Remote Sensing for Sustainable Forest Management - Chapter 7 pps

Empirical relationships between inventory variables such as canopy closure,stand volume, and species composition and airborne spectral response are typicallystronger than those obtained

Forest Structure Estimation

The … proper interpretation of remote sensor data requires a thorough understanding

of the temporal and spatial characteristics inherent in the vegetation cover types present and of the related changes in spectral response.

— R M Hoffer, 1978

INFORMATION ON FOREST STRUCTURE

The spatial and statistical output from a classification procedure comprises one of themajor information products on forest condition available by remote sensing; generally,

a second set of forestry information products is obtained by continuous variableestimation procedures Classification produces information on the features that arecontained in the list of classes imposed on the image data; the result is typically aclassification map Continuous variable estimation produces information on featuresthat vary continuously over the landscape depicted in imagery The result may be amap or an image in which the tones correspond to the level or value of the feature ofinterest and vary over the extent of the map The process can become more complexwhen continuously varying forest conditions are used in the process of classification.This is not usually a problem in conventional vegetation typing or species composition

of stands; the map is derived via the usual logic of classification (Zsilinsky, 1964;Avery, 1968) But typing and compiling species composition are only two of thestructural attributes of forest stands that are of interest, usually as part of a generalforest inventory Some of the other forest attributes of interest might include:

1 Forest crown closure,

2 Diameter at breast height (dbh),

As in species classifications, aerial photography has been instrumental in oping maps of these forest structures almost exclusively through the photomorphicapproach followed by field work, but also through direct image interpretations bymanual means (e.g., height calculations by parallax or shadows, crown closureestimation using templates, etc.) Species composition has been classified usingdigital image classification techniques — with high spatial detail imagery — butgenerally without the level of acceptance accorded the aerial photographic approach,for a wide range of reasons, not the least of which is the difficulty in generatingconventional maps with the digital methods (see Chapter 6) Digital classificationhas been used less frequently when the objective is to map other forest structures,because this type of mapping resembles more the estimation of a continuous variablerather than a discrete categorization Classification of different density or heightclasses has been described (Franklin and McDermid, 1993), but applications ofremote sensing aimed at these continuous aspects of the forest inventory have beendriven largely by empirical or semiempirical model estimation Unlike classification,which is typically driven by a statistical understanding of what the spectral responsepatterns mean, such models are based more on the relationships incorporated in afundamental understanding of the physically based radiative transfer in forests.

devel-A plethora of such studies have been reported attempting to estimate individualforest parameters such as crown closure, basal area, or volume, as independentvariables which can be predicted or estimated using a calibrated remote sensingimage The general approach is to:

1 Establish a number of field observation sites in a forest area,

2 Collect forest condition information at those sites,

3 Acquire imagery of the sites,

4 Locate the sites on the image,

5 Extract the remote sensing data from these sites,

6 Develop a model relating the field and spectral data, and finally,

7 Use the model to predict forest parameters for all forest pixels based onthe spectral data

Typically, the objective is to predict the selected field variable through modelanalysis, with the available remote sensing data as the dependent variables Then,the model is inverted to predict the independent variable (such as stand volume ordensity) over large areas of forest In other words, the spatially explicit remotesensing data are considered the predictors of the locally known field parameters sothat the remote sensing image can be used to map that parameter across the imagelandscape The remote sensing data are inverted to provide predictions of the desiredfield variables Intuitively this seems reasonable; users are aware of the fact that theremote sensing data are dependent on the field data, not the other way around Thecommon tool is model inversion; models developed through experimental or nor-mative designs are used to describe the relationships contained within a forest/remotesensing data set The aim is to generate new insights which can guide the fieldscientists and help new applications become possible

The physically based models are built mathematically on theoretical models thatare typically designed to quantify advances in the ability to predict target andradiation interactions (Jupp and Walker, 1997) The model is driven by the principles

of radiation physics to relate spectral properties to biophysical properties (Gerstl,1990) The model is derived from current experimental understanding of radiationphysics, geometry, and energy/chemical interactions The role of such models inadvancing the science of remote sensing cannot be overestimated; but typically,remote sensing data analysts and forestry users have little contact with these models.Their complex and demanding structure have meant that they will likely remain inthe domain of the remote sensing instrumentation and radiation specialists (Silva,1978; Woodham and Lee, 1985; Teillet et al., 1997) rather than the applicationsspecialists (Landgrebe, 1978b; Strahler et al., 1986; Cohen et al., 1996b; Franklinand Woodcock, 1997)

