Ferenc Csillag

Destriping multisensor imagery with moment matching

Image destriping is necessary due to sensor-to-sensor variation within instruments. This has most often been done by assuming that each sensor views a statistically similar subimage, and a histogram of each sensors response is made to match the overall histogram. Histogram matching shows sensitivity to violations of the similarity assumption. An alternative algorithm is suggested which matches the gain and offset of each sensor to typical values, and which is resistant to the effects of outliers. Tests on a sample image show the moment matching algorithm reduces the variance between sensors to a greater degree than histogram matching.

Spatial patterns of lightning-caused forest fires in Ontario, 1976-1998

The spatial pattern of forest fire locations is of interest for fire occurrence prediction and for understanding the role of fire in landscape processes. A spatial statistical analysis of lightning-caused fires in the province of Ontario, between 1976 and 1998, was carried out to investigate the spatial pattern of fires, the way they depart from randomness, and the scales at which spatial correlation occurs. Fire locations were found to be spatially clustered. Kernel estimation of the spatial pattern of lightning strikes on days when the dryness of the forest floor exceeded a designated threshold yielded clusters in the same areas as the lightning fire clusters.

Wavelets, boundaries, and the spatial analysis of landscape pattern

Abstract Recent developments in remote sensing and geographical information systems make widely available data sets covering large extents with a variety of spatial resolutions, such as digital images and geographic databases, potentially interesting for ecological applications. Such regular lattice data consisting of several million observations are now frequently analyzed for mapping, monitoring, and interpreting landscapes from an ecological, pattern-sensitive perspective. This requires adaptive data description and interpretation, which usually comes in the form of partitioning the data, for example, by the application of boundary-detection and/or classification and segmentation. Such tasks can benefit from utilizing new statistical and image processing tools, including wavelets, because they can characterize global as well as local pattern. This paper briefly introduces the wavelet representation and links it with hierarchical spatial data structures (quadtrees) and extensively used statistical techniques (nested analysis of variance, geostatistics, spectral analysis). The wavelet transformation and its variants are extremely efficient in summarizing or hierarchically approximating very large data sets while focusing on interesting subsets of the studied landscape. The use of wavelets in spatial data analysis and their relevance in characterizing and partitioning landscapes is demonstrated using our own and commercial software (S+WAVELETS) by comparing this representation with other methods. Simulated examples and real data from a grassland field study in Saskatchewan and regional net primary productivity across Ontario derived from satellite images illustrate the methodology.

When are two landscape pattern indices significantly different

Abstract.Landscape pattern indices (LPI), which characterize various aspects of composition and configuration of categorical variables on a lattice (e.g., shape, clumping, proportion), have become increasingly popular for quantifying and characterizing various aspects of spatial patterns. Unlike in the case of spatial statistical models, when either the joint distribution of all values is characterized by a limited number of parameters, or the distribution is known for certain (usually random) cases, the distributions of LPI are not known. Therefore, comparisons of LPI or significance testing of differences among various landscapes and/or studies are uncertain. This paper scrutinizes six widely used LPI, which are computed based on categories mapped onto regular lattices. We designed a simulation using Gauss-Markov random fields to establish the empirical distributions of LPI as functions of landscape composition and configuration. We report the results for stationary binary landscapes. The confidence intervals for LPI are derived based on 1000 simulations of each given combination of parameters, and further details are evaluated for three illustrative cases. We report the distributions of the LPI along with their co-variation. Our results elucidate how proportion of cover classes and spatial autocorrelation simultaneously and significantly affect the outcome of LPI values. These results also highlight the importance and formal linkages between fully specified spatial stochastic models and spatial pattern analysis. We conclude that LPI must be compared with great care because of the drastic effects that both composition and configuration have on individual LPI values. We also stress the importance of knowing the expected range of variation about LPI values so that statistical comparisons and inferences can be made.

