Products

High Resolution DEM Paper

Tools

STATEMENT OF PROBLEM

Digital Elevation Models (DEMs) are the digital representation of topographic and/or man-made features located on the surface of the earth. DEMs are widely used for hydrologic analyses, resource management, transportation planning, earth sciences, and environmental assessment. Increased use of DEMs in these wide-ranging applications has led to greater need for higher resolution and higher accuracy digital elevation data. This is especially true for local government applications such as basemap generation/registration, stormwater management, watershed mapping, water-quality assessment, and flood-risk assessment because of the fine horizontal and vertical spatial scale of most features (streets, buildings, localized watersheds, etc.).

High resolution DEMs are also essential in producing ortho-rectified image products used as digital basemaps in GIS-based local government applications. A digital image basemap is an essential component in local government GIS systems because many other vector-based data layers (streets, parcel boundaries, sewer lines, etc.) are routinely superimposed upon the image basemap so as to provide an easily understood visual frame of reference. Digital image basemaps must have a high degree of horizontal resolution (e.g. 1 m pixel size) and planar accuracy so that all other data layers can be superimposed without geometric distortions. Topographic distortions must be removed from raw airborne or satellite based digital imagery in order to provide digital orthoimages with sufficient planar accuracy for use as digital image basemaps.

PRESENT SITUATION

DEMs
There are several commonly used satellite data sources for high-resolution DEMs, and these fall into two categories: optical and radar. Stereo-correlation processing of panchromatic imagery from the SPOT satellite typically yields DEMs with a 20 m horizontal (xy) resolution. Typical absolute RMS vertical accuracies for SPOT DEMs are on the order of 8-15. DEMs available directly from SPOT Image, Inc. have a 20 m xy resolution and a 7-11 m RMS z accuracy. Similarly, DEMs produced from Interferometric Synthetic Aperture Radar (IfSAR) image processing (e.g. ERS-1/2) have yielded 25 m xy resolutions and 10-30 m RMS z accuracies. The most widely available high-resolution DEMs in the U.S. are the 7.5-minute Level 1 DEMs available from the United States Geological Survey (USGS). These are produced from stereo processing of 1:40,000-scale panchromatic aerial photographs which are acquired every 5-6 years by the National Aerial Photography Program (NAPP). The USGS 7.5-minute DEMs have a 30 m xy resolution, a 7-15 m RMS z accuracy, and are available for about 90% of the continental U.S. The horizontal resolution and vertical accuracy of all the DEMs described above are not suitable for the vast majority of local government applications.
Recently, this has prompted development of higher-resolution and/or higher accuracy DEM data products from commercial remote sensing data providers. These providers use airborne IfSAR and LIDAR (LIght Detection And Ranging) systems for topographic mapping. For example, the Intermap STAR 3i X-band IfSAR system can provide digital elevation data with a 5 m xy resolution and a 2-3 m RMS z accuracy. Commercial LIDAR systems can provide 1.5 m xy resolution and 0.4 m vertical accuracy. However, the current cost of commercial IfSAR and LIDAR data is in most cases too great for widespread adoption by city and county governments in areas other than major metropolitan cities.

OrthoImage Basemaps
The most widely available high resolution digital orthoimage dataset available to local governments throughout the U.S. are the Digital Orthorectified Quarter Quadrangles (DOQQs) produced by the USGS using the 30 m Level 1 DEMs in conjunction with the NAPP aerial photography. The DOQQs have a horizontal resolution of 1 m and a planar accuracy of 10 m. The reported 10-m planar accuracy is a circular error at 90% probability (CE90) and corresponds to National Map Accuracy Standards (NMAS). The DOQQ horizontal resolution is sufficient for local government basemap applications. The quoted planar accuracy is somewhat of a limitation for widespread use. Nevertheless, many city and county governments have utilized DOQQs as their data source for orthoimage basemaps because the cost to obtain these data is essentially zero. However, problems with displacements between individual DOQQs in the overall basemap and problems with proper registration of vector data layers are routinely experienced. Another limitation of the DOQQs is that these are produced by the USGS from NAPP aerial photographs which are only acquired about every 5-6 years. Thus, the DOQQs often are outdated in many areas and do not reflect the current situation within the city and/or county, especially in areas that experience rapid development.

