L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term...

25
LANDSAT TM IMAGERY IN LANDSCAPE ARCHAEOLOGY: DETECTION AND MODELING Landscape archaeology Landscape archaeology approaches human activity by placing it in the physical landscape. While physical landscapes are multi-scalar and can be defined in a variety of ways, they also contain discrete systems including geomorphology, hydrology, topography and vegetation. Landscape archae- ology is not merely the process of placing people on maps. Landscapes are dynamic in a variety of ways, and static views are always overly-simplistic. More over, humans are not points on the landscape, but rather points in the landscape. People change the land, and the land changes people. This is the human-environmental dynamic that is the focus of many regional studies (ZEDENO 1997). While in the post-industrial era most of us have only a tenuous direct interaction with the landscape, the relationship was much more direct for many peoples of the past. This means that any reasonable historical or ar- chaeological knowledge must account for both the physical realities as well as perceptions of past landscapes. This demands not only a large body of technical skill and knowledge on the part of researchers, but also the mecha- nisms to visualize, share and explicate multi-temporal and spatial data. This article illustrates two uses of Landsat TM multispectral imagery in archaeological investigations, highlighting the ability of remote sensing not just to find and map, but also to analyze and illustrate. Using two projects joined only by their landscape approach, an emphasis on vegetation, and the use of Landsat data, we hope to explicate techniques developed both for the detection of potential archaeological sites and the modeling of long-term human-environmental interaction. The topics under study are, of course, extremely complex, and the reader should not expect to find here the com- plete publication of our findings; we have deliberately over-generalized and omitted relevant detail to allow a focus on one discrete goal. Additionally, the discussion is intended to proivide sufficient detail about the processing of the Landsat data to allow other to adopt and modify these rechniques. In both cases we will illustrate how to define the research ques- ©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale 1

Transcript of L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term...

Page 1: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

LANDSAT TM IMAGERY IN LANDSCAPE ARCHAEOLOGY: DETECTION AND MODELING

Landscape archaeology

Landscape archaeology approaches human activity by placing it in the physical landscape. While physical landscapes are multi-scalar and can be defined in a variety of ways, they also contain discrete systems including geomorphology, hydrology, topography and vegetation. Landscape archae­ology is not merely the process of placing people on maps. Landscapes are dynamic in a variety of ways, and static views are always overly-simplistic. More over, humans are not points on the landscape, but rather points in the landscape. People change the land, and the land changes people. This is the human-environmental dynamic that is the focus of many regional studies (ZEDENO 1997).

While in the post-industrial era most of us have only a tenuous direct interaction with the landscape, the relationship was much more direct for many peoples of the past. This means that any reasonable historical or ar­chaeological knowledge must account for both the physical realities as well as perceptions of past landscapes. This demands not only a large body of technical skill and knowledge on the part of researchers, but also the mecha­nisms to visualize, share and explicate multi-temporal and spatial data.

This article illustrates two uses of Landsat TM multispectral imagery in archaeological investigations, highlighting the ability of remote sensing not just to find and map, but also to analyze and illustrate. Using two projects joined only by their landscape approach, an emphasis on vegetation, and the use of Landsat data, we hope to explicate techniques developed both for the detection of potential archaeological sites and the modeling of long-term human-environmental interaction. The topics under study are, of course, extremely complex, and the reader should not expect to find here the com­plete publication of our findings; we have deliberately over-generalized and omitted relevant detail to allow a focus on one discrete goal.

Additionally, the discussion is intended to proivide sufficient detail about the processing of the Landsat data to allow other to adopt and modify these rechniques. In both cases we will illustrate how to define the research ques-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 1

Page 2: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

tions and method of investigation, and then how to get the Landsat TM data to yield answers (or at least clues). A rudimentary understanding of multi­spectral remote sensing is assumed, but the focus is on basic techniques that have proven successful. While satellite remote sensing is one of the most discussed topics in archaeology, remarkably little has been published that details the steps researchers are using, and only a small handful of institu­tions teach remote sensing in that manner. Published manuals focus on tech­niques for other disciplines and often address scales of investigation far larger than those used by archaeologists. Here, then, we show how we have adapted those techniques for our similar, but markedly different, ends.

The Rio Plátano Biosphere, Honduras: visual detection of anomalies and potential archaeological sites

The Río Plátano Biosphere Reserve is a 5250 km2 natural sanctuary, and the largest track of natural land remaining in Honduras (Image 1, Fig. 1). As such it represents one of the last extensive inhabitable tracts of land on the planet that has not undergone extensive modification at the hands of humans. The area of the biosphere has been occupied by humans for a mini­mum of 1500 years, and is today home to a Mestizo population, as well as Miskito and Pesch (Paya) Amerindians, and Garífunas (Afro-Caribbeans). Population in the biosphere at present is, however, scarce, and the prehis­toric population may have been larger. The biosphere is composed primarily of tropical and sub-tropical rainforest and is not only heavily vegetated but also rugged in terrain. It remains largely unexplored, and little archaeologi­cal work has been conducted here.

While the precise affiliations of the prehistoric inhabitants remain un­known, they exhibit affinities with both Mesoamerican and South American cultures. The size and nature of the known remains indicate the presence of agricultural or at least semi-agricultural societies. This is the key for detec­tion of prehistoric activity in the region. Direct detection of archaeological sites is implausible both because of the small and rather indistinct nature of the remains, and because penetration of the canopy with remote sensing techniques is difficult.

The cultural characteristics of the area point to semi-permanent resi­dences on the floodplains of the rivers and along tributaries. The rivers were seemingly a very important means of transportation that served in lieu of the large Maya causeways known in other parts of Mesoamerica. These residen­tial areas would have been surrounded by fallow and active agricultural fields. Sites are not limited in proximity to major rivers, but are also found on the edges of valleys, ridge crests and along tributaries. Along the Paulaya River, two large sites, Limoncito and the Tulito Complex, are found more than a

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 2

Page 3: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

kilometer up smaller tributaries. Habitation sites along tributaries usually occur in areas of flat terrain where the steep river valleys widen (BEGLEY

1999). The limited evidence indicates that the population would not have been very large. For slash and burn agricultural practices only 200-250 peo­ple per square mile can be supported (SEVER 1998).

One should not envision sites on such a grand scale as Tikal or Copan “hiding” in the biosphere. Known sites are more modest and range from small square platforms to large, linear longhouse structures with ball courts, plaza and patio groups, and cobble paved causeways. Many of the sites consist of homogenous clay mounds, some of which are paved with cobbles. To date, no stepped pyramid structures have been recorded in the region (BEGLEY 1999).

While the prehistoric occupation and use of the area was moderate, this should not lead to the conclusion that it had no environmental impact. The rainforest as an ecosystem is extremely sensitive to disturbances. Nutri­ents are stored and redistributed through the canopy system and the soils tend to be poor and fragile. Archaeological and ethnographic evidence indi­cates that swidden (slash and-burn) agriculture predominates in this sort of environment. When areas are cleared for agriculture and then abandoned, the regeneration of vegetation is slow, limited to select species, and may never be complete. As the poor soils make it impossible to maintain fields for ex­tended periods, fields are cropped for one or two years and then abandoned. The fields are then allowed to lie fallow for five to twenty-five years before they are re-cleared and then reused. During the first five years of the fallow period, dense weeds and underbrush will colonize the land. After five years woody species will begin to grow and the underbrush will decrease. The Miskito farmers leave their fields fallow for five years, stating that it is easier to clear an earlier-utilized field that has been regenerating vegetation at a slow rate than to clear virgin forest (DODDS 1998).

The short-term impact of such activity is a clear diminution of biomass, biovigor and biodiversity. Limited cultivation with lengthy fallow periods seemingly allows a good recovery of the fertility (Image 2, Fig. 2), although it is not known if there is ever complete recovery. With the fragility of the rainforest soils, and the pressure to limit the length of fallow periods, how­ever, the more common situation seems to be a progressive diminution of fertility (Image 3, Fig. 3) (PARK 1992, GOUDIE 1984). The long-term impact of swidden agriculture is less fully understood. Scientific consideration of swidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned with other issues. At the present it is unclear whether the soil degradation caused by swidden agriculture is permanent (Image 4, Fig. 4) or eventually reversed (Image 5, Fig. 5).

The details of the regeneration of rainforest vegetation after swidden agriculture must await further study, including considerations by archaeolo-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 3

Page 4: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

gists. The parameters are, however, clear enough to indicate that at least in some cases there is a noticeable diminution of vegetation coverage. This dimi­nution is noticeable in biomass (the density of vegetation coverage), biovigor (the vitality and health of the vegetation coverage), and biodiversity (the number of species present). All three of these characteristics are detectable and measurable using multi-spectral data. Lack of biodiversity is not, how­ever, a parameter used in this study. The abundance of plant species in the rainforest (an estimated 2000 species in the Rio Plátano Biosphere) makes species identification extremely challenging, and this is complicated by the three-tiered canopy of the rainforest; the top of the canopy is all that is measured by the Landsat TM Remote Sensing platform, leaving specific spe­cies of the middle and bottom zones undetectable.

As mentioned, the archaeological sites and trails in the Rio Plátano Biosphere tend to be small and simple in comparison to other areas of Mesoamerica, and direct detection is quite difficult if not impossible. Never­theless, a detection technique that focuses on biomass and biovigor is a suit­able indirect detection technique. Structures tend to be formed either from stone blocks or by making mounds of clay, both of which support vegetation that appears to the remote sensing platform less intense than surrounding vegetation. Heavily used trails should also evidence such diminution, although none were detected in the biosphere, presumably because they are too small to affect the upper canopy and thus cannot be detected by the 30m resolu­tion of the Landsat imagery.

The use of Landsat for archaeological prospection in the biosphere be­comes, then, a search for anomalies. Discrete locations that evidence less biomass and biovigor than their surroundings are anomalous and may repre­sent areas of swidden agriculture or man-made structures. Anomalies will also include, however, current and recent villages, geological and topographic features, storm damage, landslide and tree falls. The reality remains that surety in areas of such heavy vegetation coverage can come only with ground­truthing.

Because of the multiple variables involved, as well as the uncertainty of the results, an approach based on visual detection rather than automated classification is to be preferred. The technique emphasizes developing algo­rithms that maximize the human ability to detect changes in biomass and biovigor. Because perception varies from individual to individual, there is no ideal technique, but rather a spectrum of various results, some of which offer the least possibility for misperception of data to occur. There is also a corre­sponding payoff as the same techniques can be used as a basis for automated classification, as illustrated later in this document.

Examination of this area utilized two Landsat TM scenes (Path 16, Row 50), from 1986 and 1994. Out of concern of international looting in the Biosphere, the coordinates of the sections of the scenes utilized are not indi-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 4

Page 5: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

cated here. Atmospheric correction using the “dark-water” technique has been performed on each image (RAABE, STUMPF 1997). The 1994 scene is more difficult to use than the 1986 scene as the image, for reasons unknown, suffers from a very limited dynamic range as well as banding.

COLOR COMPOSITES

Standard viewing of Landsat TM images is often done with color-com-posites; projecting band 3 as red, band 2 as green, and band 1 as blue yields a result close to what humans perceive as “true color” (Image 6) (JENSEN

1996). Other color combinations are possible and useful. Projecting band 4 as red, band 3 as green and band 2 as blue yields a pseudo-infrared image, where the intensity of red represents the vigor of the vegetation (Image 7).

A combination of bands 3,4,1 is excellent for discriminating water and vegetation (Image 8). Band 4, the near-infrared, is projected as green. Water absorbs the near-infrared almost entirely, while vegetation reflects about 50% of the near infrared, thus providing the crisp delineation. In the Rio Plátano Biosphere, the combination has been used to map small streams. Color com­posites are perfectly suitable for visual detection of major anomalies in the rainforest vegetation coverage. No further processing is necessary to detect, for example, the cultivation and habitation areas in this image.

BAND RATIOING

For detection of subtleties in the coverage however, a more vigorous approach using ratios is necessary in the rough terrain. The problem is largely one of shadows in the image. A close examination of the crest in the upper­left corner of Image 8 illustrates the problem both the problem of shadowing and the benefit of ratioing. The northwest slope is in shade; the southeast slope is not. Thus from mere visual examination of the color combination, it is not possible to determine with surety the relative intensity of the vegeta­tion or if anomalies are hidden in the shadow (Image 9).

Ratioing helps to minimize the effects of shadowing by a simple princi­ple. The pixel values for all bands except band 6 (thermal) are diminished by the same percentage in shadowed areas. An examination of the pixel values on both sides of the crest indicates this reasonably well. Vegetation coverage on both sides of the crest is about equal, but the pixel values returned are different. Thus when displayed as a 4,3,2 (pseudo-infrared) a visual interpre­tation might assume greater vegetation intensity on the southeast slope (Im­age 10). The eye is adept at determining the depth of shadow, and the prob­lem is not severe when dealing with obvious shadow as along the crest. But in areas of shadow that are small and not noticed, or in non-visual statistical interpretation, the problem can be severe. For example, a non-ratioed inter-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 5

Page 6: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

pretation would look at the higher values of bands 4 and 5 (both of which plants reflect strongly) for pixels 1 and 2, and determine that the vegetation coverage of 2 is heavier than 1 (Image 10).

Checking the ratio of band 4 to band 7 reveals another truth: for pixel 1, 54/8=6.75, for pixel 2, 120/16=7.5. In other words, the vegetation inten­sity is about the same. This is illustrated in Image 11, which displays the value of band 4 divided by band 7 and reveals relatively uniform vegetation coverage, although it should be noted that the effects of the shadow cannot be completely overcome.

The efficacy of ratios is also illustrated if we return to the area viewed in Image 9. By comparing a 3,2,1 color composite and a ratio of bands 4/7, habitation areas become glaringly apparent by their relative lack of vegeta­tion (Image 12). This can be seen well enough in the color composite, but the ratioing both solves the problems of shadows and makes the interpreters job easier in spotting these. Because, however, the ratioing removes the shadows it also removes the clues to topography human eyes rely on, so the topogra­phy of the image becomes more difficult to interprete.

Bands 3,4,5 and 7 of Landsat TM imagery are the most sensitive to vegetation, and thus are the most useful in the detection of human impact on vegetation (JENSEN 1996) The limited dynamic range of band 3 in this par­ticular image has made it rather unusable, so an emphasis has been placed on bands 4 and 5, the mid-infrared, and band 7, the infrared. A ratio of bands 4 and 5 often indicates density of canopy coverage. In forestry, the ratio is normally band 4/5, which renders denser canopy brighter. As the object of interest in this study is, however, human activity, we reverse the ratio (5/4) rendering less-canopy brighter. This is in keeping with the standard methodology of visual inspection-make the objects being searched for brighter (Image 13, Fig. 6).

The ratios 7/5 and 7/4, both of which indicate general vegetation vigor, are also extremely useful (Image 14). As the goal is to create an image to be used in locating cultural activity, the ratios render reduced vegetation as brighter. All three ratios, 5/4, 7/5 and 7/4 yield different, yet complimentary, results and it is desirable to combine them into one image. 5/4 is projected as red, 7/5 as green and 7/4 as blue. The red (5/4) and green layers (7/5) were clipped at 99% (excluding the first and last 0.5% of the data, and blue (7/4) was not clipped (Image 15). In this image areas with lesser vegetation cover­age are indicated by yellow (Image 16). Not only does the image formed by the ratios highlight the areas of cultivation and habitation well, it also reveals subtleties in the rainforest vegetation not visible in a standard color compos-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 6

Page 7: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

ite projection. It is, for example, possible to differentiate vegetation cover­age along streams and rivers and more mountainous terrain.

