Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images

VEGETATION - 2000

Lake Maggiore - Italy, 3-6 April 2000


Estimation of surface variables at the sub-pixel level for use as input to climate and hydrological models

Jean-Pierre Fortin*,Monique Bernier , Ali El Battay, Yves Gauthier and Richard Turcotte
*INRS-Eau, 2800 rue Einstein, C.P. 7500, Sainte-Foy (Québec) G1V 4C7 Canada
Tel.: (418)-654-2591; Fax: (418)- 654-2600; e-mail: jpf@inrs-eau.uquebec.ca

Paper (pdf file, 238 k)

For hydrological simulation and forecasting, estimation of the spatio-temporal variability of watershed variables like albedo, soil moisture and snow cover is very important. The needed information cannot come uniquely from gound survey data, as the number of sample sites required and the frequenxy of measurements would be prohibitive. Remote sensing can help to find out a solution.

As a new medium resolution but high frequency sensor, having spectral bands in the visible and near infrared,was to be on board the SPOT-4 satellite, together with an enhanced high resolution sensor, we have proposed an investigation having as objectives (a)the estimation at the sub-pixel level of physical variables of the surface, namely the reflectances of each land cover within the pixel, (b) the estimation at the sub-pixel level of the spatial distribution of snow cover and, finally, (c)as accurate as possible registration of the images for multitemporal input into a spatially distributed hydrological model using geocoded data.

During the pre-launch phase of the investigation, we have developed the methodology, using simulated VEGETATION and HRVIR data from TM data. Our results were very encouraging. Using a methodology based on the theory of spectral mixture, we were able to estimate from VGT pixels the reflectances of the broad land use classes present on a summer image with a very good accuracy, the estimations being well within one standard deviation from the "true" values estimated from corresponding simulated HRVIR pixels. Also, we have defined two snow indices allowing estimation of snow cover at the sub-pixel level with a very good accuracy, more than 70% of the estimations being within 10% of the snow cover value estimated from HRVIR pixels. Finally, we were able to obtain registrations of VGT pixels within less than 100m from the simulated "true" location.

For the post-launch phase, we were able to have access to 15 VGT images, from January to June of 1999, as well as to one HRVIR image and one panchromatic image from SPOT-2, both during the snowmelt period. The idea was to verify the methodology with actual VGT and HRVIR data.

In this communication, after recalling the results obtained in the pre-launch phase of the investigation, we will discuss the results obtained with the actual data. In short, in 1999, the snowmelt period in the selected region did not coincide very much with the available high resolution data necessary to test our methodology. We were however able to map the snow cover and estimate the reflectances of land covers present in the pixels. Also, the VGT images used were localised within the expected accuracy with respect to each other and with the HRVIR image.