The recent launch of the IKONOS satellite has opened up a new area for acquiring up-to-date high-resolution panchromatic digital imagery for use as digital orthoimage basemaps. The IKONOS panchromatic imagery has a horizontal resolution of 1 m and an 11-bit information content. The 1 m pixel size is identical to the DOQQ horizontal resolution while the 11-bit imagery provides image contrast and quality that is superior to the DOQQs. The lowest precision IKONOS panchromatic image product (Carterra Geo) costs $12/km2 and has a planar accuracy of only 50 m (CE90). While this product is affordable for most local government entities, the planar accuracy of 50 m renders this product unusable for basemap applications. Our interaction and discussion with basemap users within the City of Columbia and the County of Boone indicate that a planar accuracy of 3-4 m CE90 is required for useful implementation as a digital basemap. The highest accuracy IKONOS panchromatic image product (Carterra Precision) costs $66/km2 and has a planar accuracy of 4 m (CE90). While this meets the planar accuracy requirement for basemap implementation, it is 5.5 times more expensive than the Geo product. This cost is prohibitive for most local government agencies with limited budget resources. For example, Boone County covers an area of approximately 2,400 km2. Thus, for complete county coverage the Geo and Precision IKONOS datasets would cost $28,000 and $155,000, respectively. The latter figure is well beyond the budgetary capability of most local governments. The challenge then becomes to develop a methodology that would enable the use of the lower-cost Geo products for creation of digital basemaps.

NEW APPROACH AND POTENTIAL BENEFITS

In this research we develop and demonstrate the use of National Aerial Photography Program (NAPP) imagery for the creation of high resolution DEMs for local government applications. These are the same aerial photographs used by the USGS for creation of their 30 m Level 1 DEMs and the DOQQs discussed above. However, we utilize the NAPP photos to create significantly higher resolution DEMs than those generated by the USGS. Precision ground control from rapid-static and kinematic GPS surveys are used to validate the vertical accuracy of the DEMs. The results indicate that DEMs with 3-m horizontal resolution and an RMS vertical accuracy of ~ 2 m can be generated from the NAPP data.

In addition, we create orthorectified images suitable for use as digital basemaps using the NAPP-based DEMs and the raw NAPP photos. Furthermore, we demonstrate that the NAPP-based DEMs can be used to orthorectify low-precision IKONOS Geo image products to generate a suitable IKONOS-based digital basemap. A number of studies are done to demonstrate the planar accuracy of various digital basemap datasources, e.g. NAPP, DOQQ, IKONOS, etc. The results are then used to assess the cost/benefit of the various sources for digital basemap creation and use for local government applications.

The benefits of this research demonstrate several important applications of remote-sensing data products pertinent to local government needs. The data products are shown to viable in terms of both cost and utility (resolution, accuracy, etc.). The combined results indicate that there are significant opportunities for cost-effective generation of high-resolution DEMs and digital orthoimage basemaps for utilization by local government entities.

TECHNICAL APPROACH

The study area chosen for this work was in Southern Boone County (SBC), Missouri just south of the City of Columbia (population 80,000). The area is about 220 km2 (14 x 16 km) and is part of the US 63 corridor connecting Columbia and the capital of Jefferson City. The study area also contains the small town of Ashland (population 3,000). This is an environmentally sensitive area containing several state parks. The area is under pressure from rapid expansion (primarily single family housing developments) pushing south out of Columbia and north out of Ashland.

