Characterizing temperate forest structural and spectral diversity with Hyperion EO-1 data

Forest structural diversity can serve as an important indicator of biodiversity. The relationship between spaceborne hyperspectral remotely sensed data and several measures of forest structure was explored over a 625 km² coastal temperate forest landscape on Vancouver Island, British Columbia, Canada. Thirteen Hyperion bands were selected for analysis based on the documented and hypothesized importance of various spectral wavelengths to forest characterization. To aid in understanding spectral trends, measures of forest stand structural diversity (projected age, projected height, and stand species composition complexity) were derived from forest inventory data. The spectral distance between the stand mean and standard deviation of reflectance and related expectations from global equivalents for each of the 13 bands were used to relate measures of spectral diversity (N = 801 forest inventory stands).Canonical correlation analysis was then used to determine the independent and shared relationships between these selected measures of forest structural diversity (dependent variables) and spectral diversity (independent variables). The dependent variables that were most strongly correlated with the first canonical variate were projected age and projected height, with canonical loadings of 0.973 and 0.979, respectively. In contrast, stand species composition complexity had a weak, negative correlation with spectral diversity (canonical loading = − 0.025). The wavelengths contributing the most to the canonical function included: 681-740 nm, 551-680 nm, and 1401-2400 nm. There have been few studies that attempt to directly link spectral and species diversity in temperate forest environments. From this initial investigation, we posit that the complex spectral response of coastal temperate forests may confound efforts to directly link spectral and species diversity across a range of site conditions.Our results, which are constrained by the spectral and spatial resolution of the data used, our target environment, and the metrics selected for measuring forest structure, suggest that attributes that characterize forest structural conditions may have a more meaningful relationship with spectral diversity than measures of species diversity alone, and that future studies in coastal temperate forests that seek to link spectral diversity with biodiversity should include measures of forest structural diversity, in addition to measures of species diversity.     

Initial assessment of ATSR-2 data structure for land surface applications

Principal component analysis was used to assess the structure of 1 km spatial resolution data from the second Along Track Scanning Radiometer (ATSR-2). Results indicate that the nadir spectral data have a structure similar to Landsat Thematic Mapper data and could be reduced to two eigenvectors with a loss of under 10 per cent of the total variance. The dual-view spectral data could be reduced to three eigenvectors with a loss of under 12 per cent of the total variance. The thermal data may provide additional information to that contained in the spectral data at some sites. The physical basis of the data structure is interpreted. Thus, ATSR-2 data are of considerable interest for estimating land surface characteristics at a regional scale.     

Use of Landsat data for automatic classification and area estimation of sugar-cane plantation in Sao Paulo state,Brazil

Sugar-cane (saccharum spp.) plantations in São Paulo State were classified automatically using an lmage-100 system and Landsat digital data. Ten segments of size 10 × 20 km were aerially photographed and used as training areas for automatic classification. The study area was covered by four Landsat paths, 235, 236, 237 and 238. The percentages of overall correct classification for these paths range from 79·56 per cent for path 238 to 95·59 per cent for path 237.     

TM digital processing of a tropical forest region in southeastern Mexico

A humid forest in the neotropical area of Los Tuxtlas, in southeastern Mexico has been used as a test area (900km²) for classification of landscape and vegetation by means of Landsat Thematic Mapper (TM) data, aerial photography and 103 ground samples. The area presents altitudinal variations from sea level to 1640m, providing a wide variety of vegetation types. A hybrid (supervised/unsupervised) classification approach was used, defining spectral signatures for 14 clustering areas with data from the reflective bands of the TM. The selected clustering areas ranged from vegetation of the highlands and the rain forest to grassland, barren soil, crops and secondary vegetation. The digital classification compared favourably with results from aerial photography and with those from a multivariate analysis of the 103 ground data. The statistical evaluation (error matrix) of the classified image indicated an overall 84·4 per cent accuracy with a kappa coefficient of agreement of 0·83. A geographical information system (GIS) was used to compile a land unit and a vegetation map. The TM data allowed for delineation of boundaries in the land unit map, and for a finer differentiation of vegetation types than those identified during field work. Digital value patterns of several information classes are shown and discussed as an indirect guide of the spectral behaviour of vegetation of highlands, rain forest, secondary vegetation and crops. The method is considered applicable to the inventory of other forested areas, especially those with significant variations in vegetation.     

