Patterns and scale relations among urbanization measures in Stockholm,Sweden

In this study we measure urbanization based on a diverse set of 21 variables ranging from landscape indices to demographic factors such as income and land ownership using data from Stockholm, Sweden. The primary aims were to test how the variables behaved in relation to each other and if these patterns were consistent across scales. The variables were mostly identified from the literature and limited to the kind of data that was readily accessible. We used GIS to sample the variables and then principal component analyses to search for patterns among them, repeating the sampling and analysis at four different scales (250 × 250, 750 × 750, 1,250 × 1,250 and 1,750 × 1,750, all in meters). At the smallest scale most variables seemed to be roughly structured along two axes, one with landscape indices and one mainly with demographic factors but also impervious surface and coniferous forest. The other land-cover types did not align very well with these two axes. When increasing the scale this pattern was not as obvious, instead the variables separated into several smaller bundles of highly correlated variables. Some pairs or bundles of variables were correlated on all scales and thus interchangeable while other associations changed with scale. This is important to keep in mind when one chooses measures of urbanization, especially if the measures are indices based on several variables. Comparing our results with the findings from other cities, we argue that universal gradients will be difficult to find since city shape and size, as well as available information, differ greatly. We also believe that a multivariate gradient is needed if you wish not only to compare cities but also ask questions about how urbanization influences the ecological character in different parts of a city.     

Surface metrics: an alternative to patch metrics for the quantification of landscape structure

Modern landscape ecology is based on the patch mosaic paradigm, in which landscapes are conceptualized and analyzed as mosaics of discrete patches. While this model has been widely successful, there are many situations where it is more meaningful to model landscape structure based on continuous rather than discrete spatial heterogeneity. The growing field of surface metrology offers a variety of surface metrics for quantifying landscape gradients, yet these metrics are largely unknown and/or unused by landscape ecologists. In this paper, we describe a suite of surface metrics with potential for landscape ecological application. We assessed the redundancy among metrics and sought to find groups of similarly behaved metrics by examining metric performance across 264 sample landscapes in western Turkey. For comparative purposes and to evaluate the robustness of the observed patterns, we examined 16 different patch mosaic models and 18 different landscape gradient models of landscape structure. Surface metrics were highly redundant, but less so than patch metrics, and consistently aggregated into four cohesive clusters of similarly behaved metrics representing surface roughness, shape of the surface height distribution, and angular and radial surface texture. While the surface roughness metrics have strong analogs among the patch metrics, the other surface components are largely unique to landscape gradients. We contend that the surface properties we identified are nearly universal and have potential to offer new insights into landscape pattern–process relationships. Electronic supplementary material  The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

Modeling grain-size dependent bias in estimating forest area: a regional application

A better understanding of scaling-up effects on estimating important landscape characteristics (e.g. forest percentage) is critical for improving ecological applications over large areas. This study illustrated effects of changing grain sizes on regional forest estimates in Minnesota, Wisconsin, and Michigan of the USA using 30-m land-cover maps (1992 and 2001) produced by the National Land Cover Datasets. The maps were aggregated to two broad cover types (forest vs. non-forest) and scaled up to 1-km and 10-km resolutions. Empirical models were established from county-level observations using regression analysis to estimate scaling effects on area estimation. Forest percentages observed at 30-m and 1-km land-cover maps were highly correlated. This intrinsic relationship was tested spatially, temporally, and was shown to be invariant. Our models provide a practical way to calibrate forest percentages observed from coarse-resolution land-cover data. The models predicted mean scaling effects of 7.0 and 12.0% (in absolute value with standard deviations of 2.2 and 5.3%) on regional forest cover estimation (ranging from 2.3 and 2.5% to 11.1 and 23.7% at the county level) with standard errors of model estimation 3.1 and 7.1% between 30 m and 1 km, and 30 m and 10 km, respectively, within a 95% confidence interval. Our models improved accuracy of forest cover estimates (in terms of percent) by 63% (at 1-km resolution) and 57% (at 10-km resolution) at the county level relative to those without model adjustment and by 87 and 84% at the regional level in 2001. The model improved 1992 and 2001 regional forest estimation in terms of area for 1-km maps by 15,141 and 7,412 km² (after area weighting of all counties) respectively, compared to the corresponding estimates without calibration using 30 m-based regional forest areas as reference.     

Determining an appropriate minimum mapping unit in vegetation mapping for ecosystem restoration: a case study from the Everglades,USA

This paper documents the analyses that were conducted with regards to investigating an appropriate Minimum Mapping Unit (MMU) to be used to capture the potential changes in vegetation patterns for a 10,924 square km restoration project being conducted in south Florida, USA. Spatial landscape and class metrics that were shown to change predictably with increasing grain size were adopted from previous studies and applied to a multi-scale analysis. Specifically, this study examines the effects of changing grain size on landscape metrics, utilizing empirical data from a real landscape encompassing 234,913 ha of south Florida’s Everglades. The objective was to identify critical thresholds within landscape metrics, which can be used to provide insight in determining an appropriate MMU for vegetation mapping. Results from this study demonstrate that vegetation heterogeneity will exhibit dissimilar patterns when investigating the loss of information within landscape and class metrics, as grain size is increased. These results also support previous findings that suggest that landscape metric “scalograms” (the response curves of landscape metrics to changing grain size), are more likely to be successful for linking landscape pattern to ecological processes as both pattern and process in ecological systems often operate on multiple scales. This study also incorporates an economic cost for various grain dependant vegetation mapping scales. A final selection of the 50 × 50 m grain size for mapping vegetation was based on this study’s investigation of the “scalograms”, the costs, and a composite best professional judgment of seasoned scientists having extensive experience within these ecosystems.     

The influence of patch-delineation mismatches on multi-temporal landscape pattern analysis

Investigations of land-cover change often employ metrics designed to quantify changes in landscape structure through time, using analyses of land cover maps derived from the classification of remote sensing images from two or more time periods. Unfortunately, the validity of these landscape pattern analyses (LPA) can be compromised by the presence of spurious change, i.e., differences between map products caused by classification error rather than real changes on the ground. To reduce this problem, multi-temporal time series of land-cover maps can be constructed by updating (projecting forward in time) and backdating (projecting backward in time) an existing reference map, wherein regions of change are delineated through bi-temporal change analysis and overlaid onto the reference map. However, this procedure itself creates challenges, because sliver patches can occur in cases where the boundaries of the change regions do not exactly match the land-cover patches in the reference map. In this paper, we describe how sliver patches can inadvertently be created through the backdating and updating of land-cover maps, and document their impact on the magnitude and trajectory of four popular landscape metrics: number of patches (NP), edge density (ED), mean patch size (MPS), and mean shape index (MSI). In our findings, sliver patches led to significant distortions in both the value and temporal behaviour of metrics. In backdated maps, these distortions caused metric trajectories to appear more conservative, suggesting lower rates of change for ED and inverse trajectories for NP, MPS and MSI. In updated maps, slivers caused metric trajectories to appear more extreme and exaggerated, suggesting higher rates of change for all four metrics. Our research underscores the need to eliminate sliver patches from any study dealing with multi-temporal LPA.     

Impacts of road corridors on urban landscape pattern: a gradient analysis with changing grain size in Shanghai, China

Urbanization is one of the most important driving forces for land use and land cover change. Quantifying urban landscape pattern and its change is fundamental for monitoring and assessing ecological and socioeconomic consequences of urbanization. As the largest city in the country, Shanghai is now the fastest growing city in China. Using land use data set of 2002 and combining gradient analysis with landscape metrics, we analyzed landscape pattern of Shanghai with increasing grain size to study the impacts of road corridors on urban landscape pattern. Landscape metrics were computed along a 51×9 km² transect cutting across Shanghai with a moving window. The results showed that the urban landscape pattern of Shanghai was greatly changed when road corridors were merged with urban patches and the variation of patch density would alter when grain size changed. As a linear land use type, road corridors exhibited a different spatial signature comparing with other land use types and distinctive behavior with increasing grain size. Merging road and urban patches resulted in a sharp reduction in patch density, mainly caused by segmentation of roads corridors. The results suggested that grain size around 7.5 m might be optimal for urban landscape analysis. Landscape patch density is significantly correlated with road percent coverage and the most important effect of road corridors in urban landscape is increased habitat fragmentation.     

