Estimation of tree cover using MODIS data at global, continental and regional/local scales

Comparisons of MODIS inputs appropriate to mapping land cover at different scales are made using global training data and a SAFARI 2000 validation database from western Zambia. Multiple single-date images, 40-day composites and multitemporal annual metrics from the MODIS sensor are tested in mapping percent tree cover. While the metrics outperform the composites at the global scale and are comparable to composites at the continental scale, composites outperform the metrics at the local/regional scale of the Zambia test area. Multiple single-date MODIS imagery are best at mapping the test area, and this points to their utility in mapping at the local/ regional scale. However, the overall difference between inputs is less than 1% in terms of standard error. This implies that the metrics and composites, with appropriate training data, can come close to replicating the spatial detail present in single-date images. Comparing the Zambia test area data with a subset of the global MODIS percent tree cover map in a validation exercise shows that the general tree cover distribution is well represented. However, the global signatures underestimate Kalahari woodlands on sands and do poorly in overestimating tree cover in some dambos and pans. Further work will aim at refining the global signal using multiple validation sites.