178
M. Menenti
the error on estimated evaporation. Given the spatial resolution and the equations
to be computed, weighing coefficients can be estimated to minimize the error due
to spatial aggregation. Examples of this approach were presented by Pelgrum and
Bastiaanssen (1996), Bastiaanssen et al. (1996) and Hipps et al. (1996). The underlying problem remains the definition and determination of a measure of the
spatial structure oflandscapes. For this purpose Menenti et al. (1996) explored the
use of spatial statistics of very high resolution laser measurements.
The issue of spatial aggregation (up-scaling) is of significant theoretical and
practical complexity and will require significant attention in the next few years.
The underlying cause of complexity is that the two fundamental concepts of homogeneity (of each landscape element) and heterogeneity (of a landscape comprising different elements) are relative rather than absolute concepts. It is therefore
difficult to translate these two concepts into a measure of homogeneity and heterogeneity. As a consequence, it is difficult to measure the spatial structure of a specific landscape and to assess the error due to spatial aggregation of observations at
a given resolution.
8.6 Accuracy
The issue of accuracy has haunted remote sensing studies of land evaporation
since the early years. Design of a validation experiment for maps of actual evaporation computed with nearly instantaneous image data is challenging because of
fundamental and practical reasons. The very nature of direct measurements of latent heat flux density by e.g. eddy correlation systems differs from estimates obtained with measurements of spectral radiances. The latter is local in space and
instantaneous in time, while the former integrates turbulent fluxes over some distance upwind of the observation station. Due to the intermittent nature of turbulent
heat transfer, measurements of surface radiant temperature may reveal changes too
rapid to be seen in the time integrated eddy correlation measurements. In homogeneous areas these difficulties may be limited, while in drier, heterogeneous landscapes it becomes challenging to compare correctly local direct measurements of
evaporation with the estimates obtained with image data.
Literature does provide useful, albeit somewhat conflicting, indications on accuracy. Moran et al. (1989) reported errors on LE of 1 %, -3% and -3% for different
dates and crops at the Maricopa Agricultural Research Centre in Arizona. Two
studies on the Walnut Gulch rangeland basin, also in Arizona, gave slightly different results: Moran et al. (1994b) obtained a mean absolute difference of 40 Wm- 2
with a LE-range of 0 to 340 Wm- 2 observed at the eight observation stations in the
watershed, while Kustas et al. (1995) obtained an average error of 15% over three
days of observations. Other results support the general approach of combining a
resistance model with measurements of Trad to estimate heat flux densities. Choudhury (1986) achieved an accuracy of <10% on daily LE and <5% on 10 days cumulative LE, while Zhang et al. (1995) obtained 50 Wm- 2 when comparing esti-
M. Menenti
the error on estimated evaporation. Given the spatial resolution and the equations
to be computed, weighing coefficients can be estimated to minimize the error due
to spatial aggregation. Examples of this approach were presented by Pelgrum and
Bastiaanssen (1996), Bastiaanssen et al. (1996) and Hipps et al. (1996). The underlying problem remains the definition and determination of a measure of the
spatial structure oflandscapes. For this purpose Menenti et al. (1996) explored the
use of spatial statistics of very high resolution laser measurements.
The issue of spatial aggregation (up-scaling) is of significant theoretical and
practical complexity and will require significant attention in the next few years.
The underlying cause of complexity is that the two fundamental concepts of homogeneity (of each landscape element) and heterogeneity (of a landscape comprising different elements) are relative rather than absolute concepts. It is therefore
difficult to translate these two concepts into a measure of homogeneity and heterogeneity. As a consequence, it is difficult to measure the spatial structure of a specific landscape and to assess the error due to spatial aggregation of observations at
a given resolution.
8.6 Accuracy
The issue of accuracy has haunted remote sensing studies of land evaporation
since the early years. Design of a validation experiment for maps of actual evaporation computed with nearly instantaneous image data is challenging because of
fundamental and practical reasons. The very nature of direct measurements of latent heat flux density by e.g. eddy correlation systems differs from estimates obtained with measurements of spectral radiances. The latter is local in space and
instantaneous in time, while the former integrates turbulent fluxes over some distance upwind of the observation station. Due to the intermittent nature of turbulent
heat transfer, measurements of surface radiant temperature may reveal changes too
rapid to be seen in the time integrated eddy correlation measurements. In homogeneous areas these difficulties may be limited, while in drier, heterogeneous landscapes it becomes challenging to compare correctly local direct measurements of
evaporation with the estimates obtained with image data.
Literature does provide useful, albeit somewhat conflicting, indications on accuracy. Moran et al. (1989) reported errors on LE of 1 %, -3% and -3% for different
dates and crops at the Maricopa Agricultural Research Centre in Arizona. Two
studies on the Walnut Gulch rangeland basin, also in Arizona, gave slightly different results: Moran et al. (1994b) obtained a mean absolute difference of 40 Wm- 2
with a LE-range of 0 to 340 Wm- 2 observed at the eight observation stations in the
watershed, while Kustas et al. (1995) obtained an average error of 15% over three
days of observations. Other results support the general approach of combining a
resistance model with measurements of Trad to estimate heat flux densities. Choudhury (1986) achieved an accuracy of <10% on daily LE and <5% on 10 days cumulative LE, while Zhang et al. (1995) obtained 50 Wm- 2 when comparing esti-