Empirical models might be constructed using the understanding derived fromphysically based models coupled with laboratory, field, and actual or simulatedremote sensing data Empirical remote sensing studies are plentiful — image clas-sifications, for example, are almost completely empirical This is the probable way

in which most users of remote sensing data will learn and apply their experiences.The empirical approach is a data-driven approach; learning proceeds from under-standing the data, data acquisition and the specific conditions under which modelsderived from those data were inverted The form of the model can be inferred fromphysical considerations, while specific model parameters are estimated from empir-ical data Unfortunately, purely empirical models have the disadvantage of beinghighly site specific (Waring and Running, 1998; Friedl et al., 2000) This modelingsituation has given rise to an intermediate approach based on a set of semiempiricalstudies that are hybrids of the purely empirical and theoretical physical models Forexample, a statistical (empirical) model of the relationships between reflectance and

a canopy characteristics, such as leaf area index (LAI), may be augmented by aphysical understanding of the processes involved; the effect of leaf angle, leafdistribution, and leaf shape might be modeled within the larger relationship betweenreflectance and leaf area well-established through vegetation indices such as thenormalized-difference vegetation index (NDVI) (e.g., Chen and Cihlar, 1996; White

et al., 1997)

Canopy reflectance models based on geometric-optical modeling approximations

of physical processes represent an example of an emerging semiempirical method

in remote sensing; these models contain a mix of data-driven relationships andtheoretical understanding to provide answers only available in more sophisticated

or demanding experimental settings Li and Strahler (1985) developed one of thefirst such models — the geometric-optical reflectance model, commonly referred to

as the Li-Strahler model Using the model in California, Woodcock et al (1997)reported that the model appeared to confirm what had been learned in numerousempirical studies — namely, that canopy reflectance is dominated by canopy cover

— and that the advantages of using a canopy reflectance model over an empiricallyderived relationship were marginal, or at least unclear The application of forestreflectance modeling and coupling such models to physically based models that

incorporate growth and topography is in its infancy (Kimes et al., 1996; Gemmell,2000) In particular, invertible canopy models are currently scarce and impracticalfor operational use due to their complexity and our still-evolving understanding; forexample, Gemmell (2000) found that multiangle data were useful in improving theaccuracy of forest characteristics derived by inversion, but that more extensive testingand validation over larger areas and different forest conditions was essential to betterunderstand the limits of the methods With a modest investment in training, suchmodels could be used by applications specialists as well as the model developers.While specific results will vary, empirical methods used in one area to generate

a relationship between spectral response and forest conditions generally can beapplied, with few modifications, elsewhere But when using some types of remotesensing data, such as Landsat TM data, and empirical models such as linear regres-sion techniques, other difficulties arise (Salvador and Pons, 1998a,b):

1 Typically low dynamic range of the data; generally, higher correlations can

be obtained if log transformations are applied (Ripple et al., 1991; Bauliesand Pons, 1995) For example, with respect to leaf biomass, after a certaindensity is reached, doubling that parameter will not affect the spectralresponse, but a log transform can help establish linear relationships;

2 Extensive atmospheric and geometric corrections are needed;

3 Difficulty in reducing sensitivity to extraneous factors (a standard feature

of the normative remote sensing approach) (Gemmell, 2000);

4 Generally low spatial resolution relative to the objects under scrutiny —trees (Wynne et al., 2000), and;

5 Generally, small sample sizes often resulting in fewer degrees of freedomthan required for extensive use

Perhaps the most important advantage of this approach is its accessibility Thereare probably few users in forest management situations who are unable to find theresources to complete the simple normative design that is required to establish apurely empirical relationship between spectral response and, for example, canopycover All that is needed are the basic remote sensing infrastructure components, anairborne or satellite remote sensing image, and some field work The methods areslightly more demanding than classifications, but probably not by much (Franklin,1986; Walsh, 1987; Franklin and McDermid, 1993)

While the exact form and nature of the empirical relationship will not remainstable as conditions change, it is also true that the relationship will rarely differdramatically from what has already been reported or observed in an area Forexample, the normal relationship between cover and red reflectance is expected to

be expressed in a moderate negative correlation between the two variables becauseincreasing cover (larger tree crowns, more leaves) results in more red light absorption(greater photosynthesis activity) Less red light will be detected by a sensor abovethe canopy Perhaps the exact relationship is found to be an R value of –0.49 It ispossible but not likely that the correlation between red reflectance and cover will

be found to be +.49 in another, similar area More likely, the new area will have anegative relationship of approximately the same magnitude One interpretation of

this relationship might be that remote sensing images of a certain type of youngstand are almost always brighter in the visible portion of the spectrum than olderstands of that type This relationship is as likely to be found in one location as inanother If the usual (or normative) relationship between cover and reflectance isone of decreasing reflectance with age for a given species, then this will be more

or less likely to be true in New Brunswick as in Finland, Argentina, or Indonesia.The normal relationship must be established, tested, and understood in order forapplications of the relationship to be developed