Movement behavior in response to landscape structure: the role of functional grain

Landscape structure can influence the fine-scale movement behavior of dispersing animals, which ultimately may influence ecological patterns and processes at broader scales. Functional grain refers to the finest scale at which an organism responds to spatial heterogeneity among patches and extends to the limits of its perceptual range. To determine the functional grain of a model insect, red flour beetle (Tribolium castaneum), we examined its movement behavior in response to experimental flour landscapes. Landscape structure was varied by manipulating habitat abundance (0%, 10%, 30%, and 100%) and grain size of patches (fine-2 × 2 cm, intermediate-5 × 5 cm, and coarse-10 × 10 cm) in 50 × 50 cm landscapes. Pathway metrics indicated that beetles used a similar proportion of all landscape types. Several pathway metrics indicated a graded response from the fine to the coarse grain landscape. Lacunarity analysis of beetle pathways indicated a non-linear change in space use between the fine and intermediate landscapes and the coarse-grained landscape. Beetles moved more slowly and tortuously (with many turns), and remained longer in both the overall landscape and individual patches, in fine-grained compared to coarse-grained landscapes. Our research demonstrates how detailed examination of movement pathways and measures of lacunarity can be useful in determining functional grain. Spatially explicit, organism-centered studies focusing on behavioral responses to different habitat configurations can serve as an important first step to identify behavioral rules of movement that may ultimately lead to more accurate predictions of space use in landscapes.     

Impact of error on landscape pattern analyses performed on land-cover change maps

Researchers have emphasized the value of linking observed patterns of land-cover change to the processes driving changes in land-use to explain the dynamics of a land change system. The association of pattern and process requires an accurate quantification of the spatial characteristics of land-cover change. The objective of this research is to assess the impact of error on the accuracy of landscape pattern analyses performed on maps of change. Simulation was used to develop of a series of error-free and error-perturbed change maps, which varied with respect to the pattern of change occurring between the time-1 and time-2 land-cover maps and the patterns of error associated with the time-1 and time-2 land-cover maps. A variety of change and error patterns were examined. The error-free and error-perturbed change maps were compared by calculating landscape pattern metrics, which revealed the degree to which error altered the pattern of change. The introduction of error notably changed the structure of the persistent and transitioning classes, with metrics indicating a more fragmented and variable landscape under error. Agreement between the error-free and error-perturbed maps improved when a greater amount of change occurred within the time-series, change was concentrated at the boundaries of land-cover classes and when time-2 errors were increasingly correlated to their time-1 counterparts. These results have several implications for change pattern analyses given the fundamental nature of land-cover change.     

Analysing land-cover changes in relation to environmental variables in Hesse,Germany

Land-use and land-cover changes affect ecological landscape functions and processes. Hence, landscape ecologists have a central interest in a comprehensive understanding of such changes. Our study focuses on the relationships between environmental conditions and agricultural land-cover changes. We present a method to (i) characterise the major spatial-temporal processes of land-cover changes, (ii) identify the correlations between environmental attributes and land-cover changes and (iii) derive potential environmental drivers of land-cover changes in a German marginal rural landscape. The method was applied to study land-cover dynamics from 1945 to 1998 in the districts of Erda, Steinbrücken and Eibelshausen, situated in the marginal rural landscape of the Lahn-Dill Highlands, Germany. We employed land-cover data gained by the interpretation of multi-temporal aerial photographs. Various environmental variables were introduced into the analyses. We identified physical landscape attributes (elevation, slope, aspect, available water capacity and soil texture) and structural landscape dimensions (patch size, patch shape and distance between patch and nearest settlement). With the aid of GIS, K-means partitioning and canonical correspondence analysis, we investigated land-cover trajectory types, land-cover transitions at individual time intervals and their relationships to these environmental variables. Our results show that, between 1945 and 1998, land-cover changes correlated with the physical attributes of the underlying landscape. On the other hand, the structural landscape dimensions correlated with land cover only in periods of minor land-cover changes (1972–98). Greater diversity of physical landscape attributes is correlated with greater land-cover dynamics. Besides the important influence of socio-economic factors, land-cover changes in the study areas took place within the relatively stable physical constraints of the underlying landscape.This revised version was published online in May 2005 with corrections to the Cover Date.     

Scale dependence in quantification of land-cover and biomass change over Siberian boreal forest landscapes

