Modeling geospatial trend changes in vegetation monitoring data

https://doi.org/10.1016/j.compenvurbsys.2010.05.006Get rights and content

Abstract

In a constantly changing environment, monitoring supports analysis and understanding of many types of change. This paper is concerned specifically with monitoring of vegetation and describes the development and application of a formal model that supports the analysis of spatiotemporal changes in the recorded attributes of a forest/heathland environment. Typically, the monitoring points are not ideally distributed in time or space. The proposed analytical techniques are designed to deal with incomplete data sets and to reveal abnormal changes or transitions. These transitions can potentially be linked to causal events which may have not been otherwise recorded. This work distinguishes five key change types as a basis for 25 transition types present in time series of vegetation data. These are distilled from the data using a set of transition point analysis methods including spatiotemporal neighborhood and trend sequence analysis. In addition, cross-comparisons between vegetation attributes, based on the identified transitions, are illustrated. A prototype GIS-based tool VeMonA provides an analytical environment for time series data obtained through vegetation monitoring and supports understanding of dynamic geospatial ecosystems.

Introduction

Modeling dynamic geospatial ecosystems such as forests, wetlands or rivers has been a major focus of recent research in the GIScience domain (Drummond et al., 2006, Peuquet, 2002, Stewart Hornsby and Yuan, 2008). A comprehensive understanding of change enables us to analyze the history of dynamic geospatial ecosystems and to predict their future development. Monitoring is a key technique for discovering changes since it involves regular observations of properties of entities over an extended time frame. The primary aim of monitoring is to identify and understand, but also regulate or control on-going processes and changes. Monitoring is used in a wide variety of domains, ranging from stock market analysis (Huson et al., 2001, Tsay, 2005) to environmental studies (Brydges, 2004, Spellerberg, 2005).

With growing concern about environmental change, detection of changes in the environment has become crucial. In response, time and money spent on environmental monitoring has significantly increased over recent decades (Mol, Vriend, & Van Gaans, 2000). Environmental studies reveal a particular challenge: data, typically obtained through field work, is often sparse (for example, vegetation monitoring sites are usually visited only once a year) and contain spatial and temporal gaps. Moreover, monitoring projects require surveying and mapping of multiple variables at particular locations at regular time intervals. The resulting multivariate spatiotemporal data sets can be very difficult to analyze or represent, and appropriate models and tools to manage complex spatiotemporal monitoring data are particularly challenging. In addition, users analyzing monitoring data are often interested not only in the property changes of an entity, but also in the underlying events triggering these changes. For example, abrupt changes in the vegetation composition and structure of an ecosystem could be caused by a fire event. Predicting events such as wildfires based on, for example, vegetation condition, fuel moisture or lightning strike density is another challenge (Dilts, Sibold, & Biondi, 2009).

This paper shows how spatiotemporal changes observed through vegetation monitoring, are used to identify the likely events underlying changes that have occurred in a geospatial ecosystem. The approach presented in this work illustrates techniques that are useful for predicting future vegetation changes based on events that have already happened. The objective is to abstract relevant information from complex data sets by first classifying changes into qualitative impact types – a method that has already been detailed in Mau, Hornsby, and Bishop (2007) – and then analyzing transitions between these types. In this paper, an event refers to a happening that occurs in time and unfolds itself through a period of time (Galton, 2000, Galton, 2005) such as a bush fire or a storm, and the term impact is used to describe the effect of an event on an object including the change in values of properties of an object (Mau et al., 2007).

In the natural world change is normal and expected. A major focus of this paper lies in investigating transitions associated with more abrupt changes. A change profile is defined by the succession, over time, of an attribute’s values. A transition represents a considerable deviation in an attribute’s change profile through variation in the rate or direction of change. For example, an abrupt increase in vegetation height followed by a gradual decrease would be a transition involving both a new rate and new direction of change. The type of transition is determined by the change types occurring before and after the transition point. In this work, the classification of transition types, and the transition points at which they occur, enables a thorough analysis of trend changes in monitoring data. Vegetation monitoring data will be used in this research as a basis for the analysis and reasoning about change and transitions.

Data mining and knowledge discovery techniques including K-means clustering, time series analysis and change point detection are used for analyzing time series of attribute changes. Based on these techniques, a software prototype has been implemented in ArcGIS 9.2. The effectiveness of this prototype is demonstrated in a case study using vegetation monitoring data from the Royal Botanic Gardens Cranbourne in Victoria, Australia. The study area is not a garden as the name might indicate but a nature reserve that comprises roughly 360 hectares of native heathlands, wetlands and forest.

The rest of this paper is organized as follows: Section 2 presents related research within geographic information science on modeling dynamic entities. Section 3 discusses the classification of quantitative changes into qualitative impact types and introduces the notion of transitions as a new concept for analyzing change trends. Section 4 describes spatiotemporal transition analysis as a method for analyzing complex vegetation monitoring data. The application of the developed methods is shown in a case study in Section 5 and Section 6 provides conclusions and proposes future work.