The Influence of Vegetation Index and Spatial Resolution on a Two-Date Remote Sensing- Derived Relation to C4 Species Coverage

Abstract Changes in composition of plant species are expected to accompany a warming climate. In the northern mixed grass prairie, such changes are predicted to take the form of shifts in the relative ground cover of C3 and C4 photosynthetic types. In this study, we explore the feasibility of using two-date remote sensing data as a potential tool for monitoring these shifts. Our approach is based on the well-described asynchronous seasonality of C3 and C4 species. We hypothesize that the ratios of early-season to late-season aboveground live biomass (Bearly/Blate) will decrease as sites become more C4-dominated, and that if Bearly and Blate can be reliably estimated using spectral data, it may be feasible to predict C4 species coverage (%C4) from commercially available satellite information. Using spectral and botanical measurements from three upland communities in the Canadian mixed grass prairie, we (a) examined the relationship between various spectral vegetation indices and aboveground live biomass, (b) investigated the nature of the relationship between remotely sensed estimates of Bearly/Blate and %C4 at multiple sample resolutions (0.5 m, 2.5 m, 10 m, and 50 m), and (c) assessed whether these relationships were dependent on the vegetation index used to estimate biomass. We found a log-linear relationship between each spectral index and aboveground live biomass. Negative linear relationships were found between %C4 and remotely sensed Bearly/Blate at all sampling resolutions. These relationships were strongest at sampling resolutions of 10 m and 50 m. The strengths and forms of relationships were found to be partially vegetation index-dependent. Stronger relationships between variants at coarser resolutions likely result from the smoothing of fine-scale variation in aboveground live biomass and C4 species coverage. Our results suggest that commercially available satellite data at resolutions of 10 m to 50 m (e.g., Landsat Thematic Mapper) may offer the potential for estimating coverage of C4 species and that the choice of vegetation index used to estimate biomass is relatively unimportant. However, we caution that for this technique to be operationally useful, statistical model performance must be strengthened and developed to provide both temporal and spatial generality. Further investigation is needed to examine the applicability of this approach to other growing seasons, community types, and grassland regions.

Assessing the stability and uncertainty of predicted vegetation growth under climatic variability: northern mixed grass prairie

Ecosystem models which include both variability of driving variables as an input, and uncertainty and/or stability in their predictions are rare, especially outside of forest and cropland applications. Our objective is to investigate the stability of productivity levels and temporal patterns in a northern mixed grass prairie site using scenarios of varying levels of climate variability. Predictions of annual net primary productivity (NPP) are compared under a variety of global change and management scenarios. Specifically, we investigate the relative responses of C3 and C4 vegetation functional groups as a diagnostic of changes in resource availability. Scenarios of gradual temperature increase over 200 years demonstrate that warming will have different effects depending partially on the seasonal timing of that warming, but mostly on the concurrent changes in moisture availability. We propose that stability of vegetation communities may be more important than simply predicting levels of productivity for answering many questions related to the impacts of global change. This is demonstrated using frequencies of consecutive years with low productivity. Moderate increase in precipitation variability without increases to average rainfall can increase productivity and apparently increase stability. Further increase in precipitation variability decreases stability. The uncertainty in NPP predictions can be quantified by repeated simulations using stochastic variations in driving climate variables. Uncertainty in NPP predictions is found to be at the order of 20 g/m2/year, or about 25% of long-term averages. This lets us qualify our conclusions and shows that further research can reduce this uncertainty by better predictions of moisture availability, which can be obtained using finer spatial and temporal resolution representations.

Three-dimensional mapping of light transmittance and foliage distribution using lidar

The horizontal and vertical distributions of light transmittance were evaluated as a function of foliage distribution using lidar (light detection and ranging) observations for a sugar maple (Acer saccharum) stand in the Turkey Lakes Watershed. Along the vertical profile of vegetation, horizontal slices of probability of light transmittance were derived from an Optech ALTM 1225 instruments return pulses (two discrete, 15-cm diameter returns) using indicator kriging. These predictions were compared with (i) below canopy (1-cm spatial resolution) transect measurements of the fraction of photosynthetically active radiation (FPAR) and (ii) measurements of tree height. A first-order trend was initally removed from the lidar returns. The vertical distribution of vegetation height was then sliced into nine percentiles and indicator variograms were fitted to them. Variogram parameters were found to vary as a function of foliage height above ground. In this paper, we show that the relationship between ground measurements of FPAR and kriged estimates of vegetation cover becomes stronger and tighter at coarser spatial resolutions. Three-dimensional maps of foliage distribution were computed as stacks of the percentile probability surfaces. These probability surfaces showed correspondence with individual tree-based observations and provided a much more detailed characterization of quasi-continuous foliage distribution. These results suggest that discrete-return lidar provides a promising technology to capture variations of foliage characteristics in forests to support the development of functional linkages between biophysical and ecological studies.

