FEMA’s Romance with LiDAR

A 5.630Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE

In 1994, Brigadier General Gerald E. Galloway chaired the Interagency Floodplain Management Review Committee following the devastating floods of 1993. His report, widely known as the Galloway Report, recommended that FEMA evaluate newer technologies that could yield improved digital elevation models (DEMs) for floodplain modeling. Recognizing USGS as the nation’s elevation expert, The Federal Emergency Management Agency (FEMA) asked USGS to evaluate accuracies achievable from LiDAR and IFSAR. The USGS Open-File Report 96-401 entitled: "Digital Elevation Model Test for LIDAR and IFSARE Sensors," documented USGS’ evaluation of LiDAR data from a LiDAR system developed by the Houston Advanced Research Center (HARC) in cooperation with FEMA and the NASA Goddard Space Flight Center. The LiDAR was flown from an elevation of 3,000′ above ground level (AGL) over a 3 km2 area near Glasgow, Missouri which had been hard hit by the 1993 floods. The tested RMSE was 37 cm in open ground, 2.65 meters in low cover, 2.0 meters in scrub, and 3.8 meters in trees–all non-impressive by today’s standards.

Subsequently, FEMA became an early LiDAR pioneer and, to the best of my knowledge, the first federal agency to adopt LiDAR as its technology for the future with published guidelines and specifications for Flood Insurance Studies (FIS). To this day floodplain managers continue to rely more on LiDAR than any other user group. I am proud to have been part of FEMA’s engagement with LiDAR from its beginnings. My company (Dewberry) has performed independent QA/QC of more than 400,000 square miles of LiDAR, and a good chunk of that was for FEMA to ensure that the LiDAR data satisfied requirements for the National Flood Insurance Program (NFIP), which produces Flood Insurance Rate Maps (FIRMs) that delineate a community’s flood risk zones. The NFIP’s value and credibility is largely based on accurate mapping of flood risks, and LiDAR is best for accurate hydrologic and hydraulic (H&H) modeling, at affordable costs.

FEMA has progressed with the times regarding its accuracy requirements for topographic data to be used for floodplain modeling and mapping. Historically, FEMA’s elevation data requirements were based on 4′ contours from photogrammetry and "best available data." This article will walk through the timeline of LiDAR guidelines and specifications used for FEMA flood risk mapping projects, as best I can recall from happenings nearly two decades ago. FEMA has adopted higher standards as they have become technically feasible and is committed to using the best topographic data for new mapping projects.

Figures 1A and 1B could be interchangeable, the point being that LiDAR data may either show that structures previously outside FEMA’s Special Flood Hazard Area (SFHA) should be inside the SFHA–or vice versa. Without accurate LiDAR data and H&H models, homeowners will invariably question whether they really need flood insurance. In this example, many properties were removed from the SFHA, but in the many instances, LiDAR data of improved accuracy and currency shows more properties at risk of flooding than previously mapped.

Schoharie County, NY
FEMA’s actual use of LiDAR did not begin until 1998 as I can best recall. In 1998, Dewberry was tasked to prepare a LiDAR DEM Accuracy Verification report for FEMA’s first use of LiDAR, in Schoharie County, NY. We were specifically tasked to independently determine if LiDAR data, acquired from 6,000′ AGL, was equivalent to topographic maps with 1- or 2-foot contours as desired for future FEMA accuracy standards. FEMA desired DEMs equivalent to either 1- or 2-foot contours (depending on what proves achievable from this LiDAR evaluation) for complete automated hydrologic and hydraulic (H&H) modeling and analysis.

After flooding in 1996, Schoharie County planners and emergency management officials approached the New York State Department of Environmental Conservation (NYSDEC) wanting new flood maps that more accurately represented their threat from flooding. Schoharie County was planning a new GIS-based E-911 system and wanted to incorporate updated flood hazard mapping data into the system. Concurrently, FEMA was preparing a plan to modernize the flood hazard maps across the United States using the latest technologies and collaborating with state and local partners to instill local ownership of the program. NYSDEC’s needs were compatible with FEMA’s goals, and the pilot project was funded. PAR Government Systems was the engineering firm for the flood studies and EagleScan was the LiDAR provider. Dewberry developed FEMA’s LiDAR accuracy validation procedures and proposed we use the new National Standard for Spatial Data Accuracy (NSSDA) published in 1998 by the Federal Geographic Data Committee (FGDC) that required accuracy testing against checkpoints of higher accuracy to determine vertical accuracy at the 95% confidence level, computed by multiplying RMSEz by 1.9600. Unfortunately, the NSSDA’s statistics assumed that all errors followed a normal error distribution, and that proved to be an erroneous assumption for testing the accuracy of LiDAR Digital Terrain Models (DTMs) in vegetated terrain.