We investigated the influence of remote sensing spatial resolution on estimates of characteristic land-cover change (LCC) and LCC-related above-ground biomass change (Δbiomass) in three study sites representative of the East Siberian boreal forest. Data included LCC estimated using an existing Landsat-derived land-cover dataset for 1990 and 2000, and above-ground standing biomass stocks simulated by the FAREAST forest succession model and applied on a pixel basis. At the base 60 m resolution, several landscape pattern metrics were derived to describe the characteristic LCC types. LCC data were progressively degraded to 240, 480, and 960 m. LCC proportions and Δbiomass were derived at each of the coarser resolutions and scale dependences of LCC and Δbiomass were analyzed. Compared to the base 60 m resolution, the Logged LCC type was highly scale dependent and was consistently underestimated at coarser resolutions. The Burned type was under- or over-estimated depending strongly on its patch size. Estimated at the base 60 m resolution, modeled biomass increased in two sites (i.e., 3.0 and 6.4 Mg C ha⁻¹ for the Tomsk and Krasnoyarsk sites, respectively) and declined slightly in one site (i.e., −0.5 Mg C ha⁻¹ for the Irkutsk site) between the two dates. At the degraded resolutions, the estimated Δbiomass increased to 3.3 and 7.0 Mg C ha⁻¹ for the Tomsk and Krasnoyarsk sites, while it declined to −0.8 Mg C ha⁻¹ for the Irkutsk site. Results indicate that LCC and Δbiomass values may be progressively amplified in either direction as resolution is degraded, depending on the mean patch size (MPS) of disturbances, and that the error of LCC and Δbiomass estimates also increases at coarser resolutions.     

Landscape Fragmentation and Ice Storm Damage in Eastern Ontario Forests

With return times between 20 and 100 years, ice storms are a primary disturbance type for temperate forests of eastern North America. Many studies have been conducted at the forest patch and plot scales to examine relations between damage and variables describing site, composition and structure. This paper presents results from a landscape scale study of fragmentation relations with damage in eastern Ontario forests. Data previously collected for two independent and spatially non-overlapping patch level damage studies were used. A Generalized Linear Model (GLM) was used to analyse relations between damage and fragmentation metrics representing patch isolation, edge density, and the relative size and distribution of patches in the landscape. The metrics were applied using spatial extents of 1 × 1 km and 4 × 4 km, following analyses of the variability of numbers of patches and of the lacunarity of forest patterns over a range of extents. The results showed that patch isolation, as measured by the mean Euclidean distance between patches (ENN) was significantly related to damage.     

The Lorenz curve: a suitable framework to define satisfactory indices of landscape composition

Context

Patch diversity, evenness and dominance are important metrics of landscape composition. They have been traditionally measured using indices based on Shannon’s information entropy (H) and Simpson’s concentration statistic (λ).

Objectives

The main objectives of this study are: (1) to show that the Lorenz curve is an appropriate framework to understand and measure patch dominance, evenness and diversity; (2) to show that Lorenz-compatible indices have better mathematical behavior than H-based and λ-based indices.

Methods

Thirteen different hypothetical landscapes were created to assess landscape composition with the Lorenz curve and to compare the mathematical behavior of Lorenz-compatible indices with that of H-based and λ-based indices.

Results

The Lorenz curve is a suitable framework to understand and measure patch dominance, evenness and diversity due to four relevant equivalences: (1) patch dominance?=?the separation of the Lorenz curve from the 45-degree line of perfect patch evenness; (2) patch evenness?=?1 ? patch dominance; (3) patch diversity (eliminated by patch dominance)?=?patch richness?×?patch dominance; (4) patch diversity (preserved by patch evenness)?=?patch richness?×?patch evenness. Accordingly, patch diversity/patch richness?=?1???patch dominance and land-cover concentration?=?1/patch diversity.

Conclusions

Lorenz-compatible indices have better mathematical behavior than H-based and λ-based indices, exhibiting greater coherence and objectivity when measuring patch dominance, evenness and diversity.

     

Landscape patterns from mathematical morphology on maps with contagion

The perceived realism of simulated maps with contagion (spatial autocorrelation) has led to their use for comparing landscape pattern metrics and as habitat maps for modeling organism movement across landscapes. The objective of this study was to conduct a neutral model analysis of pattern metrics defined by morphological spatial pattern analysis (MSPA) on maps with contagion, with comparisons to phase transitions (abrupt changes) of patterns on simple random maps. Using MSPA, each focal class pixel on a neutral map was assigned to one of six pattern classes—core, edge, perforated, connector, branch, or islet—depending on MSPA rules for connectivity and edge width. As the density of the focal class (P) was increased on simple random maps, the proportions of pixels in different pattern classes exhibited two types of phase transitions at threshold densities (0.41 ≤ P ≤ 0.99) that were predicted by percolation theory after taking into account the MSPA rules for connectivity and edge width. While there was no evidence of phase transitions on maps with contagion, the general trends of pattern metrics in relation to P were similar to simple random maps. Using an index P for comparisons, the effect of increasing contagion was opposite that of increasing edge width.     

