Remote Sensing and GIS Accuracy Assessment - Chapter 20 (end)

CHAPTER 20 Area and Positional Accuracy of DMSP Nighttime Lights Data Christopher D. Elvidge, Jeffrey Safran, Ingrid L. Nelson, Benjamin T. Tuttle, Vinita Ruth Hobson, Kimberly E. Baugh, John B. Dietz, and Edward H. Erwin CONTENTS 20.1 Introduction...........................................................................................................................281 20.2 Methods................................................................................................................................283 20.2.1 Modeling a Smoothed OLS Pixel Footprint............................................................283 20.2.2 OLS Data Preparation..............................................................................................284 20.2.3 Target Selection and Measurement..........................................................................285 20.3 Results...................................................................................................................................286 20.3.1 Geolocation Accuracy ..............................................................................................286 20.3.2 Comparison of OLS Lighting Areas and ETM+ Areas............................................288 20.3.3 Multiplicity of OLS Light Detections......................................................................288 20.4 Conclusions...........................................................................................................................289 20.5 Summary...............................................................................................................................291 Acknowledgments..........................................................................................................................291 References......................................................................................................................................291 20.1 INTRODUCTION The Operational Linescan System (OLS) is an oscillating scan radiometer designed for cloud imaging with two spectral bands (visible and thermal) and a swath of approximately 3000 km. The OLS is the primary imager flown on the polar orbiting Defense Meteorological Satellite Program (DMSP) satellites. The OLS nighttime visible band straddles the visible and near-infrared (NIR) portion of the spectrum from 0.5–0.9 µm and has six-bit quantitization, with digital numbers (DNs) ranging from 0 to 63. The thermal band has eight-bit quantitization and a broad band-pass from 10–12 µm. The wide swath widths provide for global coverage four times a day: dawn, daytime, dusk, and nighttime. DMSP platforms are stabilized using four gyroscopes (three-axis stabilization) and platform orientation is adjusted using a star mapper, Earth limb sensor, and a solar detector. At night the OLS visible band is intensified using a photomultiplier tube (PMT), enabling the detection of clouds illuminated by moonlight. With sunlight eliminated, the light intensification results in a unique data set in which city lights, gas flares, lightning-illuminated clouds, and fires can be 281 © 2004 by Taylor & Francis Group, LLC 282 REMOTE SENSING AND GIS ACCURACY ASSESSMENT Visible Band PMT Shift Point PMT Shift Point Thermal Band Figure 20.1 Visible and thermal NIR nighttime OLS images over California.With sunlight eliminated, the OLS’s light intensification results in the detection of lights present at the Earth’s surface. observed (Figure 20.1). The OLS visible band sensor system is designed to produce visually consistent imagery of clouds at all scan angles for use by U.S. Air Force meteorologists with a minimal amount of ground processing. The visible band base gain is computed on-board based on scene source illumination predicted from solar elevation and lunar phase and elevation. This automatic gain setting can be overridden or modified by commands transmitted from the ground. The automatic gain is lowest when lunar illuminance is high. As lunar illuminance wanes, the gain gradually rises. The highest visible gain settings occur when lunar illumination is absent. The combination of high gain settings and low lunar illuminance provides for the best detection of faint light sources present at the Earth’s surface. The drawback of these high gain setting observations is that the visible band data of city centers are typically saturated. Data acquired under a full moon when the gain is turned to a lower level are generally not as useful for nighttime lights product generation since they exhibit fewer lights and have the added complication of bright clouds and terrain features. In addition to tracking lunar illuminance, gain changes occur within scan lines with the objective of making visually consistent cloud imagery, regardless of scan angle. The base gain is modified every 0.4 ms by an on-board along-scan-gain algorithm. A bidirectional reflectance distribution function (BRDF) algorithm further adjusts the gain to reduce the appearance of specular reflectance in the scan segment where the solar or lunar illumination angle approaches the observation angle. The OLS design provides imagery with a constant ground-sample distance (GSD) both along-and across-track. The along-track GSD is kept constant through a sinusoidal scan motion, which keeps the track of the scan lines on the ground parallel. The analog-to-digital conversion within individual scan lines is timed to keep the GSD constant from the nadir to the edge of scan. OLS data can be acquired in two spatial resolution modes corresponding to fine-resolution data (0.5-km © 2004 by Taylor & Francis Group, LLC AREA AND POSITIONAL ACCURACY OF DMSP NIGHTTIME LIGHTS DATA 283 Nadir 2.2 km pixel 3.0 km pixel 4.3 km pixel PMT center FOV switching position 766 km from nadir Edge of scan 5.4 km pixel Figure 20.2 The OLS fine-resolution nighttime visible band instantaneous field of view (IFOV) data starts at 2.2 km at the nadir and expands to 4.5 km at 766 km out from the nadir. After the PMT electron beam is switched the IFOV is reduced to 3 km and expands to 4.8 km at the far edges of the scan. GSD) and smoothed data (2.7-km GSD). All data are acquired in fine resolution mode, but in most cases the recorded data are converted to the smoothed resolution by averaging of 5 5 pixel blocks. While the GSD of OLS data is kept constant, the instantaneous field of view (IFOV) gradually expands from the nadir to the edge of the scan (Figure 20.2). At nadir the low-light imaging IFOV of the fine resolution data is 2.2 km and it expands to 4.3 km 800 km out from the nadir. At this point in the scan the electron beam within the OLS PMT automatically shifts to constrain the enlargement of pixel dimensions, which normally occurs as a result of cross-track scanning (Lieske , 1981). This reduces the IFOV to 3 km. The IFOV then expands to 5.4 km at the edge of the scan, 1500 km out from the nadir. Thus, the IFOV is substantially larger than the GSD in both the along-track and along-scan directions. At the nadir the smoothed OLS low-light imaging pixel has an IFOV of 5 km and at the edge of the scan the IFOV is approximately 7 km. In order to build cloud-free global maps of nighttime lights and to separate ephemeral lights (e.g., fires) from persistent lights from cities, towns, and villages, a compositing procedure is used to aggregate lights from cloud-free portions of large numbers of orbits, spanning months or even multiple years (Elvidge et al., 1997, 1999, 2001). To avoid the inclusion of moonlit clouds in the products, only data from the dark half of the lunar cycle are composited. The lights in the resulting composites are known to overestimate the actual size of lighting on the ground. The objective of this chapter is to document the area and positional accuracy of OLS nighttime lights and to examine the causes for the area overestimation of OLS lighting. We have done this using light from isolated sources located in southern California. The analyses were conducted using data from four OLS sensors spanning a 10-year time period. 20.2 METHODS 20.2.1 Modeling a Smoothed OLS Pixel Footprint A scaled model of an OLS PMT smoothed pixel IFOV at nadir was built by placing 25 fine-resolution pixel footprints onto a 5 5 grid, each displaced by a 0.5-km GSD. The number of times a light would get averaged into a smoothed pixel was tallied for each of the resulting polygon outlines (Figure 20.3). A similar model was built to show the IFOV overlap between adjacent PMT smoothed pixels. This model was constructed by placing nine of the smoothed pixel footprints from Figure 20.5 onto a 3 3 grid using a scaled GSD of 2.7 km. The number of smoothed pixel detection opportunities was then tallied for each polygon zone (Figure 20.4). © 2004 by Taylor & Francis Group, LLC 284 REMOTE SENSING AND GIS ACCURACY ASSESSMENT 1 1 1 2 2 2 2 1 4 4 4 1 3 5 6 6 5 3 6 2 5 1 4 8 2 6 1 4 9 2 6 1 4 8 2 5 6 3 8 9 8 10 11 9 110 8 9 8 6 9 1112 15 31 131512 11 9 8 14 14 12 17 12 15 11 13 14 15 15 13 1114 12 1 1513 5171 12 314 9 12 12 15 10 12 1315 412 41513 12 9 12 13 12 9 8 9 10 11 1 10 9 8 8 9 8 5 6 6 5 5 2 4 1 6 2 4 1 6 2 4 1 5 2 3 1 1 4 4 4 2 2 2 2 1 1 1 5 km 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Figure 20.3 Scaled model of a PMT smoothed pixel at the nadir composed of 25 fine-resolution pixel footprints. The overlap between IFOVs of adjacent fine-resolution pixels results in the possibility that lights present on the Earth’s surface will be averaged into the smoothed pixels multiple times.The number labels marked on the polygons indicate the number of times lights present in the polygons would be averaged into the resulting smoothed pixel. 