A comparison of three approaches for predicting C4 species cover of northern mixed grass prairie

The C4 composition of Canadian mixed-grass communities is more sensitive to environmental change than other grasslands. Reliable methods of detecting such changes are necessary if these landscapes are to be properly managed. One approach is to use satellite remote sensing systems. Various studies have shown that the asynchronous seasonality of C3 and C4 species allows the relative abundance of each photosynthetic type to be estimated using temporal trajectory indices (TTIs) of sensor-derived normalized difference vegetation index (NDVI). In this study, we compared three approaches for predicting C4 species cover at Grasslands National Park (GNP) (Saskatchewan, Canada). TTIs related to Approach I were calculated from plots of NDVI vs. day-of-year (DOY). TTIs related to Approach II were calculated from plots of normalized cumulative NDVI vs. growing degree day (GDD). TTIs related to Approach III were calculated as ratios of earlyseason NDVI to late-season NDVI. Our analyses were conducted at two separate ecological scales. Awithin-community analysis used fieldsampled data from upland grassland to compare techniques at sampling resolutions of 0.5, 2.5, 10, and 50 m. An across-community analysis compared techniques using a vegetation survey of the GNP region and TTIs calculated from Advanced Very High Resolution Radiometer (AVHRR) data (1 km). At both scales, TTIs related to the timing of specific phenological events were the best predictors of C4 species cover. While all techniques performed well in the within-community study, Approach III performed best. Here, the predictive ability of each approach was weak at a resolution of 0.5 m but stronger at 2.5, 10, and 50 m resolutions. We also found that the optimal sampling dates for Approach III fell within a certain GDD range. This is encouraging for the a priori selection of sample dates, which would make the need for full seasonal time series redundant. In the across-community analysis, the AVHRR-derived Approach II TTIs were better able to discriminate among grasslands of different C4 composition than any other technique (overall accuracy=74%). However, for some C4 cover classes, the predictive accuracy of this approach was low. While these results are encouraging for the use of spectral data in monitoring the C4 cover of northern prairie, various research issues remain. At the within-community level, these include (a) further attempts to define objective criteria for the a priori identification of sampling dates for Approach III, and (b) and the extension of such studies to other growing seasons and community types/grassland regions. At the across-community level, these include the expansion of such techniques to a larger geographical region that contains a wider range in C4 cover values and land use types (e.g. ungrazed vs. grazed grasslands). D 2003 Elsevier Science Inc. All rights reserved.

Mutual information spectra for comparing categorical maps

The continual accumulation of categorical data sets, presented as nominal categories mapped onto regular grids, provides for the increased desire to compare the patterns observed between these maps. We present a measurement scheme for the comparison of categorical maps that decomposes the differences in multidimensional nested coincidence tables according to variables that record occurrence frequencies of categories (Z), at levels of spatial aggregation (Y), on specific maps (X). Sequences of conditional entropies computed according to the specific questions asked (e.g. is there coincidence between colours and locations), characterize the correspondence between the three types of variables in common units (bits) measured by mutual information. The form of these sequences, as a variable runs from coarse to fine detail, referred to as spectra, provide meaningful characterizations of the similarities/differences between categorical maps, including their spatial configuration. We introduce the information theory‐based conceptual framework and illustrate its beneficial application by comparing a series of demonstration maps.

Hierarchical decomposition of variance with applications in environmental mapping based on satellite images

A quadtree-based image segmentation procedure (HQ) is presented to map complex environmental conditions. It applies a hierarchical nested analysis of variance within the framework of multiresolution wavelet approximation. The procedure leads to an optimal solution for determining mapping units based on spatial variability with constraints on the arrangement and shape of the units. Linkages to geostatisiics are pointed out, but the HQ decomposition algorithm does not require any homogeneity criteria. The computer implementation can be parameterized by either the number of required mapping units or the maximum within-unit variance, or it can provide a “spectrum” of significances of nested ANOVA. The detailed mathematical background and methodology is illustrated by a salt-affected grassland mapping study (Hortobágy, Hungary), where heterogeneous environmental characteristics have been sampled and predicted based on remotely sensed images using these principles.