Section snippets

Background

Modeling dynamic entities has been a central research topic within geographic information science. Dynamic entities are regarded in this research as geographic objects or units, such as a forest or a river, whose spatial or other properties are changing over time. Previous research has focused on modeling change in geographic domains (e.g., Claramunt and Theriault, 1996, Galton, 2000, Hornsby and Egenhofer, 2000) where many of the change-based approaches concentrate on alterations to an

Impact classification

In previous research, impacts have been distinguished into two basic types: gradual impacts and abrupt impacts based on two threshold values ε and δ (Mau et al., 2007). The first of these thresholds, ε, was used to discern gradual impacts that are reflected in the progressive change of an attribute’s value over several time steps. This change in attribute value must be positive for all possible intervals (in the case of an increase), or negative for a decrease, such that the overall change is

Spatiotemporal transition point analysis

The aim of this work is to detect unusual change behavior in attribute values, defined by abrupt attribute value changes (i.e., abrupt increase or abrupt decrease), and derive likely underlying events from this behavior. For this purpose, a thorough analysis of the discovered transition points is necessary. The following section details the steps that have to be taken: creating transition maps of the detected changes, analyzing neighborhood relations of changes at adjacent monitoring sites,

Study area

In order to provide evidence of the effectiveness and usefulness of this approach, the developed prototype has been applied to monitoring data of long-term vegetation surveys at a nature reserve 50 km southeast of Melbourne, the Royal Botanic Gardens Cranbourne (Fig. 6a). The Royal Botanic Gardens Cranbourne is one of Victoria’s most precious areas of native bushland. This nature reserve comprises roughly 360 hectares of native heathlands, wetlands and forest.

The site was chosen as a study area

Conclusions and future work

This paper provides methods for a comprehensive analysis of complex vegetation monitoring data, especially discovering key transitions (e.g., abrupt changes) in observed vegetation attributes and detecting possible underlying events. This work extends considerably an earlier model of qualitative change for vegetation monitoring data (Mau et al., 2007) by increasing the number of thresholds from two to four in order to allow for defining increase and decrease of the same type (i.e. abrupt of

Acknowledgements

Kathleen Stewart Hornsby’s research is supported in part by grants from the US Department of Defense HM1582-08-2001, HM1582-05-1-2039 and HM1582-08-1-0013. The authors would like to thank the Royal Botanic Gardens Cranbourne, Australia, for providing the vegetation monitoring data used in this study. Inga Kulik thanks the Cognitive Systems group at the Transregional Collaborative Research Center SFB/TR 8 Spatial Cognition, University of Bremen, Germany, for generously providing an office and

References (43)

  • R.E. Keane et al.

    A classification of landscape fire succession models: Spatial simulations of fire and vegetation dynamics

    Ecological Modeling

    (2004)
  • K. Beard et al.

    A framework for visualization and exploration of events

    Information Visualization

    (2008)
  • D. Brunner et al.

    An extended concept of voxel neighborhoods for correct thinning in mesh segmentation

    ACM

    (2005)
  • T. Brydges

    Basic concepts and applications of environmental monitoring

  • H. Cao et al.

    Periodic pattern discovery from trajectories of moving objects

  • Cheal, D. (1994). Fire succession in heathlands and implications for vegetation management. Paper presented at fire and...
  • N.R. Chrisman

    Beyond the snapshot: Changing the approach to change, error, and process

  • Claramunt, C., & Theriault, M. (1995). Managing time in GIS. In An event oriented approach. Recent advances in temporal...
  • C. Claramunt et al.

    Toward semantics for modelling spatio-temporal processes within GIS

  • Cole, S., & Stewart Hornsby, K. (2005). Modeling noteworthy events in a geospatial domain. Paper presented at...
  • T.E. Dilts et al.

    A weights-of-evidence model for mapping the probability of fire occurrence in Lincoln County, Nevada

    Annals of the Association of American Geographers

    (2009)
  • J. Drummond et al.

    Dynamic and mobile GIS investigating changes in space and time

    (2006)
  • A. Galton

    Qualitative spatial change

    (2000)
  • A. Galton

    Fields and objects in space, time, and space-time

    Spatial Cognition and Computation

    (2004)
  • Galton, A. (2005). Dynamic collectives and their collective dynamics. Paper presented at COSIT 2005, LNCS (Vol. 3693,...
  • Galton, A., & Worboys, M. (2005). Processes and events in dynamic geo-networks. In Paper presented at geospatial...
  • Gill, A. M., Woinarski, J. C. Z., & York, A. (1999). Australia’s biodiversity – Responses to fire: Plants, birds and...
  • M.F. Goodchild et al.

    Representation and computation of geographic dynamics

  • K. Hornsby et al.

    Identity-based change: A foundation for spatio-temporal knowledge representation

    International Journal of Geographical Information Science

    (2000)
  • K. Hornsby et al.

    Modeling moving objects over multiple granularities

    Annals of Mathematics and Artificial Intelligence

    (2002)
  • M.R. Huson et al.

    Internal monitoring mechanisms and CEO turnover: A long-term perspective

    The Journal of Finance

    (2001)
  • Cited by (0)

    View full text