Mapping invasive species and spectral mixture relationships with neotropical woody formations in southeastern Brazil

Author(s)

Amaral, Cibele H. ^[1] ; Roberts, Dar A. ^[2] ; Almeida, Teodoro I. R. ^[1] ; Souza Filho, Carlos R. ^[3]

Affiliation

^[1] Univ Sao Paulo, Inst Geosci, BR-05508080 Sao Paulo, SP - Brazil

^[2] Univ Calif Santa Barbara, Dept Geog, Santa Barbara, CA 93106 - USA

^[3] Univ Estadual Campinas, Inst Geosci, BR-13083870 Campinas, SP - Brazil

• Document type: Journal article
• Source: ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING; v. 108, p. 80-93, OCT 2015.
• Web of Science Citations: 10

Abstract

Biological invasion substantially contributes to the increasing extinction rates of native vegetative species. The remote detection and mapping of invasive species is critical for environmental monitoring. This study aims to assess the performance of a Multiple Endmember Spectral Mixture Analysis (MESMA) applied to imaging spectroscopy data for mapping Dendrocalamus sp. (bamboo) and Pinus elliottii L (slash pine), which are invasive plant species, in a Brazilian neotropical landscape within the tropical Brazilian savanna biome. The work also investigates the spectral mixture between these exotic species and the native woody formations, including woodland savanna, submontane and alluvial seasonal semideciduous forests (SSF). Visible to Shortwave Infrared (VSWIR) imaging spectroscopy data at one-meter spatial resolution were atmospherically corrected and subset into the different spectral ranges (VIS-NIR1: 530-919 nm; and NIR2-SWIR: 1141-2352 nm). The data were further normalized via continuum removal (CR). Multiple endmember selection methods, including Interactive Endmember Selection (IES), Endmember average root mean square error (EAR), Minimum average spectral angle (MASA) and Count-based (CoB) (collectively called EMC), were employed to create endmember libraries for the targeted vegetation classes. The performance of the MESMA was assessed at the pixel and crown scales. Statistically significant differences (alpha = 0.05) were observed between overall accuracies that were obtained at various spectral ranges. The infrared region (IR) was critical for detecting the vegetation classes using spectral data. The invasive species endmembers exhibited spectral patterns in the IR that were not observed in the native formations. Bamboo was characterized as having a high green vegetation (GV) fraction, lower non-photosynthetic vegetation (NPV) and a low shade fraction, while pine exhibited higher NPV and shade fractions. The invasive species showed a statistically significant larger number of spectra erroneously assigned to the woodland savanna class versus the alluvial and submontane SSF classes. Consequently, the invasive species tended to be overestimated, especially in the woodland savanna. Bamboo was best classified using the VSWIR(CR) data with the EMC endmember selection method (User's accuracy and Producer's accuracy = 98.11% and 72.22%, respectively). Pine was best classified using NIR2-SWIR(CR) data with the IES selected endmembers (97.06% and 62.26%, respectively). The results obtained during the two-endmember modeling were fully translated into the three-endmember unmixed images. The sub-pixel invasive species abundance analysis showed that MESMA performs well when unmixing at the pixel scale and for mapping invasive species fractions in a complex neotropical environment, at pixel and crown scales with 1-m spatial resolution data. (C) 2015 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved. (AU)

• FAPESP's process: 10/51758-2 - Use of hyperspectral images of high spatial resolution in remote recognition of species of flora in Southeastern Brazil: methodological development and application potential in geobotany
• Grantee: Teodoro Isnard Ribeiro de Almeida
• Support Opportunities: Regular Research Grants