LiDAR-Landsat data fusion for large-area assessment of urban land cover: Balancing spatial resolution, data volume and mapping accuracy Kunwar K. Singh ⇑ , John B. Vogler, Douglas A. Shoemaker, Ross K. Meentemeyer Center for Applied Geographic Information Science, Department of Geography and Earth Sciences, University of North Carolina, 9201 University City Blvd., Charlotte, NC 28223, USA article info Article history: Received 18 February 2012 Received in revised form 20 September 2012 Accepted 20 September 2012 Keywords: LiDAR Landsat Fusion Land cover Large-area assessment Mapping accuracy Managed clearings abstract The structural characteristics of Light Detection and Ranging (LiDAR) data are increasingly used to clas- sify urban environments at fine scales, but have been underutilized for distinguishing heterogeneous land covers over large urban regions due to high cost, limited spectral information, and the computational dif- ficulties posed by inherently large data volumes. Here we explore tradeoffs between potential gains in mapping accuracy with computational costs by integrating structural and intensity surface models extracted from LiDAR data with Landsat Thematic Mapper (TM) imagery and evaluating the degree to which TM, LiDAR, and LiDAR-TM fusion data discriminated land covers in the rapidly urbanizing region of Charlotte, North Carolina, USA. Using supervised maximum likelihood (ML) and classification tree (CT) methods, we classified TM data at 30 m and LiDAR data and LiDAR-TM fusions at 1 m, 5 m, 10 m, 15 m and 30 m resolutions. We assessed the relative contributions of LiDAR structural and intensity surface models to classification map accuracy and identified optimal spatial resolution of LiDAR surface models for large-area assessments of urban land cover. ML classification of 1 m LiDAR-TM fusions using both structural and intensity surface models increased total accuracy by 32% compared to LiDAR alone and by 8% over TM at 30 m. Fusion data using all LiDAR surface models improved class discrimination of spec- trally similar forest, farmland, and managed clearings and produced the highest total accuracies at 1 m, 5 m, and 10 m resolutions (87.2%, 86.3% and 85.4%, respectively). At all resolutions of fusion data and using either ML or CT classifier, the relative contribution of the LiDAR structural surface models (canopy height and normalized digital surface model) to classification accuracy is greater than the intensity sur- face. Our evaluation of tradeoffs between data volume and thematic map accuracy for this study system suggests that a spatial resolution of 5 m for LiDAR surface models best balances classification perfor- mance and the computational challenges posed by large-area assessments of land cover. Ó 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved. 1. Introduction Land-cover change in rapidly urbanizing regions is significantly changing the way societies experience their environment, with wide-ranging economic and ecological implications, including lo- cal climate change (Guneralp and Seto, 2008), biodiversity loss (Gagne and Fahrig, 2011), flood hazards (White and Greer, 2006), fragmentation (Irwin and Bockstael, 2007), and the degradation of ecosystem services (Turner, 2010) and esthetic value (Sander et al., 2010). The speed at which urbanization generates spatial heterogeneity and landscape fragmentation (Irwin and Bockstael, 2007) makes it difficult to accurately track land-use and land-cover (LULC) changes at a desired scale and reasonable cost (Esch et al., 2009). Remotely-sensed data and imagery provide a comprehen- sive, scalable means for detecting and quantifying LULC change, and its use in mapping urban growth, estimating population den- sity, and modeling sustainability and quality of life is becoming increasingly popular as the scale, cost, and spatial-temporal cover- age improve (Rogan and Chen, 2004). However, the spatial hetero- geneity inherent to urban environments represents substantial challenges to discriminating LULC types using remotely sensed data. Spectral mixtures of vegetation and impervious surfaces common in transitory urbanizing landscapes challenge the ability of spectral-based, hard classification algorithms, such as maximum likelihood (ML), to detect unique signatures and accurately assign pixels to a probable dominant class (Lo and Choi, 2004). Standard LULC classification schema for moderate-resolution data at regio- nal and greater scales often lack the specificity (e.g., ‘‘mixed’’ class) and completeness (e.g., ‘‘other’’ class) necessary for accurate repre- sentation of complex urbanizing landscapes. For example, the 0924-2716/$ - see front matter Ó 2012 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.isprsjprs.2012.09.009 ⇑ Corresponding author. Tel.: +1 704 359 7139; fax: +1 704 687 5966. E-mail addresses: ksingh9@uncc.edu (K.K. Singh), john.vogler@uncc.edu (J.B. Vogler), d.shoemaker@uncc.edu (D.A. Shoemaker), rkmeente@uncc.edu (R.K. Meen- temeyer). ISPRS Journal of Photogrammetry and Remote Sensing 74 (2012) 110–121 Contents lists available at SciVerse ScienceDirect ISPRS Journal of Photogrammetry and Remote Sensing journal homepage: www.elsevier.com/locate/isprsjprs