Keywords

1 Introduction

Landscape pattern characterization aims to map, quantify, and interpret landscape spatial patterns, and is therefore critical to address the spatial interaction between landscape patterns and ecological processes (Turner 2005). An increasing number of indices or metrics have been developed to quantify various landscape aspects (e.g. McGarigal and Marks 1995; McGarigal et al. 2009). These metrics can be derived from categorical maps that are predominately produced through remote sensing (e.g. Yang and Liu 2005a; Wang and Yang 2012). And various landscape elements are normally represented as digital maps, and geographic information systems (GIS) can be used to relate these elements in search of their causal relationship (e.g. Serneels and Lambin 2001; Hietel et al. 2004). Furthermore, spatially explicit modeling techniques can be used to examine the underlying landscape dynamics being linked with various biophysical and socio-economic conditions (e.g. Yang and Lo 2003; Hepinstall et al. 2008; Paudel and Yuan 2011).

Although remote sensing, GIS, and spatial analysis have been widely used for landscape pattern characterization (c.f. Turner and Gardner 1991; Steiniger and Hay 2009), there have been some major limitations when landscape ecologists utilize these techniques in their specific applications. First, with an increasing use of remote sensing and GIS in landscape ecological studies, many do not address some key technical issues, such as the potential uncertainty or error relating to landscape pattern measure, analysis, and modeling. For example, according to a recent study conducted by Newton et al. (2009), among 438 research papers published in the journal Landscape Ecology for the years 2004–2008, more than one third of these studies explicitly mentioned remote sensing but there was a frequent lack of important technical details, with approximately three-quarter failing to provide any assessment of uncertainty or error in image classification and mapping. Without these critical technical details, the quality and credibility of the scientific research would be open to question.

On the other hand, landscape ecologists were among the earliest groups who benefited from the use of remote sensing and geospatial techniques, as attested by Carl Troll’s pioneering work in African savannah landscape analysis through the use of aerial photographs (Troll 1939). However, few scholars in landscape ecology are fully aware of the latest development in these techniques. For example, existing landscape ecological studies that incorporated a remote sensing component have predominately used aerial photographs and Landsat imagery, with few targeting new types of data acquired in optical and microwave portions of the electromagnetic spectrum (Cohen and Goward 2004; Newton et al. 2009).

This chapter will provide an overview on the utilities of remote sensing and GIS techniques for landscape pattern characterization. While remote sensing allows a direct observation of landscape pattern and process at various scales, GIS provides a platform for integration and synthesis of theories and technologies in support of landscape pattern analysis and modeling (Fig. 11.1). The chapter is organized into several major sections, beginning with a discussion of the research status identifying some gaps in the use of remote sensing and GIS techniques in landscape ecology. Section 11.2 will examine remote sensing and landscape metrics for landscape pattern mapping and quantification. Section 11.3 will discuss some GIS-based spatial statistical analysis and modeling techniques for examining patterns, relationships, and emerging trends and for simulation and prediction. The last section will summarize the major findings. While the topics covered in this chapter span the entire spectrum in landscape analysis, our emphasis is not on a comprehensive review but on some methodological issues highlighting caveats and cautions when using geospatial techniques in landscape ecology. We believe the issues identified here can help landscape ecologists to better utilize remote sensing and GIS techniques in their specific applications.

Fig. 11.1
figure 1

A framework guiding the use of remote sensing and geospatial analysis for landscape pattern characterization. While remote sensing provides an indispensable source of data for landscape pattern mapping and quantification, GIS offers a platform for spatial data integration and synthesis that can help link observed landscape patterns with socio-ecological processes and for predicting landscape dynamics

Full size image

2 Landscape Pattern Mapping and Quantification

Landscape patterns consist of two components: composition and configuration (McGarigal and Marks 1995). The composition of a landscape refers to the number and occurrence of different types of landscape elements, which is a nonspatial measure of the landscape. And the configuration of a landscape refers to the physical distribution or structural arrangement of landscape elements, which is spatial by nature as it mainly deals with such an aspect as dimension, shape, or orientation of landscape elements. Together the spatial configuration and composition of landscape elements define the spatial pattern or heterogeneity of landscapes and play an important role in the ecological functionality and biological diversity (McGarigal and McComb 1995).