The subtleties detected are, in fact, probably too many for purposes here. A close-up of an area of habitation becomes as cacophony of visual data (Image 17). This can be remedied, however, by applying a smoothing filter of 3×3 pixels (Image 18). Filtering in such a manner reduces the resolution of the Landsat imagery so much that effectively nothing smaller than 90m in size will be noticed. The reality is, however, that in the massive variation of the rainforest, higher resolution is too confusing, and it is better to detect something than nothing.

At this stage the image is quite interpretable, but one problem remains. The ratioing and smoothing has removed much of the visible clues to topog­raphy. While not a critical problem, in this approach emphasizing visual in­terpretation, it is desirable to restore this topography, especially as the solu­tion is simple. The visible bands 1,2,3 suffer from moderate to severe shad­owing. By using any of these as an intensity layer, the shadowing can be used to restore a sense of topography without adding any noticeable distortion to the relative pixel values indicating evidence of human impact on vegetation coverage (Image 19). In this Landsat scene, band 1 contains the greatest dy­namic range and thus is most suitable; band 3, less subject to atmospheric distortion, is usual a good choice as well.

The introduction of band 1 as an intensity layer also restores some of the detail lost via the smoothing filters. Any suitable georeferenced imagery could be used as the intensity layer. The higher resolution of a SPOT scene, or aerial photography would be ideal, if available. Likewise a digital eleva­tion model could be used to render a three dimensional representation, al­though this would not restore the lost resolution.

SUPERVISED CLASSIFICATION AS A VISUAL SIGNPOST

The final problem present in visual interpretation is the still substantial amount of variation on the image. While recent cultivation and habitation is clearly evident and easily locatable, possible archaeological sites and past cultivation could be difficult to locate. The easiest solution to this is a super­vised classification based on the three ratios used. Because the ratios have already been coordinated to yield higher pixel values for areas of interest, it is easy and quick to perform, in fact, a manual supervised classification (effectively a parallelpiped classification) (JENSEN 1996). When the pixel values of the three ratios are stretched from 0 to 255 (min-max linear contrast stretch), the val­ues that indicate areas of cultivation and habitation are determined, and a formula for a new (red) layer is written that incorporates all of these values: IF b4/5>201 AND b7/b5>190 AND b5/b4>210 THEN 255 ELSE null (Image 20).

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 7

Page 8: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

The classification is far from fool-proof or reliable. It identifies areas that are probably not anthropogenic in origin, and is not vigorous enough to allow for older, more subtle vegetation diminution. But lowering the thresh­old value is not a practical solution. The formula IF b4/5>180 AND b7/ b5>180 AND b5/b4>200 THEN 255 ELSE null introduces so many possi­bilities it merely populates the landscape with noise (Image 21, Tav. VIa).

A supervised maximum likelihood classification conducted with soft­ware using training regions from other areas of the image also works ad­equately, but populates the image with some noise (Image 22, Tav. VIb). The preference for the manual method is largely one of ease – the manual classi­fication can be easily adjusted by changing the threshold variables. The auto­mated classification can be adjusted by redefining training areas and classifi­cation techniques – a much more laborious (and disk-space consuming) proc­ess.

At this stage we have reached the limit of identification of past human activity in the landscape using an automated or semi-automated method, and close visual interpretation compounded with reasonable archaeological and historical knowledge is the next step. It is for this very reason that the meth­odology used here from the start emphasized the visual rather than purely numeric solutions. This can be illustrated using two example areas, one a suspected archaeological site, the other a known archaeological site.

DETECTING PREHISTORIC ACTIVITY

The prehistoric peoples associated with archaeological sites utilized the landscape in much the same manner as those practicing contemporary swidden agriculture. The dense vegetation of the rainforest combined with the rough terrain makes waterways the only easy form of transport. The beginning rule is, therefore, to focus on remote sensing anomalies in proximity to water­ways. A difficulty, of course, is determining where past rivers and streams may have been, and open-minded consultations with geologists is a sine qua non for this sort of work.

The illustration of the method will begin with a suspected archaeologi­cal site. The anomaly sits along a stream feeding a major river. Contemporary swidden agriculture is practiced along this stream within one kilometer of the anomaly, but the area around the anomaly is not currently active. The anomaly seems to be on a small rise, adjacent to lowland areas (Image 23, Tav. VIc).

The pattern of modern swidden agriculture is one of sporadic expan­sion from existing plots-isolated plots are not encountered, and thus the anomalies must be something else. The position of this anomaly fits the model of prehistoric swidden agriculture as well as our interpretation of the region of one where civilization has contracted; i.e. what may be evidenced here is

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 8

Page 9: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

a prehistoric exploitation of a remote region not currently in use. Closer examination of the anomaly reveals it to be a triangular shaped area with sides of approximately 100m×100m×200m (Image 24, Tav. VId). Normally such linear features can confidently be declared anthropogenic, but in this fault-riddled region this is never certain. While in the 3,2,1 color composite the anomaly is whitish, the detection algorithm indicates that it may be an area of diminished vegetation. An angular area of diminished vegetation in the right location is a prime candidate for an archaeological site.

The information is not conclusive, and this highlights the desirability of multiple datasets. The above detection is based on 1986 imagery. With that alone, serious questions would remain as to the nature of the anomaly-it could be a small cloud, or an odd-shaped tree fall. The anomaly appears in its trian­gular form, however, in a 1996 Landsat image, and a 1995 JERS1 radar image (Image 25, Fig. 7). This multiple appearance rules out data errors, clouds, and diminishes the likelihood that we are looking at a tree fall or landslide.

Unfortunately, the radar data is so cluttered it cannot assist in identifi­cation. But two main possibilities remain: a geological feature, or an anthro­pogenic feature. There is no reason to anticipate a geological anomaly of this shape in this spot, so archaeological site becomes the primary interpretation. Nevertheless, ground-truthing will be necessary for verification.

Our final example, a known archaeological site, is more difficult to deal with because the 1996 Landsat scene available to the project is much poorer than the 1984 scene used above. The problem is one of dynamic range. For reason unknown, all of the bands have pixel values that are lim­ited, for the most part, to ranges below 100. Additionally, the pixel values were clumped, so that a max-min linear contrast stretch helps, but leaves gaps (Image 26).

While undoubtedly there is an explanation for this, such concerns were beyond the expertise and more importantly time of the researchers. Visual interpretation of the image is plagued by lack of detail and streaky data (Im­age 27). While processing techniques exist to reduce streaking, none were applied here, as they involve complicated filters. As the investigation of ar­chaeological sites involves analysis of sometimes just a few pixels, a minimalist approach is preferred. Filters can introduce data artifacts that can be mis­taken for reality by those who do not fully understand how the filters work. In other words, we did not use a fast-Fourier transformation to remove streak­ing, because we do not feel sufficiently familiar with the functioning of this transformation.

Several archaeological sites are known in this region in the biosphere, although they have not been professionally investigated. One of these sites is located on the 4km long plateau illustrated in Image 27. Because of the prob­lem with dynamic range, the ratios of the detection algorithim are not fully effective, and fail in regions of heavy shadow (Image 28). While more difficult

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 9

Page 10: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

to read than earlier examples, the image is still nevertheless understandable. The plateau exhibits strong variation in vegetation density, especially along the east-west stream that bisects it. There is a concentration of this disturbance near the hills. Once again, this fits the pattern of prehistoric exploitation well, and the vegetation variation may be a result of past cultivation and use.