Statistical models of landscape pattern metrics,with applications to regional scale dynamic forest simulations

Forest managers in Canada need to model landscape pattern or spatial configurationoverlarge (100,000 km²) regions. This presents a scalingproblem, as landscape configuration is measured at a high spatial resolution,but a low spatial resolution is indicated for regional simulation. We present astatistical solution to this scaling problem by showing how a wide range oflandscape pattern metrics can be modelled from low resolution data. Our studyarea comprises about 75,000 km² of boreal mixedwoodforest in northeast Alberta, Canada. Within this area we gridded a sample of 84digital forest cover maps, each about 9500 ha in size, to aresolution of 1 ha and used FRAGSTATS to compute a suite oflandscape pattern metrics for each map. We then used multivariate dimensionreduction techniques and canonical correlation analysis to model therelationship between landscape pattern metrics and simpler stand table metricsthat are easily obtained from non-spatial forest inventories. These analyseswere performed on four habitat types common in boreal mixedwood forests: youngdeciduous, old deciduous, white spruce, and mixedwood types. Using only threelandscape variables obtained directly from stand attribute tables (totalhabitatarea, and the mean and standard deviation of habitat patch size), ourstatistical models explained more than 73% of the joint variation in fivelandscape pattern metrics (representing patch shape, forest interior habitat,and patch isolation). By PCA, these five indices captured much of the totalvariability in the rich set of landscape pattern metrics that FRAGSTATS cangenerate. The predictor variables and strengths of association were highlyconsistent across habitat classes. We illustrate the potential use of suchstatistical relationships by simulating the regional, cumulative effects ofwildfire and forest management on the spatial arrangement of forest patches,using non-spatial stand attribute tables.This revised version was published online in May 2005 with corrections to the Cover Date.     

Neutral models for testing landscape hypotheses

Neutral landscape models were originally developed to test the hypothesis that human-induced fragmentation produces patterns distinctly different from those associated with random processes. Other uses for neutral models have become apparent, including the development and testing of landscape metrics to characterize landscape pattern. Although metric development proved to be significant, the focus on metrics obscured the need for iterative hypothesis testing fundamental to the advancement of the discipline. We present here an example of an alternative neutral model and hypothesis designed to relate the process of landscape change to observed landscape patterns. The methods and program, QRULE, are described and options for statistical testing outlined. The results show that human fragmentation of landscapes results in a non-random association of land-cover types that can be describe by simple statistical methods. Options for additional landscape studies are discussed and access to QRULE described in the hope that these methods will be employed to advance our understanding of the processes that affect the structure and function in human dominated landscapes.     

Landscape fragmentation,land-use legacy and propagule pressure promote plant invasion on coastal dunes: a patch-based approach

Coastal dunes and sand areas are reported to be among the habitats most invaded by alien species in Europe. Landscape pattern could be a significant driver in invasion processes in parallel with land-use legacy. Fragmentation of natural habitats combined with the availability of propagules from the surrounding matrix may enhance the invisibility of ecological communities. Based on multitemporal land cover maps (1954–2008) and a floristic database, we analyzed how habitat fragmentation, propagule pressure and land-use legacy have affected alien plants’ presence and richness on natural dune patches along the Lazio Coast (Central Italy). Floristic data were derived from an existing geo-database of random vegetation plots (64 m²). A set of landscape patch-based metrics, considered to be adequate proxies of the main processes affecting alien invasion and richness, was calculated. First, we fit a generalized linear model (GLM) with binomial errors to assess which landscape metrics are influencing patch invasion. Second, we extracted invaded patches and, with GLMs, we investigated how landscape metrics affect average alien species richness. Alien invasion and alien richness seem to be affected by different processes: although alien invasion of each patch is strongly associated with its land-use legacy, the richness of aliens is more affected by landscape fragmentation and by the propagule pressure to which patch is exposed. By integrating spatial and temporal landscape metrics with floristic data, we were able to disentangle the relations of landscape fragmentation, propagule pressure and land-use legacy with the presence and richness of alien plants. The methodological approach here adopted could be easily extended to other alien species and ecosystems, offering scientifically sound support to prevent the high economic costs derived from both the control and the eradication of aliens.     

