INTEGRATING DIFFERENTIAL GPS, GIS, AND SONAR MEASUREMENTS TO MAP THE BATHYMETRY OF TOPOCK MARSH, ARIZONA

Integrating Differential GPS, GIS, and Sonar Measurements to Map the Bathymetry of Topock Marsh, Arizona

Natural resource scientists routinely need bathymetric maps of inland water bodies to answer environmental questions. This study integrated differential GPS, GIS, and sonar measurements to construct a bathymetric map and model of Topock Marsh, Arizona, a shallow water impoundment and wetland. Our procedures are summarized as follows: (1) create a georeferenced base-map using USGS orthophotoquads; (2) locate a base station near the marsh; (3) collect bathymetric measurements; (4) post-process GPS location data; and (5) convert, analyze and display results. This technique was ideal for small- to medium-sized lakes and ponds where sub-meter accuracy (x,y) is desired. Our aim was to calculate the monthly change in storage of the marsh for a water balance calculation (Guay, 1999). These procedures generated resource maps, a bathymetric model, and volume-surface area-depth relationships.

Proper water management of fresh water reservoirs, lakes and ponds often requires the storage capacity or the bathymetry of the water-body be known. For instance, bathymetric data has been used to determine reservoir sedimentation rates (Leonard and Coughlan, 1997), the distribution and occurrence of submerged vegetation (Lehman, 1998), and to develop an environmental management plans (Sherin, 1997). The aim of this study was to determine the monthly change in storage of a shallow reservoir. These results were incorporated into a larger hydrological investigation that included an annual water balance. ARC/INFO was used to analyze the bathymetric data and provided volume-surface area-depth relationships.

The approach was to integrate a differentially corrected Global Positioning System (GPS), a Geographic Information System (GIS) and sonar measurements. There already exist commercially available navigation devices (e.g., Eagle’s UltraMap^TM) that combine and automate these functions, and even provide background maps. These units rely on a differentially corrected Coast Guard navigation signal (Radio Technical Commission for Maritime - RTCM protocols). Originally the RTCM signal was broadcast along the coast, major rivers, and major lakes (e.g., Great Lakes), but there are plans to expand its range (pers. comm., Navigation Information Service, 703-313-5900). The principle limitation to these systems for natural resource mapping are their: (1) horizontal accuracy of 10 meters or less, (2) inefficiency in attributing (annotating) a position with complex data information (e.g., plant or animal sightings, water quality measurements, etc.), (3) inability to use and display multiple layers of background data, (4) line-of-site signal interference, and (5) relatively small memory.

Similar to previous investigations of inland water bodies (Leonard and Coughlin, 1997), our procedures were aimed primarily at overcoming these limitations. Cost and time never became a critical concern for this project, in part, because we used low-cost graduate student labor, in-house technical expertise (University of Arizona, Advanced Resources Technology Group- ART), and less than state-of-the-art receivers and laptops. Our procedures can be briefly summarized as follows: (1) create a georeferenced base-map from USGS orthophotoquads; (2) determine coordinates of the base station antennae; (3) collect bathymetric measurements; (4) post-process the GPS location data, and (5) convert, analyze and display results in the ARC/INFO environment. The last procedure (#5) consumed an estimated half the man-hours of the project because data augmentation (explained below) was needed to generate a realistic model of the marsh bathymetry, and there are some complex physiographic features at the site.

The study site was Topock Marsh, located on Havasu National Wildlife Refuge, in northwestern Arizona (fig 1). The 13 km-long marsh occupies a former channel of the Colorado River that was converted to a 1638 ha (4045 ac) shallow water reservoir and wetland with dikes and levees in the 1960's. Most of the marsh is shallow (1 to 1.5 m deep) with gradually changes in relief, but along the western shore there are a series of man-made dredged channels (1.5 to 4 m deep) with sharp breaks in slope. In addition, the shallowness, strong winds, and prolific stands of tangled deadwood and cattails precluded navigation and data collection in some areas (fig 2a,b). The marsh also experiences a phenomenon known as seiches, or wind-induced stacking of water. On windy days, estimates of the marsh elevation had added uncertainty because the elevation shifted at either end by as much as 0.15 m (0.5 ft).