Patterns of covariance between forest stand and canopy structure in the Pacific Northwest

In the past decade, lidar (light detection and ranging) has emerged as a powerful tool for remotely sensing forest canopy and stand structure, including the estimation of aboveground biomass and carbon storage. Numerous papers have documented the use of lidar measurements to predict important aspects of forest stand structure, including aboveground biomass. Other papers have documented the ability to transform lidar measurements to approximate common field measures, such as cover, stand height, and vertical distributions of foliage density and light transmittance. However, only a small number of existing works have thoroughly examined relationships between comprehensive assemblages of forest canopy and forest stand structure indices. In this work, canonical correlation analysis of coincident lidar and field datasets in western Oregon and Washington is used to define seven statistically significant pairs of canonical variables, each defining an axis of variation that stand and canopy structure have in common. The first major axis relates mean stand height, and related variables, to aboveground biomass. The second relates canopy cover and volume to leaf area index and stem density. The third relates canopy height variability to mean stem diameter and the basal area of deciduous species. Of the four remaining axes, three are related to contrasts between mature and old-growth stands. Canonical correlation analysis provides a method for ranking the importance of these effects, and for placing both canopy and stand structure indices within the overall covariance structure of the two datasets. In this sense, and for the study area involved, the first three factors (mean height, cover or leaf index area, height variability) represent the same kind of enhancement of lidar data that the tasseled cap indices [Crist, C.P., R.C. Cicone, 1984. A physically-based transformation of thematic mapper data—the TM tasseled cap. IEEE Transactions on Geoscience and Remote Sensing 22, 256-263.] represent for optical remote sensing.     

A Review of: “Applied Geomorphology: Geomorphological Surveys for Environmental Development.” By H. TH. VERSTAPPEN. (Amsterdam: Elsevier Science Publishers, 1983.) [Pp. 437.] Price $95.75 (U.S.A. and Canada) Dfl. 225.00 (Rest of World).

Abstract

This research was conducted to assess the utility of Landsat Thematic Mapper thermal data (10.4 to 12.5/μm) for urban feature analysis. Seven band thematic mapper data collected on 25 October, 1982 for the Chicago metropolitan area were used to determine the relative importance of the thermal band as a component of an urban land use classification.|Analysis consisted of: (1) the use of a transformed divergence measure (separability) to test the interclass spectral variability of selected features for different channel combinations and (2) a comparison of classification results, with and without thermal data using a supervised layered classification algorithm. The use of thermal data resulted in an overall 9.3 per cent increase in classification accuracy. Elimination of classes such as forest, asphalt and railroad resulted in an increase of 12.0 per cent in classification performance. Examination of the separability analysis and classification results indicated there were specific areas of success resulting from the use of thermal data. Increased classification accuracy occurred in several classes such as shadow (+ 36.7 per cent), residential ( + 41.1 per cent), active industry ( + 20.0 per cent) and commercial industrial fringe ( + 77.1 per cent). These results demonstrated that the information content of thermal data was valuable for urban area analysis.     

