Satellite images acquired after Hurricane Katrina can be used to measure the extent of flooding and to estimate the volume of water for defined areas. Images from multiple sensors (Landsat 5, Landsat 7, SPOT, and RADARSAT) were used to delineate inundated areas (flood polygons). These flood polygons can then be combined with a high resolution digital elevation models (DEM) derived from 5-m lidar data to estimate water depths and volumes in the defined areas. The lidar data were collected in 2002 and the resulting DEM has a spatial resolution of 10 m.

Flood polygons for the 30 Aug 05 (Landsat 7) and 2 Sep 05 (SPOT) dates were provided by the Dartmouth Flood Observatory (http://www.dartmouth.edu/~floods/). Flood polygons for the 7 Sep 05 (Landsat 5) and 15 Sep 05 (Landsat 7) dates, as well as the RADARSAT scenes (2 Sep 05, 5 Sep 05) were developed at USGS EROS. Before calculating flood volume estimates, the Landsat and SPOT flood polygons were adjusted with GIS techniques to account for flooded houses/structures that had rooftops above the waterline – i.e., houses and structures not included in the original flood polygons (because they were not identified as water pixels) that were surrounded by water and within the larger flood polygons were re-categorized as flood pixels, and included in the flood polygons.

In order to calculate flood volume estimates, pixels within the flood polygons were filled with elevation values from the DEM. Four “cells” were created based on USACE (U.S. Army Corps of Engineers) water pumping units, and elevation values were analyzed for each of the cells individually. The maximum elevation values within these cell-based flood polygons were analyzed to remove any erroneously high-valued pixels. High resolution IKONOS imagery was used to spot-check the margins of the flood polygons from the same day to determine the error in the flood polygons. Pixels greater than 2.5 standard deviations above the norm for the SPOT and Landsat scenes were eliminated as error. This produced greater agreement between the flood polygons and the visible flooding in the IKONOS imagery. The RADARSAT-derived polygons had a slightly higher error, thus all pixels greater than 2.0 standard deviations were eliminated from these polygons. This affected very few pixels, but the inclusion of the pixels would have skewed the analysis for each of the cells, causing over-estimation of flood depth (by pixel) and flood volume (by cell). By eliminating these error pixels, our maximum inundated elevation went from an unreasonable average of 12 m to approximately 1-3 m depending on the cell and date. The corrected maximum inundation elevation was used as the height of the water within each cell. Only a small number of pixels (black pixels in Figure 1) were excluded from the analysis.

Figure 1: Elevations values for 2 Sep 05 SPOT-derived flood polygons

For each of the cells defined on the flood depth/volume maps, depths were created by subtracting the maximum elevation value determined in the previous step (circled value in the legend in Figure 1) from all elevation values equal or less than that value and taking the absolute value. This set the depths to zero for all elevation values higher than the corrected maximum inundation elevation.

Figure 2: Water depth used for calculating water volume per cell.

Water volume was then calculated for each pixel by multiplying the depth by the pixel size. Pixel volumes were summed up within each cell to obtain the total water volume within each of the hydrographic units.

	L7 - Aug 30	Spot - Sept 2	L5 - Sept 7	L7 - Sept15
1	101,197,000	163,977,000	155,685,000	62,954,100
2	170,089,000	196,888,000	121,830,000	78,156,000
3	0	50,579,400	38,864,700	43,851,200
4	50,589,000	83,555,800	58,111,200	27,731,300
	-----------------	-----------------	-----------------	-----------------
Total	321,875,000	495,000,200	374,490,900	212,692,600

Figure 3: USACE cells, and water volume for individual sensors/dates (in cubic meters).

The maximum volume inundating the city on 2 Sep 05 was nearly a half billion cubic meters (Figure 3). As of 15 Sep 05 more than half of the water had been drained from New Orleans. The various rates of draining are due in part to the number of operational pumps in each of the cells. See http://www.hq.usace.army.mil/cepa/katrina/pumps/pumps.html for additional information on the status of the pumps.

The analysis of the RADARSAT data provided information very similar in extent (of flood polygons) and magnitude (of water volume) to that provided by the analysis of the Landsat and SPOT data. However, due to the difficulty in interpreting the radar data, and the very different sensors, RADARSAT-derived results are not included here for direct comparison with the Landsat- and SPOT-derived results.

These data/results are provisional. They have not been verified or validated with ground-based information, and are subject to revision. Nevertheless, this application demonstrates the utility of mid-resolution satellite data, in conjunction with high resolution DEMs, analyzed over a large spatial extent.

High-resolution, high-accuracy elevation data derived from lidar data collected in 2002 have been used extensively during the first weeks of recovery from Hurricane Katrina. The precise topographic measurements provided by lidar have proven very useful for various mapping activities in the low relief environment of New Orleans.

The graphic below was produced at USGS/EROS using the 2002 lidar data. Because an accurate delineation of the inundated area from remote sensing data was not yet available, the flood water level within the city was derived from a lake level gage on Lake Pontchartrain, with the assumption being that the level of the lake and the flood waters had equalized by the afternoon of Friday, September 2, 2005. Comparison with subsequent imagery has shown the flood delineation to be a reasonable depiction.

The lidar data have also been used to derive estimates of flood water volume. The work to produce a flood capacity curve was done at the request of the USGS Office of Surface Water, which was responding to an inquiry from the Corps of Engineers to provide independent corroboration of their estimates. Accurate estimates of flood volume are needed to project the length of time required to remove the water from the city.

The graph below displays the estimated volume and surface area of the flood waters at 1 foot increments. Note that the volume and area estimates are only for the areas shown as inundated on the above graphic. The depths are relative to the water surface as of the afternoon of Friday, September 2, 2005. The volume and area of flood waters have decreased since that time as pumping operations have been initiated and are ongoing.

These data are provisional, have not been verified or validated with ground-based information, and are subject to revision. However, this application demonstrates the usefulness of highly detailed topographic information for inundation mapping and analysis.

By combining the precise elevation information from lidar with accurate ground-based water level information and remote sensing derived inundation delineations, a complete history of flooding and de-watering can be reconstructed. Such a history will be useful for assessing the effect of the flooding on urban land cover. For example, the effects of depth and length of inundation upon urban vegetation land cover classes can be assessed. Likewise, the effects of varying inundation conditions on different types of structures can be documented.

Another use of an understanding of the effects of inundation over time on the urban environment will be to help in planning reconstruction of infrastructure. If a detailed hydrologic analysis shows specific inundation duration patterns, then structures can be rebuilt in a way that mitigates impacts from future storms. The inundation history and conditions can also be used to test how accurate pre-storm planning simulations have been, and to make appropriate modifications to future modeling scenarios.

Download the [PDF] [WRD]