Wikindx Resources

Journal Article: BibTeX citation key:  Philippon2009b
Philippon, N., Martiny, N. & Camberlin, P. (2009) Forecasting the vegetation photosynthetic activity over the Sahel: a Model Output Statistics approach. IN International Journal of Climatology, 29. 1463–1477.
Added by: Devic 2009-09-16 11:20:57    Last Edited by: Fanny Lefebvre 2010-10-25 18:19:09
Categories: General
Keywords: Vegetation
Creators: Camberlin, Martiny, Philippon
Collection: International Journal of Climatology
Bibliographies: Prior150410

Peer reviewed
Number of views:  835
Popularity index:  42.62%
Maturity index:  published

The predictability of the mean August-September photosynthetic activity of vegetation over the Sahel for the period 1982-2002 is explored through a Model Output Statistics approach using ECHAM4.5 retrospective forecasts. Given the poor ability of Atmospheric General Circulation Models (AGCMs) to correctly simulate rainfall over the Sahel, the stress is put on using atmospheric dynamics alone. The mean July-September predicted fields of zonal wind at 600 hPa, and humidity flux at 850 hPa, are selected because of their key role in the West African Monsoon system and their consistency in AGCMs. Coupled modes of NDVI/atmospheric dynamics are extracted using Canonical Correlation Analyses performed in leave-one-out cross-validation. The most relevant modes (using NCEP/DOE 2 or ECHAM4.5 atmospheric dynamics) associate enhanced greenness to a weakened African Easterly Jet displaced northward, and strengthened moisture fluxes from the Guinean Gulf. They are linked with increased rainfall over the Sahel and positive (negative) SST anomalies over the Mediterranean (the eastern equatorial Pacific).
Used as predictors in a Multiple Linear Regression (MLR) model, the third cross-validated canonical coefficient (CC) derived from ECHAM4.5 simulations added to the August-September NDVI value of the previous year enable to explain 30% of the variance of a Sahelian regional index with a 2-month lead time. Applied at an 8-km spatial resolution, the statistical model possesses a usable skill (>.5) for 24% of the pixels analysed. The NDVI of pixels covered by open grassland appears as the most predictable.
Last Edited by: Fanny Lefebvre