Spectral texture for improved class discrimination in complex terrain

Abstract

A spatial co-occurrence algorithm has been used to derive image texture from Landsat Multispectral Scanner (MSS) data to increase classification accuracy in a moderate relief, boreal environment in eastern Canada. The aim was to investigate ‘data-driven improvements’, including those available through digital elevation modelling. Overall classification accuracy using MSS data alone was 59·1 per cent when compared to a biophysical inventory of the area compiled primarily by aerial photointerpretation. This increased to 66·2 per cent with MSS plus texture and to 89·8 per cent when MSS data were analysed with geomorphometry extracted from a digital elevation model (DEM). The introduction of MSS texture resulted in statistically significant increases in individual class accuracies in classes that were also well defined using the geomorphometric and integrated data sets. This suggested that some of the additional information provided by geomorphometry was also contained in spectral texture. It was also noted that individual texture orientations resulted in higher class accuracies than average texture measures; this is probably related to structural (slope/aspect) characteristics of specific vegetation communities.     

Retrieving forest stand parameters from SAR backscatter data using a neural network trained by a canopy backscatter model

It was possible to retrieve the stand mean dbh (tree trunk diameter at breast height) and stand density from the Jet Propulsion Laboratory (JPL) Airborne Synthetic Aperture Radar (AIRSAR) backscatter data by using threelayered perceptron neural networks (NNs). Two sets of NNs were trained by the Santa Barbara microwave canopy backscatter model. One set of the trained NNs was used to retrieve the stand mean dbh, and the other to retrieve the stand density. Each set of the NNs consisted of seven individual NNs for all possible combinations of one, two, and three radar wavelengths. Ground and multiplewavelength AIRSAR backscatter data from two ponderosa pine forest stands near Mt. Shasta, California (U.S.A.) were used to evaluate the accuracy of the retrievals. The r.m.s. and relative errors of the retrieval for stand mean dbh were 6.1cm and 15.6 per cent for one stand (St2), and 3.1cm and 6.7 per cent for the other stand (St11). The r.m.s. and relative errors of the retrieval for stand density were 71.2 treesha-1 and 23.0 per cent for St2, and 49.7 treesha-1 and 21.3 per cent for St11.     

Estimating the age and structure of forests in a multi-ownership landscape of western Oregon,U.S.A.

Forests of the Pacific Northwest region of the U.S.A. are part of an ongoing political debate that focuses on the trade-offs between commodity and non-commodity values. A key issue in this debate is the location and extent of closed canopy mature and old-growth forest remaining in the region. Remote sensing can play a major part in locating mature and old-growth forests, but. several challenges must be overcome to do so with acceptable accuracy. Conifer forests of the region have high leaf area indices. Thus, most incident solar energy is absorbed, making these forests difficult targets for discrimination of classes. Additionally, spectral characteristics can be affected more by the effects of steep topography than condition of the closed canopy forest.

Experimenting with a number of techniques, we estimated and mapped forest age and structure in 1988 over a 1 237 482 ha area on the west side of the Oregon Cascade Range with an overall accuracy of 82 per cent. Unsupervised classification enabled several forest classes to be defined in terms of per cent cover: open (0-30 per cent), semi-open (30-85 per cent), closed mix (> 85 per cent, of which at least 10 percent is comprised of non-conifer species), and closed conifer (> 85 per cent, of which less than 10 per cent is non-conifer). These classes represented nearly distinct spectral groups. Within the closed canopy conifer class, between two and three age and structural classes could be distinguished using regression analysis (e.g., young, mature, and old-growth). Defining more classes seriously degraded map accuracies. The Tasseled Cap wetness index was not sensitive to topography, and yielded more accurate results in closed canopy conifer stands than Tasseled Cap brightness or greenness, even when regression models using these indices were based on solar incidence angle stratification.

The multi-ownership study area consisted of 76 per cent forestland. Of the total forestland, 70 per cent was closed canopy conifer, with 42 per cent being in a mature or old-growth state. Forests administered by the USD1 Bureau of Land Management (BLM) and the USDA Forest Service, but protected by congressional and administrative mandates from harvest, were 10 per cent of the total forestland. Of the protected category, only 60 per cent was mature and old-growth forest, Unprotected BLM and Forest Service lands accounted for 53 per cent of the forestland in this study (8 and 45 per cent, respectively). Of the unprotected category, the BLM had 63 per cent, and the Forest Service had 49 per cent, respectively, of their holdings in a pre-canopy closure and young conifer condition. Thirty-five per cent of the forestland was privately owned, and consisted of 73 per cent pre-canopy closure and young conifer forest stands. Of all mature and old-growth forest, 22 per cent was found on private land, 7 per cent on unprotected BLM land, 55 per cent on unprotected Forest Service land, and 15 per cent on protected land.     