Similar logic and approaches have been reported using SAR imagery In ular, Ahern et al (1996) exhaustively tested for relationships between SAR back-scatter and boreal forest stand structure measures, but none of the statistical relation-ships were strong enough to suggest that C-band backscatter might be capable ofproviding reliable estimates of stand structural parameters Different species differed

partic-in the strength and significance of the relationships (Table 7.1) Wilson (1996), using

a sample of stands from the same data set, took a different approach First, multipleregression equations were developed that included SAR backscatter and texturemeasures to predict mean height of spruce and pine stands; standard errors were lessthan 15% Then, the stands were grouped by forest inventory classes for height andcrown closure SAR data could provide discrimination of broad height and crownclosure classes at reasonable accuracies (Table 7.2) So, despite low correlationbetween spectral response and a forest variable on a pixel-by-pixel basis, high levels

of classification accuracy could still be generated over broader classes and areas.This is one approach to achieving a more successful (i.e., more useful) remote sensingestimation of a continuously varying forest condition; create logical classes andreduce the problem to a classification decision After all, 42 to 57% classificationaccuracy of crown closure into four classes is not high; under most circumstances,however, this would be considered much better (more useful!) than nothing.The success of this empirical inversion idea has generated a vast literaturecomprised of specific studies and experiments Many of these studies can be seen

as contributing insights to satisfy the growing need to understand the appropriaterole of remote sensing in providing information to sustainable forest managementgoals (Franklin and McDermid, 1993) A number of early empirical studies haveserved to demarcate the boundary for the use of airborne (Irons et al., 1987, 1991;

TABLE 7.1 Relative Importance of Forest Variables in Explaining Airborne C-Band SAR Backscatter in 93 Alberta Boreal Forest Stands

Hardwood Volume/ha, biomass/ha, mean age, pine cover Pine Hardwood cover, pine cover, crown volume, crown closure Spruce No statistically significant relationships were found

Source: Adapted from Ahern et al (1996).

Neville and Till, 1991; Miller et al., 1991) and satellite remote sensing (Franklin,1986; Butera, 1986; Danson, 1987; Walsh, 1987) in forest inventory assessmentbeyond which correlations are probably too tenuous — or too far from the normative

— to support the endeavor These studies were followed by a number of systematicattempts to integrate satellite remote sensing into forest inventory compilations oflarge areas (De Wulf et al., 1990; Brockhaus and Khorram, 1992; Bauer et al., 1994)and detailed studies of smaller areas designed to confirm or refine the empiricalrelationships for certain species or forest types of interest (Ripple et al., 1991; Dansonand Curran, 1993; Franklin and Luther, 1995)

Empirical relationships between inventory variables such as canopy closure,stand volume, and species composition and airborne spectral response are typicallystronger than those obtained from satellite sensors (Franklin and McDermid, 1993).This is probably because of the higher dynamic range and smaller pixel sizescommonly acquired by airborne sensors But new satellite sensor data with improvedcharacteristics are increasingly available and will continue to be tested What is ofinterest here is a general assessment of remote sensing in estimating the kinds offorest variables that are of interest in compiling a forest inventory Currently remotesensing is limited to the following generally successful applications (discussed ingreater detail in following sections):

• Remote sensing data can be used to stratify forest covertypes at the broadlevel into classes of density, biomass, or volume; such strata are morepronounced in areas of significant topographic relief, which can be used

to enhance the spectral differences and the actual differences in the targetvariable likely to be more related to topographic (ecosystematic or envi-ronmental factors) differences than to forest spectral response conditions;

• Remote sensing data can be used to stratify forest canopy (crown) acteristics; this procedure will be more successful in large (perhaps exten-sively managed) areas with a simple canopy structure and few species (e.g.,plantations) which are relatively flat; this works well because the differ-ences in the reflectance recorded by the satellite sensor will be dominated

char-by changes in crown closure and density rather than char-by topography;

TABLE 7.2 Discrimination of Height and Crown Closure Classes Using Airborne SAR Imagery and Texture Variables in 66 Conifer Stands

in Alberta

Classification Accuracy (%) Conifer Type Height Classes Crown Closure Classes

Source: Modified from Wilson (1966).

• Remote sensing data can be used to construct composite structural indicesthat can be used to differentiate forest stands, and to understand spectralresponse, in order to better employ the predominately L-resolution satel-lite imagery in forest inventory assessment (e.g., in classifications).

FOREST INVENTORY VARIABLES

F OREST C OVER , C ROWN C LOSURE , AND T REE D ENSITY

Several early studies established that Landsat and SPOT satellite remote sensingdata were related to forest cover, stand age, and crown closure (Walsh, 1980; Poso

et al., 1984; Franklin, 1986; Butera, 1986; Horler and Ahern, 1986) The relationshipswere similar to those understood to be in effect with small-scale (high-altitude) aerialphotographs; for example, decreasing visible reflectance (darker image tones) would

be associated with increasing crown development As a stand grows and ages theareas between the crowns are no longer visible, and the shadows cast by the crowns

on each other deepen (Figure 7.1) Larger crowns would absorb more light, butreflect more strongly in the infrared (Butera, 1986; Franklin, 1986) The strongestcorrelations were typically found with the infrared bands (De Wulf et al., 1990)because greater atmospheric penetration would create deeper shadows from largertrees, and because of the large contrast and greater dynamic range