Dewberry’s Special Project Report, dated December 31, 1998, documented systematic issues pertaining to LiDAR sensor calibration, relative accuracy (inaccuracy) between overlapping swaths, data processing errors, and large data voids in forested areas where most points had been filtered out. Concurrently, we determined that it was unfair to the LiDAR data provider to use QA/QC checkpoints on steep slopes or bridge abutments where LiDAR interpolation would unfairly make the LiDAR look worse than it really was. After removing unfair checkpoints, the data initially tested at 4.44 feet at the 95% confidence level (approximately equivalent to 7.5 ft contour accuracy). EagleScan determined that systematic below-ground errors required reprocessing of LiDAR data with new atmospheric refraction models, and they revised procedures for vegetation removal.

Dewberry’s first conclusion/recommendation stated: "For the Schoharie FIS, the reprocessed LiDAR DEMs should be adequate to support automated hydrologic modeling of watersheds, but they should not be used alone for automated hydraulic modeling of floodplains. As already planned by PAR, detailed cross section surveys should be performed to augment the DEMs for hydraulic modeling."

Pinellas County, FL
Dewberry’s second accuracy testing of LiDAR was in 1999 when we evaluated LiDAR data of Pinellas County, FL. FEMA wanted to determine if LiDAR DEMs have the resolution and accuracy necessary for automated hydraulic modeling, without the need for expensive, ground-surveyed cross sections. At that time, our goal was RMSEz of 15 cm within floodplains because we thought this accuracy was achievable. The county provided 677 checkpoints in five major land cover categories determined to represent the floodplain area. The tested RMSEz values for these five categories were as follows:
1. Woods: RMSEz = 25.3 cm
2. Tall weeds and agricultural fields: RMSEz = 14.0 cm
3. Short grass or weeds, or bare earth, sand or rocks: RMSEz = 10.6 cm
4. Mangrove: RMSEz = 57.8 cm
5. Sawgrass: RMSEz = 66.3 cm

We learned that the mangrove and sawgrass was so dense that the LiDAR last returns were not penetrating to the ground. Because the surrounding terrain was very flat, we recommended that the LiDAR points within mangrove and sawgrass polygons be removed as ground points and filled in with interpolation from surrounding ground points; today, those would be called Low Confidence Areas. As with Schoharie County, the wooded areas in Pinellas County proved that LiDAR elevation errors in forested areas do not follow a normal error distribution, that a few outliers cause the RMSEz values to exaggerate the inaccuracies, and a better way needed to be found to assess the accuracy of LiDAR in densely vegetated terrain. Yet, after correction for systematic errors, the LiDAR data nearly satisfied FEMA’s traditional requirement for 4-foot contour accuracy or better in most vegetation categories (with the exception of mangrove and sawgrass), and it was clear to me that photogrammetry could not have done any better. We concluded that LiDAR could not be used for automated hydraulic modeling without the need for surveyed cross sections to accurately determine stream channel geometry both above and below the water level.

FEMA 37, Appendix 4B, Airborne Light Detection and Ranging Systems
I worked with Karl Mohr and Mary Jean Pajak at FEMA headquarters, and Stan Hovey of Michael Baker Jr. Inc., to publish Appendix 4B, Airborne Light Detection and Ranging Systems, to FEMA 37, "Guidelines and Specifications for Study Contractors" in 1999, with minor revisions in 2000. Based on what was then believed to be achievable, it specified a maximum 5-meter post spacing and 15 centimeter RMSEz requirement for all major vegetation categories that predominate within the floodplain being studied, with DEM accuracy reported for up to five representative land cover categories. Contractors were required to select a minimum of 20 test points for each major vegetation category, with a minimum of 60 test points for a minimum of three major vegetation categories. In addition to performance standards, it provided guidelines for system calibration, flight planning, GPS base stations, LiDAR post-processing, QA/QC, and deliverables.

The North Carolina Floodplain Mapping Program (NCFMP)
In 2000, twenty-two Federal and local community entities joined North Carolina (as FEMA’s first Cooperating Technical State) in an agreement to work together to produce accurate, up-to-date flood hazard data for the State of North Carolina, using statewide LiDAR and automated H&H modeling to the degree possible. Dewberry worked closely with John Dorman, Gary Thompson and others at the state (and FEMA) to help ensure the program would be successful.

Various issue papers developed LiDAR specifications for data acquisition, calibration, required accuracy and nominal point spacing, generation of bare-earth ASCII files, generation of TINs and breaklines, tile sizes, and development of DEMs in four different file formats. Quality control (QC) procedures were developed for accuracy testing, visual QC of cleanliness, QC of LiDAR-derived cross sections, evaluation of different methods for generating breaklines, comparison of datasets between different LiDAR and H&H contractors, and comparison with other existing LiDAR datasets. Another issue paper addressed maintenance, archiving and dissemination of the NCFMP LiDAR data.

In 2000-2001, Dewberry and NCFMP participants were struggling to determine what LiDAR accuracies were achievable, especially in forests where we knew (and later confirmed) that DTM errors would not follow a normal error distribution. We required calibration data to be collected during each flight over a calibration course established at each airport. The NCFMP required a vertical RMSEz of 20-cm for coastal counties and 25-cm for inland counties, computed after discarding the worst 5% of the checkpoints as there was no written guidance from anyone as to what to do when errors do not follow a normal distribution. The NCFMP chose five land cover categories (bare-earth and low grass, weeds and crops, scrub, forested, and built-up) and surveyed 120 QA/QC checkpoints per county, i.e., 20 per county in each of four land cover categories plus 40 per county in forested areas.

Based on the 120 checkpoints per county provided by Gary Thompson of the North Carolina Geodetic Survey, Dewberry performed the accuracy assessments and submitted 100 LiDAR Accuracy Assessment Reports, one for each of the 100 counties in the state. Some counties were better than others, but when we aggregated all 12,000 checkpoints into a single spreadsheet, the average RMSEz was 18.5-cm which satisfied the criteria for 2-foot contour accuracy. However, this statistic only pertained to the best 95% of the checkpoints, after the worst 5% had been removed from the calculations. Although not statistically defensible in terms of NSSDA requirements (that erroneously assumed all errors followed a normal distribution), this was still better than the original National Map Accuracy Standards (NMAS) of 1947 which essentially reported the accuracy for the best 90% of points tested. Using 2 ft contours as an example, the NMAS would require that no more than 10 percent of the elevations tested be in error by more than 1 ft (one half the contour interval), but the 10% outliers had no limitations at all. Thus, the NCFMP’s LiDAR would easily have passed NMAS requirements for 2-ft contour accuracy.

Appendix A: Guidance for Aerial Mapping and Surveying, to FEMA’s "Guidelines and Specifications for Flood Hazard Mapping Partners"
Working with Paul Rooney at FEMA headquarters, Dewberry provided the primary authors for FEMA’s Appendix A approved in February of 2002 and revised in April of 2003. Appendix A set FEMA’s requirements in terms of the NSSDA though it also provided contour interval equivalents, i.e., specifying FEMA’s requirements for elevation data to have 2-ft equivalent contour accuracy for flat terrain and 4-ft equivalent contour accuracy in rolling to hilly terrain– regardless of the technology used. It also included guidelines and specifications for base maps; horizontal accuracy; data requirements for different forms of flood studies; different data models (mass points, breaklines, TINs, DEMs and/or contours); file size, tile size and buffers; mapping areas; cross sections; treatment of hydraulic structures (bridges, culverts, dams, weirs); datums, projections and coordinate systems; data formats; hydro-enforced elevation data; and ground surveys for photogrammetric control, cross sections, hydraulic structures and checkpoints. In addition to FEMA’s extensive guidance for photogrammetric surveys, it also included section A.8 on LiDAR surveys.

The LiDAR portion of Appendix A addressed LiDAR system definitions; general guidelines for use; performance standards (including data voids, artifacts, outliers, system calibration, flight planning, GPS base stations); post-processing of LiDAR data to include breakline requirements; QA/QC of LiDAR to include accuracy testing and analysis of errors by land cover categories, locations, dates, and sensors; verification of airborne GPS, inertial measurement units (IMUs), and laser ranges; correction of systematic errors; cross flight verification; and various forms of deliverables. The LiDAR portion of Appendix A remained the LiDAR industry’s de facto standard until Karl Heidemann presented the draft USGS LiDAR Guidelines and Base Specification Version 13 at the International LiDAR Mapping Forum (ILMF) in 2010.

Appendix A required field surveyed cross sections immediately upstream and downstream of bridges and culverts, to include channel invert elevations for the deepest part of the channel. It directed Mapping Partners to survey intermediate cross sections where bridges and culverts are more than 1,000 feet apart, especially where a significant change in conveyance occurs between cross sections. These intermediate cross sections could be "cut" from stereo-photogrammetric or LiDAR datasets so long as there was no significant change in stream bed geometry below the water level. With automatic H&H and LiDAR datasets, the cross sections could be more numerous and truly representative of shorter reaches.

Other LiDAR Standards, Guidelines and Specifications
In 2004, I was the primary author of the NDEP Guidelines for Digital Elevation Data, version 1.0, and the ASPRS Guidelines: Vertical Accuracy Reporting for Lidar Data, both of which documented procedures for using the 95th percentile to estimate the accuracy of LiDAR data in vegetated and forested areas at the 95% confidence level, based largely on additional lessons learned from the NCFMP. In 2010, Harold Rempel and I were the principal authors of FEMA’s Procedure Memorandum No. 61–Standards for LiDAR and Other High Quality Digital Topography that adopted the 95th percentile methodology and aligned FEMA’s LiDAR specifications with USGS’ draft v13 specifications, but with vertical accuracy requirements linked to the level of flood risk identified by FEMA.

National Enhanced Elevation Assessment (NEEA)
In 2012, Dewberry authored the National Enhanced Elevation Assessment (NEEA) sponsored by USGS, FEMA, NRCS, NOAA, NGA and others, that directly led to USGS’ current 3D Elevation Program (3DEP) based on QL2 LiDAR or better (2 or more points/m2 with RMSEz 10-cm). With input from 34 Federal agencies, 13 private and non-profit organizations and all 50 states, the NEEA estimated annual flood risk management benefits between $295M and $500M from nationwide coverage of QL2 LiDAR. FEMA had identified many areas throughout the country where the flood hazards shown on the older FIRMs understated the true risk of flooding; those who built to standards on those older maps were in fact subject to a higher probability of flooding. FEMA also found that in some areas flood hazards on the older maps overstated the true risk (Figure 1A) which means those properties insured under the NFIP were paying more than they should be paying. Overall, FEMA’s experience indicates that map updates tend to be fairly even between adding and removing homes from the SFHA. Older data do not consistently overstate or understate flood risk; there is some of both. Figure 1B illustrates the improvements in map accuracy attained by the use of LiDAR data; more than 300 structures, some of which are outlined in red, were removed from the SFHA in Towns County, GA, through the use of higher accuracy LiDAR data. In other locations nationwide, additional structures are added to SFHAs as a result of more-accurate data and modeling. It works both ways. Whether flood risks are currently understated or overstated, higher accuracy LiDAR yields the following benefits: (1) structures insured at appropriate levels; (2) more consistent insurance ratings through better information about risk; and (3) more insurance purchased because of improved understanding of risk.

FEMA’s Elevation Guidance (November, 2015)
FEMA’s latest Elevation Guidance for Flood Risk Analysis and Mapping (November 2015) retired FEMA’s Appendix A (2003) and PM 61 (2010) while aligning FEMA’s requirements with the latest USGS LiDAR Base Specification, V1.2 from Karl Heidemann at USGS and the ASPRS Positional Accuracy Standards for Digital Geospatial Data (Nov, 2014) for which Karl Heidemann and I authored the LiDAR portions. This requires LiDAR point density of at least 2 points/ m2 with RMSEz 10-cm, meaning that FEMA’s requirements have improved from 4 ft contour accuracy in 1998 (from photogrammetry) to the current 1 ft contour accuracy (from LiDAR). I am proud to have been part of FEMA’s romance with LiDAR for nearly 20 years and am thrilled that FEMA is actively supporting the 3DEP which is vital to all of us in the LiDAR profession.

Dr. David Maune is an Associate Vice President at Dewberry where he is an elevation specialist and manages LiDAR, IFSAR and photogrammetric projects for USGS, NOAA, FEMA, USACE, and other federal, state and county governments. He specializes in independent QA/QC of LiDAR data produced by others. He is a retired Army Colonel, last serving as Commander and Director of the U.S. Army Topographic Engineering Center (TEC), now the Army Geospatial Center (AGC). In 1998, he authored NOAA’s National Height Modernization Study on how to modernize the National Height System in the U.S. based on Continuously Operating Reference Stations (CORS), differential GPS, LiDAR and IFSAR. Between 1998 and 2010, he authored all major FEMA guidelines for LiDAR. In 2004, he authored the Guidelines for Digital Elevation Data published by the National Digital Elevation Program (NDEP). In 2001 and 2007, he was the editor and principal author of the 1st and 2nd editions of Digital Elevation Model Technologies and Applications: The DEM Users Manual, published by ASPRS, with the 3rd edition planned for 2017. In 2012, he authored the National Enhanced Elevation Assessment (NEEA) report that provided the blueprint for today’s 3D Elevation Program (3DEP). In 2014, he co-authored the ASPRS Positional Accuracy Standards for Digital Geospatial Data. In 2015, he was the editor and principal author of USACE EM 1110-1-1000, Photogrammetric and LiDAR Mapping. In 2016 he won the ASPRS Photogrammetric Award for outstanding achievement in the field of photogrammetry. Dr. Maune earned his PhD in Geodesy and Photogrammetry from The Ohio State University in 1973.

A 5.630Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE