Interpretation of ERS-SAR images over urban surfaces

Abstract

Large variations of backscatter intensities were observed in between and within ERS-SAR images of the Los Angeles metropolitan area. Low intensities backscatter were found over smooth pavement and asphalt, and higher ones over industrial and commercial areas identified using a SPOT-HRV image classified for land-use. Highest backscatter anisotropy occurred when the flight direction of the SAR was parallel to the alignment of streets. These contaminations mask the backscatter variations due to surface roughness and to soil moisture, and indicate that the retrieval of moisture availability from ERS-SAR images is contingent upon eliminating backscatter contamination from processes inherent to the urban structure.

Keywords: ERS-SAR, soil moisture, surface roughness, anisotropy effects, urban climatology

1. Introduction

Urbanization modifies the nature and properties of the surface and atmosphere. It alters the radiative, thermal, hydrological and aerodynamics components of the energy fluxes and their interaction with the surface. The understanding of urban microclimate perturbations relays on our ability to model the alteration of these components and of their interaction. In particular we need to better understand the physical processes that determine the partition of radiative energy into sensible and latent heat fluxes.

Energy fluxes components such as Land Surface Temperature (LST), albedo, vegetation density, aerosols and cloudiness can efficiently be collected by radiometers on board polar orbiting satellites. For example, urban LST have been derived from infrared sensors such as NOAA-AVHRR (Dousset 1989), ERS-ATSR or LANDSAT-TM.

Soil moisture is an important forcing factor. As shown in Fig. 1, there is a strong negative correlation between averaged summer afternoon LST and the Normalized Difference Vegetation Index over Los Angeles. In fact, the moisture availability of residential areas allows for a great fraction of radiant energy to be converted into latent heat flux, lowering the sensible heat flux, hence the surface temperatures (Dousset 1991). Such properties could be applied to mitigate heat island effects by a proper spatial distribution of permeable materials, vegetation and urban parks.

Figure 1. Joint distribution of the summer Los Angeles Land Surface Temperatures at 14:50 PDT (averaged over 15 NOAA-AVHRR IR images), versus the Normalized Difference Vegetation Index computed from NOAA-AVHRR ch. 1 and 2

The application of microwave sensing to the retrieval of moisture content of bare soil and vegetation canopy, such as forest or crops, have been extensively studied through physical observation and theoretical modeling. Conversely, there have been few applications over urban areas due in part to the heterogeneity and roughness of their surfaces. Moisture availability is still inferred indirectly from a combination of thermal remote sensing and inverse modeling, as described in Carlson et.al, (1981), and Gillies et al., (1995). The lack of moisture monitoring from remote sensing limits our ability to close the urban surface energy balance, and to model the LST amplitude.

The retrieval of soil moisture from Synthetic Aperture Radar systems such as ERS-SAR is based on the dielectric properties of the surfaces, influencing the intensity of backscattered electromagnetic waves. Sensing of soil moisture arises from the large difference between the dielectric constant of water and that of dry soil (Dobson and Ulaby, 1986).

Here, we present a preliminary analysis of ERS-SAR series of images collected over the Los Angeles basin to explore the applicability of microwave remote sensing to the study of the energy balance at the urban surface.

2. Test site and data acquisition

The test site is the Los Angeles metropolitan area, located in a coastal plain bounded by mountains, centered at 118 W-34 N. It is characterized by a high degree of urbanization and a low density population of 15 million. The climate of Los Angeles is semi-arid, with sporadic periods of high rainfall associated to global climatic cycles, resulting in a large annual and interannual amplitude of soil moisture. The topography of the Los Angeles basin is rather uniform, simplifying the interpretation of SAR backscatter.

The analysis is based on the following co-located data:

• 19 ERS-SAR images in C band (6cm) VV polarization, sensed in between July 1992 and January 1996 spanning all seasons.
• 1 SIR-C image in L band HH and HV polarizations, and C band HV polarization, sensed in October 1994.
• 1 Multispectral 20 m resolution SPOT-HRV image, sensed in June 1986.
• Near-continuous monitoring of Land Surface Temperature from NOAA-AVHRR imaging.