20.2.2 OLS Data Preparation Nighttime DMSP-OLS data from 2210 orbits acquired between April 26, 1992, and April 4, 2001, were processed to produce georeferenced images of lights and clouds of the southern California region. The data were initially processed for the NOAA National Marine Fisheries Service to determine the locations and temporal patterns of squid fishing activities conducted using heavily lit boats offshore from the Channel Islands. Data were included from four day–night DMSP satellites: F-10, F-12, F-14, and F-15. DMSP data deliveries to the archive were irregular during 1992, resulting in gaps in the early part of our time series. Orbits were selected from the archive based on their acquisition time to include nighttime data over California. The orbits were automatically suborbited based on the nadir track to 32˚–42˚ north latitude. Lights and clouds were identified using the basic algorithms described in Elvidge et al. (1997). The next step in the processing was to geographically locate (geolocate) the suborbits. The geolocated images covered the area from 32˚–36˚ north latitude and 117˚–122˚ west longitude. The OLS geolocation algorithm uses satellite ephemeris (latitude, longitude, and altitude at nadir) generated by the SPEPH (Special Ephemeris) orbital model developed by the U.S. Air Force specifically for the DMSP platforms. The orbital model was parameterized by bevel vectors derived from daily RADAR sightings of each DMSP satellite. Ephemeris data were calculated for each © 2004 by Taylor & Francis Group, LLC AREA AND POSITIONAL ACCURACY OF DMSP NIGHTTIME LIGHTS DATA 285 1 2 3 4 Number of Overlapping Smooth Pixels Figure 20.4 Scaled model of a three-by-three block of smoothed OLS PMT pixels. The dashed line indicates the boundary of a single smoothed pixel IFOV at the nadir, as modeled in Figure 20.3. Because of the substantial overlap between adjacent smoothed pixel IFOVs it is possible for point sources of light to show up in more than one smoothed pixel. The number of overlapping smoothed OLS pixel IFOVs for each polygon is indicated using the grayscale. Additional levels of overlap are encountered in actual OLS imagery as the pattern is extended beyond this three-by-three example and as the IFOV expands at off-nadir scan angle conditions. scan line. The geolocation algorithm calculates the position of each OLS pixel center using the satellite ephemeris, a calculation of the scan angle, an earth geode model, and a terrain correction using GTOPO30. The pixel center positions were used to locate the corresponding 30 arc second grid cells, which are filled with the OLS DN values. This generates a sparse grid, having DN data only in cells containing OLS pixel centers. The complete 30 arc second grids were then filled to form a continuous image using nearest-neighbor resampling of the sparse grids. 20.2.3 Target Selection and Measurement A composite of cloud-free light detections was produced using data from the entire time series. The composite values indicated the number of times lights were detected for each 30 arc second grid cell. These were then filtered to remove single-pixel light detections, a set that contains most of the system noise (Figure 20.5). The cloud-free composites were then used to identify persistent light sources (present through the entire time series) for potential use in the study. Two types of persistent lights were selected: (1) isolated point sources with lighting ground areas much smaller than the OLS pixel, such as oil and gas platforms in the Santa Barbara Channel (Figure 20.6) and (2) isolated lights with more extensive areas of ground lighting. We identified five point sources: four oil and gas platforms (Channel Islands 1–3 and Gaviota 1) and a solitary light present at an airfield on San Nicolas Island. Calibration targets with more extensive area of lighting included a series of cities, towns, and facilities found on land (Table 20.1). © 2004 by Taylor & Francis Group, LLC ... - tailieumienphi.vn