ÉTICA HACKER

“Research over the past 15 years has conclusively established that the E arth’s surface is non-Lambertian, and any future analyses, excepting perhaps the most simple and crude o f approximations, cannot legitimately assume Lambertian properties. Reflectance anisotropy is significant. ’’

Conclusion drawn by Diane E. Wickland (Manager, NASA Terrestrial Ecology Programme) following NASA workshop on multiangular remote sensing for environmental applications, Jan. 29-31 1997, Univ. Maryland, (in Privette et a i, 1997).

It has long been known that natural surfaces do not in general reflect incident radiation equally in all directions but instead tend to display varying degrees of anisotropy (Minnaert, 1941; Nicodemus et a l , 1977; Hapke, 1981). As mentioned previously, surface reflectance is not only a function o f the spectral and spatial properties o f the incident radiation and target respectively, but also o f the direction from which the surface is illuminated and viewed (Ross, 1981; Goel, 1988; Privette et a l, 1997). The reflectance anisotropy o f a surface is determined by two principal factors. Firstly, the intrinsic directionality o f the spectral reflectance, transmittance and absorptance o f the scattering material. Secondly, anisotropy is a function o f surface roughness, or structure. The latter property is determined by the density and arrangement o f objects on a surface and hence the nature o f the shadowing caused by these objects as a function of illumination and observation angles (Torrance and Sparrow, 1966; Otterman and Weiss, 1982; Li and Strahler, 1986, 1992; Roujean et a l, 1992). Surface structure will tend to cause incident radiation to be reflected more strongly in some directions than others. This dependence o f surface reflectance on viewing and illumination geometry is described by the Bidirectional Reflectance Distribution Function (BRDF). Knowledge o f the surface BRDF can therefore potentially be exploited to provide information regarding the structure of the scattering surface. Clearly, knowledge o f surface reflectance is also vital in order to calculate albedo, which is intimately related to BRDF. This thesis investigates models o f surface reflectance which have been developed recently in response to the need for accurate characterisation o f surface reflectance at global scales.

It is clear that the anisotropic nature o f the Earth's surface will have an effect on the reflectance observed from a remote sensing platform and such effects can derive from a variety o f sources. Sensors possessing wide viewing swaths, such as AVHRR and M ODIS, have large variations in view angle across recorded scenes (Cihlar et a l, 1994; Leroy and Roujean, 1994). The same is true o f sensors with along-track scanning or off- nadir pointing capabilities such as MISR, and the Along Track Scanning Radiometer instruments (ATSR and ATSR-2). Variations in the solar zenith (illumination) angle over a point on the Earth’s surface will also tend to have the same effect as varying the viewing zenith angle. This occurs if a particular point is imaged at different times o f day or at the same time o f day throughout the year. The anisotropy o f surface reflectance makes direct comparisons between such data impossible. Apparent observed changes in the nature o f the surface may in fact be caused purely by variations in viewing and illumination conditions from one image to another. As the availability o f time-series of

reflectance data has increased, variations caused by inconsistent viewing and illumination angles has come to be recognised as a serious problem (Leroy and Roujean, 1994; Cihlar

e t a l , 1994).

Problems may also arise within individual scenes. Land surface process and climate modelling requirements have directed many new sensors towards global coverage at lower spatial resolution (of the order o f km per pixel), and away from the high resolution observations o f limited spatial coverage that have been the norm (10s o f m per pixel). This is necessary in order to keep the quantities o f data produced manageable. The wide swaths o f such sensors such as AVHRR, MODIS and POLDER mean that there may be large variations in view angle across any single swath. MODIS for example has a swath width o f approximately 2330km, with a consequent variation in view zenith of around 110° from one side to the other. So there are likely to be directional effects in observed reflectance within MODIS data simply because the viewing zenith angle varies from 0° at nadir to nearly 55° at either edge o f the image. The orbital characteristics o f the sensor also mean that on repeated orbits a particular point on the surface will be observed at a wide range of viewing (and illumination) angles.

Directional effects in BO data were typically ignored until relatively recently. This is exemplified by the widespread use o f AVHRR maximum-value composite (MVC) products (Holben, 1985). It is now recognised that in order to make comparisons of measured surface reflectance across or between scenes, then directional effects must be accounted for (Roujean et al., 1992; Wu et al. 1994; Privette et al., 1997). In examining changes in cover type for example, comparisons must be made between data that may be separated in time by many months, and which may have been measured at significantly different viewing and illumination angles (Cihlar et a l, 1994). In addition to the effects mentioned above, which are often considered as obstacles to be overcome (e.g. Roujean

et a l, 1992; Leroy and Roujean, 1994), the directional nature o f surface reflectance can

also be exploited. The directional dependence o f the surface reflectance on its structural properties implies that observations o f the directional component o f reflectance will contain information relating to these properties (Goel and Strebel, 1983; Goel, 1988; Goel and Reynolds, 1989; Myneni et a l, 1989). It may be possible to relate variations in observations at different viewing and illumination angles to the surface features causing these variations (Asrar, 1989; Pinty and Verstraete, 1991; Myneni et a l, 1995).

Analogous to the manner in which spectral variations in reflectance are exploited via multi-band data, the directional component can be exploited using multi-angle data.

The fact that the directional signature o f remotely sensed surfaces will contain structural information has led to the development o f new missions designed specifically to explore and exploit the directional signal. This signal is not available to sensors that have no directional sampling capability, either through a wide field o f view, a pointing capability or orbital characteristics. Airborne instruments such as the NASA Advanced Solid state Array Spectrometer (ASAS) (Irons et a l, 1991) have been in use for some time (Barnsley et a l, 1997b; Lewis et a l, 1999). On a much larger scale, the MODIS and MISR instruments on-board the Terra platform are both specifically designed to exploit directional variations in surface reflectance (W anner et a l, 1997; Knyazikhin et a l,

1998a,b; Martonchik et a l, 1998a,b). MODIS data exploit the instrument's wide swath and rapid repeat coverage to provide viewing and illumination variation, whilst MISR is has nine fixed cameras, giving four look angles in the along-track direction fore and aft as well as a nadir view. An intermediate instrument is CHRIS (Compact High Resolution Imaging Spectrometer) aboard the PROBA (Project for On Board Autonomy) platform launched in autumn 2001 (Barnsley et a l, 2000; w w w [l.l 1]). CHRIS is a high-resolution (25 to 50m) optical instrument which (uniquely) combines (19) selectable radiometric bands with a pointable multi-angle capability. This variety o f instruments highlights the well-established (and growing) interest in multi-angle remote sensing.