A method for the use of landscape metrics in freshwater research and management

Freshwater research and management efforts could be greatly enhanced by a better understanding of the relationship between landscape-scale factors and water quality indicators. This is particularly true in urban areas, where land transformation impacts stream systems at a variety of scales. Despite advances in landscape quantification methods, several studies attempting to elucidate the relationship between land use/land cover (LULC) and water quality have resulted in mixed conclusions. However, these studies have largely relied on compositional landscape metrics. For urban and urbanizing watersheds in particular, the use of metrics that capture spatial pattern may further aid in distinguishing the effects of various urban growth patterns, as well as exploring the interplay between environmental and socioeconomic variables. However, to be truly useful for freshwater applications, pattern metrics must be optimized based on characteristic watershed properties and common water quality point sampling methods. Using a freely available LULC data set for the Santa Clara Basin, California, USA, we quantified landscape composition and configuration for subwatershed areas upstream of individual sampling sites, reducing the number of metrics based on: (1) sensitivity to changes in extent and (2) redundancy, as determined by a multivariate factor analysis. The first two factors, interpreted as (1) patch density and distribution and (2) patch shape and landscape subdivision, explained approximately 85% of the variation in the data set, and are highly reflective of the heterogeneous urban development pattern found in the study area. Although offering slightly less explanatory power, compositional metrics can provide important contextual information.     

Ecosystem services of regulation and support in Amazonian pioneer fronts: searching for landscape drivers

Landscape dynamics result from forestry and farming practices, both of which are expected to have diverse impacts on ecosystem services (ES). In this study, we investigated this general statement for regulating and supporting services via an assessment of ecosystem functions: climate regulation via carbon sequestration in soil and plant biomass, water cycle and soil erosion regulation via water infiltration in soil, and support for primary production via soil chemical quality and water storage. We tested the hypothesis that patterns of land-cover composition and structure significantly alter ES metrics at two different scales. We surveyed 54 farms in two Amazonian regions of Brazil and Colombia and assessed land-cover composition and structure from remote sensing data (farm scale) from 1990 to 2007. Simple and well-established methods were used to characterize soil and vegetation from five points in each farm (plot scale). Most ES metrics were significantly correlated with land-use (plot scale) and land-cover (farm scale) classifications; however, spatial variability in inherent soil properties, alone or in interaction with land-use or land-cover changes, contributed greatly to variability in ES metrics. Carbon stock in above-ground plant biomass and water infiltration rate decreased from forest to pasture land covers, whereas soil chemical quality and plant-available water storage capacity increased. Land-cover classifications based on structure metrics explained significantly less ES metric variation than those based on composition metrics. Land-cover composition dynamics explained 45 % (P < 0.001) of ES metric variance, 15 % by itself and 30 % in interaction with inherent soil properties. This study describes how ES evolve with landscape changes, specifying the contribution of spatial variability in the physical environment and highlighting trade-offs and synergies among ES.     

A factor analysis of landscape pattern and structure metrics

Fifty-five metrics of landscape pattern and structure were calculated for 85 maps of land use and land cover. A multivariate factor analysis was used to identify the common axes (or dimensions) of pattern and structure which were measured by a reduced set of 26 metrics. The first six factors explained about 87% of the variation in the 26 landscape metrics. These factors were interpreted as composite measures of average patch compaction, overall image texture, average patch shape, patch perimeter-area scaling, number of attribute classes, and large-patch density-area scaling. We suggest that these factors can be represented in a simpler way by six univariate metrics - average perimeter-area ratio, contagion, standardized patch shape, patch perimeter-area scaling, number of attribute classes, and large-patch density-area scaling.     

The influence of scale and the spatial characteristics of landscapes on land-cover mapping using remote sensing

Statistical analyses provide a means for assessing relationships between landscape spatial pattern and errors in the estimates of cover-type proportions as land-cover data are aggregated to coarser scales. Results from a multiple-linear regression model suggest that as patch sizes, variance/mean ratio, and initial proportions of cover types increase, the proportion error moves in a positive direction and is governed by the interaction of the spatial characteristics and the scale of aggregation. However, the standard linear model does not account for the different directions of scale-dependent proportion error since some classes become larger and others become smaller as the scene is aggregated. Addition of indicator variables representing class-type significantly improves the performance by allowing the model to respond differently to different classes. A regression tree model provides a much simpler fit to the complex scaling behavior through an interaction between patch size and aggregation scale. An understanding of the relationships between landscape pattern, scale, and proportion error may advance methods for correcting land-cover area estimates. Such methods could also facilitate high-resolution calibration and validation of coarse-scale remote-sensing-based land-cover mapping algorithms. Ongoing initiatives to produce global land-cover datasets from remote sensing, such as efforts within the IGBP and the EOS MODIS Land-Team, include significant emphasis on high level calibration and validation activities of this nature.