Satellite estimation of tropical secondary forest above-ground biomass: Data from Brazil and Bolivia

This paper reports on a test of the ability to estimate above-ground biomass of tropical secondary forest from canopy spectral reflectance using satellite optical data. Landsat Thematic Mapper data were acquired concurrent with field surveys conducted in secondary forest fallows near Manaus, Brazil and Santa Cruz de la Sierra, Bolivia. Measurements of age and above-ground live biomass were made in 34 regrowth stands. Satellite data were converted to surface reflectances and compared with regrowth stand age, biomass and structural variables. Among the Brazilian stands, significant relationships were observed between middle-infrared reflectance and stand age, height, volume and biomass. The canopy reflectance-biomass relationship saturated at around 15.0 kg m^-2, or over 15 years of age (r > 0.80, p < 0.01). In the Bolivian study area, no significant relationship between canopy spectral reflectance and biomass was observed. These contrasting results are probably caused by a low Sun angle during the satellite measurements from Bolivia. However, regrowth structural and general compositional differences between the two study areas could explain the lack of a significant relationship in Bolivia. The results demonstrate a current potential for biomass estimation of secondary forests with satellite optical data in some, but not all, tropical regions. A discussion of the potential for regional extrapolation of spectral relationships and future satellite imagery is included.     

Vegetation spatial variability and its effect on vegetation indices

Abstract

Landsat MSS data were used to simulate low resolution satellite data, such as NOAA AVHRR, to quantify the fractional vegetation cover within a pixel and relate the fractional cover to the normalized difference vegetation index (NDVI) and the simple ratio (SR). The MSS data were converted to radiances from which the NDVI and SR values for the simulated pixels were determined. Each simulated pixel was divided into clusters using an unsupervised classification programme. Spatial and spectral analysis provided a means of combining clusters representing similar surface characteristics into vegetated and non-vegetated areas. Analysis showed an average error of 12·7 per cent in determining these areas. NDVI values less than 0·3 represented fractional vegetated areas of 5 per cent or less, while a value of 0·7 or higher represented fractional vegetated areas greater than 80 per cent. Regression analysis showed a strong linear relation between fractional vegetation area and the NDVI and SR values; correlation values were 0·89 and 0·95 respectively. The range of NDVI values calculated from the MSS data agrees well with field studies.     

Classification of SPOT HRV imagery and texture features

Abstract

Spatial co-occurrence matrices were computed for a SPOT HRV multispectral image for a moderate-relief environment in eastern Canada. The texture features entropy and inverse difference moment were used with the spectral data in landcover classification, and substantive increases in accuracy were noted. These range from 10 per cent for exposed bedrock to over 40 per cent in forest and wetland classes. The average classification accuracies were increased from 511 per cent (spectral data alone) to 86.7 per cent (spectral data plus entropy measured in band 2 and inverse difference moment in band 3). Classes that are homogeneous on the ground were characterized adequately by spectral tone alone, but classes containing mixed vegetation patterns or strongly related to structure were characterized more accurately by using a mixture of spectral tone and texture.     