In addition to the variation in vegetation coverage, two very strong anomalies, about 100-200 meters on their long axes exist in this region (Im­age 29, Fig. 8). The lack of dynamic range in the Landsat image stops the investigation at this point, and the anomalies would remain questionable were it not for the availability of other datasets. Fortunately, there exists excellent aerial photography from 1961 (Honduran Geographical Institute), and JERS 1 data (Image 30, Fig. 9). Both illustrate the variation in vegetation coverage and confirm disturbances to the vegetation at the site of anomalies (Image 31, Fig. 10).

Even with the photographic documentation, however, that the anoma­lies are archaeological in nature cannot be certain. In the western-most anomaly, a shadow of a tall tree can be seen. This clearing corresponds with a bend in the river, and could be hydrological in nature. This, of course, does not mean that prehistoric peoples did not utilize and maintain this natural clearing, or place structures upon it. The anomaly to the east lacks a precise identification, but its high degree of linearity implies that it is part of one of the known archaeological sites. Unfortunately, a ground-truthing expedition to verify these anomalies had to be postponed due to Hurricane Mitch until after this document was prepared.

SUMMATION OF DETECTION TECHNIQUES

Detection attempts must be conducted with a clear understanding of the archaeological and environmental parameters of the region under inves­tigation. This need is amplified in the investigation of prehistoric cultures, which in general leave a subtle imprint on the landscape. Direct detection need not be the only or even first option; with available remote-sensing plat­forms, environmental indicators of human activity rather than the activity itself are more easily identified. Likewise, in a regional approach it is neces­sary to develop techniques that allow for broad coverage. By utilizing a tech­nique that focuses on known types of impact, the need or tendency to ad­dress all anomalies or curiosities in an image is avoided.

In the area of the rainforest, where vegetation coverage is particularly susceptible to even very mild anthropogenic activity, the obvious indicators are variations in biomass and biovigor. But like all environments, the rainfor­est is dynamic and such variations are normal. Imagery of the Rio Plátano biosphere post Hurricane Mitch, for example, is full of anomalies that have nothing to do with human activity past or present. Because of these varia-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 10

Page 11: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

tions, a vigorous methodology that diminishes unnecessary investigation of anomalies is called for.

Ratioing is preferred to the use of color-composites, especially in areas of rough terrain. Not only does ratioing allow for topographic correction, it allows a more acute understanding of what varying intensities in an image represent. Also important is the use of multi-temporal and multi-scalar im­agery to confirm potential anthropogenic anomalies and exclude transient environmental phenomena and data errors. Remote sensing detection of ar­chaeological sites is an aid to discovery and interpretation that allows the field archaeologist to focus research areas and goals, and to understand re­gional human-environmental interaction. Traditional methodologies, how­ever, including field expeditions, are necessary to ground-truth and precisely identify activity noted from a remote platform.

Mille Lacs Kathio State Park, Minnesota, USA: vegetation coverages and environmental history

RESEARCH AGENDA

The area of Lake Mille Lacs has been home to multiple population beginning with some of the oldest inhabitants of north America, to pre-Da-kota, Dakota, and Ojibwe peoples, as well as American logging companies, Northern European farmers, ranchers and homesteaders, and recreational industries. People have clustered at southwestern shore of lake Mille Lacs and along the headwaters of the Rum River to take advantage of the wetlands, rich soils, abundant natural foods, access to rivers, and beautiful scenery. An intense process of human-environmental interaction has been ongoing for, at a minimum, 1500 years. As part of historical and archaeological research a consideration of past manifestations of the landscape, and the relationship of humans to those manifestations must be made. The research agenda at hand is singular, yet complicated: what was the landscape of Mille Lacs in the past?

GEOMORPHOLOGY AND HYDROLOGY

The landscape of Mille Lacs resulted from the glaciers that passed re­peatedly over the land at various periods. With each advance and retreat of the many glacial stages the terrain was eroded, sediment was redeposited, vegetation was swept away, rocks were pulverized, and the landscape was forever changed. Humans first entered this area at the conclusion of the final glaciation of the Wisconsin Ice Age.

The massive ice sheets that covered Minnesota during the Quaternary period originated in a basin near Hudson Bay (Image 33). The most recent glacial stage, the Wisconsin, was divided into the Des Moines and Superior

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 11

Page 12: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

lobes. These lobes, channeled by the lowlands, descended and retreated at different locations, times, and directions (TESTER 1995; CUSHING 1967; ALBERT

1995). Between 10,000-12,000 years ago, the climate of Minnesota became warmer by about two to four degrees and drier than the previous thousands of years. The Wisconsin ice sheets began to melt and the slow recovery from the glacial period began (TESTER 1995; CUSHING 1967; ALBERT 1995).

The region of Lake Mille Lacs was formed by these glacial processes. The northern portion of Mille Lacs county is hilly and rough due to a Wis­consin moraine that follows south to west and includes some parts of the shores of Lake Mille Lacs. The central parts of the county are a glacial till plain, with ridges of sand and gravel 1.5 to 7.5 meters in height. To the extreme south­eastern corners of Mille Lacs county are sand hills. The south central parts of the county are sandy outwash plains. Elsewhere in Mille Lacs county the landscape is flat and gently undulating. Between the sandy regions and the Wis­consin glacial till plain there are plains that were gullied out by older lakes (ALBERT

1995; MHRSD Mille Lacs 1942; MHRSD Aitkin 1942; ECRDC 1980). The glacial processes that formed this terrain defined the initial form of

the lakes and wetlands. Lake Mille Lacs was formed when melt water was collected behind the terminal moraine of the Superior lobe that forms its west and south sides (ALBERT 1995; MHRSD Mille Lacs 1942; ECRDC 1980). Lakes Onamia and Shakopee were probably ice basins in glacial till where stagnant ice blocks left depressions in the moraine’s till which were later filled by water (ZUMBERGE 1952).

The Mille Lacs Watershed provides drainage for approximately 400 square miles. There are fifty-eight standing bodies of water, and in addition to the thirteen perennial streams, there are numerous smaller, intermittent streams. Three main geological formation types, based on the contour or relief of the parent soil material, mark the Mille Lacs Watershed. The Mille Lacs Moraine Complex has rolling to hilly kettle and knob topography with small wet depressions and peat bogs. The water holding capacity is high to low; with the water table of the peat bogs averaging six feet while the wet depression’s average is over ten feet on the knobs. In the western and south­ern portions of the watershed the soil is acidic reddish brown till with sandy gravely pockets. In the northern and eastern parts of the watershed lies the west to southwest running Automba Drumlin Area. The drumlins are sepa­rated by poorly drained peat and mineral, non-limy, reddish brown fine sandy loam soils. The water holding capacity is high in this area. In the Crow Wing outwash plain, along the extreme northeastern edge of the watershed, the water holding capacity is moderate to low. There is a level to gently rolling topography with small to medium sized peat bogs. The outwash sediment is usually sand but sometimes gravel is present. The surface and subsurface of this area has sandy loam soils.

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 12

Page 13: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

Currently Lake Mille Lacs covers 207 square miles and is relatively shallow; the maximum depth is 35 feet. According to remnant beach shores around the northern limits of the lake, it is believed the water levels were at one time at least fifteen feet higher. Ancient beach ridges found around Lake Ogechie indicate levels for this lake were approximately ten feet higher as well. There are no indications, however, that these high stands occurred si­multaneous with human presence in the area.