Although landscape elements can be represented as biotopes or habitats, they are commonly described in a more aggregated way, as land cover classes. Such land cover categories represent the interface between biophysical conditions and anthropogenic influences through time. Land cover types can be mapped through field surveys or remote sensing. The former can achieve an excellent mapping accuracy, especially for a relatively small geographic area. But this method can become less efficient when the study area is quite large or poorly accessible. Remote sensing, through sensors mounted on various aerospace platforms, can acquire photos or images over the visible, infrared and microwave portions of the electromagnetic spectrum within a short time period. They can revisit and acquire data over a specific study site, suitable for monitoring the dynamics of a landscape.

2.1 Land Cover Mapping

Critical to the entire landscape pattern analysis is the production of a land cover map by remote sensing, which relies upon the acquisition of remote sensor data and the identification of information extraction methods that are appropriate to the landscape characteristics under investigation (Yang and Lo 2002).

2.1.1 Remote Sensor Data Acquisition

Over the past several decades, data from various remote sensors have been used for land cover mapping. Earlier attempts were largely built upon the use of aerial photography. The acquisition of information on regional, national and global land cover has been the subject of numerous studies and evaluations since the early 1970s (e.g. Gaydos and Newland 1978; Jensen 1981; Haack et al. 1987; Yang and Lo 2002; Seto and Fragkias 2005; Lackner and Conway 2008; Bagan et al. 2012), which were largely stimulated by the launch of Earth Resources Technology Satellite-1 (ERTS-1; later renamed as Landsat) in 1972. Images acquired by the US Landsat program and French SPOT satellites are the principal sources of data for land cover mapping. In addition, large volumes of valuable data have been acquired by the Indian remote sensing satellites (IRS), the NASA Terra satellite, the China-Brazil Earth resources satellites (CBERS), and several European, Canadian, and Japanese satellites carrying active remote sensing devices. Moreover, remote sensor data with high spatial resolution acquired by several commercial satellites have become available since the late 1990s, which allow a substantial proportion of the basic land cover units to be distinguished.

Although various types of remote sensor data are widely available, it can be quite difficult to identify an appropriate type of imagery in a land cover mapping application. To help ease this data acquisition process, one should carefully consider several important issues. Firstly, before actual data acquisition and processing, an appropriate land cover classification scheme should be identified or established considering specific research objectives and the characteristics of landscapes under investigation. There are many different land cover classification schemes available, and for the sake of data interoperability, one should consider those being widely used whenever possible, such as the one developed by United States Geological Survey (USGS) for use with remote sensor data (Anderson et al. 1976).

Secondly, when deliberating over a range of remote sensor data, one should consider different aspects of image characteristics, such as spatial, spectral, radiometric, and temporal resolutions. Traditionally, land cover mapping emphasizes the importance of image spatial resolution but recent studies suggest that the other three resolutions are also critical (Jensen 2005). The choice of spatial resolution should be linked with the thematic accuracy specified in the land cover classification scheme. In other words, images with a coarse spatial resolution can be only used for mapping broad land cover categories, while higher resolution images should be used for detailed land cover mapping. Images with higher spectral resolution can be quite useful for mapping different types of vegetation. And multi-temporal images can not only be used to analyze landscape changes but also map some land cover categories that are affected by seasonal or phonological variations.

