Introduction

Over the last two decades, satellite remote sensing has become an increasingly important tool in archaeological research. Remote sensing has the advantage of providing the researcher with a regional view and access to land cover information contained in wavelengths of the electromagnetic spectrum beyond the range of the human eye. The computerized manipulation of remote sensing data, such as spectral filtering and stretching methods, allows detection of archaeological features that are not visible from the ground or in aerial photographs (Kouchoukos 2001; Showalter 1993).

The use of remote sensing data has proven particularly useful in the investigation of buried or obfuscated linear features since the “bird’s eye” perspective provided by satellite imagery improves the scope for their detection. In this study, we define “obscured” features as those which have been subject to postdepositional burial by alluvium or slopewash, impacted by plowing or driving over the ground surface and/or afforested with new vegetation growth. Such forces inhibit the identification of features through pedestrian reconnaissance but may still be visible through remote sensing. CORONA imagery has proven to be a valuable remote sensing dataset to detect large, ancient river courses, irrigation networks, and road systems, particularly in arid regions of southwestern Asia (Alizadeh and Ur 2007; Hritz 2010; Pournelle 2007; Ur 2003, 2005). CORONA imagery provides high-resolution panchromatic images taken on 50-mm film from the 1960s to 1980s at low cost and provides a record of landscapes that have now been disturbed by human activities. Despite the many advantages of CORONA imagery, the spatial resolution varies across one image tile due to an inherent S-shaped distortion and makes detection difficult of archaeological features less than 4-m wide, such as canals (Parcak 2009: 55).

The availability of high-resolution multispectral imagery such as Ikonos (since 1999), QuickBird/Google Earth imagery (since 2001), and SPOT imagery (since 2001) ranging between 0.61- and 0.8-m spatial resolutions has opened new possibilities in the detection of smaller archaeological features (De Laet et al. 2007; Lasaponara and Masini 2007; Tømmervik et al. 2010). The high cost of most high-resolution multispectral imagery has considerably limited their use in small-scale archaeological research projects, such as pilot studies or projects with budget constraints. However, the recent availability of high-resolution Google Earth imagery downloaded from Google Earth Pro has opened up possibilities for archaeological projects with smaller budgets since unrestricted amounts of imagery can be downloaded for free as of 20 January 2015. The QuickBird satellite was launched in October 2001 and collects multispectral high-resolution imagery. The Google Earth imagery used in this study (downloaded in 2009) is a 0.6-m resolution pan-sharpened version of the original 2-m color Quickbird image that Google has processed into a tiled RGB image. Google Earth Pro imagery is a three-band pan-sharpened image (R,G,B), with a resolution ranging between 0.61 and 0.72 m. This is the fourth highest spatial resolution satellite image currently available to the public. In 2011, Google Earth 6 was launched, which includes a feature that allows users to access historical imagery, including aerial photographs. Thus, with this feature, Google Earth Pro has potentially become an even more valuable source of free high-resolution imagery.

Studies using Google Earth imagery for archaeological research are becoming more widespread (Contreras and Brodie 2010; Contreras 2010; Lasaponara and Masini 2006, 2007; Scollar and Palmer 2008; Ur 2006), but until recently, there were few systematic comparisons weighing the relative effectiveness of Google Earth imagery downloaded from Google Earth Pro (cf. Hritz 2013; Subias et al. 2013), and there remain none from the American Southwest. This paper discusses the results of a pilot study on the potential of high-resolution pan-sharpened Google Earth imagery to investigate prehistoric irrigation canals of the Hohokam period (a.d. 450–1450) in the Middle Gila River Valley (MGRV), Arizona, in 2009 (Rost et al. 2010). The data were compared to CORONA and Landsat imagery, aerial photographs, historic maps (Southworth 1914; General Land Office plat maps), and archaeological data (e.g., site locations and Woodson 2009). The conclusion of the research was that Google Earth imagery provided the best information for detecting obscured irrigation channels from this landscape, and so, the results of the investigation are summarized as a potential model for application in other dryland settings.

Research goals

The prehistoric Hohokam built a series of 13 to 15 canal systems along the MGRV between a.d. 450 and 1450 (Woodson 2010). The main canalsFootnote 1 in these systems had a cumulative length of over 220 km (137 mi), and they could have irrigated between 12,000 and 19,000 ha (or about 30,000 to 47,000 ac). Knowledge of these irrigation systems has so far been based on data recovered from archaeological surveys and excavations, the interpretation of aerial photographs, and the study of historical documents (Woodson 2003, 2010).

Decades’ worth of survey efforts, beginning with Southworth’s (1914–1915) attempts to document the Gila River irrigation networks in southern Arizona, have resulted in our gaining exceptional knowledge on the location, course, dimension, and construction design of ancient canals. The relative accuracy and precision of investigations have improved with time as significant technological advances have been made in mapping tools such as the Global Positioning System (GPS), Geographical Information Systems (GIS), and the availability of remote sensing data.

However, despite the considerable information on aboriginal canal systems, there remain a number of unsolved issues. The number of branch and distribution canals in most irrigation systems and the endpoints of canals have yet to be determined. These gaps in knowledge impede the determination of whether or not specific canal segments belong to one or multiple irrigation systems and/or canals (Woodson 2003, 2010). These uncertainties have also limited understandings of relationships between settlement and canals, such as whether there is a correlation between settlement size/population density and the sizes of canals within a single irrigation system. Further, longitudinal and diachronic details of the evolution of irrigation systems in the MGRV from the Pioneer period (ca. a.d. 450–750) through the Classic period (a.d.1150–1450) have so far not been fully established.