Regenerating boreal forest structure estimation using SPOT‐5 pan‐sharpened imagery

Forested stand structure is an important target variable within the fields of wildlife ecology. Remote sensing has often been suggested as a viable alternative to time consuming field and aerial investigations to determine forest structural attributes. In this study, 44 stands of recently harvested, regenerating, and old growth forest within the Foothills Model Forest in west‐central Alberta were selected to test the ability of pan‐sharpened SPOT‐5 spectral response to classify stand structure. For each stand, a Structural Complexity Index (SCI) was calculated from field data using principal components analysis. To complement the spectral response data set and further increase accuracy, the normalized difference moisture index (NDMI) and three window sizes (5×5, 11×11, and 25×25) of first‐ (mean and standard deviation) and second‐order (homogeneity, entropy, contrast, and correlation) textural measures were calculated over the pan‐sharpened image. Stepwise multivariate regression analysis was used to determine the best explanatory model of the SCI using the spectral and textural data. The NDMI, first‐order standard deviation and second‐order correlation texture measures were better able to explain differences in SCI among the 44 forest stands (r² = 0.79). The most appropriate window size for the texture measures was 5×5 indicating that this is a measure only detectable at a very high spatial resolution. The resulting classified SCI values were comparable to the actual field level SCI (r² = 0.74, p = 0.01) and were limited by the strong variability within stands. Future research may find this measure useful either as a separate parameter or as an indicator of forest age for use in wildlife habitat modelling.     

Mapping land-cover over large areas using multispectral data derived from the NOAA-AVHRR: A case study of Nigeria

Land-cover information for Nigeria was obtained from a countrywide, low-level aerial survey conducted in 1990. A range of spectral vegetation indices (SVIs) and ground surface temperature estimates were calculated for Nigeria using daily data throughout 1990 from the National Oceanic and Atmospheric Administration's (NOAA) Advanced Very High Resolution Radiometer (AVHRR) data. A supervised classification of the land-cover classes was then performed using a modified discriminant analysis in which predictor variables were selected from the mean, maximum, minimum and standard deviation of the raw waveband AVHRR data, AVHRR derived products and a digital elevation model (DEM). With a 60 per cent threshold coverage by any one of eight major vegetation types the analysis correctly predicted land-cover type with producer accuracies (excluding 'bare ground' with only a few points) of between 48 per cent (cultivation) and 100 per cent (mangrove) (average 74.5 per cent).     

Requirements for labelling forest polygons in an object-based image analysis classification

The ability to spatially quantify changes in the landscape and create land-cover maps is one of the most powerful uses of remote sensing. Recent advances in object-based image analysis (OBIA) have also improved classification techniques for developing land-cover maps. However, when using an OBIA technique, collecting ground data to label reference units may not be straightforward, since these segments generally contain a variable number of pixels as well as a variety of pixel values, which may reflect variation in land-cover composition. Accurate classification of reference units can be particularly difficult in forested land-cover types, since these classes can be quite variable on the ground. This study evaluates how many prism sample locations are needed to attain an acceptable level of accuracy within forested reference units in southeastern New Hampshire (NH). Typical forest inventory guidelines suggest at least 10 prism samples per stand, depending on the stand area and stand type. However, because OBIA segments group pixels based on the variance of the pixels, fewer prism samples may be necessary in a segment to properly estimate the stand composition. A bootstrapping statistical technique was used to find the necessary number of prism samples to limit the variance associated with estimating the species composition of a segment. Allowing for the lowest acceptable variance, a maximum of only six prism samples was necessary to label forested reference units. All polygons needed at least two prism samples for classification.     

Structural attributes of three forest types in central Spain and Landsat ETM+ information evaluated with redundancy analysis

Structural attributes of forest, such as canopy crown closure, stand height, stem density and basal area, derived from the third Spanish National Forest Inventory (IFN‐3) were used in combination with spectral information derived from Landsat Enhanced Thematic Mapper Plus (ETM+) imagery and topographic information to evaluate their relationships. To deal with the variability found in the literature, three different types of vegetation, dominated by conifers, evergreen sclerophyll and broad‐leaved deciduous trees, were analysed. In addition, the analyses were performed using three sets of plots filtered to be successively more homogeneous. A multivariate canonical ordination method, redundancy analysis (RDA), was used to enable the simultaneous evaluation of the two data sets and provide a useful graphical output highlighting the relationships between response (structural attributes) and explanatory (spectral and topographic) variables. Rank correlation analyses were also performed. The low percentage of explained variance at the multivariate analyses and low rank correlation coefficients made it difficult to derive practical empirical models. The strong influence of vegetation type on the results was confirmed, given that each type was sensitive to a different kind of spectral information. Finally, the results did not allow validation of the hypothesis that the relationship should be better when using a more homogeneous set of plots.     

