elib
DLR-Header
DLR-Logo -> http://www.dlr.de
DLR Portal Home | Imprint | Privacy Policy | Contact | Deutsch
Fontsize: [-] Text [+]

Estimating the fractional cover of growth forms and bare surface in savannas. A multi-resolution approach based on regression tree ensembles

Gessner, Ursula and Machwitz, Miriam and Conrad, Christopher and Dech, Stefan (2013) Estimating the fractional cover of growth forms and bare surface in savannas. A multi-resolution approach based on regression tree ensembles. Remote Sensing of Environment, 129, pp. 90-102. Elsevier. DOI: 10.1016/j.rse.2012.10.026

[img] PDF
2MB

Abstract

Evaluations of existing land cover maps have revealed that high landscape heterogeneity and small patch sizes are a major reason for misclassification. These problems globally occur in biomes of mixed vegetation structure and are particularly relevant for African savannas. This paper presents a multi-resolution approach to derive fractional cover of vegetation growth forms at sub-pixel level, aiming at an improved mapping of land cover in the African grassland, savanna and shrubland biome. Fractional cover is delineated for woody growth forms (trees and shrubs), herbaceous growth forms, and bare surface. The approach incorporates very high resolution (QuickBird/IKONOS, 0.6–1 m), high resolution (Landsat TM/ETM+, 30 m), and medium resolution data (MODIS, 250 m). While QuickBird/IKONOS data are classified into discrete classes, at Landsat and MODIS resolutions, sub-pixel cover is delineated using non-parametric ensemble regression trees from the random forest family. The propagation of errors in the hierarchical multi-resolution approach is assessed with Monte Carlos simulations. The multi-resolution approach allows the adequate description of the heterogeneous vegetation structure in selected study regions of Southern Africa. The RMSE of the delineated fractional cover values range between 3.1% and 8.2% when compared with higher resolution data and between 4.4% and 9.9% when compared with field surveys. Errors at the Landsat resolution show minor influence on the accuracy of the MODIS results. Regarding the inter-resolution error propagation, for 90% of the Monte Carlo simulations, errors at the Landsat resolution resulted in RMSEs for MODIS increased by less than 4% (woody vegetation), 3.5% (herbaceous vegetation) and 2% (bare surface).

Item URL in elib:https://elib.dlr.de/79611/
Document Type:Article
Title:Estimating the fractional cover of growth forms and bare surface in savannas. A multi-resolution approach based on regression tree ensembles
Authors:
AuthorsInstitution or Email of AuthorsAuthors ORCID iD
Gessner, Ursulaursula.gessner (at) dlr.deUNSPECIFIED
Machwitz, MiriamCentre de Recherche Public — Gabriel Lippmann, Department of Environment and Agro-Biotechnologies (EVA), 41, rue du Brill, 4422 Belvaux, LuxembourgUNSPECIFIED
Conrad, ChristopherUniversität Würzburg, Lehrstuhl für Fernerkundung, Am Hubland, 97074 WürzburgUNSPECIFIED
Dech, Stefanstefan.dech (at) dlr.deUNSPECIFIED
Date:2013
Journal or Publication Title:Remote Sensing of Environment
Refereed publication:Yes
Open Access:Yes
Gold Open Access:No
In SCOPUS:Yes
In ISI Web of Science:Yes
Volume:129
DOI :10.1016/j.rse.2012.10.026
Page Range:pp. 90-102
Publisher:Elsevier
Status:Published
Keywords:Sub-pixel fractional cover; Savanna; Grassland and shrubland biome; Africa; Vegetation structure; Land cover; Multi-resolution analysis
HGF - Research field:Aeronautics, Space and Transport
HGF - Program:Space
HGF - Program Themes:Earth Observation
DLR - Research area:Raumfahrt
DLR - Program:R EO - Erdbeobachtung
DLR - Research theme (Project):R - Vorhaben Fernerkundung der Landoberfläche (old)
Location: Oberpfaffenhofen
Institutes and Institutions:German Remote Sensing Data Center > Land Surface
Deposited By: Gessner, Ursula
Deposited On:14 Feb 2013 10:19
Last Modified:31 Jul 2019 19:38

Repository Staff Only: item control page

Browse
Search
Help & Contact
Information
electronic library is running on EPrints 3.3.12
Copyright © 2008-2017 German Aerospace Center (DLR). All rights reserved.