Since remote sensing data have yet not been fully utilized in the study of Hohokam canals of the MGRV, a pilot study was carried out in the summer of 2009 with the goal of determining whether or not remote sensing applications have potential to contribute to the understanding of the evolution of Hohokam irrigation systems. The reasoning was that the regional or “bird’s eye” perspective as provided by remote sensing data would increase the chances of detecting remains of prehistoric canal alignment and allow for surveying a larger area with considerable efficiency relative to traditional archaeological approaches (Kouchoukos 2001; Masini et al. 2008; Parcak 2009; Showalter 1993).

Moreover, GIS data processing software allows integration of disparate data sets (i.e., survey and excavation data, or data from historical sources) onto a single platform (Hritz 2010: 189). The integral analysis of different data sets including remote sensing resources suggested that new insights could be gained into the development of prehistoric irrigation system in the MGRV beyond traditional pedestrian survey.

A further goal of this study was to determine the kinds of skills and knowledge needed for a successful interpretation and usage of remote sensing data. Our study sought to determine the relative necessity of ground truthing, which has implications for the use of remote sensing data from areas inaccessible for archaeological research, as is currently true for parts of southwestern Asia. From the outset, this project was designed to determine the validity of unchecked remote sensing data and determine the amount of false positives as a result of the high resolution of now publically available satellite imagery. The results derived from this study are important for similar projects in dryland environments seeking to employ similar methods for cultural resource management or academic investigations of water distribution networks (canals) or transportation corridors (roads).

Research areas

The research project was carried out in cooperation with the Cultural Resource Management Program of the Gila River Indian Community (GRIC-CRMP) in conjunction with the Pima-Maricopa Irrigation Project (P-MIP) in Sacaton, Arizona (map, Fig. 1). Two study areas were delineated for a detailed investigation: the Santan Area, which includes the zone extending west from Olberg Butte to Gila Butte on the north side of the Gila River, and the Casa Blanca Area, which encompasses an area south of the river and west of Interstate-10 to a point north of the modern town of Casa Blanca. These two areas were chosen to be able to compare relatively well-known archaeological areas with those that are relatively poorly understood. The canal alignments of the Santan Area are well studied due to extensive previous archaeological reconnaissance (backhoe cuts and archaeological surveys; see Woodson 2009). Thus, information retrieved from the satellite image interpretation was able to be quickly cross-checked with the archaeological information which facilitated the assessment of the degree of visibility of prehistoric canals. The experience gained in the Santan study area was then applied to the Casa Blanca Area, which has not been subject to as extensive archaeological excavations.

Fig. 1
figure 1

Map of portion of Middle Gila River Valley showing the Santan and Casa Blanca study areas

Full size image

The project area is situated in the floodplain and adjacent terraces and sandsheet of the MGRV in the Sonoran Desert of southern Arizona (see Fig. 1). The portion of the MGRV that includes the GRIC is composed of three erosional fluvial terraces with Aridisols and Entisols as the primary surface soils (Waters and Ravesloot 2000; Wright and Waters 2011). Sandsheets and dunes have buried many irrigation canals and Hohokam archaeological settlements (Wright et al. 2011). Overall, the surface soils are sandier on the distal terraces and sandsheet and have a clay- or silt-rich matrix on the proximal terraces. Irrigation canals are primarily concentrated on the second distal terrace (T-2) of the Gila River, the proximal parts of which have episodically been flooded since the 19th century, but the primary deposition of T-2 occurred between ca. 3000 b.c.e. and a.d.1500 (Waters and Ravesloot 2000). The project area has been extensively cultivated for over 1000 years with irrigation canals guiding water away from the Gila River to the agricultural fields (Ravesloot et al. 2009). The GRIC is in the process of expanding agricultural production within their jurisdiction, which has involved reclaiming fields and canals that have been abandoned for decades or centuries.

For the purpose of this paper, we have separated the discussion of irrigation features in this manuscript into three primary categories. Prehistoric canals are those that date from the Hohokam Period from after a.d. 450 until the time of the Euroamerican settlement in which thematic documentation of irrigation networks began. Historic canals are those that were built and used since the advent of written historical documents around a.d.1700. Many historic canals were documented by cartographers beginning in the 1870s (Bandelier 1892; Robinson and Moody 1940; Southworth 1914; USGLO 1870, 1919; USGS 1914, 1952) and have varying degrees of accuracy. Nevertheless, they represent a historical source of information about potential canal alignments, which are unavailable for the preceding period. Finally, extant canals are those that are part of the modern San Carlos Irrigation Project (SCIP) irrigation system that presently delivers water to the GRIC for commercial agricultural pursuits.

Methods

The methods for this research can be divided into two primary components: image analysis and fieldwork. Image analysis was carried out in two different locations: Stony Brook University in New York and at GRIC-CRMP in Sacaton, Arizona (June 24 through July 10, 2009). Image analysis included first the acquisition (either downloading or scanning) of the relevant imagery, aerial photographs, and maps. Access to Google Earth Pro was provided by the Anthropology Department of Stony Brook University. The imagery was downloaded at the maximum spatial resolution of 0.6 m that resulted in individual tiles covering an areaFootnote 2 of approximately 85 km2. These tiles were georeferenced using the imagery processing software, ArcGIS9.1®. The imagery was further processed with a number of enhancement methods, such as contrast stretching using histogram equalizer and applying various high, low, and edge detection filters. Vector files of visible linear features were created based on the visual inspection of the imagery in raw and processed form.

It should be noted that the Google Earth image mosaic for the worldwide coverage introduces distortions that need to be corrected. In the course of this study, a major mismatch along latitude approximately 33° 8.75′ N was encountered and corrected by means of GPS control points of known landmarks visible on the Google Earth imagery. The images to the north of the aforementioned latitude concur with the WGS84 datum coordinate system; however, the images to the south of it are off by ca. x = 11.6 m to the west and ca. y = 17 m to the south in the Santan Area. The xFootnote 3 distortion value increases further to the west; thus, in the Casa Blanca Area, the images are offset by x = 21.5 m to the west.