DEMs
Ten aerial photos (B/W film positives) were obtained from the NAPP archive at a cost of only $100 ($10/photo). The photos were acquired by the NAPP program in leaf-off conditions in April, 1996. The photos were oriented in two N/S strips containing 5 photos/strip. The overlap in the photos is approximately 60% N/S and 30% E/W. Stereo-coverage from the ten photos covered an area of approximately 325 km2 (15 x 20 km). The B/W film positives were precision scanned at 1200 dpi (0.85 m pixel size) by a third party vendor at a total cost of $200 ($20/photo). Precision scanning is required to preserve the geometric integrity of the B/W photos. A rapid-static GPS survey was conducted in the study area to obtain 35 ground control points (GCPs) for DEM registration/generation. In addition, kinematic GPS data were collected between GCPs to provide independent check points (ICPs) for DEM validation. The accuracy of the rapid-static GCPs were found to be 3-5 cm RMS for all three coordinates (x, y, and z). The vertical accuracy of the kinematic GPS data used as ICPs was found to be 10 cm RMS.

The NAPP photos were processed using commercially available software (PCI APEX v7.0) for DEM extraction. Thus, the results generated in this study could be easily reproduced by other parties (e.g. third-party consulting companies, GIS specialists in city/count governments, etc.). The GCPs are used to generate a highly accurate coordinate reference frame for triangulation and registration of the NAPP photos. DEMs with 3, 10, and 30 m horizontal resolutions were extracted via automated stereo-correlation processing. DEMs with the different horizontal resolutions were created to evaluate the impact of resolution on vertical accuracy and on output products derived from this and other ICREST/Synergy I projects using the DEMs as input (e.g. digital basemaps, hydrologic models for water quality assessment, etc.). A dataset of 1000 ICPs derived from the kinematic GPS survey was used to assess the vertical accuracy of the DEMs. The ICP dataset was a small subset of the kinematic GPS data selected first by eliminating low-accuracy kinematic positions (due to obstructions caused by trees and buildings during survey). In addition, the ICP data were selected: a) for uniform distribution throughout the study area, and b) for a minimum separation between ICPs of 30 m.

OrthoImage Basemaps
Digital orthorectified image basemaps were produced using a variety of data sources and techniques. All orthoimage basemap products were produced using commercially available software (PCI OrthoEngine v7.0). Thus, just as was the case for the DEM generation, these results can be easily reproduced by other parties as needed. This is an important issue for widespread adoption by state and local government entities. Proprietary algorithms and software are in impediment for widespread adoption of remote-sensing information products.

First, the NAPP-based DEMs with 3 m, 10 m, and 30 m horizontal resolution were used to orthorectify the ten raw NAPP aerial photos to produce three different NAPP digital basemaps. In this way the effect of DEM resolution and accuracy could be evaluated in terms of the impact on the visual quality and planar accuracy of the resulting orthoimage basemap products. Note that these digital basemaps are produced from the same NAPP data used by the USGS to produce the 30 m Level 1 DEMs and the DOQQs. However, the approach adopted here should produce better results because the horizontal resolution and accuracy of the NAPP DEMs produced in our research are significantly better than the USGS DEMs. This should then produce better overall digital basemaps, and this assumption is validated by the results shown in the next section.

Next, the 10 m and 30 m NAPP DEMs were used to orthorectify a low-precision IKONOS georeferenced image. There are several options available in the orthorectification software for establishing an accurate coordinate reference frame for the georeferenced IKONOS data using GCP input. These options are simple polynomial (SP), rational function polynomial (RFP), and the more rigorous sensor model (SM). The SP option was not used as this is known to produce very poor results compared to the other two options. Both the RFP and SM options were evaluated using the 10 and 30 m NAPP DEMs.

The visual quality and the planar accuracy of the all these digital basemaps were evaluated. The visual quality was subjectively measured by examining the resulting linearity of known linear features (primarily roads). The planar accuracy was assessed by using a dataset 20 ICPs whose positions were derived from the rapid-static GPS survey noted previously. The ICPs were selected a priori at locations in the imagery that were sharp and distinct point features. This then enables the identification of these points in the digital basemaps and subsequent comparison with the known locations. The planar accuracy of the USGS DOQQs was also evaluated in this manner using the same ICP dataset for comparison purposes.