In 28 stands of Corsican pine (Pinus nigra var maritima) in England, a poor

relationship between SPOT HRV near-infrared reflectance and forest canopy coverwas found (Danson, 1987) The explanation was that, rather than a function ofvegetation amount, the variation in the amount of shadow within the canopy influ-enced the strength of the relationship Few significant relationships between SPOTHRV measured reflectance and lodgepole pine stands in Alberta were found (Franklinand McDermid, 1993); much stronger relationships with reflectance measured athigher spatial resolution by an airborne sensor were thought to be a result of thehigher dynamic range in the data and the smaller pixel size Again, shadowing effectswere thought to be the dominant influence on the spectral response A stepwisemultiple regression predicting cover and density using seven variables of tone andtexture extracted from red, green, and near-infrared bands of a 2.5 m spatial reso-lution airborne image yielded adjusted R2-values between 0.63 and 0.66 in 14lodgepole pine stands; this was reduced to 0.45 in the satellite data

After a fire in lodgepole pines stands in Yellowstone National Park, Jakubauskas(1996a,b) found that TM spectral response was dominated by soil reflectance As astand progressed to later successional stages, the spectral response was increasinglydominated by the forest canopy, until maximum canopy closure occurred at approx-imately 40 years post-fire As stands developed further, the overstory densitydeclined, but live basal area, height, LAI, and site diversity increased The largergaps in the canopy, species composition, and the canopy size of individual treesbegan to dominate the spectral response Stands thinned by beetle-induced mortalityoccupied a middle position in that spectral response, and stands that had been opened

up were again influenced largely by understory and soil characteristics Correlations

to height, basal area, and biomass were reasonably strong between lodgepole pine

stand conditions and Landsat TM data (Jakubauskas and Price, 1997); correlations

to density, size diversity, mean diameter, and number of species were moderate;correlations to understory measures (number of seedlings, understory species, totalcover) were weak

These and other studies have led to the understanding that the effect of increasing

or decreasing age, dbh, height, volume, and so on are all second- or third-ordereffects on remotely sensed image data; the principal influence on the spectralresponse is the illumination geometry (target-sensor-solar conditions) followed bythe amount of vegetation viewed from above As cover approaches full crown closure,the correlation between reflectance and these biophysical variables approaches zero;

“… stand reflectance is primarily dependent on the density, size, and arrangement

FIGURE 7.1 The geometrical-optical modeling approach considers that spectral response,

in areas where the pixel size is larger than the objects (trees), is a combination of shaded and sunlit components Here, the influence of the relationships is shown with (a) randomly located small trees and different sun angles and again with (b) different tree crown sizes The amount

of shadow and sunlit tree crown and the amount of area visible between the trees varies with the modeled characteristics The ideal use of the GO model would be to construct a lookup table using all possible variations in the area of interest and then to examine the actual data relative to the modeled data to determine correspondence If there were marked differences between the predicted and actual spectral response, then perhaps the area had been subjected

to an unidentified change (e.g., canopy had been reduced by disturbance) (From Jupp, D L.

B., and J Walker 1997 The Use of Remote Sensing in the Modeling of Forest Productivity,

pages 75–108, Kluwer, Dordrecht With permission.)

Incidence Angle = 30 o

of crowns and the reflectances of illuminated and shadowed components in the stand,and indirectly on other attributes (site quality, species composition, age) throughtheir effects on these former characteristics” (Gemmell, 1995: p 296).

The main problem seems to be a fundamental one (Holmgren and Thuresson,1998): the sensor detects reflectance only from the top of the canopy If the canopy

is open, the reflectance can be correlated with other attributes, such as understorycharacteristics which may be indirectly related to the target variables; if the canopy

is closed the extent to which other parameters can be predicted seems to depend onthe extent to which a closed canopy can predict them In Oregon forests there was

“little predictability in the spectral response of conifer forests beyond about 200years of age, or once old-growth characteristics are attained … forest stand condi-tions continue to evolve, but spectral changes appear uncorrelated with that devel-opment” (Cohen et al., 1995a: p 727) In many forests, crown closure will reach amaximum (perhaps reaching 100%) and basal area and structural complexity willcontinue to increase, but the remotely sensed signal is not significantly affected bythese increases (Franklin, 1986)

C ANOPY C HARACTERISTICS ON H IGH S PATIAL D ETAIL I MAGERY

Shadowing related to tree size may be the dominant influence on stand reflectancewhen high spatial resolution imagery are considered (St-Onge and Cavayas, 1995)

A pixel in this type of image will characterize only a small part of a tree crown,shadow, or understory The texture of the forest stand is generated by the light anddark tones created by individual tree crowns Texture thus holds the most promisefor automated forest cover or density estimation (Eldridge and Edwards, 1993; St-Onge and Cavayas, 1997) Customized texture windows — based on the rangederived from image semivariograms calculated over the stand — were useful forestimating canopy coverage in one study in Alberta (Franklin and McDermid, 1993).More frequently, as we have seen, image texture has been used to help classifyindividual species in a stand (Fournier et al., 1995) and subsequently, to classifyspecies composition (Franklin et al., 2000a)