Fieldwork consisted of making site visits to ground truth the traced lines from satellite imagery with direct observation of landforms and artifacts that could confirm or refute the preliminary identification. Site checks utilized vector data transferred to a Topcon GMS-2 GPS receiver, which contained points and lines of select locations identified in aerial imagery. Following completion of the fieldwork phase of the project, the results from the field were analyzed at Stony Brook University, and a comparative study of the relative strengths and weaknesses of different types of satellite imagery was conducted (Rost et al. 2010).

Imagery used

The present study utilizes a number of remote sensing resources, which are widely available in the public domain (see Table 1). The primary focus of this study involved the interpretation of pan-sharpened Google Earth imagery; however, aerial photographs, CORONA Satellite imagery, as well as Landsat Enhanced Thematic Mapper Plus (ETM+) were also utilized.

Table 1 Imagery used
Full size table

The main reason for focusing on Google Earth imagery as the main dataset was to test the potential of open access/low cost, high-resolution imagery in archaeological research, in general, and in detecting ancient canal alignments, in particular. Alternative satellite images of comparable spatial resolution are cost intensive to acquire and frequently beyond the funding parameters of small-scale pilot studies. Further, as the average width of Hokokam canals is typically ca. 1 to 4 m (Haury 1976: 139-140), it seemed unlikely that prehistoric canals would be visible on satellite imagery with a lower spatial resolution, unless canal segments of considerable length are preserved.

Thirteen archival aerial photograph tiles provided by the GRIC-CRMP were part of the remote sensing data set used in this study. These aerial photographs were taken in March 1936 as part of the Pima-Papago Project, which was conducted as a Special Project Survey by the US Soil Conservation Service. The reason for making use of this data source was to test its potential use in the study, as they documented a much less disturbed landscape particularly in regards to intensive, mechanized agricultural cultivation. In addition, as the quality of Google Earth imagery varied, aerial photography was used as a substitute. In particular, the available imagery of the Casa Blanca was bright, which made the discrimination of subtle color differences necessary for the detection potential prehistoric canals difficult (Rost et al. 2010).

CORONA and Landsat imagery were integrated as a comparative dataset in order to evaluate the relative effectiveness of various types of aerial imagery in the detection of ancient canals. As has been mentioned above, previous studies (Hritz 2005; 2010; Hritz and Wilkinson 2006; Pournelle 2007; Ur 2003, 2005) have shown that they are well suited for detecting linear archaeological features, such as ancient road systems, river and canal systems even the outline of fields (Hritz and Pournelle 2015; Ur 2005). For this study, CORONA image (DS1104-2145DF035) was analyzed with the spatial resolution of 7 and 9 m in the respective study areas. The image had a greater S-shaped distortion at the center of the image strip than anticipated which resulted in a lower resolution (Hritz 2013; Subias et al. 2013).

In this study, Landsat ETM+ imagery was chosen to explore the potential of multispectral imagery (in particular, the infrared spectrum) to detect prehistoric canals. Landsat imagery is widely used in archaeological projects due to the global coverage it provides at low or no cost, and its multispectral properties that allows for a wide range of image enhancing and image classification techniques to be applied (Parcak 2009: 58-63). We chose to use Landsat over ASTER imagery (despite ASTER’s higher spatial resolution without a resolution merge) due to its low cost, easy availability, and high degree of confidence in the flexibility of Landsat imagery. In particular, the technique of combining the seven image bands in three-band patterns allows making certain aspects of a landscape visually distinct. The so-called IR-R-G band combination (infrared, red, and green, each band viewed in the red, green, and blue bands) is used to visualize reflectance of healthy vegetation in red and is the most commonly used band combination in remote sensing work done for archeological projects (Parcak 2009: 91-92). Archaeological sites and features do affect local vegetation growth, which can result in differential remote sensing signatures between natural and anthropogenic landscape attributes displayed in multispectral imagery (Parcak 2009: 91-92).

Two Landsat scenes dating to 2003 were downloaded from the University of Maryland’s Global Land Cover Facility (GLCF, http://www.landcover.org/index.shtml) Web site. Both tiles derived from the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) satellite, collecting multispectral images with 30-m resolution and overlaid with the infrared band that had a 60-m resolution. A high-resolution merge was performed with the panchromatic Band 8, which enables the researcher to view all bands in 15-m spatial resolution. This merge was preferred over other resolution merging methods since it allows the retention of spectral information in all seven bands. A principle component analysis (PCA) was performed to reduce the spectral redundancy inherent in the information provided by the 7 bands of a Landsat ETM+ image. Showalter (1993) was able to detect obscured Hohokam canals in the neighboring Salt River Valley close to Phoenix based on the analysis of Landsat Thematic Mapper (TM) imagery with 30-m resolution. The detection of these prehistoric canals, despite the relatively low spatial resolution of the used imagery, was possible due to the preservation of a considerable length (≥1 km) of the individual canal segments.

Imagery analysis

Prior to the ground truthing, linear features that might represent potentially obscured canals were traced based on visual inspection of the imagery. At the time of initial digitization of suspected irrigation features, information on the exact location of known prehistoric canals was not considered so as not to bias the initial classification. After the initial digitization phase was completed, the analysis of the preliminary results of the imagery analysis was evaluated in cooperation with the GRIC-CRMP staff and based on archaeological data archived in the GRIC-CRMP GIS database.

Historic and, in some cases, extant canals follow the alignment of prehistoric canals due to hydrological parameters constrained by topography (Darling et al. 2004). Thus, it became necessary to attempt to discriminate between extant, historic, and prehistoric canal alignments as part of the data analysis. The preliminary identification of the traced lines was done by superimposing georectified raster maps and AutoCAD files from the GRIC-CRMP GIS database converted to shape files of canals in the historic Akimel O’odham irrigation systems, the SCIP irrigation system (1924), and the more recent and ongoing P-MIP improvements to the system (1985). In addition, General Land Office (GLO) plat maps of the area of interest were used, which document the alignments of historic canals and roads.