Forestry information content of Thematic Mapper data

This paper provides an initial investigation of the spectroradiometric data structure and information content of Thematic Mapper (TM) data for some Canadian forest-cover types. The Dryden-Lac Seul region in western Ontario is an important commercial and tourist area containing mainly boreal forest. A LANDSAT-4 TM scene of this region was analysed in conjunction with a considerable amount of ancillary data. The data were reduced to a manageable volume by selectingsubscenes;a preliminary attempt at atmospheric correction and radiometric calibration was then carried out; polygons representing a wide variety of cover classes were defined and, finally, the spectroradiometric information available for class discrimination was analysed using several techniques. Principal component analysis and feature selection reveal that the spectral data can be reduced to three eigenvectors with a loss of less than 10 per cent of the scene variance, and that the best three TM bands perform almost as wellas the first three principal components; for general cover-type discrimination, these bands areTM 3,4 and 5.TM bands 1,4 and 5 are marginally better for separating a set of softwood classes, although TM I has a very small dynamic range. The shortwave infrared (SWIR) spectral region, represented byTM bands 5and 7, seems to be particularly sensitive to forest vegetation density, especially in the early stages of clearcut regeneration. Shadowing is suggested as a factor at least as important as leaf moisture content in influencingthe spectral reflectanceof forests in this region. The first three principal components are related to the scene brightness (PC I), greenness (PC 2) and the contrast between the SWIR and the visible and near-infrared regions. We propose the name ‘swirness’ for the third component until a more complete understanding of its properties is achieved.     

Sparse CCA using a Lasso with positivity constraints

Canonical correlation analysis (CCA) describes the relationship between two sets of variables by finding linear combinations of the variables with maximal correlation. A sparse version of CCA is proposed that reduces the chance of including unimportant variables in the canonical variates and thus improves their interpretation. A version of the Lasso algorithm incorporating positivity constraints is implemented in tandem with alternating least squares (ALS), to obtain sparse canonical variates. The proposed method is demonstrated on simulation studies and a data set from market basket analysis.     

Radiometric resolution for monitoring vegetation How many bits are needed?

Abstract

The number of radiometric quantizing levels required for satellite monitoring of vegetation resources was evaluated by using in situ collected spectral reflectance data, an atmospheric radiative transfer simulation model and a satellite sensor simulation model. Reflectance data were converted to radiance data; passed through a model atmosphere to an altitude of 706 km; and subsequently quantized at 16,32,64,128,256 and 512 digital count levels for Thematic Mapper bands TM3 (0·63-0·69 μm) and TM4 (0·76-0·90 μm), The simulated digital count data were regressed against in situ biological data to quantify the relationship(s) between quantizing levels.

Results of the analysis demonstrated that solar zenith angle had an effect on QEΔρ, that 256 quantizing levels gave a 1-3 per cent improvement per channel over 64 quantizing levels, and that 256 quantizing levels gave a 1 per cent improvement per channel over 128 quantizing levels. No improvements were found for 256 versus 512 quantizing levels.     

DALASS: Variable selection in discriminant analysis via the LASSO

The objective of DALASS is to simplify the interpretation of Fisher's discriminant function coefficients. The DALASS problem—discriminant analysis (DA) modified so that the canonical variates satisfy the LASSO constraint—is formulated as a dynamical system on the unit sphere. Both standard and orthogonal canonical variates are considered. The globally convergent continuous-time algorithms are illustrated numerically and applied to some well-known data sets.