Image understanding techniques have been developed to delineate tree crowns(Gougeon, 1995; Brandtberg, 1997) and then build a better estimate of crown closure,stem density, and species composition (Gerylo et al., 1998) (Chapter 7, Color Figure1*) This idea was preceded by attempts to better estimate species proportions, andhence cover, on digitized aerial photographs (Meyer et al., 1996; Magnusson, 1997)

A typical process might resemble the three-step procedure applied by Gougeon(1997) to airborne multispectral data acquired with spatial resolutions ranging from

30 to 100 cm:

• Individual tree crown delineation: Using the areas of shade between trees,

an algorithm was designed to find local maxima (bright spot assumed to

be the crown apex) and local minima (dark spot assumed to be the deepestshadows between crowns) Then, by following the valleys of dark areas

* Color figures follow page 176.

between bright areas, the tree crowns were delineated A rule-base of treecrown sizes was invoked to separate each crown completely from adjacentbright areas (e.g., impossibly large crowns were separated) Comparisons

of the resulting tree crown sizes to field crown estimates were within 7%.Once crowns were separated, estimates of stem density and canopy closurecould be generated with fixed or geographic windows

• Individual tree crown classification: Spectral signatures were used in astandard supervised classification procedure to identify the species asso-ciated with each delineated crown The key here was to treat each isolatedtree crown, delineated in the first step, as an object rather than to applyindividual per-pixel classification Options for classification included theuse of the brightest pixels, the average of the sunlit portion of the crown,

or the mean value within the delineated crown Classification accuracieswith four or five coniferous species were typically in the 72 to 81% range,depending on the original image spatial resolution

• Forest stand delineation: Using the delineated and classified tree crowns,

an algorithm was designed to regroup crowns into stands based on threederived variables in fixed windows: stem density, canopy closure, andspecies concentrations An unsupervised classifier was applied to reducethe crown groupings still further, based on a minimum stand area criterion.The results were converted to a vector base to obtain polygons whichclosely resembled those mapped using aerial photography in a traditionalapproach to forest inventory

This image understanding approach based on individual tree crown delineationappeared to work best with images having a spatial resolution <1 m, but conceivablywill work well enough with 1 m satellite data to justify more extensive use anddevelopment (Wulder, 1999) Most of the current work has been reported in pure

or mixed coniferous stands, or stands with only simple combinations of one or twoconifer and deciduous tree species (Gerylo et al., 1998) Deciduous tree crowns aremuch less simple (typically with multiple maxima and less distinct edges), and arethus more difficult to delineate using existing procedures (Warner et al., 1999).Despite tremendous growth and recent successes in specific areas, such as crownsegmentation algorithms (Gougeon, 1995; Hill and Leckie, 1999), this area of imageanalysis in forestry appears surprisingly poorly developed; a poor relation to thebetter tested and better understood algorithms for use with L-resolution data Imagerywith 1 m or better spatial resolution have been available for decades, and are nowavailable from satellite platforms These remote sensing data are ready-made forforestry applications, and are ideally suited to help answer some of the same ques-tions now addressed by extensive field and aerial photographic work, at lower costand higher accuracy Yet comparatively few successful projects using these imagerysystems for species composition mapping, crown closure estimation, age, and struc-ture mapping have been reported Perhaps the difficulty in geometric correction andregistration has prevented more widespread use of the data Perhaps the preoccupa-tion with the relatively coarse resolution satellite imagery has prevented a moreconcerted effort in the airborne arena

Certainly, the image processing tools for use in high spatial detail applications,with or without high spectral resolution (hyperspectral imagery), are increasing insophistication and value; after considerable development, the application of highspatial detail multispectral imagery in forestry remains underexploited, but the poten-tial is being recognized.

F OREST A GE

It would be difficult to argue that forest age can be remotely sensed directly; manyforests are hundreds of years old, but there are not many leaves hundreds of yearsold growing on trees! But forest age is surprisingly difficult to specify directly, even

in the field Typical measurements are age since establishment for young forests, orbasal area weighted age in traditional forest inventories; this latter measure may notincrease by a simple one-year-per-year For example, with a thinning treatment thebasal area weighted age can actually decrease What is the age of a shelterwoodstand, for example? Site productivity can be a major confounding factor in trying

to estimate age directly, even when the species and density are reasonably uniform.Direct correlation of stand age and remote sensing spectral response approaches

a classic “nonsense” correlation (De Wulf et al., 1990) Age is a descriptor orsurrogate of forest conditions but not itself an attribute (Cohen et al., 1995a); inessence, a third-order effect on spectral response is what is of interest — the changes

in physical structure and composition — such as the size and density of tree speciesare captured over time in a variable called age In conifer forests in Oregon, Cohen

et al (1995a) related these changes to satellite spectral response in broad age classes(e.g., <80, 80 to 200, and >200 years), but it is important to remember that thesedifferences in physical structure and composition that are characteristic of agingstands are not directly sensed — rather, it is the differences in illumination, absorp-tion, and shadows which are related to the different sizes and density of trees(Gemmell, 1995) Thus, there is only a small possibility that the relationshipsbetween age and spectral response will be strong or invariable across a forested area