The results of the preliminary identification showed that most of the traced lines were, in fact, historic or extant canals and were ruled out as possible prehistoric canals, unless they represented reused ancient canals (Fig. 2). Of the features identified in the initial phase of data collection, 219 of the 448 features were determined to have been historic canals. During fieldwork, five historic canals were ground truthed in order to correlate the physical traces of historic canals with the remote sensing signatures (see section “Historic canals”).

Fig. 2
figure 2

Historic canals documented in the Casa Blanca project area. Red lines were traced from Google Earth imagery. Base map by Southworth 1914

Full size image

Overlying shape files of documented Hohokam canals from the GRIC database (Woodson 2009) onto the Google Earth imagery made it clear that the signature of prehistoric canals is, in most cases, faint and much less clearly visible than the historic alignments (see section “Prehistoric canals”). Whereas the focus prior to fieldwork was placed on delineating clearly visible lines, lines were added to the database after the initial phase of fieldwork on the basis of comparing GPS points taken on known prehistoric canal alignments with analogous remote sensing signatures. Remote sensing signatures of the known prehistoric canals varied depending on ground surface conditions, such as vegetation, previous disturbance, wet or dry, and overlaying linear features (e.g., footpaths, horse trails). The goal was to ground truth a wide variety of remote sensing signatures and test if specific signatures can be singled out to identify prehistoric canals.

Ground truthing

The term “ground truthing” refers to the process of going into the field and classifying the features that have been identified using the satellite imagery. It must be noted, however, that ground truthing procedures vary greatly from project to project and depend on the subject matter under question. Although actual subsurface reconnaissance (i.e., excavation) would be the most accurate means of confirming the type of feature in question, pedestrian survey is a much more commonly used method of ground truthing. The latter method is more time- and cost-efficient than excavation, and under the right conditions, it can be sufficient for identifying features on the landscape. The ground truthing method used in this project consisted of determining the precise location on the ground of features identified using the imagery (in these case, traced lines) and then following these traces on foot with the assistance of a GPS unit.

In order to ground truth the previously traced lines, six field trips were conducted in the Santan and Casa Blanca areas. Prior to each field trip, shape files of suspected obscured canal alignments were loaded into a Topcon GMS-2 GPS unit. In addition, a laptop was taken in order to facilitate locating the lines under question and to directly compare the signatures of certain lines on the ground with the remote sensing imagery.

Traced lines representing potential prehistoric canals were sampled for ground truthing. Decisions concerning what lines to ground truth were primarily judgmental and were made in consultation with members of the GRIC-CRMP staff. Selected areas were visited and identified with varying degree of confidence (numerically coded from 1—least confident—to 4—absolute confidence; see Table 2). An additional classificatory scheme was developed to separate traced features into the following categories: potential prehistoric canals, historic canals, historic or modern roads, plow furrows, animal paths, natural drainages, or unidentifiable features. The immediate description/identification of remotely sensed features was documented in the field, and the information was stored in the attribute table of the individual shape files. Photos were taken of some of the surface features, and a sample of them will be discussed and shown in the “Results” section. After each field trip, a summary report was written describing the nature of the signature of each line on the remote sensing imagery and during ground truthing and documenting the provisional classification of the feature.

Table 2 Remote sensing identification confidence intervals based on signatures and ground expressions of known and unknown prehistoric canal alignments
Full size table

Results

The results of the analysis conducted to this point indicate that Google Earth imagery is able to detect a number of prehistoric canals within the MGRV. The high resolution of the imagery, however, leads to a significant amount of false positives, which need to be evaluated and eliminated to produce archaeologically informative results. The false positives consisted of features, such as historic and modern dirt roads, natural drainages, traces of agricultural activities such as plowing or seeding and animal tracks. In some instances, the traces of a single motor vehicles crossing were visible on the imagery. Out of the 138 traced and visited lines, only three lines had a high probability of representing a prehistoric canal. Another seven lines had indications of prehistoric canals (artifact scatters, an unvegetated stretch of compact clay, and/or elevated berms), which were not sufficiently diagnostic to make a clear decision on whether they were canals or other types of linear features. Historic canals, on the other hand, had a clear and distinct signature on the Google Earth imagery.

Historic canals

Historic canals have different remote sensing characteristics depending on the canal type in question. Historic secondary (distribution) canals are clearly visible on Google Earth imagery as fine, clearly visible, pronounced dark lines (Fig. 3a, blue lines, and b, red arrows). However, tertiary canals (field laterals) were not as easy to discriminate in the imagery. Historic distribution canals in the Santan area typically run from the southeastern to the northwestern corner of historic/modern fields and subdivide the fields into equal segments. The field shape in the project area is usually constrained within the township, section, and range plats dating from the 19th century GLO surveys of Arizona. The historic field boundaries resemble modern field boundaries, although changing land tenure practices have resulted in aggregation of previously smaller 16.2-ha (40-ac) plots into larger land holdings.

Fig. 3
figure 3

Satellite (Google Earth) and ground truthed views of historic canals (blue lines in a; yellow lines false positives), in the Santan area

Full size image

Diverging field laterals have a different signature on the Google Earth imagery, which shows up as a bright, unvegetated line mainly in the uncultivated fields. Historic field laterals extend from distribution canals at a 90° angle and traverse the modern/historic fields from northeast to southwest. Differences in the remote sensing signature between historic distribution canals and field laterals can be attributed to differences in channel morphology. The historic field laterals were shallower and narrower than distribution canals and show up only as a lighter, more diffuse line in opposition to precisely defined and clearly visible dark line of distribution canals (see Fig. 3b).