The base map was created by mosaicing four USGS orthophotoquads into one image for subsequent ARC/INFO use as a background map. The USGS orthophotoquads were standard 7.5 minute quadrangle, georeferenced, monocolor, photographic quality, 1 meter resolution image maps. The marsh shore line, cattail islands, and other key resources were digitized on-screen in Arcview (as shape files), converted into ARC/INFO coverages, then "ungenerated" into ASCII coordinate files for use as GPS background maps during field work (fig 3).

The Refuge headquarters, located 15 km west of the marsh, was chosen as a secure base station. The coordinates of the base station antennae were determined by making concurrent position measurements at three known geodetic control points (e.g., registered benchmarks) in both Arizona and California (fig 4). This task required the maximum three sets of equipment (laptop, antennae, receiver). The receivers (Geolink Inc., Motorola, 8 channel) were utilized because of their low cost and ability to generate carrier-phase position data for several post-processing software packages. Carrier-phase post-processing software, GRAVNAV by Waypoint Ltd., was able to calculate the base station antenna position with a horizontal accuracy of less than 0.5 meters. Unfortunately, this task was more complex than usual because the benchmarks, marsh, and base station were located in adjacent UTM zones, they used different reference datums (NAD 27 vs. WGS 84) and historical measuration techniques, and one of the benchmarks appeared to be displaced from its original position (fig 5).

A 15-foot Glastron boat was equipped with: (1) a laptop with Geolink data collection software, (2) an Oncore Receiver with GPS antenna, (3) a commercial-grade sonar unit (Eagle Magna III^TM), and (4) a plastic tube that indicated the sonar draft on a staff gage inside the stern of the boat (figs 6a,6b,6c). The elevation of the marsh was recorded hourly by a pressure transducer calibrated to a staff gage at the outlet of Topock Marsh. The measurement error of the pressure transducer, draft indicator, and sonar are estimated or cited as ± 0.03 m (0.1 ft). With the GPS base station operating, a two-person team traversed the marsh at trolling speed and manually entered sonar, draft, and field observations into the GEOLINK data collection software. Sonar measurements were entered about 5 to 30 meters apart depending on the bottom terrain, while draft readings were recorded every few minutes. Measuring water depth with a calibrated survey rod and comparing to our computed depth tested system accuracy.

Data collection occurred in half-day intervals to reduce file sizes, minimize possible data loss, and to facilitate data post-processing techniques. The GPS raw data files were collected and attributed using the Geolink Data Collector software module, post-processed together with the base station files using Post-Point of GRAFNAV differential correction software, then converted from Motorola file formats to an interim ASCII format utilizing Geolink's Data Management Software. Using Post-Point software, the horizontal accuracy of boat’s position was estimated to be 2 to 4 meters. The Data Management Software generated all the necessary geo-relational files and "sml" programs to automatically build coverages in a PC-ARC/INFO format in the field. Eventually all coverages were exported to ArcView so that the data could be easily viewed by refuge managers and researchers. Each new coverage was reviewed daily to expose major data collection or post-processing errors. Since each hour of data collection used about 1.5 to 2.0 megabytes of disk space, files were transferred to Iomega ZIP disks for archival storage.

After GPS post-processing and coverage attribution, the bathymetric data were normalized in ArcView database environment to a reference elevation (456.5 ft msl) by correcting for sonar draft and daily marsh elevation The reference elevation coincides with the upper water elevation that Refuge managers use most of the year (Havasu NWR, 1987) and the digitized shore line. The shore line was delineated on the orthophotoquads by using recent airphotos, vegetation types, and site knowledge. Depths for each point were re-computed using the reference elevation as the "zero depth." These points were added as an attribute to the coverage, making feet below the reference level as the mapping unit. The areal distribution of the corrected depth values are shown in Figure 7. Not surprisingly, the data are concentrated in the most accessible boating areas. The original field location data contained 2,888 attributed points. The resulting shape files were then used to create a point coverage in ARC/INFO.