It is more likely that remote sensing data will be suggestive of age, or age classes,because of the differences in tree size and density, understory, canopy development,nutrient status, and species among young and old forest stands

In very young (<35 years) homogeneous Douglas-fir stands in Oregon, a strongrelationship between age and TM reflectance was found (Fiorella and Ripple, 1993).Using a neural network model in this same region, Kimes et al (1996) consideredvariability within regenerating clearcuts (<50 years) In predicting the year logged,the model network yielded an RMSE of approximately 8 years using the image dataalone; adding topograpic data decreased these errors to approximately 5 years.Further decreases in errors in predicting stand age from Landsat TM and topographicdata were obtained using site-specific information such as planting year, site prep-aration, and species planted These decreases in error were considered evidence thatthe Landsat imagery contained unexplained variability at the scale of the sites related

to the number of replantings, density of plantings, variations in site preparation, andsoil conditions When that variability was accounted for (using texture measures),the neural network model appeared to produce satisfactory inference of forest age

for young stands in this region The performance of the neural network modelprovided a baseline or standard against which the more desirable physically basedinvertible reflectance and growth models could be developed and compared.Another approach to infer forest stand age is to define the relationship betweenage and vegetation structure, development, or type in a classification (Niemann,1995) Using this approach, stand development was classified as a surrogate of standage in airborne imagery acquired over a forest area on Vancouver Island, BritishColumbia The measured reflectance and age differences among the stands werecreated as a result of management activities, such as planting and harvesting Standsgreater than 60 years were generally mixed species with a significant understory;stands between 20 and 60 years were typically more uniform, with a closed canopywithout the understory; stands less than 20 years were more open (plantations)(Figure 7.2) Reported classification accuracies for the corresponding age differenceswere on the order of 70% correct In intensively managed areas, such as plantations,age may be a surrogate for a thinning cycle or other management treatment (De Wulf

et al., 1990)

In another example, Hall et al (1991a) used an understanding of boreal forestsuccession following a fire to classify different forest stand ages on satellite imagery.The use of successional age classes in forest sites in Puerto Rico, Costa Rica, andMississippi was reported by Sader et al (1989) The investigation was designed to

FIGURE 7.2 Mean spectral response curves suggest that as forests age, spectral response

changes even if pixel sizes are small Here, for a few Douglas-fir forest age classes, the differences were large enough that classification of age can be accomplished with a high degree of accuracy Age differences were largest in the younger age classes (From Niemann,

K O 1995 Photogramm Eng Rem Sensing, 61: 1119–1127 With permission.)

Wavelength ( m)

500 0.00

explore the possible link between NDVI, canopy foliage biomass, and woody mass in main bole and tree branches Confusing the analysis were the topographiceffects on image and forest conditions, and the multiple successional pathways thatcould be interpreted from periodic historical aerial photography (e.g., grass to brush

bio-to forest regeneration; row crops bio-to grass bio-to forest regeneration) Across a wide range

of forest conditions, weak or no significant correlation was found between sional age class and various transformations of Landsat TM data, and only weak tomoderate correlation to similar bands of 10 m spatial resolution airborne multispec-tral scanner data As in Oregon (Fiorella and Ripple, 1993; Kimes et al., 1996) and

succes-in British Columbia (Niemann, 1995), stand age only succes-in the very young stages ofregrowth could be predicted reasonably well These young areas were almost alwaysbrighter (less shadows) and more variable (more openings through which the back-ground could be observed) than older forest areas

Peterson and Nilson (1993) used successional age trajectories of forest tance and calibrated these with field observations of stand age An age trajectorycould be constructed in two ways (Nilson and Peterson, 1994):

reflec-1 Collect multiple measurements at the same location over time (the classicLocation Through Time or LTT) design;

2 Collect observations at different locations thought to represent the ent conditions which exist over time — this approach is sometimes known

differ-as the Location Through Condition (LTC) design, and is similar to theclassic chronosequence

The long-lived and highly variable nature of most forests would preclude sive use of LTT sampling An appropriate use of LTC sampling, though, requiresthat airborne or satellite spectral response patterns from numerous sites be available

exten-to represent distinct stages of development as the forest ages; in some ways this issimilar to the earlier spectral library concept and suffers from the same problems

— in an almost infinite variety of conditions, how to acquire such a huge number

of observations? One way is to model the reflectance that would be generated if asensor was there to record it Fewer observations are needed, but there may be asacrifice in precision and accuracy Using airborne radiometer data, Nilson andPeterson (1994) found that the primary controls on reflectance in their forest standswere changes in canopy closure, tree/canopy LAI, species composition, and back-ground reflectance Because the age dependence of reflectance was strong, thereflectance of a stand at any time during its existence could be predicted by thesuccessional age trajectory Then, the important tasks were to periodically checkdifferent stands with airborne or satellite images in which the effects of sun angle,view angle, and phenology were controlled