During pedestrian reconnaissance, historic distribution canals often appear as unvegetated linear features with blocky, high-clay fraction sediment (see Fig. 3c). Based on the results of numerous field investigations over the course of the last 20 years, this phenomenon can result from the deposition of fine-grained suspended load sediments that settled in the distal ends of the canal during its operation. On a level to convex ground surface, the clays are not conducive to infiltration of moisture into the subsurface; so, vegetation cover tends to be less on the distribution canals compared to the noncanal areas. A further diagnostic attribute of historic canals on the landscape is the presence of a subtle linear swale. In some cases, the swale improves the ability of the sediments to retain more moisture, and vegetation is therefore denser along the course of historic distribution canals. Historic field laterals are much harder to discern on the ground and can only be distinguished by a sediment texture difference under favorable lighting conditions. The sediment texture of the field laterals is loamier than the structure of the surrounding soil due to the deposition of silty suspended load during the lateral’s operation. In cases where they are visible, these features have similar surface expressions as distribution canals with unvegetated sediments composed of blocky, clayey sediments within subtle swales.

Prehistoric canals

Although they are frequently difficult to detect, the remote sensing signatures of what have been provisionally identified as prehistoric canals can vary. In the following section, the remote sensing signatures of known prehistoric canals will be discussed and used to inform the possible identification of potential prehistoric canals in areas that are not documented as thoroughly. Table 2 presents an overview of the results for both known and potential prehistoric canals.

Prehistoric canals that remain visible on the surface can be identified either as linear, unvegetated stretches of compact, fine silt-clay sediments with high-density artifact scatters. Or, where there are no physical remains of a canal, linear artifact scatters frequently indicate the presence of a prehistoric canal (Wells et al. 2004; Woodson 2007). These linear artifact scatters resulted from the use of prehistoric canals as a water source for domestic use (i.e., cooking, washing cloth, and bathing) or as navigation aids between landmarks. The cumulative and variable uses of the canals through time have resulted in the deposition of disproportionately high numbers of artifacts adjacent to the canal margins. The visible density of artifact scatters varies depending on the postdepositional processes, such as overbank alluvial or eolian deposition, deflation, and man-made impact such as construction and agricultural activities. The results of the ground truthing showed that linear artifact scatters could not be detected on the Google Earth imagery. However, unvegetated lines, which in three cases most likely do represent prehistoric canals, were easily discerned. In one instance, relict berms were also detected that had not been recorded in previous pedestrian surveys.

The remote sensing signature of known prehistoric canals (Degree of Confidence 4) on the Google Earth imagery varies between two distinct signature types. One of the signatures is a faint dark line of varying width (see Table 2). The other signature is a broad band of washed out white color with a highly reflective center that appears on the ground as a wide, unvegetated strip of land. The latter signature also has been observed in other surface features provisionally identified as locations of prehistoric canals (see below). In certain instances, such as with prehistoric canal GR-445, Feature 8, a feature may appear on the Google Earth imagery having both signature types (see Fig. 4). Thus, the southeastern portion of the visible segment of this prehistoric canal is a faint dark line, while the northwestern part appears as a broad band of washed out white color (see Fig. 4b). Prehistoric canal GR-445, Feature 8, was documented in subsurface contexts in advance of the construction of the P-MIP Santan main-stem canal (Loendorf et al. 2007; Rodrigues et al. 2003).

Fig. 4
figure 4

Satellite (Google Earth) and ground truthed views of a known prehistoric canal, see blue line in a) (yellow lines false positives) in the Santan area (Confidence Interval 4). Linear artifact scatter can be seen in c

Full size image

During the ground truthing process, differences in the reflectivity of a single feature on the Google Earth imagery were attributed to the variable states of preservation of the prehistoric canals. Whenever the berms of prehistoric canals are preserved, the remote sensing signature is that of a faint dark line, representing the slightly depressed canal bed. This specific signature was also observed for other relict prehistoric canals in the vicinity such as the Gila Butte Canal (Neily et al. 2000), the Granite Knob Canal (Gregory 1994), and the northern branch of the Casa Blanca Canal (Miles et al. 2008). The remote sensing signature of a broad band of washed out white color with a highly reflective center, however, indicates that the prehistoric canal is preserved as a linear, unvegetated strips of land composed of compact, clayey sediments with a high density of artifacts identified on the ground surface (see Fig. 4c). On the ground, the feature appears as a single or parallel raised earthen ridge(s). This signature has been observed at the upper end of the Gila Butte Canal (Neily et al. 2000).

Many prehistoric canals known from archaeological excavations, however, do not have any discernable remote sensing signature. Relatively few of the documented prehistoric alignments in Casa Blanca area are visible on the Google Earth imagery. This finding is linked to the fact that many known prehistoric canals either no longer have a surface trace or linear artifact scatter, or have only a subtle trace, and thus, they will not be reflected on a pan-sharpened RGB image. The destruction of a canal’s surface expression is related to the alteration of the landscape in which they are located. Postdepositional processes, such as geomorphologic landscape changes (e.g., alluviation, eolian deposition/deflation) and human impact (e.g., plowing, development), can erase or cover any surface traces of prehistoric canals.

Three linear features visible on the Google Earth imagery represent highly probable locations of prehistoric canal alignments (Degree of Confidence 3). The signature of the linear features observed in the remote sensing imagery consists, in all three cases, of a light-colored line which is washed out in places (see Table 2 and Figs. 5, 6 and 7). In one case (see Fig. 7), there is also a faint, darker line surrounded by a light-colored, washed-out line attributed to the fact that the berms of this possible prehistoric canal were still preserved. Thus, the ground truthing result of the potential prehistoric canal in Fig. 7 strongly resembled the surface signature of known prehistoric canals with dense concentrations of artifacts located atop an unvegetated alignment of compact, clayey sediments with visible relict berms on the margins. The ground truthing result of the potential prehistoric canal seen on Fig. 5 consisted of an unvegetated strip of land with a moderate amount of artifacts. The surface expression of the potential prehistoric canal seen on Fig. 6 was similar but included a slightly more significant quantity of artifacts scattered on the ground surface, which is most likely related to the close proximity to the prehistoric site. This potential prehistoric canal follows an alignment that intersects with another inferred canal alignment to the northwest of another prehistoric site.