HUMAN-ENVIRONMENTAL INTERACTION

Humans have inhabited the region of Lake Mille Lacs for over 9000 years, and the main reason for both prehistoric and historic habitation and use has been the abundance of resources created by the varied landscapes. While in contemporary thought, wetlands (swamps) are to be avoided, the reverse is true for almost all other periods of history. The abundance and variety of plant and animal resources make the wetlands a combination “su­permarket” and “hardware store” (VILEISIS 1997). The prehistoric and early historic populations of Kathio depended heavily on the water resources of the area. Wetlands supplied reeds for basket making, shrubs for gathering berries, and waterfowl for food. The lakes and rivers supplied the popula­tion with fish as well as wild rice-an excellent and storable food source. The vegetation that surrounds these ecological communities supplied many re­sources as well, such as the maple tree, used for making maple sugar, birch for making vessels, and basswood for making twine (DENSMORE 1979).

The first humans appeared in the region c. 7000 B.C. These Palaeo-Indians are known only through isolated finds of projectile points. The sparse archaeological record prevents any understanding of their precise subsist­ence patterns. The general North American picture for this period, however, indicates that these people were probably “big-game” hunters, following the movement of large animals in the transitional period following the end of glaciation (JOHNSON 1988).

Permanent settlements are certain by c. 1500 B.C. during the Late Ar­chaic period. A lack of excavated sites again limits our knowledge, but the eareliest and best known site, Petaga Point, lies along the Rum River. Stone tools and use of cold-hammered and annealed copper indicate technological development and specialization, and the subsistence pattern seems a combi­nation of large game hunting well as multiple resource exploitation (BLEED

1969). A cultural shift occured c. 500 B.C. with the initiation of the Middle

Woodland period. A ceramic culture is attested at a multitude of sites in the region. The use of copper diminished and a more sophisticated stone tool technology took its place. The presence of permanent settlements and ce­ramic technology indicates a combination of hunting and gathering of the

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 13

Page 14: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

abundant food products of the region. This is the first period with which burial mounds can be associated, and these people likely were the ancestors of the Mdewakanton Dakota.

The late Woodland period, beginning c. A.D. 600 and lasting until con­tact (c. A.D. 1680) is the best known period. The period shows marked con­tinuity with the Middle Woodland and there is no reason to suspect a popu­lation shift. The evidence, indicates, however, the wholehearted adoption of wild rice as a primary food source. The late Woodland culture at Mille Lacs is comparable to the corn-growing Woodlands culture of slightly farther south, with the substitution of rice for corn. Notable, however, is that corn requires much more intensive cultivation than rice, which was readily available and required only intermittent and mild intervention to increase the harvest.

The entire prehistoric period of lake Mille Lacs is noted by the close relationship to the natural environment. While the extreme variations in cli­mate and topography create hard living conditions, they also create abun­dant and seasonally varying supplies of food. Moderate agriculture may have been practiced, but intensive agriculture was never adopted. The environ­ment provided, it seems, an embarrassment of riches for those who knew how to exploit them, and this sedentary lifestyle based largely on gathering is attested by Europeans who encountered the indigenous peoples.

The first written accounts of the Mille Lacs region came from explorers and missionaries. Daniel Greysolon, Sieur de Luth, explored the area in 1679. Louis Hennepin, a French priest, visited Mille Lacs in 1680. On his explora­tions, he encountered and lived with a Dakota tribe near Lake Mille Lacs (MHRSD Mille Lacs 1942). Upon their return to their homelands, Europe was washed with curiosity and awoke to prospective economic endeavors; expansion into the region soon followed

The interests of the Europeans was also linked to the environment of lake Mille Lacs. With the pelt-bearing animals of the east having suffered greatly from depletion, the fur trade pushed farther and farther west. Mille Lacs, with its abundant wetlands, was home to an abundant variety of pelt­bearing animals. Initial expansion into the area was slow but by 1806 Mille Lacs was frequented by traders. Contact with the Europeans had a strong impact on the life ways of the Ojibway, the predominate group in the area at the time. But while metal replaced ceramic and stone tools, the acquisitions of foodstuffs followed similar patterns.

The intrusion of the timber industry, however, changed all patterns of human-environmental interaction and instigated sudden landscape change. As the timber supplies of the eastern United States were depleted, the indus­try set upon the expansive resources of Michigan, Wisconsin, and Minnesota to meet the growing nations demand for lumber, especially white pine. Com­mercial logging of Minnesota’s forests began in the St. Croix Delta around the 1830’s spreading into northern Minnesota. Using the waterways of the

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 14

Page 15: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

Mississippi to explore new timber sites, timber cruisers soon came upon the Mille Lacs region. Daniel Stanchfield, a “land-looker” or timber-cruiser, came to the Mille Lacs region on 1 September 1847, in search of good logging sites. He followed the Rum River’s sloping pine covered valleys from St. Anthony Falls to Lake Mille Lacs, stopping every six to eight miles to climb the tallest tree on the highest hill and record the amount of white pine. Upon his return, Stanchfield claimed, «70 mills could not cut all of that in that many years» (SWANHOLM 1978; MORRISON 1974; LARSON 1949). With such prom­ising prospects of heavily forested areas, the Foley-Bean and Crockston Lumber Companies soon claimed the area. The booming logging industry brought many white European settlers into the area as early as 1855 (ECRDC 1980).

What the loggers particularly wanted was the white pine of the Rum River. These trees stood about 200 feet tall and had diameters of five feet. Like other white pine, it is strong, buoyant, straight, resinous, plentiful, and easy to work with (LARSON 1949). In recent historical times, white pine was used for lumber, laths, shingles, matches, sashes, doors, blinds, woodenware, and telegraph poles. There were also other types of conifer logged in the area, including red pine and tamarack, a particularly durable wood. Some deciduous trees were logged as well, including aspen, birch, and balsam for use as pulp wood and maple was used for flooring lumber. But the initial prize was white pine, and the other trees were harvested either incidentally, or after the white pine was gone (HIENSELMAN 1974; SWANHOLM 1978; LARSON

1949; Minnesota Chippewa Tribe 1985). Timber extraction tended to be clear-cuts, especially in the initial log­

ging ventures. The target areas were those where white pine was abundant, as that maximized profit, but once the process was begun it was easy enough to extract the other timber. The other trees would have to be cut down any­way, to make it possible to move and carry off the huge white pines. After the taking of all valuable trees, the stumpy and rocky cut over land was often burned to reduce future fire hazards, but no reclamation was attempted. These degraded cut-over lands were then turned into reservations for native Americans, or sold to immigrant farmers (TESTER 1995). Evidence indicates that the areas clear-cut first regenerated as aspen and poplar groves, which were themselves in turn clear-cut.

By the 1920’s most of the usable lumber for Mille Lacs had been ex­tracted, and some areas had been clear-cut more than once. Although the whole process took less than 75 years, this most severe form of human-envi-ronmental interaction shaped all following interaction. The subsistence pat­terns of the prehistoric peoples, which lasted some 8000 years, were no longer feasible in a devastated landscape, and a period of reservation living largely divorced from traditional environmental practices began that is only now coming to an end. A new pattern of habitation and exploitation followed after the loggers left: homesteading.

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 15

Page 16: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

The agents of the timber companies and the railroads exulted the won­derful soils of central Minnesota as they sold-off the cut-over land. The set­tlers, however, soon found that agriculture was not always a productive in­dustry in the region (Image 34). The soils of the southern region of the lake were low in nutrients and very acidic, because of the previous conifer forest ecosystems and the sediment deposits of previous glaciers. With the barren­ness of the cut-over lands, logging destroyed the further potential of agricul­ture and increased level of soil erosion. This does not mean, however, that good agriculture could be found in locations near Lake Mille Lacs and within the surrounding area. The tracts just north of Mille Lacs are recognized as being easily cultivated and rich agricultural lands (MHRSD Aitkin 1942). But in the waste of the cut-over, such differentiations could not easily be made by newcomers. Most homesteaders quickly learned that they had to farm and also supplement their income, often through logging.