The different inclination of the ERS satellite orbits and both ascending and descending passes allows for the analysis of backscatter variations with illumination azimuth. The images cover the entire Los Angeles basin, however the observations were principally focussed on the area comprised between downtown Los Angeles and Long Beach harbor (Fig.2).

The SPOT-HRV image was classified and the Normalized Difference Vegetation Index computed. The classification (Fig. 2) arises from the joint distribution of the visible HRV-1 and near-infrared HRV-3 channels, and the vegetation index results from their normalized difference.

Fig. 2 Classified SPOT-HRV image of Los Angeles. White: clouds or high reflective materials; red: dry soils, densely built areas, transportation, commercial and industrial areas; pink:low clouds over the latter areas; dark blue :residential areas; turquoise:lighter soil and thin clouds over the latter areas; green: urban parks and golf-courses; yellow: natural vegetation.

3. ERS-SAR difference and mean images

An ERS-SAR image difference (not shown here) was derived from a pair of descending images. The images were aligned using the coastline and reference points from the harbor structures in Long Beach. The texture of the image difference revealed distortions amounting to 4-5 pixels, precluding interferometry without first remapping the images by a bilinear or higher order transform to a reference image.

The ERS-SAR mean image (not shown here) was derived from the same pair of images. It was remapped to a smaller SPOT- HRV image, and corresponds to an area with little distortion, ~1-2 pixels. The mean image emphasizes the urban structures such as highways, roads, airport runways, bridges upon the Los Angeles River, and the delineation of land covers such as parks and asphalt by reducing speckle.

Fig. 3, shows the joint distribution of the ERS-SAR mean image and SPOT-HRV Vegetation Index. The distribution displays two clusters: the small one at the bottom left corresponds to the ocean, the larger one at the center left corresponds to land surfaces devoid of vegetation. These latter exhibit the highest backscatter, probably due to surface roughness and contamination inherent to the urban structure. The points comprised between 0.1-0.5 vegetation index and 0-2000 backscatter intensity indicates the relatively large amount of vegetation in this urban area, and the relationship of backscatter intensity to vegetation, related to moisture availability and evapo-transpiration. This figure also demonstrates the need to mask pixels with high backscatter from urban surfaces devoided of vegetation, (industrial, commercial or densely built) and from moving vehicles, before attempting to retrieve soil moisture. Therefore, I focus here on the backscatter anisotropy and its relationship to the morphology of different urban land-uses/covers.

Figure 3. Joint distribution of Backscatter Intensity from the mean ERS-SAR image and the Normalized Difference Vegetation Index from the SPOT-HRV image.

4 . Observation over different urban land-uses

Large variations of backscatter intensity were observed among and within the images. The response of each land-use class (Fig.2) was examined separately and is shown in Figs. 4 and 5.

Figure 4. ERS-1 SAR band C image of Los Angeles, November 24, 1993 at 18:31 UTC, descending pass.

Figure 5. ERS-1 SAR band C image of Los Angeles, November 5, 1992, at 06:13 UTC, ascending pass.

Commercial and industrial

The areas classified as devoid of vegetation based on the SPOT image, correspond actually to three different subclasses in the SAR images: smooth paved surfaces, industrial metallic structures, commercial and industrial buildings. Very low backscatter occurs on asphalt, pavement and bare soil, such as airport runaways (Long Beach airport, A), freeways and storm channels (Los Angeles river, B). Low backscatter results both from the surface smoothness, that reduces Bragg scattering, and from low soil moisture, that reduces subsurface scattering. Random contamination occurs along many freeways, possibly due to vehicles acting as crude corner reflectors. Some may be Doppler-shifted, explaining the lateral displacement seen in enlarged images (Fig.6).

Figure 6. Lateral displacement of vehicles along Los Angeles Freeway 10.