Fig. 5
figure 5

Satellite (Google Earth) and ground truthed views of a likely prehistoric canal alignment (blue lines in a, yellow lines false positives) in the Santan area (Confidence Interval 3)

Full size image
Fig. 6
figure 6

Satellite (Google Earth) and ground truthed views of a likely prehistoric canal alignment (blue line in a, yellow lines false positives) in the Santan area (Confidence Interval 3). Other unmarked linear features are relict historic canals, as shown in Fig. 3

Full size image
Fig. 7
figure 7

Satellite (Google Earth) and ground truthed views of a likely prehistoric canal alignment (blue lines in a, yellow lines false positives) in the Santan area (Confidence Interval 3)

Full size image

Based on the results presented above, it can be posited that a clear remote sensing signature of well-preserved, larger main prehistoric canals is generally represented by a wide, washed-out, light-colored band surrounding a highly reflective white center. However, by comparing this signature to the overall investigated research area, it is clear that this signature is not only restricted to prehistoric canals but also applied to historical lateral and distribution canals (see Figs. 3 and 7). At the same time, it is noteworthy that the signature described above is not found in other natural or man-made linear features. A check of historic thematic, plat, and quadrangle maps was effective in determining whether the canals were likely historic or prehistoric prior to an actual site visit.

The signature of a potential prehistoric canal identified with a confidence ranking of 2 is quite different from those signatures discussed above. This signature consists of a faint line of slightly brighter color that contrasts slightly with the surrounding landscape (see Table 2). The remote sensing signature is more visible in cultivated fields but is still faintly recognizable in uncultivated fields (e.g., Fig. 8a, b). In the three cases investigated in this category, the ground truthing did not yield clear ground surface expressions common to prehistoric canals such as linear artifact scatters, unvegetated tracts of land, and raised berms. Dispersed artifacts in the three areas investigated did not include linear patterning suggesting the presence of a canal. However, in the cases of potential prehistoric canal alignments as presented on Fig. 8, slightly darker sediment with a high gravel fraction was observed during ground truthing, which may have produced the fainter color signature on the Google Earth imagery (see Fig. 8c). Gravel content is high in the alluvial subsoil in some areas of the GRIC (Waters and Ravesloot 2000), which is preserved within the backfill of relict prehistoric canal berms and later spread out during modern plowing activities.

Fig. 8
figure 8

Satellite (Google Earth) views of potential prehistoric canal alignments (blue lines in a, yellow lines false positives), and ground truthed photo of a potential canal alignment (in c) in the Santan area (Confidence Interval 2)

Full size image

Archaeological test excavations conducted after the completion of this pilot study provided evidence that supports the identification of the northeastern alignment (marked in blue on Fig. 8), as a prehistoric canal (CRMP project 2011.12 x 3). Also, a canal feature also was found along the westernmost linear feature shown as a yellow line in Fig. 8a (Fertelmes and Loendorf 2011).

Most linear features ranked with the least degree of confidence (1) as potential prehistoric canals were located in the Casa Blanca Area. The signatures of potential prehistoric canals are taken from the 1936 aerial photographs. The reason for making use of this data source was to test its potential use in the study, and as a substitute to the Google Earth imagery in the Casa Blanca Area. The Google Earth imagery of this area was much attenuated, which made the discrimination of subtle color differences necessary for the detection potential prehistoric canals difficult. The high contrast in the imagery is possibly related to the fact that agricultural fields in the Casa Blanca are laid fallow for a longer period of time than fields in the Santan Area, which are intensively cultivated every year. This may have resulted in the evolution of a more homogenous landscape surface masking surface expression of prehistoric, historic canals, and other linear features. This could also be the reason why most cases of potential prehistoric canals with the Degree of Confidence 1 are found in that area. The signatures of probable prehistoric canals on the 1936 aerial photographs were a range between different degrees of gray tones (see Fig. 9). The difference in the tone might be attributed to whether a linear feature traverses a cultivated versus an uncultivated field. The signature seen on Fig. 9 suggested being the result of the presence of a prehistoric canal. There was adispersed distribution of artifacts in the areas identified as the locations of potential prehistoric canals (see Fig. 9c).

Fig. 9
figure 9

1936 aerial photograph and ground truthed views of a probable prehistoric canal alignment (blue line in a, yellow lines false positives) in the Casa Blanca area (Confidence Interval 1)

Full size image

Result from the imagery analysis of CORONA and Landsat ETM+ imagery

As outlined in the “methods” section above, two types of imagery were included in the study and analyzed after returning from conducting fieldwork. The CORONA image arrived 1 month after returning from fieldwork and the software necessary for manipulating multispectral images as Landsat ETM+ imagery was not available during the fieldwork phase of the project.

CORONA imagery

CORONA imagery is panchromatic and comparable with the properties of aerial photographs at slightly lower spatial resolution. CORONA was used in this study since previous studies (Alizadeh and Ur 2007; Ur 2003, 2005) have shown that they are well suited for detecting linear archaeological features (i.e., road system as well as ancient canals). The image consists of one strip of film, which was sent by the USGS/EROS data center in four tiles. Only one tile (DS1104-2145DF035_35_c, taken August 16, 1968) was relevant for the present study, and only this one was georeferenced relative to control points determined from the road system shape file from the GRIC database. After georeferencing the image, the shape file of prehistoric Hohokam canals generated from Woodson (2009) was superimposed onto the CORONA image. The potential of CORONA imagery in the detection and study of prehistoric canal was tested by checking whether or not the documented and known prehistoric canals could be seen on the image.