With the end of logging, the arrival of the drought of the 1930’s, and an economic depression, many of the new settlers were forced to leave the area in search of more profitable lands. Those who stayed, however, found a new form of human-environmental interaction. Less than 100 miles from the ma­jor cities of Minneapolis and St. Paul, Mille Lacs was marketed as a resort area. Rather then taking physical items from the landscape, the interaction was one of enjoyment. Until the 1970’s the emphasis was on quiet retreats, sport fish­ing and hunting (Image 35). In the 1980s, the use of motorized recreational crafts, including jet-skis and snowmobiles was introduced (ECRDC 1980).

Key to the recreational use of the area was the foundation of Mille Lacs Kathio State Park in 1957. This park comprises an area of 10,747 acres and includes many important archaeological sites, some excavated by the Univer­sity of Minnesota and examined through surface survey by St. Cloud State University. This protected area is the study region for this project, and the Minnesota Department of Natural resources attempts to protect and con­serve this part of the natural heritage of the Mille Lacs region.

VEGETATION HISTORY

The soil and vegetation coverage of Mille Lacs was initially formed through a combination of erosional processes and the climatic influences of the gla-cier’s retreat. After the retreat, the landscape was very barren and tundra like. This barren landscape probably had moist glacial till deposits, enormous ice blocks melting and forming depressions into the Earth, and no vegetation. The first vegetation to invade the land was probably similar to the current Cana­dian tundra vegetation, with mosses, sedges, and various grasses (WATTS 1967; TESTER 1995). Because both the environmental and archaeological records are incomplete, it is unknown whether humans occupied the landscape at this point. Most likely they arrived at the end of the tundra period.

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 16

Page 17: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

After the initial tundra like conditions, the area seems to have devel­oped into regions of varying vegetation coverage. The overall picture is prob­ably a wondrous landscape of mixed hardwoods with conifer groves. Pure white pine groves, as well as occasional stands of mixed red and white pine, would have dotted the landscape. In some areas one should imagine giant white pines dominating the super-story with smaller groups of sugar maple in the under-story, and in other places we would expect hardwood and pine forests with a mix of aspen, birch and red and white pine. Throughout the region conifer bogs and black ash swamps would have existed in kettle de­pressions. And quite importantly, the rivers edges and lakeshores would have been composed of a variety of wetlands. It is this diversity that encouraged the growth of the prehistoric population.

While the enthusiasm of the logging industry might lead one to think the area was predominately white pine, this is an obvious simplification, as areas of pure conifer were particularly unsuited to the support of human life. The resources of the native population that did not come from the wetlands came from hardwood forests, including items such as maple syrup, berries and game animals. The abundance of white pine encountered may have been a climatic anomaly at a fortuitous time for the lumber industry, or more likely our sources may be describing what they want to see.

The logging industry certainly altered the vegetative patterns of the region, but we know remarkably little of the extent or long-term effect. For our study area, we are uncertain of what areas were clear-cut and what type of forest was in the areas that were cut. We are uncertain whether logging effects detected are from the first or second wave of logging activity. The main reason for this uncertainty is two-fold. One is that the lifespan of cer­tain deciduous trees tends to be less than 40 years, placing three “tree-gen-erations” between us and logging. The second is the influence of other fac­tors on the regeneration of forest following logging.

There are a multitude of complications in understanding regeneration that cannot be addressed here. There seems to have been a population explo­sion of white-tailed deer in the absence of large predators, and the deer love to eat saplings, especially pine. New diseases have entered the region, in­cluding devastating white pine rust. But probably the greatest factor com­plicating issues is the suppression of fire, which has been the standard policy of Euro Americans as well as the Minnesota Department of Resources.

Fire is an important aspect in the succession and maintenance of the forest communities of the region, and has the power to create the most dra­matic, widespread, and significant short-term vegetation changes within a region. While it is not unlikely that the prehistoric inhabitants used fire to modify the landscape, there is no direct evidence of this activity. Following the logging period, however, fire suppression became the norm, as expected wherever Euro Americans settled. Fire is a key to forest succession, and the

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 17

Page 18: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

entire area began a new succession cycle following logging. With suppres­sion the more fire sensitive species, such as maple and basswood, can sur­vive and thrive, forcing out the fire tolerant forests such as oak-aspen scrub (DAUBENMIRE 1935). The simple policy of fire suppression may have allowed the expanse of the deciduous forests into pine and prairie environments. Moreover, suppression prevents the succession of deciduous forest into pine. Periodic fires, which pine are tolerant of, destroy the saplings of sugar maple, basswood, balsam fir, white spruce, and paper birch. If these saplings grow unhindered, they prevent the growth of pine (TESTER 1995; UPHAM 1991).

The tremendous impact of clear-cut logging and fire suppression in quick succession compromise the largest anthropogenic landscape change in the history of the region, and also creates a barrier to a more sophisticated understanding of human-environmental interaction. Because the landscape has been so heavily altered, it is difficult to understand the environment of the prehistoric peoples, as well as envision the area as the logging companies encountered it. With out specific records from these periods, it is necessary to extrapolate the past landscape from available evidence. What cannot be assumed is that the vegetation of Mille Lacs today is the same as what was there in earlier periods.

The problem facing the research, however, is not one of establishing this general picture, but of gaining more precise indicators of the vegetation of specific areas at specific times. While a daunting task where certainty will never be achieved, the goal is not unfeasible if a model sufficient to create a research agenda can be created. Landsat TM data was utilized to provide the basis of such a model.

OUTLINE OF THE MODEL

The proposed model of vegetation history is markedly anthropocentric and is more suited to historical and archaeological than an environmental research. While acknowledging room for substantial variation with the model, the vegetative history of the Mille Lacs region can be divided into discrete periods marked by specific processes of change.

We have at least a rudimentary understanding of forest succession, and the processes that have effected this succession. Thus it is possible to begin with present vegetation and model backward in time to recover past vegeta­tion in discrete areas of the region of Mille Lacs. The task is categorizing vegetation coverages into discrete units that facilitate without overwhelming the model.

RELEVANT VEGETATION COVERAGES

Vegetation categories have been designed specifically for this research

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 18

Page 19: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

project. Much more finessed categories are in use by other researchers, and we use a more detailed categorization when documenting in the field. The reality is, however, that our model is very coarse, and with the multitudinous variables will never reach a great level of subtlety. As the model develops there may be a desire to sub-divide categories and bring techniques from other disciplines to bear, but at the moment the model cannot produce more detailed results, and thus there is no reason to multiply categories meaning­lessly. Additionally, in an area of high biodiversity, we desired categories that were workable with a limited amount of satellite imagery, and limited process­ing time.

Using the model and these categories, it is possible to chart with some success a general pattern of landscape and vegetation development in the park back to pre-contact vegetation. At the present, the model has not been extended earlier than this period. For ease of explication, wetlands have been omitted from this presentation, as the factors in wetland disturbance and generation involve a detailed consideration of hydrology and current agri­culture.

CLASSIFICATION PROCESS FOR MILLE LACS VEGETATION COVERAGE

After the relevant vegetation coverage types were defined, classifica­tion was done with a Landsat TM image (28 March 1985). The image was recorded in very early spring before vigorous growth began, as is indicated by relative lack of intensity of the vegetation coverage. The only vigorous growth is conifer, which appears as green (Image 42, Fig. 12). A March scene was purchased because initial plans were to analyze hydrology and soils, and a lack of foliage was desirable for these goals. Unfortunately, a scene from later in spring would have functioned better for vegetation analysis, but budg­etary constraints encouraged the use of this early spring scene.