The management implications of this work suggest that imagery be acquiredperiodically, in the same stands, to determine if there is a difference between theexpected reflectance (from the successional age trajectory model coupled with agrowth model) and the observed reflectance (Jupp and Walker, 1997) Note that themodel would be constructed using LTC sampling, but the ongoing monitoring would

be done by LTT sampling Controlling the differences not due to actual changes

would be possible If no differences were found between predicted and observedreflectance, then the stand would be thought to be experiencing normal forestdevelopment However, if significant differences were found, then perhaps a distur-bance agent or large-scale change in growth conditions should be considered likely.This would require detailed investigation In the Oregon Transect project, a similarlogic was used in forest stand LAI estimation (Peterson and Waring, 1994) Theremote sensing imagery were used to determine if the ecosystem process modelFOREST-BGC predictions of LAI were reasonable (Running, 1994); if the remotesensing imagery suggested that LAI was higher or lower on a given site than themodel prediction (given the various assumptions of climate and soils), then (1)further investigation on the ground was warranted or (2) model parameterizationand functioning could be subjected to additional testing and potentially improved

to bring predictions in line with observations In essence, once remote sensingobservations confirmed that there was a discrepancy in the observations vs modelpredictions, a new task for remote sensing and field visits was to try to identify thecause of the differences

T REE H EIGHT

The uses of aerial photography in tree height estimation are well known to forestmanagers and remote sensing scientists (Titus and Morgan, 1985; Avery and Berlin,1992; Sader et al., 1989; Kovats, 1997), as are the uses of height in the development

of other information of interest to forest management; for example, in using allometricrelationships with the crown diameter to predict other forest variables such as volume(Hall et al., 1989b; Hall et al., 1993) Digital airborne and satellite remote sensinghas not been very successful in producing reliable estimates of tree or canopy height;

in essence, the biophysical relationships between height and spectral response arerarely strong enough to justify model development The few exceptions can be found

in highly site-specific studies designed (1) to relate standard photogrammetric ciples to shadows on imagery; for example, Shettigara and Sumerling (1998) and (2)

prin-to classify or estimate height as a relative attribute in a few general height classes

In this latter instance, for example, two height classes of semideciduous forestwere mapped from Landsat TM data in an Amazon study area — Class 1: semide-ciduous forest, and Class 2: tall semideciduous forest (Marsh et al., 1994) However,retaining these two classes did not necessarily represent a particularly logical clas-sification structure for the area, and combining them into a single class improvedthe overall classification accuracy Active sensors such as radar and lidar represent

a more promising solution to remote tree height estimation (Hyyppä et al., 1997).Airborne SAR image data, under certain specific conditions (such as pure coniferstands with simple structure) have been found to be significantly correlated withtree heights (Weishampel et al., 1994) In Alberta, subsequent predictive models for

an independent sample of stands indicated that standard errors were similar to thosecontained with the existing GIS forest inventory (derived by air photo parallaxmeasurements) (Wilson, 1996)

Of the available remote sensing instruments, it appears that lidar measurements

of tree height have the greatest potential Since the early 1980s, lidars have been

used experimentally to improve estimates of stand forest biomass and volume(Maclean and Krabill, 1986; Nelson et al., 1988) Early problems included the factthat the laser profiler obtained heights from the shoulder of the tree crown, as well

as the peak; comparisons to field measurements showed that the spot lidar wouldsystematically underestimate tree height The system worked better in softwoodstands where tree crowns were more distinct Tree height variability was greater inthe lidar data of hardwood stands when compared with field measurements Arefinement is to include lidar estimates of canopy density (or porosity); such anestimate can be produced by considering the number of times the laser pulses directly

to the ground The lidar-generated tree heights could be used in estimates of biomassand volume, but tree diameter variation accounted for much of the variation in site-to-site biomass estimates because tree diameter is far and away the most importantcomponent in biomass and volume equations Combining lidar sensors with a spa-tially explict remote sensing device, such as a digital camera or spectrograph, willprovide the ideal solution to the problem of remote height determination (Means etal., 1999; Lefsky et al., 1999a,b)

S TRUCTURAL I NDICES

Structural indices based on field measurements (Lahde et al., 1999; Latham et al.,1998) and remote sensing measurements (Cohen and Spies, 1992) have been soughtfor a variety of forest management applications that require information on a com-posite or summary measure of forest stand structure In general, structure is animportant factor in affecting many ecological responses (Lindenmayer et al., 1999;Spies, 1997) From a remote sensing perspective, the importance of structure lies in

the use of linear combinations of field data interpreted as structural indices It is

hoped that these indices can be used to replace individual forest attributes — withtheir generally low correlation with spectral response — with a composite of fieldmeasurements that represents adequately the differences amongst forest stands ofinterest, but increases the strength of the overall relationships (Gemmell, 1995).Cohen and Spies (1992) first developed a structural complexity index to capture thestructural diversity and upper canopy conditions of closed canopy (young to old-growth) stands in the Pacific Northwest The idea was to compare field-determinedstructure with data obtained from image semivariograms (Cohen et al., 1990) Thesesemivariograms were hypothesized to capture subtle structural characteristics overthe area of the stand, or a transect sample through the stand Support for theirinterpretation was obtained when higher spatial detail imagery (SPOT 10 m pan-chromatic) showed stronger correlations to the structural index than did lower spatialdetail imagery (Landsat TM) (Cohen and Spies, 1992)

