Composition in false color of Formosat-2 image series of Auradé region, Gers, France.
The active and covering vegetation is represented in red. The areas that evoluve from blue/green to red
between June 5 and June 29 correspond to summer crops, mainly sunflower.
Conversely, the areas evolving towards the blue are winter crops, wheat and colza. The wheat harvest
causes a sudden increase of the reflectances which is visible on the image of June 29 (circled in white)
Multi-temporal data acquired by Formosat-2 in 2005 over the Southwest site have been used to get a detailled lan cover map. This map has been obtained by a supervised method (D. Ducrot et P. Gouaux, CESBIO), by using as input a database of more than 1000 terrain observations obtained in 2006 by CESBIO. One half of these observations was used to supervise the classifications, the other half to validate the results. The results obtained (table below), clearly show the interest of the multi-temporal information for the realisation of detailled classifications (distinction of the various crops). The number of poorly classified pixels drops from 34% when only one date is used (the best of the available dates, in March) to 1,3% when 7 dates are used.
|Number of dates||Dates used|
(after test of all possible combinations,
the best date combination is retained)
well classified pixels
|2||14/3 - 27/5||85.9|
|3||14/3 - 23/6 - 10/8||92.5|
|4||14/3 - 27/5 - 23/6 - 10/8||95.6|
|5||14/3 - 27/5 - 23/6 - 2/7 - 9/9||97.0|
|6||14/3 - 27/5 - 14/6 - 23/6 - 14/7 - 31/8||97.9|
|7||14/3 - 27/5 - 14/6 - 23/6 - 2/7 - 31/8 - 9/9||98.7|
By using the 7 dates, the following ground use map has finaly been obtained:
Result: D. Ducrot, P.Gouaux
The future work, at CESBIO, will address the realisation of classifications "in real time", which,for instance, will enable to answer the following questions:
- From which date is it possible to identify the winter crop plots?
- From which date is it possible to identify the corn plots?
In the same way, Formosat data enable to monitor: