Fig. 1. (a) General location (geographic projection) and (b) the Tel-Aviv test area.

Fig. 2. Spectral library of the 10 land-cover classes, selected for this study. ETM+ wavebands are numbered and overlaid on top. Reflectance values were offset for clarity.

Fig. 3. Four kernels used for convolution in the synthesis process.

Fig. 4. Processing steps: (a) ETM+ image, (b) classification by maximum likelihood, (c) simulated image (convolution operator #2), and (d) a corresponding GoF image (band #1). Note the lowest GoF values along the Yarkon stream (see also Fig. 1). Black pixels in (b–d) are those classified as “un-known” and were not considered in the convolution process. (For interpretation of the references to colour inthis figure legend, the reader is referred to the web version of this article).

Fig. 5. Performance by GoF per band: (a) means, and (b) general consistency trend.

Fig. 6. Performance by GoF of land-cover (LC) classes: (a) means and (b) general consistency trend. The dashed line represents the shift from relatively small objects to larger objects.

Fig. 7. Band-wise performance according to the correlation determination coefficient (R²).

Fig. 8. Land-cover (LC) class-wise performance according to the correlation determination coefficient (R²).

Fig. 9. Coefficients of correlation determination of land-cover classes in the Tel-Aviv case study: water, grass and buildings (a–c, respectively).

Fig. 10. Synthesis performance according to the correlation determination coefficient (R²). Naïve corresponds to no convolution, and columns 1–4 correspond to convolution operators 1–4, respectively (see discussion at Section 2.3.1).

Fig. 11. A subset of band 4 of the reference (real) image (a), corresponding simulation outputs of convolution operators 1 (b) and 2 (c) and of a “naïve image” simulation that includes no mixing (d). Respective correlation values are attached.

Fig. 12. A forecast for land-cover distribution in two imaginary future states: the “grey” scenario (a), the “green” scenario (b), and several extracted spectra (c) from either generated spectral imagery. Colors correspond to the legend in Fig. 4 (For interpretation of the references to colour inthis figure legend, the reader is referred to the web version of this article).

Fig. 13. Absolute frequency changes for selected land-cover classes.

Fig. 14. Quantitative added value products derived from two imaginary development scenarios. The central column indicates the current state, the left column expresses the green scenario, and the right column expresses the grey scenario. Lines of figures from top to bottom indicate reflectance (band 5), SAVI, albedo and absorbed short-wave energy (E_a), respectively. See full discussion in the text.

Fig. 15. Layout of some main features of the simulation software (within the ENVI 4.1 image visualization and analysis package).

Fig. 16. Workflow of processing steps and outputs for simulation of imagery.

Main outputs of two assessment tests for class performance

n is the number of pixels classified as class i, R² is the calculated correlation determination coefficient of that class with reference data, μ and σ are the mean and standard deviation of the GoF image, respectively.

Quantitative measures extracted from spectral properties of land-cover classes: overall summation (∑) and mean (μ)