The results show that despite the much lower spatial resolution (7 m for Casa Blanca area and 9 m for Santanarea) than the 1.8-m maximum resolution for CORONA imagery, the prehistoric Casa Blanca canal and one of its distribution canals were clearly visible on the image (see Fig. 10a, b). Further to the north of the area shown in Fig. 10, the tail end of the prehistoric Gila Butte Canal and Snaketown Canal had also a clearly discernable trace on the CORONA imagery. These canals can also be seen on the Google Earth imagery. Unfortunately, in the Santanarea, the prehistoric canals, which were partially visible on the Google Earth imagery, were faint on the CORONA image. This is most likely due to the lower spatial resolution (9 m) available for this area.

Fig. 10
figure 10

CORONA image analysis of prehistoric canal alignments in the Casa Blanca area

Full size image

In sum, it can be said that CORONA imagery is a data set that has potential applications for identifying obscured irrigation channels. Google Earth Pro provided at the time when the study was carried out only Google Earth imagery from one date; thus, the impact of seasonality on the visibility of prehistoric canals could not be tested in this study. However, that has changed in the meantime since Google Earth Pro now provides imagery from different dates that can be viewed using the time slider and downloaded at no cost. This makes Goggle Earth Pro an even more valuable source for open access high-resolution imagery than was the case when this study was carried out. There are usually four CORONA Images available for a specific area taken in different seasons. A more extensive analysis of a still affordable larger set of CORONA imagery might thus yield better results despite the slightly lower spatial resolution than Google Earth imagery.

Landsat ETM+ imagery

Landsat ETM+ imagery was chosen as the medium to evaluate the potential of multispectral imagery to detect prehistoric canals in the MGRV. A number of image enhancing methods were applied in the analysis of Landsat ETM+ imagery such as high-resolution merge, principal component analysis, NDVI, and several filtering and stretching methods in order to test the potential of multispectral imagery to detect prehistoric canals. In each case, the manipulated image was projected, and the mapped canals (Woodson 2009) were superimposed onto the Landsat ETM+ images. The latter revealed that only prehistoric canals with a Degree of Confidence 4 were visible while those of lower intervals were not, which is most likely related to the considerably low spatial resolution of 15 m of Landsat ETM+ imagery. This result shows that image’s spatial resolution is a determining factor in detecting obscured prehistoric canals.

In contrast, Showalter (1993) documents canal segments ≥1 km in length but with widths measuring ≤5 m on 30-m resolution images. The differences in success of the analysis of Landsat imagery could be attributed to the current study not having fully exhausted all manipulation possibilities so that the “right” combination of techniques was not discovered despite making numerous attempts to discover the appropriate formula. It is also possible that the spectral properties of prehistoric canals differ according to local environments due to local soil conditions such that the radar sensor is unable to detect all obscured irrigation channels in all depositional environments. Further, the thermal properties also vary according to season due to different degrees of saturation related to precipitation. Ambient sediment moisture across the Sonoran Desert tends to be <4 % of the total volume of the matrix during the dry seasons (Schade and Hobbie 2005); thus, subtle differences in moisture-related thermal conductivity of different sediment fractions will not necessarily be apparent. It is possible that the selected dates of the Landsat ETM+ images were not optimal for displaying contrasts in the ground surface that would reveal obscured irrigation features.

Discussion

The results of this analysis show that the spatial resolution of the selected imagery to be the most significant factor determining whether or not images were effective for detecting obscured irrigation features, even more so than whether the imagery contained multiple or single spectra. Additionally, we interpret Google Earth imagery as having had better results than the available 1936 aerial photographs, which documented a much less disturbed landscape particularly in regards to intensive, mechanized agricultural cultivation, but were lower in spatial resolution than the Google Earth imagery. The fact that Google Earth imagery is a color image with >3x the resolution to other media gave it a higher rate of success than multispectral imagery in detecting obscured irrigation features.

This being said, the spatial resolution of the imagery was by no means the only determining factor in the success of detecting prehistoric canals. As became apparent in this study, the high resolution of Google Earth imagery and imagery of comparable spatial resolution leads to a high degree of information overload. The pronounced visibility of historic canals as was shown earlier (see Fig. 3) is just one example of this phenomenon. Linear features caused by animal and/or human traffic, historic or modern vehicles, or by agricultural activities such as plowing (see Fig. 11) or seeding are clearly visible and can be misleading. Thus, retrieving archaeologically informative results from the information overload depends greatly on the researcher’s familiarity with the local archaeology, history, and environment. This entails not only an understanding of the nature of surface traces of the archaeological features under question but also a knowledge of the postdepositional processes, such as geomorphologic landscape changes (i.e., alluviation) or anthropomorphic impacts (i.e., plowing and development) that may have altered the surface traces. This latter point is especially important in terms of deciding which type of imagery will be most useful for a specific remote sensing study. The successful untangling of the information overload in this study was the result of a fruitful cooperation with knowledgeable GRIC staff. Without the interface with people who have extensive knowledge of the cultural and natural processes of the landscape, it is unlikely that the project would have been successful. The collaboration with GRIC-CRMP personnel allowed for the rapid development of a typology of signature types of modern traces, historical, and prehistoric canals. The application of the typology in the analysis of the Google Earth imagery proved to be an effective method for evaluating the information overload and removing noise and allowed for producing archaeological informative results. The detection of three previously unknown prehistoric canal segments of a total length of approximately 2 km shows that the integrative use of Google Earth imagery has great potential to inform the investigation of prehistoric canal systems (see Figs. 12 and 13).