Lastly, remote sensor data acquisition cost is another important consideration. It can vary greatly, from unaffordably expensive to virtually free. While most of the airborne products (e.g. AIRSAR, ATLAS, and LIDAR) and high-resolution satellite images are quite expensive, certain licenses could lead to the availability of these products to selected users at a substantially reduced price or even for free. For example, Terra Image USA, the master distributor for the U.S. market of satellite imagery from the SPOT constellation of high resolution Earth imaging satellites, and the University of Californian at Santa Barbara have formed a research partnership since 2005 that allows the students and faculty members at the campus to freely access the entire SPOT image dataset with spatial resolution comparable to aerial photography (http://www.ia.ucsb.edu/pa/display.aspx?pkey=1311). Images with coarse resolution (e.g. MODIS and AVHRR) have been freely available, and beginning in early 2009, the entire Landsat image archive collected by USGS EROS Data Center over nearly the past four decades can be downloaded via internet at no charge (http://landsat.usgs.gov). This moderate-resolution image archive is unmatched in quality, details, coverage, and length, and has been an invaluable resource for examining natural and anthropogenic changes on Earth’s surface (Yang 2011a).

2.1.2 Information Extraction from Remote Sensor Data

Both visual interpretation and computer-based image classification techniques can be used to extract land cover information from remote sensor data. Through the combined use of various image elements with human intelligence, visual interpretation can achieve excellent mapping accuracy. And this technique can be implemented through on-screen digitizing in a GIS environment. However, it is manual by nature, and can be very much labor intensive for land over mapping over a large area. On the other hand, computer-based image classification can automate the entire land cover mapping process although some further research is still needed towards the operational use of this promising technique. In general, image classification is preferred over visual interpretation for land cover mapping over large areas (Jensen 2007).

Remote sensor image classification is largely based on the manipulation of statistical characteristics of one or more multispectral scenes. A variety of classification methods have been developed for remote sensing applications. By using specific criteria, these methods can be grouped in different ways: a parametric or a nonparametric classifier; a supervised or an unsupervised classifier; a “hard” or a “soft” classifier; a spectral or a spatial classifier; a sub-pixel, a pixel or an object-based classifier (Jensen 2005). Each of these methods has its own advantages and disadvantages, and there is no any single classifier that can be superior to another in all aspects (Duda et al. 2001).

Among all existing image classifiers, some are considered as advanced ones. However, most of the mapping applications have relied upon the use of a conventional classifier that largely manipulates a single image element (e.g. color or tone) in the multispectral pattern recognition (Jensen 2005). Conventional pattern recognition methods are largely based on the use of parametric statistics, which generally work well for medium-resolution scenes covering spectrally homogeneous areas, but not in heterogeneous regions or when scenes contain severe noises due to the increase of image spatial resolution. For years, substantial research efforts have been made to improve the performance of pattern classification for working with different types of remote sensor data and with spectrally complex landscapes. Some strategies have been developed as a result of such efforts: (1) the identification of various hybrid approaches that combine two or more classifiers, or incorporate pre- and post-classification image transformation and feature extraction techniques (e.g. Yang and Liu 2005b); (2) the development of ‘soft’ classifiers by introducing partial memberships for each pixel to accommodate the heterogeneous and imprecise nature of the real world (e.g. Shalan et al. 2003); (3) the decomposition of each pixel into independent endmembers or pure materials to conduct image classification at sub-pixel level (e.g. Verhoeye and Wulf 2002); (4) the incorporation of the spatial characteristics of the neighboring (contextual) pixels to develop object-oriented classification (e.g. Walker and Briggs 2007); (5) the fusion of multi-sensor, multi-temporal, or multi-source data for combining multiple spectral, spatial, and temporal features and ancillary information in the image classification (e.g. Tottrup 2004); and (6) the use of artificial intelligence technology, such as rule-based classifiers (e.g. Schmidt et al. 2004), artificial neural networks (e.g. Zhou and Yang 2011), and support vector machines (e.g. Yang 2011b), for pattern classification.

