Plomoo

int. j. remote sensing, 2001 , vol. 22, no. 13, 2633–2639

Monitoring land-cover changes in West Africa with SPOT Vegetation: impact of natural disasters in 1998–1999
F. LUPO, I. REGINSTER and E. F. LAMBIN
Department of Geography, Universite catholique de Louvain, 3, Place Louis ´ Pasteur, B-1348 Louvain-la-Neuve, Belgium; e-mail: lupo@geog.ucl.ac.be (Received 20 October 2000; in nal form 22December 2000) Abstract. A multi-temporal change-vector algorithm has been applied to SPOT Vegetation data of West Africa over two growing seasons. Change maps displaying the impact on surface attributes of natural disasters such as oods and droughts were produced, based on diVerent vegetation indices and compositing periods. The accuracy assessed on the basis of independent data on naturaldisasters, was high (76% correct for the NDVI-based map). This land-cover change product provides important information to evaluate the spatial and temporal distribution of rapid changes in land surface attributes.

1.

Introduction The impacts of global change on land surface attributes are not uniformly distributed geographically. Assessing the regions aVected by rapid land-cover changes andnatural disasters is a priority for global change research and is a vital input for policies aimed at mitigating the impact of these changes. By May 2000, two full years of consistent global Earth observation data at a 1 km spatial resolution were available from the SPOT Vegetation ( VGT) sensor. Natural disasters modify surface moisture and/or vegetation activity, which impacts on the re ectedradiances measured by the satellite sensor. The objective of this study was to monitor over West Africa the impact on land surface attributes of natural disasters and land-cover change ‘hot spots’. 2. Data 2.1. Study area The study area extends from 0° to 18° N in latitude and ­ 18° W to 18° E in longitude (i.e. from the coastal zones of the Gulf of Guinea to the Sahel in West Africa). The rainy seasonoccurs from June to November. From south to north, the vegetation types are equatorial evergreen forests, open forests dominated by deciduous and semi-deciduous species with an herbaceous stratum, tree to shrub savannahs, steppes and semi-desertic vegetation. 2.2. Remote sensing data We used the ten-day synthesis product (VGT-S10) which corresponds to the highest value of top-of-atmospher eNormalized DiVerence Vegetation Index (NDVI )
Internationa l Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online © 2001 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/01431160110053202

2634

F. L upo et al.

for every pixel during a ten-day period. The radiance data were atmospherically corrected with the Simpli ed Method for the Atmospheric Correction (SMAC)procedure. The VGT-S10 products were delivered in the Plate carree 1 km map ´ projection. The data cover two growing seasons (June to November 1998 and 1999 ). Over this short period, one expects to detect mostly the in uence of interannual climatic variability and natural hazards. An additional cloud screening was necessary as some clouds remained, particularly along the Gulf of Guinea. The cloudmask provided with the data was expanded by a neighbourhood analysis. Firstly, all pixels in a 5×5 moving window were added to the cloud mask when one pixel of the window was a cloud. Then, a 10×10 majority lter was applied. Thirdly, the cloud mask was widened to include the pixels with a high change magnitude (see below) which were neighbours of cloudy pixels. 2.3. Reference data A database onnatural disasters was assembled to develop an independent estimate of change. Decadal rainfall data were extracted from the Africa Data Dissemination Service (WWW1). Daily data on res detected from NOAA AVHRR were taken from the World Fire Web (WWW2). Information on the major natural disasters in the region was assembled from FAO’s Global Information and Early Warning System (WWW3), the Famine...