Elsevier

Remote Sensing of Environment

Volume 132, 15 May 2013, Pages 159-175
Remote Sensing of Environment

A comprehensive change detection method for updating the National Land Cover Database to circa 2011

https://doi.org/10.1016/j.rse.2013.01.012Get rights and content

Abstract

The importance of characterizing, quantifying, and monitoring land cover, land use, and their changes has been widely recognized by global and environmental change studies. Since the early 1990s, three U.S. National Land Cover Database (NLCD) products (circa 1992, 2001, and 2006) have been released as free downloads for users. The NLCD 2006 also provides land cover change products between 2001 and 2006. To continue providing updated national land cover and change datasets, a new initiative in developing NLCD 2011 is currently underway. We present a new Comprehensive Change Detection Method (CCDM) designed as a key component for the development of NLCD 2011 and the research results from two exemplar studies. The CCDM integrates spectral-based change detection algorithms including a Multi-Index Integrated Change Analysis (MIICA) model and a novel change model called Zone, which extracts change information from two Landsat image pairs. The MIICA model is the core module of the change detection strategy and uses four spectral indices (CV, RCVMAX, dNBR, and dNDVI) to obtain the changes that occurred between two image dates. The CCDM also includes a knowledge-based system, which uses critical information on historical and current land cover conditions and trends and the likelihood of land cover change, to combine the changes from MIICA and Zone. For NLCD 2011, the improved and enhanced change products obtained from the CCDM provide critical information on location, magnitude, and direction of potential change areas and serve as a basis for further characterizing land cover changes for the nation. An accuracy assessment from the two study areas show 100% agreement between CCDM mapped no-change class with reference dataset, and 18% and 82% disagreement for the change class for WRS path/row p22r39 and p33r33, respectively. The strength of the CCDM is that the method is simple, easy to operate, widely applicable, and capable of capturing a variety of natural and anthropogenic disturbances potentially associated with land cover changes on different landscapes.

Highlights

► A comprehensive change detection method (CCDM) includes MIICA, Zone, and Combination. ► MIICA can capture various disturbances well with low commission errors. ► Zone is sensitive to forest-related subtle changes, e.g. regrowth. ► Combination reduces commission and omission errors simultaneously. ► CCDM presents a framework for national land cover updating & change monitoring.

Introduction

Global and regional assessments on land cover and land use status and changes are fundamentally important for climate and environmental change studies (Foley et al., 2005, Matthews et al., 2004, Turner et al., 2007). While some changes in land cover, such as long-term changes, are due to natural causes, human activity increasingly plays an important role in changing the land cover and land use throughout the world. The importance of characterizing, quantifying, and monitoring these changes through remotely sensed and geospatial data as a key component of the land change science has been widely recognized by global and environmental change studies (Turner et al., 2007).

Digital change detection is a process of determining and quantifying changes based on co-registered, multitemporal remotely sensed data (e.g., Green et al., 1994, Loveland et al., 2002, Yang and Lo, 2002, Yang et al., 2003). Many remote sensing change-detection methods have been developed (e.g., Hansen et al., 2008, Healey et al., 2005, Huang et al., 2010, Jensen et al., 1995, Jin and Sader, 2005, Kam, 1995, Kennedy et al., 2009, Latifovic and Pouliot, 2005, Lunetta et al., 2006, Ridd and Liu, 1998, Sader and Winne, 1992, Sohl, 1999) and reviewed since the late 1980s (e.g., Gong et al., 2008, Jensen et al., 1995, Kam, 1995, Lu et al., 2004, Ridd and Liu, 1998, Singh, 1989). In general, two principal approaches are commonly used for change detection: 1) a spectral-based approach by which simultaneous analysis of multitemporal and/or multispectral data is conducted, and 2) a post-classification based approach when independent classifications are made and compared to derive information on changes. A hybrid approach using both 1) and 2) can also be adopted in land cover change study.

A general consensus among researchers is that there is no single method/algorithm that can be universally applicable for change detection and analyses. This is especially the case when large area and regional scale land cover change detection is involved. The main challenge is how to accurately extract land cover changes while distinguishing them from other non-land cover changes caused by natural variability (e.g., weather and climate conditions) and other extraneous factors. A pressing need is to develop robust, efficient, and accurate automated or semi-automated methods necessary for cost effectively monitoring land cover changes at the regional to global scales. This is an ongoing and challenging research topic primarily because using remote sensing data alone sometimes falls short in detecting land cover changes over large geographic regions due to the “ill-defined” problem (e.g., spectral similarity of different land cover classes) similar to that faced by remote-sensing-based biophysical parameter inversion. Hence, it is reasonably concluded that land cover changes can be better detected and quantified at the global and regional scales if multi-source data that cover the temporal, spectral, and thematic domains are to be integrated, analyzed, and interpreted simultaneously.

With a few exceptions, most of the national and regional land cover change projects detect change by using only one pair of images acquired from a growing season, so they lack information on the persistence of changes within and across a season (Pouliot et al., 2009, Zhan et al., 2000). Recent developments using a trajectory-based change detection method using a high temporal frequency Landsat imagery stack had some success (Huang et al., 2008, Kennedy et al., 2010). The approach is promising but was developed primary for detection of forest changes and disturbances and has not been tested for other land cover types. Another challenge is that change detection based on spectral data alone is often not sufficient to detect many types of land cover changes over a large geographic area. It is often desirable to incorporate prior knowledge about land cover and land cover change trajectory along with detected spectral change to improve detection and analysis of land cover change (Gong et al., 2008, Latifovic and Pouliot, 2005).