Figure 1. NAPP 10 m DEM for Southern Boone County study area. Location of GCP’s indicated by red triangles. The RMS vertical accuracy of the DEM is 2.3 m.

Finally, it should be noted that another ICREST Synergy I project produced two other digital basemaps using alternate approaches. The first product was produced for the entire area of Boone County by combining together approximately sixty USGS DOQQs into one seamless mosaic. The second product was produced from a low-precision IKONOS georeferenced image covering about 40% of the county. The IKONOS digital basemap was created by performing an image-to-image registration of the IKONOS georeferenced image to the DOQQ mosaic through selection of several hundred tie points. The planar accuracy of these two digital basemaps was also evaluated using a dataset of several hundred ICPs derived in a similar manner as described above. The planar accuracy of these two products will also be shown for comparative purposes.

PRODUCTS AND RESULTS

(1) DEMs with horizontal resolutions of 3 m, 10 m, and 30 m were generated for the Southern Boone County (SBC) study area using the ten NAPP aerial photos and the approach described in the previous section. Figure 1 shows the NAPP DEM with a 10 m horizontal resolution. The RMS vertical accuracies of the 3 m, 10 m, and 30 m DEMs were 3.0, 2.3, and 2.2 m, respectively. A sample histogram of the measured vertical error using 1000 ICPs is shown in Figure 2 for the 10 m NAPP DEM. The 3 m DEM is noisier (larger RMS error) because stereo-correlation errors are the largest at the highest resolution. This is caused by a reduced number of pixels used in the final iteration of the stereo-correlation matching process. Nevertheless, the RMS vertical errors of all the DEMs are significantly better than the quoted 7-15 m RMS vertical accuracy of the 30 m USGS Level 1 DEMs. In addition, the 3 m and 10 m horizontal resolutions are a substantial improvement relative to the 30 m horizontal resolution USGS DEMs. This is especially important for city/county governments that must resolve features with fine spatial detail. These DEMs were utilized as a key input layer in another ICREST Synergy I project involving the use of hydrologic models for water quality assessment. In addition, these DEMs were used to orthorectify the NAPP and IKONOS images for digital basemap creation.

(2) The ten raw NAPP aerial photographs were orthorectified using the 3 m, 10 m and 30 m NAPP DEMs to produce three seamless digital basemaps for the study area with a pixel resolution of 1 m (all three mosaics). While the horizontal resolution of all three mosaics was the same, the visual quality of the mosaics produced using the 10 m and 30 m NAPP DEMs was much better than the mosaic produced using the 3 m NAPP DEM. Linear features (e.g. streets) in the mosaic produced from the 3 m NAPP DEM were slightly warped and distorted compared to the other two mosaics because of the larger RMS vertical error of the 3 m NAPP DEM. An assessment of the planar accuracy of the three mosaics is presented shortly.

(3) A low-precision IKONOS georeferenced image (Carterra Geo Product) was orthorectified using the 10 m and 30 m NAPP DEMs to produce several different digital basemaps for the study area. The 3 m NAPP DEM was not utilized because of the larger vertical error and its affect on the orthoimage visual quality noted above. Digital basemaps were created for each DEM using either the RFP or the SM software option for orthorectification. In addition, the 30 m USGS Level 1 DEM was used as well to generate digital basemaps from the IKONOS image for comparative purposes. All digital basemaps were generated with a pixel resolution of 1 m. Figure 3 shows the IKONOS digital basemap created using the SM option and the 10 m NAPP DEM. Comparison of the visual quality and planar accuracy of the various IKONOS basemaps are given next.

(4) The visual quality and planar accuracy of the NAPP and IKONOS digital basemaps generated using different DEMs and/or coordinate reference frame approach (RFP or SM for IKONOS) were evaluated. The visual quality was assessed subjectively based upon the ability to preserve the linearity of known linear features (e.g. streets and buildings). The planar accuracy was computed using an ICP dataset as described in the technical approach section. A summary of these results is given in Table I. The raw IKONOS georeferenced image and the USGS DOQQs were also included in this evaluation for comparative purposes.