2.2 Landscape Pattern Quantification

Once a land cover map is available, landscape patterns can be quantified using landscape indices or metrics. These metrics are algorithms measuring the diversity, homogeneity or heterogeneity of a landscape. Although many earlier efforts have been made to identify various metrics that are meaningful for ecological functionality and biological diversity, it was the first primary release of the FRAGSTATS software package in 1995 that helped landscape ecologists to revolutionize the analysis of landscape structure (Kupfer 2012). FRAGSTATS defines eight major groups of structural or functional metrics: area/density/edge, shape, core area, isolation/proximity, contrast, contagion/interspersion, connectivity, and diversity (McGarigal and Marks 1995). Structural metrics measure the physical composition or configuration of a landscape without explicit reference to an ecological process, while functional metrics explicitly measure landscape pattern that is functionally relevant to the organism or process under consideration (McGarigal 2002). These metrics can be commonly measured at three levels: patch, class, and landscape. Patch-level metrics characterize the spatial context of patches, class-level metrics integrate over all patches of a given land cover type, and landscape-level metrics synthesize over the entire landscape. With the development of GIS software engineering, some metrics originally defined in FRAGSTATS have been incorporated into other software packages, such as Patch Analysis (http://www.cnfer.on.ca/SEP/patchanalyst/) and LANDISVIEW (http://kelab.tamu.edu/standard/restoration/restoration_tools.htm).

Landscape metrics offer an intuitive tool to measure landscape structure that can be linked with specific ecological processes. Because the input data (i.e. categorical land cover maps) can be derived from remote sensor imagery and the software toolkit is readily available, landscape metrics have been widely used for landscape pattern analysis. Nevertheless, there have been well documented concerns in the use of landscape metrics. Firstly, despite well-documented guidelines being given on the use of various metrics (e.g. McGarigal and Marks 1995; Haines-Young and Chopping 1996; McGarigal et al. 2002) and more than two decades of extensive research, interpreting landscape metrics beyond the simple quantitative description of landscape pattern remains difficult due to lack of generalizations concerning pattern-process relationships for many metrics (McGarigal 2002; Li and Wu 2004; Turner 2005; Fu et al. 2011).

Secondly, given the scale dependence of spatial heterogeneity, the statistic characteristics of landscape metrics can be affected by spatial extent and scale (e.g. Turner et al. 1989; Wu et al. 2002; Corry and Nassauer 2005). Turner et al. (1989) investigated the effects of changing grain size and extent of land cover data on spatial patterns. Landscape patterns were compared using metrics measuring diversity, dominance and contagion. They found that the diversity index decreased linearly as grain size increased, but dominance and contagion did not show such a linear relationship. While dominance and contagion increased with increasing spatial extent, diversity showed an erratic response. Corry and Nassauer (2005) found that the amount of linear habitat patches increased with increasing spatial resolution of land cover data. When many linear patches are present, metrics measuring landscape configuration may not be reliable; after resampling the land cover data into a coarse grain size, landscape configuration metrics became ecologically relevant. They found that composition metrics can be more useful in highly fragmented landscapes. A more comprehensive study was conducted by Wu et al. (2002) who examined how some common metrics responded to changing grain size and extent. Their found that the responses fell into three general categories: Predictable responses to changing scale with definable, simple scaling relationship; less predictable, staircase-like responses; and erratic responses without consistent scaling relationships.

Since landscape metrics are generally derived from categorical maps, the thematic resolution of land cover data and the classification scheme can also affect the statistical properties and behavior of landscape metrics. For example, Corry and Nassauer (2005) found that aggregation of land cover classes can reduce the number of patch types and thus increase the likelihood of contiguity. Huang et al. (2006) examined the sensitivity of two dozens of metrics to a number of land cover classes with different spatial patterns. They found that many metrics behaved predictably with increasing classification detail. At lower class numbers, metrics were quite sensitive to increasing classification detail. Their studies suggest the importance of land cover classification scheme in landscape pattern analysis.

Moreover, the quality of input data (i.e. land cover map) can affect the statistical characteristics of landscape metrics. For example, Wickham et al. (1997) tested the sensitivity of three common metrics (i.e. patch compaction, contagion, and fractal dimension) to land cover misclassification and differences in land cover composition. They found that differences in land cover composition need to be larger than the misclassification error in order to be confident that differences in landscape metrics are not due to misclassification. Corry and Nassauer (2005) noted that several data conversion procedures (e.g. vector to raster conversion, digitization of analogue data, and resampling) often introduce errors in land cover maps that can further affect the computed metric values. Langford et al. (2006) found that land cover classification error is not always a good predictor of errors in landscape metrics but maps with low misclassification rates can yield errors in landscape metrics with much larger magnitude and substantial variability. They also noted that certain image post-processing procedures such as smoothing might result in the underestimation of habitat fragmentation.