One potential promising approach is to detect land cover changes using a strategy that integrates a remote sensing technique with a knowledge-based system. The knowledge-based system embodies expert opinion and rules on certain types of land cover changes. Expert knowledge can be expressed as rules and/or attributes derived and assembled from the spectral, spatial, and temporal domains, and the geospatial knowledge about land cover change and trajectories can be built into the system. Within the system, multiple rules and hypotheses can be linked together that ultimately describe the target land cover change classes (Shafer and Logan, 1987, Srinivasan and Richards, 1990).

The primary goal of the research is to develop and evaluate a Comprehensive Change Detection Methodology (CCDM) as a key component for development of a new generation of the National Land Cover Database (NLCD) 2011. The objective of NLCD 2011 is to capture the land cover land use change since previous NLCD 2006 and update the national land cover map. The objective of the CCDM is to detect areas of spectral changes between 2006 and 2011 where either a land cover change or a land disturbance, caused by either a natural or anthropogenic agent, has occurred. The CCDM integrates spectral information from multi-date Landsat images, information on land cover status, and prior knowledge about the trajectory of land cover trends. It is important for the CCDM to minimize the spurious spectral changes caused by variation in vegetation phenology and/or the interannual variability of weather condition, rather than by land cover changes or land disturbances. For NLCD 2011, the product generated by CCDM is regarded as the Maximum Potential Change (MPC) that captures all potential land cover change areas rather than only areas where actual land cover change occurred, that is, only a portion of the MPC is related to actual land cover and land use changes (e.g. land conversion). The final land cover change product of NLCD 2011 is derived by integrating the MPC with the NLCD 2006 and NLCD 2011 land cover classification results. Through this integration, only those pixels within MPC that observed a real land cover change between 2006 and 2011 will be retained in the final land cover change product. Together, the MPC and the final land cover change are the two separate yet complimentary products of the NLCD 2011.

Given the objective of the NLCD 2011, the performance of CCDM is to be measured and evaluated by comparing change area detected by CCDM against the independent data of actual land cover and land use change. The CCDM is a general and robust method designed to be applicable to all the WRS path/rows across the conterminous United States. The CCDM method has been gone through several-stage of test and evaluation, which involved five Landsat path/row during an initial strategy development and eight Landsat path/rows during two operational tests. Those Landsat path/rows cover several ecoregions of the conterminous United States where a variety of natural and anthropogenic induced land cover change and disturbances occurred. In this paper, we describe in detail our comprehensive change method for updating NLCD 2011 with NLCD 2006 as a base, and illustrate the method with two case studies.

Section snippets

Background

Over the past two decades, one major effort in developing national land cover characterization and land cover change products for the United States was made by the Multi-Resolution Land Characteristics (MRLC) consortium. Through the MRLC (http://www.mrlc.gov/), there have been three major products released to the public named National Land Cover Database (NLCD): a circa NLCD 1992 for the conterminous United States with one thematic layer of land cover (Vogelmann et al., 2001); a circa NLCD 2001

Overview of the comprehensive change detection method (CCDM)

The CCDM (Fig. 1) includes three major components: MIICA (see Section 3.2), Zone (Section 3.3), and Combination (Section 3.4) of change results from two Landsat image pairs. The MIICA is a change detection method that uses four spectral indices to capture a full range of land cover disturbance and land cover change patterns between two dates. Zone is a change detection method that is specifically designed to detect the changes related to forest such as forest regeneration, forest fire, and

Study area, data, and general procedures for change detection

To demonstrate the performance of the comprehensive change detection method, we showcase the results from two Landsat path/rows (p22r39 and p33r33), which are different in landscapes and represent the two different yet complete set scenarios (i.e. decision rules) designed in the CCDM. The first Landsat path/row (22/39) is located in Louisiana and Mississippi (Fig. 6), and the center coordinates of the scene are 30° 17′ 42.48″ N and 90° 08′ 43.50″ W. According to NLCD 2006, the area is dominated

Discussion

From this research, a comprehensive change detection method (CCDM) was designed and implemented to capture a full range of land cover land use changes, including disturbances and alteration of land use practices, which can lead to many land cover changes. Furthermore, we also intend to capture the gradual or subtle changes (e.g., regeneration after forest harvest) for two reasons. First, some of those gradual changes could eventually lead to land cover changes under certain conditions. Second,

Conclusions

For development of NLCD 2011, we designed, implemented, tested, and evaluated a new Comprehensive Change Detection Method (CCDM). The CCDM integrates spectral information from multi-date Landsat images, information on land cover status, and prior knowledge about the trajectory of land cover trends to obtain a maximum potential spectral change map with the goal of minimizing omission error without introducing large commission error. The CCDM method has been tested covering a footprint of

Acknowledgments

The authors thank Thomas Adamson for his great editing. We also thank Shengli Huang and Lei Ji for their insightful reviews and comments to the manuscript. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

References (35)

  • M.K. Ridd et al.

    A comparison of four algorithms for change detection in an urban environment

    Remote Sensing of Environment

    (1998)
  • G. Shafer et al.

    Implementing Dempster's rule of hierarchical evidence

    Artificial Intelligence

    (1987)
  • G. Chander et al.

    Developing consistent Landsat data sets for large area applications—The MRLC 2001 Protocol

    IEEE Geoscience and Remote Sensing Letters

    (2009)
  • J.A. Foley et al.

    Global consequences of land use

    Science

    (2005)
  • J. Fry et al.

    Completion of the 2006 National Land Cover Database for the Conterminous United States

    Photogrammetric Engineering and Remote Sensing

    (2011)
  • J. Gong et al.

    A review of multi-temporal remote sensing data change detection algorithms

  • K. Green et al.

    Using remote sensing to detect and monitor land-cover and land-use change

    Photogrammetric Engineering and Remote Sensing

    (1994)
  • Cited by (0)

    1

    Work performed under USGS contract G08PC91508.

    2

    Work performed under USGS contract G10PC00044.

    View full text