The DOQQs for our Southern Boone County (SBC) study area had a planimetric accuracy of 3.4 m (CE90). This is significantly better than that 10 m planimetric accuracy quoted by the USGS. The county-wide seamless mosaic produced in another ICREST Synergy I project from about three dozen DOQQs was found to have a planimetric accuracy of 3.6 m (CE90). Thus, this is consistent with our findings for the smaller SBC study area. Our evaluation of DOQQs from two other study areas within the State of Missouri on a NASA-funded initiative found similar results. Thus, it appears that the planimetric accuracy of the USGS DOQQs are often times much better than the USGS quoted accuracy of 10 m CE90, which is probably a worst-case estimate. Our interaction and discussion with basemap users within the City of Columbia and the County of Boone indicate that a planar accuracy of 3-4 m CE90 is required for useful implementation as a digital basemap. Thus, the DOQQs are a viable low-cost (essentially zero) data source for digital basemap applications. The disadvantage is that the DOQQs are often not up to date for most areas of the country. The latest DOQQs available for our study are based on NAPP data acquired in 1996 and is therefore nearly 5-years out-of date.

The NAPP digital basemaps have planimetric accuracies on the order of 3.1 m, which is slightly better than the DOQQs which are based on the same NAPP aerial photos. The better planimetric accuracy is probably due to the more accurate DEMs produced as part of this project. While the planimetric accuracy of all three NAPP basemaps are nearly the same, the digital basemap created using the 3 m NAPP DEM had significant distortions in the linear image features and is therefore inferior to the basemaps generated using the 10 m and 30 m NAPP DEMs.

The raw georeferenced IKONOS Geo image had a planimetric accuracy of nearly 24 m (CE90) prior to orthorectification. After orthorectification, the IKONOS-based digital basemaps had planimetric accuracies that varied between 2.7 and 4.6 m. The orthorectification coordinate reference frames produced using the RFP method yielded the lowest errors (2.7-3.1 m CE90). However, the RFP solution also caused a significant amount of distortion in the linear features and is therefore inferior to the IKONOS-based basemaps produced using the SM solution which yielded slightly worse planimetric accuracies (3.2 – 4.6 m CE90). Of the SM IKONOS basemaps, the one created using the 10 m NAPP DEM had a much better planimetric accuracy (3.2 m CE90) compared to the basemaps created using either the 30 m NAPP DEM (4.6 m CE90) or the 30 m USGS DEM (4.6 m CE90). The IKONOS-based digital basemap created on another ICREST Synergy I project using the image-to-image co-registration method and the USGS DOQQs yielded a planimetric accuracy of 4.1 m CE90. As expected, this is slightly worse than the planimetric accuracy of the DOQQs which was found to be about 3.5 m CE90 in two independent analyses.

Considering both the planimetric accuracy and the visual quality of the IKONOS-based digital basemaps, we believe the best IKONOS-based digital basemap is the IKONOS – SM created using the 10 m NAPP DEM. This yielded a very good image quality and a planimetric accuracy of 3.2 m. The accuracy is comparable to the DOQQs but the image is recent and of much higher quality (e.g. 11-bit). These are important and unique advantages relative to the DOQQs.

Finally, we note that the planimetric accuracies of the IKONOS-based digital basemaps are on the order of the 3-4 m CE90 specified by our user communities within the City of Columbia and the County of Boone. These planimetric accuracies are achievable using the lower-cost IKONOS Geo products in conjunction with a limited amount of GCPs (ten) and a viable DEM. The IKONOS Geo products are 5.5 times less expensive than the highest precision IKONOS products that have a planimetric accuracy of 4 m CE90. Thus the results presented here demonstrate that this level of precision can be achieved using the lower-cost IKONOS Geo product. This represents a significant opportunity to expand the market of high-resolution IKONOS imagery for digital basemap creation for local government entities who would otherwise be unable to afford the highest accuracy IKONOS products.