Fig. 11
figure 11

Satellite (Google Earth) and ground truthed views of plow furrows (blue lines in a and c, yellow lines false positives) in Santan area

Full size image
Fig. 12
figure 12

Google Earth imagery mosaic of Santan area with known and potential prehistoric canals

Full size image
Fig. 13
figure 13

Google Earth imagery mosaic of Casa Blanca area with known and potential prehistoric canals

Full size image

The results of the CORONA analysis, on the other hand, have shown that degrees of landscape disturbance as well as the impact of seasonality on the visibility of obscured prehistoric canals might at times overrule the importance of image’s spatial resolution. In this study, CORONA imagery proved to be a useful tool for detecting obscured irrigation features despite the much lower spatial resolution. This finding might be attributed to the fact that this particular CORONA image was taken in 1968 and therefore documents a less disturbed landscape than the present day. Another factor might be that the CORONA imagery was taken in the month August and during a period of high precipitation, which might have enhanced the visibility of obscured archaeological features.

Google Earth Pro provides only one single-dated image tile per specific area. Thus, the impact of seasonality and difference in soil moisture levels on the visibility of potential prehistoric canals could not be tested. Studies in an arid landscape of Mesopotamia have shown that as changes in moisture levels in the soil, obscured archaeological features are visible components of the surface spectra (Ur 2005). Thus, at times, CORONA images might provide better results despite poorer image’s spatial resolution and its grayscale properties since in most cases, multiple CORONA images dating to different years and seasons can be acquired for a single area at an affordable price.

The results of this study can serve as a template for the investigation of linear irrigation features elsewhere in the world. Aridisols cover 12 % of the earth’s surface (United States Department of Agriculture 2010), and this is a soil taxon that has highly reflective surfaces as a result of limited vegetation growth. The lack of moisture also encourages the use of irrigation-supported agriculture, which has become a common feature of drylands since the middle Holocene (Dale 1997). However, the detection of obscured irrigation features is not a simple task because highly reflective surfaces with low vegetation growth preserve human and animal paths, which can appear identical to irrigation features in an aerial or satellite image (Rost et al. 2010). This study has attempted to discriminate the identification of irrigation features from false positives, which is a function of the degree of ground surface disturbance, geomorphic processes such as deflation or sediment accretion, and inherent soil formation processes. All of these factors affect the contrast visible on the remote sensing imagery and limit the interpretative value of the data.

Our proposed model of detecting obscured irrigation features can be summarized as follows:

  1. (1)

    Obtain high-resolution satellite imagery to begin the identification process.

  2. (2)

    Obtain historic aerial photographs of a given project area to check for the potential influence of modern anthropogenic activities in creating scars on the image.

  3. (3)

    Digitize as many available thematic maps as a cross-checking mechanism for potential disturbances or historic features.

  4. (4)

    Develop a ground truthing protocol early in the digitization process so that target features can be discriminated from false positives.

  5. (5)

    Be prepared to scale features on a confidence interval for data management purposes.

Conclusion

The present study evaluates the potential of remote sensing to detect and investigate obscured Hohokam canals in the MGRV. Even though the Google Earth images proved to be more useful in the detection of potential prehistoric canals than the open-access multispectral imagery, its analytical properties are comparable to an aerial photograph with a 2-m spatial resolution or higher. Pan-sharpened RGB and panchromatic imagery can only document those traces that are already visible on the surface. Thus, the greater visibility of historical canals is related to the scale of historic earthen works, which were larger than the prehistoric features. More importantly, the fact that they were constructed and abandoned more recently than prehistoric canals has left better surface traces. The surface traces of prehistoric canals are, in most cases, faint or invisible and therefore more difficult to detect using remote sensing imagery. The image enhancing methods that can be applied to pan-sharpened RGB and panchromatic images are limited and centuries of bioturbation, deflation, and cultivation on top of the relict alignments has eroded their surface expressions.

However, the success of this study suggests that the analysis of high-resolution multispectral imagery would prove to be even more successful since infrared bands do allow for detecting subsurface features (cf. Showalter 1993). Analysis of subsurface irrigation features using SPOT or Pleiades multispectral imagery would likely result in more positive identifications of even older canals. Therefore, the open access, high-resolution Google Earth imagery is well suited for gathering pilot data sufficient to predict the degree of success of a more cost intensive remote sensing project.

In terms of value for money spent, identification of irrigation features using remote sensing applications has the potential to be far more cost effective than manually surveying large swaths of territory, but cannot stand alone. Remote sensing to detect prehistoric canals provides a “bird’s eye perspective,” allowing a much larger area to be surveyed in a shorter period of time than it would be possible using pedestrian reconnaissance (Hackenberg 1974). The ability to detect prehistoric canals on satellite imagery allows for targeted pedestrian reconnaissance efforts as opposed to broad-brush surveys that are labor and resource intensive. However, sampling based on satellite image interpretation does not appear to be effective as the only or even primary means to detect prehistoric canals. It should also be stated that after 100+ years of intensive archaeological research into the alignments of Hohokam canals, there may be few opportunities remaining to detect unknown features. As discussed above, linear artifact scatters resulting from intensive human use of prehistoric canals are not detectable using satellite imagery and can only be discovered by using pedestrian reconnaissance. Further, the regional view allows for defining target areas more easily and effectively. This study also shows that remote sensing, at no matter what spatial resolution, will not replace more traditional archaeological approaches to investigate ancient irrigation.

An additional positive attribute of remote sensing technology is that it works in combination with a GIS. GIS software such as ArcMap, QGIS, or GRASS allows layered data displays (i.e., shapefiles of traced canals, backhoe cuts, known locations of canals, roads, maps of historic irrigation systems, or maps geomorphology of the landscape), which brings a large suite of analytical tools useful for identifying obscured linear features. QGIS and GRASS are open source software and can be downloaded for no cost. As has been shown in this study, the combination of remote sensing technology with GIS software allows for evaluating the relationship of prehistoric irrigation canals and systems with the natural features of the surrounding landscape. In the end, the proposed combination of field and analytical techniques has the potential to resolve many outstanding archaeological questions pertaining to the evolution of irrigation and social systems in the American Southwest as well as other dryland regions of the world.