The detection algorithm discussed above works well in this landscape to detect Euro-American activity (Image 43, Tav. VIIa). Prehistoric activity left far too subtle a trace for detection from remote platforms. The existing aerial photography from 1939, however, makes the use of Landsat for detec­tion of sites superfluous. Aerial photography, ranging in date from 1939 to 1994 also served well in other analysis of this Landsat scene, including clas­sification, by that reducing the amount of time necessary for ground-truthing data. The detection algorithm did, unexpectedly, reveal several lineations that seem to have been logging roads (Image 44, Tav. VIIb). Logging roads were used during the winter to move fallen trees over roadbeds of ice. Some logging roads turned into more permanent pathways, but many did not, es­pecially as they crossed terrain passable only in winter. The layer of ice and repeated passing of heavy loads presumably has compacted the soil suffi­ciently to prevent regrowth of vegetation at the same vigor as surrounding

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 19

Page 20: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

areas, and thus the roads appear lighter in this image. The research objective for Mille Lacs was not detection of archaeologi­

cal sites, but rather modeling of past and present landscapes. This necessi­tated a classification of the image for the desired vegetation coverage types. A maximum likelihood classification was deemed best, and training regions were defined by field examination and the use of aerial photography (Image 45, Fig. 13). A min-max linear stretch was performed on all bands of the image before classifications were conducted. Traditionally classification is conducted on all bands except band 6 (thermal), or using principal compo­nents. Neither technique, however, proved particularly effective at differen­tiating the desired coverage types. Classification using all bands fared par­ticularly poorly at identifying Conifer Groves, and had only moderate suc­cess with the other categories, sometimes including incorrect areas, other times excluding correct areas. Classification using Principal Component 1 fared better, but did poorly recognizing Hardwood Swamp, Shrub Wetland, judged almost everything to be Mixed Hardwood and Conifer, Deciduous, Shrub Wetland and Wetland, and failed miserably at identifying Conifer Groves (Image 46).

Comparison of results from these classifications to ground-truthed data, and then final remote sensing analysis revealed them to range from tolerable to useless. The precise reasons for the failure of the classifications are uncer­tain, but major contributions were the relative uniformity of the vegetation at the time the image was taken and the sharp terrain of the moraine. Two ratios 4/5 and 5/7 were noted as being the most sensitive to vegetation subtle­ties and were chosen to correct for these errors. A two-band data set, assigning 4/5 as band 1 and 5/7 as band 2, was created and the classification was per­formed using these two pseudo-bands. This classification worked very well, but tended to be over-generous in attribution to classes. The classification failed in classifying Conifer Swamp, Shallow Lake, and Conifer Groves (Image 47).

The failure of these classification techniques was addressed by a consid­eration of what might effectively differentiate vegetation types combined with some trial and error techniques. A classification utilizing both PC1 and PC3 proved to be extremely successful except in differentiating Openings, Shallow Lake, Shrub Wetland and Wetlands. This is probably because the early spring coverage of all these areas tend to be quite similar-dead reeds and grasses that have lain beneath the winter snow. Introduction of one more ratio, band 6/ band 7 increased the ability to differentiate these coverages in classification. The resolution of the thermal band 6 is only 120m, but its ability to identify colder (and in this case wetter) areas was essential (Image 48, Fig. 14).

Best classification results were achieved using these data sets: It should be noted, however, that the research agenda preferred results

with limited overlap between vegetation classes, and a sharp differentiation between wet and dry areas. In effect, we have mixed vegetative and topologi-

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 20

Page 21: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

cal criteria, separating the white birch in very wet soil from the dry birch in better-drained soil. The choice of classification techniques chosen, and the determination of what results are “better,” reflects this research agenda. In sum, the preferred technique should not be defined as more “accurate,” but rather more useful.

After achieving acceptable results, the classification scheme was saved as a dataset, with each class acting as an independent band. As a maximum likelihood classification was used, it was desirable to save the typicality infor­mation, which records how typical each pixel is of the class, and thus renders a usable spectrum of certainty of identification and class overlap. This infor­mation also allows for later manipulation of acceptable class membership boundaries by adjusting the output histogram or filtering without requiring a new classification. Each classification band was saved then not as 0 (not in class) or 1 (in class), but rather with values from 1 to 255 (achieved with a linear stretch).

Two interpretive interventions and one cosmetic interventions are re­quired before the final data set is saved. The first involves a definition of the acceptable threshold of typicality. In ERMapper, typicality is measured from 1 to 100, with 1 being least typical and 100 being most typical. The question is, at what level of typicality does a pixel cease to belong to our desired class. The instinctual move, to accept everything over 50, is not correct, as the classification works differently and with greater and lesser sensitivity for our various vegetation types. Rather the threshold can be determined only by visual inspection, ground-truthing and the use of aerial photographs.

For our dataset, the typicality limit was left at 1-100 for all coverages, except Mixed Hardwood and Conifer (30-100), Shrub Wetland (30-100), Wetland (40-100), and Hardwood Bog (30-100). This is illustrated by the Conifer Groves; all likely pixels need to be included to preserve all conifer identified by the classification (Image 49). The threshold for Shallow Lake was set at 70-100, but this coverage will not be used anyway, due to the failure of Landsat to identify this subtle type.

A cosmetic step is added here. Two classes, Hardwood Swamp and Mixed Hardwood and Conifer, are particularly fragmented. Ground-truthing and examination of aerial photography indicates that the classification is con­servative in identifying these types, and the interstices between pixels often belong in the representative class. Attempts to fix this through the classifica­tion or adjusting the threshold became problematic, so a simper solution was adopted. These two classifications were processed with a 3×3 low pass filter to smooth them (Image 50). While this does allow for the introduction of erroneous data, such concerns are minimal; the classification by its nature is imperfect, and many of the pixels introduced by the filter are removed by the next step, meaning that they tend to fill only null cells in the final data set. Removal of these null cells is an important cosmetic consideration, as will be

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 21

Page 22: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

illustrated below. Despite all this, both an unfiltered and filtered version of the classification are used, in order to err on the side of caution.

The final step before saving as a dataset is to remove overlap between coverages. This is necessitated by software limitations, not physical reality. The Conifer Swamp is, of course, both Wetland and Conifer Swamp (Image 51). Most remote sensing and GIS software has only a limited ability to deal with these overlaps, and it can become difficult to control which classifica­tion gets preference in display.

If the data were important for the research agenda, the software limita­tions would need to be overcome. In this case, however, we are dealing largely with human perception and use; this hierarchy functions at the level of percep­tion where the primary characteristic of an area is the largest woody vegeta­tion; a Conifer Swamp is first a Conifer Swamp, and second a wetland. As this knowledge is inherent in the classification it poses far less a problem then if, for example, a change in wetland area were being assessed. This hierarchy is represented by the order of the vegetation coverages presented in the table above.

In ERMapper the final data set is created by creating classification lay­ers for each vegetation type. These bands were saved as a hierarchy of cover­ages established to remove overlap: IF band 1 > band 2 AND band 1> band 3 AND band 1>band 4... THEN i1 ELSE null where band 1 is the vegetation class to be saved as an individual band and bands 2 and following are the bands below that band in the hierarchy listed above. This formula ensures that in areas of overlap, only the data from the top of hierarchy is kept (Image 52, Tav. VIIc).

At this stage the classification is completed and ready to be integrated into a GIS (ArcView 3.2) for further examination. As most GIS cannot incor­porate images with more than 3 bands, the classification is saved as a BIL with pixel values representing classes (1=Conifer Swamp, 2=Conifer Grove...). This BIL can easily be converted to a grid, and a legend created (Image 53, Tav. VIIIa). While the results are perfectly acceptable, one more cosmetic embellishment is desirable. The classification as it stands is still frag­mented enough that it is quite difficult to read. A boundary clean can be performed with the GIS program to additionally smooth this data set (Image 54, Tav. VIIIb).