Table I. Comparison of the Visual Quality and Planar Accuracy of Various Digital Basemap Products with 1 m Pixel Resolution

(5) The IKONOS – SM basemap created using the 10 m NAPP DEM has been delivered to the user communities in the City of Columbia and County of Boone. These have been incorporated into a web-based data delivery and decision support system developed by another ICREST Synergy I project. User evaluation of this basemap and the county-wide DOQQ basemap created by another ICREST Synergy I project is ongoing at this time.

(6) Based upon the results summarized in Table 1, we created an orthorectified digital basemap of a georeferenced IKONOS multispectral (MS) image. The 10 m NAPP DEM was used along with the SM coordinate solution. Figure 4 shows the orthorectified MS basemap. The pixel size of the MS basemap is 4 m. The planimetric accuracy of the IKONOS-MS basemap was evaluated and found to be 5.6 m. The IKONOS-MS basemap is currently being used by the ICREST Synergy I hydrologic modeling project to derive land-use classifications

Figure 2. Histogram of vertical error in NAPP 10 m DEM for Southern Boone County study area. A dataset of 1000 Independent Check Points (ICPs) derived from a kinematic GPS survey was used. The RMS vertical accuracy of the DEM is 2.25 m.

SUMMARY

The following summarizes the key results from this project:

1) NAPP-based DEMs can be generated with horizontal resolutions that are a ten-fold improvement over the most widely available high resolution USGS 30 m DEMs. The ~ 2-3 m RMS vertical accuracy of these DEMs are significantly better than the 7-15 m RMS vertical accuracies quoted for the USGS DEMs.
2) The NAPP DEMs are created using commercial, off-the-shelf, software using data that is currently available for 90% of the U.S. at very low cost ($10/photo where each photo covers an area of about 80 km2).
3) The benefits listed in 1) and 2) demonstrate that low-cost DEMs can be generated at resolutions and accuracies useful for many local government applications using an existing dataset available for much of the U.S. Thus, the NAPP imagery represent a viable low-cost data source, already in existence, that could be reasonably exploited and utilized for many local government applications.
4) The NAPP-based orthoimages and the USGS DOQQs with 1 m horizontal resolution are found to have planar accuracies on the order of 3-3.5 m (CE90). This is superior to the highest precision IKONOS orthoimage product (Carterra Precision – 4 m CE90). These therefore represent viable low-cost digital orthoproducts suitable for use as digital basemaps in local government applications. The main disadvantage is that the NAPP and DOQQ images are acquired in any given state only once every 5-6 years.
5) Low precision IKONOS Geo imagery can be orthorectified using commercial, off-the-shelf software and the NAPP and/or USGS DEMs to yield high precision digital basemaps with planar accuracies on the order of 3.2-4.6 m (CE90). This is comparable to the highest precision IKONOS product (Carterra Precision). This effectively demonstrates that high-planar accuracy digital basemaps can be created using the lower-cost, lower-precision IKONOS Geo imagery. This could make the IKONOS products affordable for many smaller local governments for digital basemap generation.

Taken altogether, the results above demonstrate the high-resolution, high accuracy DEMs and digital basemaps can be generated at costs that are affordable for local government applications. These represent significant opportunities for effective use of remote-sensing data products by various user communities within city and county government entities.

Figure 3. Ikonos Panchromatic orthoimage produced from low precision georeferenced Space Imaging product using NAPP 10 m DEM. GCPs and ICPs are shown as red and yellow triangles respectively.

Figure 4. Ikonos Multi-Spectral orthoimage produced from low precision georeferenced Space Imaging product using NAPP 10 m DEM. GCPs are shown as red triangles.

PROJECT PARTICIPANTS

Dr. Curt Davis, Dr. Xiangyun Wang, and Mr. Brian Baker of the Department of Electrical Engineering carried out this project for ICREST under the Raytheon Synergy I program.