A high temporal resolution NDVI time series to monitor drought events in the Horn of Africa

Abstract

This study investigates the reconstruction of climatological patterns and vegetation dynamics in the Horn of Africa region using high temporal resolution (i.e. daily) Normalized Difference Vegetation Index (NDVI) datasets. The analysis compares a straight-forward processing approach to derive a daily vegetation index from a geostationary (SEVIRI) satellite with existing NDVI series from geostationary or polar-orbiting (MODIS, MetOp-AVHRR) satellites, highlighting the impact of cloud contamination on data quality in high temporal resolution datasets. Using a smoothing process designed to reconstruct the upper envelope of the vegetation status series, we obtained a daily vegetation dataset that effectively mitigated cloud-induced fluctuations, outperforming polar-orbiting (e.g. MODIS) satellite-derived dataset in capturing regional climatology. We demonstrated this through statistical analysis, including autocorrelation and mean absolute difference between consecutive observations. We showed that cloud contamination significantly affects high temporal resolution NDVI series, particularly in forest areas, which makes it difficult to identify a suitable dataset to validate our approach. Therefore, we mitigated this problem using a Maximum Value Compositing technique, designed to remove cloud-induced biases and further compared our results with another independent vegetation index at coarser temporal resolution derived from AVHRR. We found that our vegetation index closely relates with MODIS 10-day composites after removing cloud-contaminated pixels. Furthermore, the study evaluates the sensitivity of the selected NDVI datasets to drought events, demonstrating the strength of the proposed SEVIRI dataset in capturing the intensity and persistence of vegetation anomalies. In conclusion, the study presents an innovative strategy for deriving daily-resolution NDVI datasets in cloud-prone regions, validating it with independent datasets at different sub-monthly temporal scales.