Within the same SPOT class, pixels with very high backscatter intensity correspond to industrial structures, for example chemical industries (fuel tanks, pipelines, oil wells) in Carson (C), and harbor structures (containers, cranes, ships) in Long Beach (D). Comparing the different images, the high reflectance appears relatively isotropic, presumably because many of these structures have either curved surfaces (tanks, ships) or are randomly oriented (cranes, oil wells).

In contrast, distinctly anisotropic properties are found where streets and tall buildings are regularly aligned. For example, downtown Los Angeles (E) and commercial districts along the main boulevards (Alameda, F) have intensities depending on the illumination azimuth. In these areas, the regular geometry of reinforced concrete walls, metalized roofs and steel beams provides numerous dihedral reflectors, resulting in greatly enhanced backscatter when the illumination azimuth is orthogonal to the buildings.

Parks and vegetation

Areas classified as vegetation such as urban parks, golf-courses, grass in undeveloped blocks and airports have in general low isotropic backscatter. Examples in Fig. 4 and 5 are the Victoria park and golf course in Carson (G), and Seal Beach Naval Weapon Station (H).

Residential

Areas classified as residential have considerable speckle, with average backscatter larger than the vegetation class. The additional soil moisture due to irrigated lawns is unlikely to be the cause of this increased intensity. Instead, the presence of numerous sub-pixel metallic scattering targets such as vehicles, utility lines, electrical wiring and appliances in houses is likely to increase backscatter to a level higher than vegetation. Some residential blocks feature a significant imaging anisotropy. A striking example is provided by the districts of Huntington Park and South Gate (J). These districts are alternately brighter (Fig. 4) and darker (Fig. 5) than average in the pair of ERS-1 images.

A close examination of city maps reveals that the streets of these districts are nearly exactly aligned with the flight direction for Fig. 4, but not for Fig. 5. A similar enhanced backscatter correlated with street alignment can be seen in the SIR-C image over Burbank and Santa Monica (not shown). This anisotropy is somewhat unexpected, and suggests that even light constructions based on wood and stucco walls may be favorable to dihedral reflections.

5. Conclusions

In many areas of the Los Angeles basin, imaging anisotropies mask the variations of backscatter due to other processes, illustrating the complexity of extracting quantitative information from SAR images of urban surfaces. Over industrial and commercial areas, high backscatter from concrete and metallic structures appears to hopelessly contaminate the signal, precluding the extraction of information on the ground surface. Over some low density residential areas, the increases of backscatter intensity is correlated with the angle between the radar illumination and the streets. With no further corrections, the estimation of urban surface properties such as roughness and soil moisture should be limited to areas devoid of constructions. Selecting those areas requires a priori knowledge of land use which can be derived from multispectral SPOT or LANDSAT images.

Future work will attempt to model the anisotropy based on a series of images with different illumination azimuths and a database on urban topography, and to derive a backscatter intensity corrected for these contaminations.

6. Acknowledgments

The author thanks the European Space Agency for providing the ERS-SAR images, and the National Aeronautics and Space Administration for support under grant NAGW-4940. The work was performed at the Oceanography Satellite Laboratory of the University of Hawaii.

7. References

Carlson, T.N., J.K. Dood, S.G. Benjamin, and J.N. Cooper, "Satellite estimation of surface energy bal- ance, moisture availability and thermal inertia," J. Applied Met., vol. 20, pp. 67-87, 1981.

Dobson, M.C and F.T Ulaby, "Active microwave soil moisture research," IEEE Trans. Geosci. Remote Sens., vol. GE-24, no. 1, pp. 23-36, 1986.

Dousset, B., "Surface temperature statistics over Los Angeles: the influence of land use," in Proceedings of IGARSS-91, pp. 367-371, I.E.E.E.,1991.

Dousset, B., "AVHRR-derived cloudiness and surface temperature patterns over the Los Angeles area and their relationship to land use," in Proceedings of IGARSS-89, pp. 2132-2137, I.E.E.E., 1989.

Gillies, R.R. and T.N. Carlson, "Thermal Remote Sensing of Surface Soil Water Content with Partial Vegetation Cover for Incorporation into Climate Models," J. Applied Met., vol. 34, pp. 745-756, 1995.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry