We are preparing maps of Floyd-related flooding in North Carolina based on three orbital sensors: NOAA Polar Orbiter-hosted AVHRR, NASA's Landsat 7 ETM, and Radarsat (both scansar and fine resolution modes). We are also comparing our results to the FEMA Q3 1% exceedance probabily floodplain maps.
Preliminary results include the following:
1) Visible and infra-red band change detection (before- and during-flood scenes) using AVHRR sensitively detects transient surface water. Due to the coarse spatial resolution, new areas of flooding can be located but the extent of flooding is overestimated: many changed pixels are mixed land/water pixels.
2) Single Landsat or Radarsat scenes cannot detect flooding under forest cover, which includes perhaps half of the flood-affected lands in North Carolina. However, both sensors easily discriminate unforested flooded lands and at resolutions ranging from 15-50 m/pixel.
3) We expect change detection methods using multispectral Landsat and C-band Radarsat scenes to better detect flooding under forest cover. There is some suggestion of enhanced C-band backscatter from flooded forest land.
4) FEMA Q3 Maps (digitized versions of the National Flood Insurance county floodplain maps) are shown to be in error on some Landsat scenes. The mapped floodplain should include some areas not presently mapped even where, generally, flooding occurred within the map limits. Along some river reaches, it appears that the flood magnitude was higher than that used in the mapping program: the maps were not in error but the size of discharge used was too small given the new experience of rainfall generated before and after Hurricane Floyd. Along these river reaches, the flooding generally exceeded the 100 yr and 500 yr recurrence interval flood as calculated prior to Floyd.
Comparison of Radarsat fine resolution and AVHRR mapping near Tarboro
Radarsat mapping in green; AVHRR in red. Forest cover obscures much flooded land from Radarsat, but AVHRR change detection, as illustrated on other maps at this site, accurately recorded radiance changes induced by flooding, though at coarse (1.1 km pixels) spatial resolution. Note: The AVHRR mapping was done from data obtained on September 17, 1999 and the Radarsat mapping was done from data obtained on September 23, 1999. Therefore, this map reflects the general subsidence of water levels between September 17 and 19th, in addition to the variations due to the use of different instruments to map the extent of inundation.
Comparison of FEMA Floodplain map (Wayne County) and Landsat 7 Panchromatic Image Map
This map of the the area just southeast of Goldsboro, NC along the Nuese River is based upon Landsat Panchromatic data from September 30, 1999. The red polygons represent non-forested flooded areas and the white lines are the data from the FEMA Q3 maps. This is an overview of the in-progess comparisons of mapped floodplain and imaged flooding that are underway for this and nearby counties.
Comparison of FEMA floodplain map and Landsat 7 Panchromatic Image Map
This is a very small subscene but illustrates a floodplain map error. The area featured is the confluence of Core Creek and the Nuese River in Craven County, NC; it is also based on the Landsat Panchromatic data from September 30, 1999. In some cases, this flood event exceeded the size represented in the FEMA 100 yr exceedance probability map, but in this case and many other cases the Landsat image records a simple map error: the land flooded should have been included in the floodplain)
