Vegetation Mapping of Africa Using Machine Learning

The accurate monitoring and evaluation of vegetation phenology is of principal importance for understanding and unravelling vegetation dynamics, climate change impacts, and sustainable resource management. Here we present an on-going investigation into the vegetation phenology of Africa, leveraging unsupervised machine learning techniques. The main objective is to discern patterns, trends, and variations in vegetation dynamics across diversely different regions of the continent. The Global Inventory Modeling and Mapping Studies (GIMMS) Normalized Difference Vegetation Index (NDVI) dataset is used as a measure of vegetation phenology. We apply unsupervised machine learning algorithms, specifically KMeans, Hierarchical Clustering (HC), and Partition Around Medoids (PAM) clustering to partition the extensive range of vegetative conditions into coherent clusters or zones. Utilizing the elbow method, we found the optimal number of clusters to be eight (8). To evaluate the unsupervised results, a comparison was carried-out against the traditional Koppen-Geiger (KG) classification using the V-measure, yielding scores that range between 0.1 and 0.2. This level of alignment underscores the efficacy of the employed clustering techniques in capturing the inherent details of vegetation patterns. In addition, spatial-temporal variability is thoroughly investigated. In particular, the modes of spatial-temporal variability are investigated through Empirical Orthogonal Function (EOF) analysis, which aids in elucidating dominant spatial patterns and their temporal evolution. The first two modes explain about 65% of the total variance, with the first mode being associated with seasonality of the intertropical convergence zone. This strategic partitioning of Africa's vegetative conditions offers insights into regions showing similar temporal variability, laying the background for elaborate comparative analyses and trend assessments we are going to perform. The fusion of advanced data analysis methodologies with comprehensive spatial and temporal assessments leads to an enhanced understanding of vegetation dynamics, facilitates the identification of hotspots for conservation purposes, and informs policy formulations regarding sustainable land management practices in the face of changing climates.