Danson and Curran (1993) developed a composite variable called canopy volume

to describe the surfaces of leaves and branches in three dimensions Canopy volume

is a synthetic variable, constructed from field measures of density, dbh, height, and

to a lesser extent, cover, using principal components analysis In a plantation forest,they hypothesized that canopy volume would be greatest for the older, thinned standswith a few large trees (low density) The composite variable would be directly related

to remotely sensed response because a large canopy volume would result in greater

interaction with the canopy and a lower spectral response By reducing the field data

to a single composite variable with a stronger hypothetical relationship than theoriginal individual field variables to spectral response, they sought to clarify thecausal relationships between forest structure and spectral response (Danson andCurran, 1993: p 61–62):

In the young stands, tree density was high, there were few gaps within the canopy, and the volume of the canopy was low The LAI and biomass of these young stands had, however, already reached high levels In this environment there will have been little penetration of radiation into the canopy, and as a result the level of reflected radiation for the canopy as a whole was relatively high.

The interaction of near infrared radiation with the young stands will have been in the form of multiple scattering due to the higher reflectance and transmittance of the needles at these wavelengths However, there will have been little penetration of radiation deep into the canopy because of the absence of gaps, and the level of reflected radiation would therefore again be relatively high.

In the older stands the tree density was low because of the removal of trees by thinning The individual trees and therefore the canopy volume were large, and there were many gaps within the canopy However, LAI, biomass, and canopy cover were maintained

at a high level In this environment, the initial penetration and subsequent absorption

of red radiation will have been great, giving rise to a larger amount of canopy shadow and lower levels of reflected radiation Similarly, the penetration of near infrared radiation will also have been high with multiple scattering and absorption taking place deep within the canopy A smaller percentage of the incident near infrared radiation will therefore have emerged from the top of the canopy giving rise to lower recorded near infrared radiance.

It is proposed that it was this set of mechanisms that gave rise to the observed dependence of … radiance on the age of the stands …

In a British Columbia study area, a structural complexity index obtained byapplying principal components analysis to field-measured stand parameters such asbasal area, dbh, stem density, and crown diameters was augmented with estimates

of variability of some of the parameters (Hansen et al., 2001) A strong loading by

field measurements of stand basal area, stem counts, and crown diameter wasexpected, and was in fact necessary in order to interpret the meaning of the compositeindex However, the index was also highly correlated to stand height (R = 0.75),and stand age (R = 0.81), which were not source variables, but were independentlyrelated to variables in the structural complexity index Strong correlation coefficientsbetween the index and each of the individual structural variable supported thesuggestion that the value of the index lies in the ability to capture the variance found

in multiple stand parameters in a single attribute (Cohen and Spies, 1992) Thestrong correlation between structural complexity and stand age, for example, sup-ported the use of a structural complexity index as a surrogate variable for stand age

In this area, as is sometimes common elsewhere, age sampling by increment corescan be unreliable

The generation of composite field indices, such as canopy volume (Danson and

Curran, 1993) and structural complexity (Cohen and Spies, 1992; Hansen et al.,

2001) has been accompanied by the search for a similar composite index in

multi-spectral remote sensing imagery Why relate individual multi-spectral bands to the tural complexity index when the same logic used to create the index can be applied

struc-to the image data? The NDVI transformation was examined by several authors (Sader

et al., 1989; Cohen et al., 1995a), but in many forests NDVI does not appear to be

a good predictor of stand structure variables One problem with the NDVI is that ituses only the red and near-infrared bands, and the shortwave infrared bands wouldappear to have important information that would thus not be included in the index.Earlier work has shown that TM data transformed into the brightness, greeness,and wetness data space with the Tasseled Cap coefficients (Crist and Cicone, 1984)were sensitive to structural characteristics of forests (Horler and Ahern, 1986; Cohen

et al., 1992) TM wetness (not necessarily an interpretation of water content) isheavily weighted by the contrast between shortwave-infrared and visible bands InFigure 7.3 the TM wetness index is plotted against the structural complexity indexfor the 38 conifer stands sampled in British Columbia by Hansen et al (2001) Thecorrelation coefficient for this relationship was significant and moderate (R = 0.58)

FIGURE 7.3 Landsat TM wetness index values are plotted against a field-based structural

complexity index (SCI) comprised of measures of basal area, height, and crown dimensions for 38 conifer stands sampled in British Columbia (Hansen et al., 2001) The position of the stands on the graph indicates stands known to differ by geographically (altitudinal zonation) different “structural” and “wetness” positions The forest stands differ in their structural complexity measured on the ground and can be distinguished spatially using the TM wetness index In areas without detailed forest inventory information, such as age classes, the wetness index may be a useful surrogate measure of stand development.

1.0 2.0 3.0

R = 0.59**

R = 0.34 2 ICH Climax ICH/ESSF Seral ESSF Climax

ICH Climax

ICH/ESSF SeralESSF Climax