistent time series of spectral indices through proper transformation models. Meanwhile, it suggests the generation of specific class(es) based on interest instead of including all classes simultaneously.

Highlights

Since the free and open data policy implemented in 2008 [1], increasing applications of the Landsat archive have been publicly reported [2,3]
AVIRIS hyperspectral data, the consistency issues among the Landsat sensors were investigated in terms of channel reflectance, spectral indices, and classification
It appeared that the ETM+ and Thematic Mappers (TM) observations showed relatively high agreement on channel reflectance and the two derived spectral indices (i.e., normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI))

Summary

Introduction

Since the free and open data policy implemented in 2008 [1], increasing applications of the Landsat archive have been publicly reported (especially in earth’s surface dynamics) [2,3]. Time series analyses based on Landsat data have increased significantly due mainly to the free and open data policy as well as other advancements in data processing [3,4]. Landsat sensors have developed from having four broad channels (of MSS) initially to having eight multispectral channels with narrow and well-positioned wavelength ranges (of OLI) [5], which may challenge the time series analyses of the Landsat archives for which data consistency is required. A relatively simple and widely used method in time series analyses (e.g., for land surface change detection), the differencing method, is highly dependent on the consistency of the data (e.g., classification, spectral reflectance or indices at different times) to be compared [3]

Objectives

Methods

Results

Discussion

Conclusion