The adjusted x,y,z point data were then used as subsequent input for elevation models using ARC/INFO CREATTIN and TOPOGRID tools and procedures. However, both procedural models incorrectly interpolated many areas of the marsh based upon our site experience. The problems were most acute where point density was low and along the dredged channels. For example, dredged channels were represented as discontinuous features with alternating deep and shallow topography, rather than as continuous u-shaped man-made channel (fig 8a,b).

To achieve a representative model of the marsh, based on site-specific experience and others physical indications of depth, we performed four types of "data augmentation" or enhancement. Figure 9 shows the location of a typical dredged channel area that is used to illustrate before and after effects of data augmentation with TOPOGRID tool. First, a relatively few data points that initially fell outside the shore line boundary were shifted as a cluster to their probable position. This was a simple adjustment since the boat tracked from shore to shore in a predictable sinusoidal path (fig 10). Next, we constrained the computer interpolation by hand drawing contour lines on-screen. This was done using the existing data, mostly along the steep slopes of dredged channels and parallel to the deep-water edge of cattail islands. The added contour lines were incorporated into the elevation model. A third type of augmentation relied on existing data and "expert opinion." That is, very important data points (VIP’s) and contour lines were added in data poor areas, usually between actual measurements, or, where expert opinion could confidently estimate depth. For example, in areas where the embankment was steep-faced and the closest sonar reading was a 50 meters or more off-shore, the computer would interpolate a shallow shelf that was known to be incorrect from field experience. This problem was especially true along the eastern shore between Five-Mile Landing and Golden Shores (fig 3). In this situation, three parallel contour lines were added to modify the bottom profile to better represent this former river meander (see fig 12). Always, the added depth values were consistent with nearby points and site experience. In three smaller backwater areas (e.g., Beal Lake) where boating was not feasible, depth values had to be estimated from shore observation while the area was drained. The overall profile of these features was a reasonable combination of measurement and judgement. The final type of data augmentation was used specifically along the bottom of the dredged channels. The channels were not recognized as continuous trench-like features (fig 10). Therefore, we used the stream option in TOPOGRID tool to create a more realistic model. Later we found segmenting this feature into shorter sections at 2-meter resolution produced a smoother representation at 10-meter resolution (fig 11a,b).

Figure 12 shows the augmented data compiled with the original data. Following the data augmentation, the point coverage (now 3,363 data points) and shore-line coverage were again used to generate a bathymetric model of the marsh. Both the TIN and TOPOGRID models were recreated at 10-meter resolution. The volume-surface area-depth relationships were determined using an AML that combined map algebra and the GRID function ZONALSUM at 2-meter (x, y) resolution for 0.13 m (0.5 foot) depth intervals.

The results of the study were encouraging. They provided various resource maps and reasonable estimates of the marsh’s volume, surface area, and depth. Figure 13 is the final bathymetric map made using TOPOGRID tool. Table 1 shows the volume estimates of the marsh using four methods. It is interesting that our final method (#3) is only about 9% less than the classical topographic map estimate (#1). And there was near parody in the two elevation models (methods #3 and #4), with only about a 2 % difference.

Figure 14 illustrates the relationship of the marsh’s volume, surface area, and depth. The average marsh elevation at the end of each month was used to determine the surface area and volume; these quantities were later used to compute evaporative losses from the open-water marsh and the month to month change in storage, respectively.

Table 1. Comparison of four volume estimate methods for Topock Marsh assuming a surface elevation of 456.5 ft above msl.

