16.05.2014 Views

PBL rapport 550026002 Calibration and validation of the land use ...

PBL rapport 550026002 Calibration and validation of the land use ...

PBL rapport 550026002 Calibration and validation of the land use ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Simulated 1993 l<strong>and</strong> <strong>use</strong> using discrete model Figure 4.1<br />

Observed 1993 l<strong>and</strong> <strong>use</strong> (left) <strong>and</strong> its simulation with <strong>the</strong> discrete model, using 1993 l<strong>and</strong> <strong>use</strong> as input (A) <strong>and</strong> <strong>the</strong><br />

multinomial regression specification (B) as input for <strong>the</strong> suitability maps.<br />

4.2<br />

Continuous model<br />

4.2.1 1993 l<strong>and</strong> <strong>use</strong> as input<br />

To <strong>use</strong> <strong>the</strong> 1993 l<strong>and</strong> <strong>use</strong> as input for <strong>the</strong> continuous model,<br />

<strong>the</strong> per-cell suitability value <strong>of</strong> each l<strong>and</strong>-<strong>use</strong> type is specified,<br />

as <strong>the</strong> natural logarithm (ln) <strong>of</strong> <strong>the</strong> area claimed by that l<strong>and</strong><br />

<strong>use</strong> in that particular cell. This definition corresponds with<br />

<strong>the</strong> original algorithm described in Section 2.1. The approach<br />

leads to an exact representation <strong>of</strong> <strong>the</strong> 1993 l<strong>and</strong> <strong>use</strong> (Table<br />

4.1), showing that <strong>the</strong> continuous model, just as <strong>the</strong> discrete<br />

model, is working properly. The map resulting from <strong>the</strong>se<br />

simulations (Figure 4.2A) also shows this correspondence.<br />

Please note that two different types <strong>of</strong> comparison are <strong>use</strong>d<br />

for <strong>the</strong> continuous model. The first type compares maps <strong>of</strong><br />

dominant l<strong>and</strong>-<strong>use</strong>, for both observed <strong>and</strong> simulated l<strong>and</strong><br />

<strong>use</strong>. This approach is identical to <strong>the</strong> one <strong>use</strong>d for <strong>the</strong> discrete<br />

model, but has <strong>the</strong> disadvantage <strong>of</strong> over- or underestimating<br />

<strong>the</strong> presence <strong>of</strong> certain l<strong>and</strong>-<strong>use</strong> types. The second comparison,<br />

<strong>the</strong>refore, relates to <strong>the</strong> actual quantities (or ratios) <strong>of</strong><br />

l<strong>and</strong> <strong>use</strong>, per cell (See <strong>the</strong> map comparison description in<br />

Section 3, methodology).<br />

4.2.2 Multinomial regression specification<br />

The second calibration attempt <strong>use</strong>s <strong>the</strong> results from <strong>the</strong><br />

multinomial regression analysis presented in Section 4.1.2. A<br />

comparison with <strong>the</strong> observed l<strong>and</strong>-<strong>use</strong> quantities (Table 4.1)<br />

shows that <strong>the</strong> continuous model performs reasonably well,<br />

with a correspondence <strong>of</strong> 58% to 97%. This result is remarkable,<br />

considering that <strong>the</strong> statistical analyses were performed<br />

on <strong>the</strong> discrete data sets, ra<strong>the</strong>r than <strong>the</strong> continuous data set<br />

28<br />

<strong>Calibration</strong> <strong>and</strong> <strong>validation</strong> <strong>of</strong> <strong>the</strong> L<strong>and</strong> Use Scanner allocation algorithms

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!