Lastly, with the development of landscape metrics over the past several decades, the choice of landscape metrics seems to be quite rich. But most higher-level metrics are derived from the same patch-level attributes, implying that many metrics can be partially or perfectly correlated with each other. This results in information redundancy. Moreover, a large number of metrics would be difficult to interpret and analyze. Practically, a small set of metrics that are not redundant but capture the major properties of a landscape are more desirable for a specific application. Ideally, these metrics should be consistent across spatial scale and time (e.g. Kelly et al. 2011). Recent studies suggest that landscape pattern can be characterized by using a small number of metrics but consensus has not reached on the choice of individual metrics (McGarigal 2002). For a specific application, one has to identify his/her own list of metrics to be used. This could be done through the use of landscape ecological principles or statistical methods or a combination of both. Many applications were based on the use of landscape ecological principles in selecting metrics (e.g. Zhang et al. 1997; Fuller 2001; Li et al. 2001; Baskent and Kadiogullari 2007). This approach may work well to reduce inherent redundancy but not empirical redundancy. On the other hand, statistical methods, such as principal component analysis, can help reduce data redundancy and select a parsimonious suite of independent metrics for landscape pattern analysis (e.g. Riitters et al. 1995; Herzog and Lausch 2001; Yang and Liu 2005a; Wang and Yang 2012).

Despite the above concerns, landscape metrics remain popular as they are seen by land managers and stakeholders as a simple tool for exploratory and descriptive analysis of spatio-temporal landscape pattern (Kupfer 2012). Encouragingly, recent advances in remote sensing and GIS software engineering allow more reliable land cover maps to be produced from multi-resolution imagery, and landscape metrics to be easily calculated through readily available software packages. The latest release of FRAGSTATS (4.0) extends the calculation of landscape metrics beyond categorical maps and into continuous maps through a moving window approach (McGarigal et al. 2012). This technical breakthrough can help minimize the variation of statistical properties of landscape metrics due to the modifiable area unit problem (MAUP). Moreover, the development of metrics rooted in graph, network, and circuit theory offers the promise of a more ecologically oriented approach to quantifying landscape pattern and process (Kupfer 2012).

3 Landscape Pattern-Process Analysis and Modeling

Landscape pattern characterization not only aims to map and quantify landscape spatial patterns, but also seeks to interpret them in relation to specific socio-ecological processes. Therefore, landscape pattern characterization serves as the centerpiece to address the spatial interaction between landscape patterns and ecological processes that can help understand the intrinsic causality and underlying landscape dynamics. On one hand, landscape patterns are strongly influenced by ecological processes that include all aspects of biological, chemical, physical, hydrological, and human-dimensional processes of the ecosystem. On the other hand, landscape patterns can significantly affect ecological processes and landscape dynamics across spatial and temporal scales. To assess the two ways of relationship, one needs to integrate the ecologically-oriented, vertical approach with the geographically-oriented, horizontal approach that incorporates aggregated, integrative environmental parameters (Bastian 2001). While a rich pool of landscape ecological literatures have discussed specific pattern-process relationships (e.g. Turner 1989; McGarigal and McComb 1995; Wu 2006), here we direct our attention on some generic methodological issues for integration and synthesis in a GIS environment.

3.1 Linking Landscape Patterns with Processes

Relating landscape spatial patterns with ecological processes involves the integration and synthesis of theories and technologies across spatial and temporal scales, which can be pursued through either the qualitative or quantitative approach. Given the scope of this chapter, we herewith limit our further discussion to some technical issues when using the quantitative approach, especially multivariate statistical analysis, to address the pattern-process linkage at the broad scale level. Multivariate statistical methods including linear and nonlinear multiple regression can be used to examine the pattern-process relationship. When constructing a multivariate regression model to assess how ecological processes would influence landscape patterns, each of the landscape metrics should be treated as a dependent variable, while various biological, chemical, physical, hydrological, and human-dimensional variables as independent variables (e.g. Lo and Yang 2002). On the other hand, when examining how landscape patterns would affect ecological processes, landscape metrics should be treated as independent variables, while each of specific ecological indicators as a dependent variable (e.g. Yang 2012). Several statistical methods, such as ordinary least squares (OLS) regression and logistic regression, can be used to determine the pattern-process relationship.

There are several issues one should pay close attention to when using the above empirical method to study the pattern-process relationship. Firstly, since all dependent and candidate independent variables are usually aggregated by areal units, how these units are defined in terms of scale and zoning systems can affect the results of parameter estimates in multivariate statistical analysis. This is actually the modifiable areal unit problem (MAUP) that has been extensively discussed in geography and spatial science literature (e.g. Openshaw 1984; Fotheringham and Wong 1991; Jelinski and Wu 1996; Dark and Bram 2007). The modifiable areal unit problem is considered as a fundamental problem inherent in the studies using spatially aggregated data because the results are always affected by the areal units used (Openshaw 1984). It can be essentially unpredictable in its intensity and effects in multivariate statistical analysis, and is therefore a much greater problem than in univariate or bivariate analysis (Fotheringham and Wong 1991). While the variations of statistical analysis due to the aggregation of smaller areal units into regions are generally well understood (e.g. Fotheringham and Wong 1991), the zoning problem is much less well understood (Jelinski and Wu 1996).

The number of observations or sample size is very important for multivariate statistical analysis. For geographically referenced data, we normally consider the number of spatial observation units being equivalent to sample size from a statistical perspective. In general, the number of observation units should be 5–10 times the number of candidate independent variables (Brace et al. 2012). For example, if a specific multivariate statistical model intends to include 5 independent variables, the number of observation units should be at least 25–50. Too many or too few units could lead statistical models to be over-fitting.

Because the total number of observation units is actually quite limited for many applications, one should be careful when selecting candidate independent variables to be included in multivariate statistical analysis. As mentioned before, many landscape metrics are perfectly or partially correlated with each other, which can cause information duplication. Therefore, when using landscape metrics as candidate independent variables to assess their impacts upon specific ecological processes, it is important to identify a small number of landscape metrics that are not duplicated but capture the major landscape properties.

A preprocessing procedure should be conducted for all dependent or independent variables. Because multivariate statistical analysis is sensitive to the variance of samples and data distribution, one should avoid using the raw data directly, particularly for those variables with a large statistical variance. For some environmental variables, such as water or air quality, one should use their average measurements by month, quarter or year. For landscape composition metrics, one should use the relative proportion rather than the total number. Before actual statistical analysis, raw data should be logarithmically transformed to improve their normality.

Before a statistical model is established, one should check the normality, multicollinearity, and spatial autocorrelation of independent variables. Data normality can be checked through the Kolmogorov-Smirnov test or the graphic approach using histograms and QQ plots. For some variables that do not show a clear normal distribution, one can transform the raw data logarithmically to improve the data normality. Any statistical models that show strong multicollinearity among the independent variables should be used with caution. The spatial autocorrelation can be computed by using Moran I or Geary C. If a strong spatial autocorrelation exists, one should use the strategies suggested by Legendre (1993) to reduce the spatial dependence.

When assessing the performance of different statistical models, one should pay attention on the number of independent variables included. In general, the explanatory power tends to be higher for a model that includes more independent variables. This suggests that any meaningful comparison of model performance should be based on the identical number of independent variables.

Finally, multiple regression analysis can make use of all or a subset of the sample in parameter estimation when building a statistical model. Although most of existing studies have built upon the use of all sample data in parameter estimations, recent development in spatial statistics suggests using a subset of the sample data can help reveal the variation of the cause-effect relationship across space (Fotheringham et al. 2002). This localized regression technique called geographically weighted regression has been included in some leading GIS software packages, such as ArcGIS 10. Although the software tool is readily available, it can be very much data demanding, particularly for some environmental data that can be only acquired through in situ measurements.

3.2 Modeling and Predicting Landscape Dynamics

Models developed to simulate and predict landscape dynamics as a physical process have become quite popular in recent years. This is necessary to understand the dynamics of complex ecosystems and to evaluate the consequences of landscape change on the environment (Yang and Lo 2003; Sutherland 2006). Ecologists were among the earliest groups who have demonstrated a strong interest in developing spatially explicit models to predict the impacts of different landscape configurations on plant and animal populations (Kareiva and Wennergren (1995). While many methods have been developed to predict the impacts at the species or population level (c.f. Sutherland 2006), here we direct our attention on the spatially explicit, dynamic models that are designed to work at the community or ecosystem level.

Over the past several decades, a variety of spatially explicit models have been developed by different communities, which can be either stochastic, such as the logit (e.g. Hu and Lo 2007), Markov (e.g. Myint and Wang 2006), cellular automata (e.g. Hagen-Zanker and Lajoie 2008), and agent-based models (e.g. Robinson et al. 2012), or processes based, such as dynamic ecosystem models (e.g. Euskirchen et al. 2006). Although these models are different by their underlying mechanism, they share many commonalities. The common approaches are the use of transition probabilities in a class transition matrix, multinominal logit methods, cellular or agent-based modeling, and GIS weighted overlay approach. These models consider different constraints by various biophysical, economical, and social parameters. Some of these parameters include land transition probabilities, topography, environmental protection, forest properties, transportation, population, economic indicators, human behavior, and policy. The role of remote sensing and GIS is indispensible in the entire model development process from model conceptualization to implementation that includes input data preparation, model calibration, and model validation. Comprehensive reviews on various spatial modeling approaches are beyond our scope in this chapter, and readers should refer to several other relevant publications (e.g. Agarwal et al. 2002; Parker et al. 2003; Verburg et al. 2004; An 2012).

While spatially explicit modeling approaches can help extend landscape analysis beyond quantitative pattern description and into the area of forecasting and prediction, there is a need to accept the limitations of prediction (Sutherland 2006). Most of the modeling efforts are technically driven, the justification and verification of ecological concepts and theories are not adequate. Some game-like simulators consider only a few untested factors, which often perform poorly when predicting for the future. On the other hand, some models tend to be too ambitious, considering too many variables, which are not easily parameterized. Many simulators do not contain components of model calibration and verification, and hence their results are generally not good enough for prediction and forecasting.

4 Concluding Remarks

In this chapter, we have reviewed the utilities of remote sensing and geospatial analysis for landscape pattern characterization, a fundamental pursuit in landscape ecology. While landscape ecologists were among the earliest groups benefiting from the use of remote sensing and GIS techniques, few of them are fully aware of the latest development in these technical areas. Essentially global coverage of remote sensor data with individual pixels ranging from sub-meters to a few kilometers can help make connections across various levels of landscape pattern analysis. The development of advanced image classification techniques and GIS software engineering has helped landscape ecologists to revolutionize the analysis of landscape structure by using pattern metrics. Nevertheless, the statistical properties and behavior of landscape metrics across a range of classification schemes and landscapes, as well as their sensitivity to changing landscape patterns, are still not fully understood. While GIS-based spatial analysis and modeling techniques can help examine patterns, relationships, emerging trends, and dynamics, landscape ecologists should also pay attention on some outstanding issues that we have discussed in this chapter. And landscape ecologists should fully understand both the strengths and weaknesses of remote sensing and geospatial techniques in order to better utilize these techniques in their specific applications. Finally, there is an increasing need to collaborate between the disciplines of landscape ecology and geospatial science that would not only lead to landscape ecology being taken more seriously but also help expand the inferential capabilities of geospatial research.