(1)

Traditional topographic map and assumed 1.22 m (4 ft) depth

(2)

Original Data w/ TOPOGRID tool

(3)

Original Data, w/ data augmentation and TOPOGRID tool

(4)

Original Data w/ data augmentation and TIN

19736000 m³

(16,000 ac-ft)

17310154 m³

(14034 ac-ft)

17894870 m³

(14508 ac-ft)

17534328 m³

(14215 ac-ft)

The study allows Refuge managers and researchers to make some informed water management decisions in the future. For example, the Refuge learned that the surface area changes little until the water elevation drops below 454.0 ft msl, when it begins to fall dramatically. Suppose the aim of water operations were to expose shoals to migratory birds that feed on benthic invertebrates, then lowering the water elevation another foot to 453 ft msl would expose an additional 1279 acres of feeding habitat.

In agreement with Leonard and Coughlan (1997), the integration of differential GPS, GIS, and sonar measurements was successful in producing a bathymetric map and, in this work, a model of Topock Marsh. Data analysis was able to provide estimates of the monthly change in storage in the marsh. At the reference elevation, the overall accuracy of our volume and surface area estimates, while unknown, are expected to be near ± 10 percent. The greatest sources of error seem to result from estimating the marsh elevation and from sonar measurements over unconsolidated bottom material (organic mud). The elevational error was greatly reduced if data collection occurred on windless days and during periods when the overall marsh level was being held constant. The sonar error was reduced by adjusting the sensitivity but was never completely overcome in backwater areas.

Several factors need consideration when following these procedures. For example, it was important to have a background map and real-time GPS display of the boat’s position. This procedure aided in navigation and eliminated certain gross errors. Notwithstanding, mosaicing imagery (100 MB memory per orthophotoquad) requires substantial computing power (we used an UltraSparc 1000 with 4 GB Ram). To locate the base station, we needed three sets of equipment concurrently operating at well defined, first-order (if possible) benchmarks, within a reasonable radius (< 10 km) of the base station, to yield the highest position accuracy with the receivers and software we used. It was a tremendous asset to have some site knowledge for planning data collection traverses and, later, data augmentation. For instance, the track of the boat over dredged channels required a 45 degree heading to the axis of the channel or much of the contour was not identified. For more regularly shaped, or more nature water bodies, a simple grid pattern seems a reasonable approach. The amount of hours spent enhancing the data was surprising large. The algorithms for producing the bathymetric models have some inherent biases. For example, TOPOGRID tool has a terrestrial bias the assumes a hydrologically active surface that has more "peaks than sinks." In addition, the stream option in TOPOGRID tool creates a continuous depression or trench that forces the overall feature to descend along its length. If the trenches were not segmented, the channels were too deep and the volume was overestimated. However, it is reassuring that the augmented data (using TOPOGRID tool) only increased by 3.4% the volume of the marsh when compared to the original data (Table 1).

Nevertheless, and even with all the data augmentation, we felt our final bathymetric model was very representative of Topock Marsh and provides a powerful water management tool for the Refuge.

The authors would like to acknowledge Andrew Hautzinger, Hydrologist, USFWS Region 2, for his patience and generous funding of this project; Havasu National Wildlife Refuge for personnel and logistical support; Dr. Eugene Maughan and Dr. Carol McIvor of the University of Arizona's Arizona Cooperative Fish and Wildlife Research Unit for encouragement to take on this project; The University of Arizona's School of Renewable Natural Resources Advanced Resources Technology Group personnel, with special thanks to Mr. Craig Wissler for GIS and modeling assistance and Mr. Cliff Hathaway for computer systems administration and support. We would like to thank Mr. Douglas Richardson of Baker-Geo-Research Inc. (Billings, Montana) for Geolink Software and the excellent technical support received from his staff and all the folks at Waypoint Consulting Inc. (Alberta, Canada) for creating the Grafnav Post-Processing Software.

Guay, B., (1999), Preliminary hydrologic investigation of Topock Marsh, Arizona. 1999. Report to USFWS, Region 2, 150p.

Havasu NWR, (1987), Havasu water management plan, Refuge document. See also Guay (1999).

Lehmann, A., (1998), GIS modeling of submerged macrophyte distribution using Generalized Additive Models: Plant Ecology, v. 139, p. 113-124.