TOWARD A VEGETATIVE HISTORY

This vegetation coverage is now quite sufficient to allow for a consid­eration of the landscape history of Mille Lacs based on the parameters previ­ously discussed. With a current vegetation coverage map complete, it is an easy step to restore the hypothetical vegetation coverages of the past, noting that for this publication, the focus is on multispectral imaging and we are

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 22

Page 23: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

ignoring a variety of important supplemental data sets that are used in the final analysis. The task of generating vegetation maps simply required a re­classification of the present vegetation coverage changing the vegetation cov­erages as elaborated in the model.

Present vegetation coverage illustrates a predominance of Mixed Hard­wood and Conifer and Conifer Swamps, with notable Hardwood Swamp present (Image 54, Tav. VIIIb). Few pockets of pine are to be found in the Mixed Hardwood and Conifer areas, and the large pine groves are culti­vated. This seems to be primarily a result of fire suppression and the rela­tively young age of this forest.

Regeneration vegetation coverage illustrates a situation similar to present forest succession, with the understanding that the composition of the Mixed Hardwood and Conifer would have been different, and the growth at a very early stage (Image 55, Tav. VIIIc). Presumably there would have been more openings than illustrated in the coverage map, but these are not identifiable at present.

Cut-over vegetation coverage illustrates the period of most dramatic (although perhaps not-longest lasting) human-induced landscape change, with great portions of the forest having been removed (Image 56, Tav. VIIId). The remaining vegetation would have been predominately the Conifer and Hard­wood Swamps, and the grasses and shrubs that would have followed shortly after the clear-cut. Additional evidence, including remnant logging roads in­dicate that the entire area illustrated may not have been clear-cut, but the details have not yet been determined.

Pre-contact vegetation coverage illustrates a situation similar to present vegetation coverage (Image 57, Tav. VIIIe). What is missing is a graphic repre­sentation of the much larger, perhaps even dominating, presence of pine in the Mixed Hardwood and Conifer areas. We should envision large stands of mostly pine interspersed in this forest coverage. As such restoration is beyond the capability of this model, it has not been graphically represented in the map.

The model and Landsat TM data has served to provide a usable, if general, understanding of the vegetative history of Mille Lacs. Equally im­portant the model has provided hypothetical reconstructions that can now be tested in the field. Obvious limitations of funding and time will not allow a complete field investigation of the region. What can be accomplished, how­ever, is sampling from the various vegetation coverage types. With these clear objectives in mind, other techniques, including, dendrology, palaeobotany and sedimentology can be brought to bear on the important task of under­standing human-environmental interaction in this area.

RICHARD M. ROTHAUS, AMBER A. DE MORETT

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 23

Page 24: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

Bibliography

D. ALBERT, 1995, Regional Landscape Ecosystems of Michigan, Minnesota, and Wis­consin: A Working Map and Classification, Michigan Natural Features Inven­tory by request of the Upper Great Lakes Biodiversity Committee, St. Paul: USDA North Central Forest Experiment Station Forest Service, Minnesota.

C. BEGLEY, 1999, Elite Power Strategies and External Connections in Ancient Eastern Honduras, Dissertation, University of Chicago.

P. BLEED, 1969, The Archaeology of Petage Point: The Preceramic Component, St. Paul: Minnesota Historical Society Press.

E. CUSHING, 1967, Late-Wisconsin Pollen Stratigraphy and the Glacial Sequence in Minnesota, in Quaternary Paleoecology, edited by Cushing and Wright, New Haven: Yale University Press, pp. 59-88.

R. DAUBENMIRE, 1935. The “Big Woods” of Minnesota: Its Structure, and Relation to Climate, Fire and Soils, Dissertation, University of Minnesota.

F. DENSMORE, 1979, Chippewa Customs, St. Paul: Minnesota Historical Society Press. D.J. DODDS, 1998, Population Growth and Forest Cover Change in the Rio Platano

Biosphere Reserve, Honduras, Indiana: Center for the Study of Institutions, Popu­lation, and Environmental Change. http://cipec.org/demography/dodds_ppr.html

East Central Regional Development Commission (ECRDC) 1980, Mille Lacs Lake Watershed: An Evaluation as an Area of Critical State Concern (SPA/80-2392), Minnesota.

A. GOUDIE, 1984, The Human Impact, Oxford: Blackwell. M.L. HEINSELMAN, 1974, Interpretation of Francis J. Marschner’s Map of the Original

Vegetation of Minnesota, St. Paul: North Central Forest Experiment Station. J.R. JENSEN, 1996, Introductory Digital Image Processing: A Remote Sensing Perspec­

tive, New Jersey: Prentice Hall. E. JOHNSON, 1988, The Prehistoric Peoples of Minnesota, St. Paul: Minnesota His­

torical Society Press, A.M. LARSON, 1949, History of the White Pine Industry in Minnesota, Minneapolis:

University of Minnesota Press. The Minnesota Chippewa Tribe, 1985, Against the Tide of American History: The

Story of the Mille Lacs Anishinabe. Minnesota Historical Records Survey Division of Community Service Programs Work

Projects Administration (MHRSD), 1942, Inventory of the County Archives of Min­nesota No. 48 Mille Lacs County (Milaca), St. Paul: Minnesota Historical Society.

J. MORRISON Jr., 1974, Never a Dull Moment. In Mainly Logging, edited by Charles Vandersluis, Minnesota: Minnesota Clinic, pp. 41-118.

C. PARK, 1992, Tropical Rainforests, London: Routledge. E.A. RAABE, R.P. STUMPF, 1997, Image Processing Methods: Procedures in Selection,

Registration, Normalization and Enhancement of Satellite Imagery in Coastal Wetlands, USGS Center for Coastal Geology. http://coastal.er.usgs.gov/ wetlands/ofr 97-287/.

T. SEVER, 1998, Validating Prehistoric and Current Social Phenomena upon the Land­scape of the Peten, Guatemala, in People and Pixels: Linking Remote Sensing and Social Science, edited by D. Liverman, E. Moran et al., Washington DC: National Academy Press, pp. 145-163.

M. SWANHOLM, 1978, Lumbering in the Last of the White-Pine States, Minnesota Historic Sites Pamphlet Series No. 17, St. Paul: Minnesota Historical Society.

J.R. TESTER, 1995, Minnesota’s Natural Heritage: An Ecological Perspective, Minneapolis: University of Minnesota.

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 24

Page 25: L TM Landscape archaeology - unisi.itswidden agriculture is too recent to have a long-term perspective, and the oral history of the practitioners is too short as well as concerned

W. UPHAM, 1991, Minnesota’s Native Vegetation: A Key to Natural Communities, Version 1.5, St. Paul: Minnesota Department of Natural Resources.

A. VILEISIS, 1997, Discovering the Unknown Landscape: A History of America’s Wetlands, Washington DC: Island Press.

W.A. WATTS, 1967, Late-Glacial Plant Microfossils From Minnesota, in Quaternary Paleoecology, edited by Cushing and Wright, New Haven: Yale University Press, pp. 89-98.

M.N. ZEDENO, 1997, Landscapes, Land Use, and the History of Territory Formation: an Example from the Puebloan Southwest, «Journal of Archaeological Method and Theory», 4, pp. 67-103.

J.H. ZUMBERGE, 1952, The Lakes of Minnesota: Their Origin and Classification, St. Paul: University of Minnesota Press.

©2003 Edizioni all’Insegna del Giglio - vietata la riproduzione e qualsiasi utilizzo a scopo commerciale – 25