Development of a MODIS tree cover validation data set for Western Province, Zambia

One of the annual land cover products to be made from Moderate Resolution Imaging Spectroradiometer (MODIS) data is the vegetation continuous fields layers. Of these fields, one is a global percent tree cover map. Using field measurements, IKONOS data, Enhanced Thematic Mapper Plus (ETM+) data, and ancillary map sources, a tree cover map was made and validated for two WRS path/rows in Western Province, Zambia. This map will be used in validating the 500-m global MODIS tree cover product. The map was made at the 30-m Enhanced Thematic Mapper Plus (ETM+) resolution and also scaled up to 250- and 500-m resolutions. Five IKONOS images were classified into crown cover/no crown cover maps at 4-m resolution. These maps were aggregated to 30 m to create a continuous training data set of percent crown cover. Three dates of ETM+ data were acquired to predict percent crown cover using a regression tree algorithm. Comparisons of training accuracies and field data to ETM+ tree estimates yielded root mean square errors (rinse) of similar to +/- 10% crown cover. When aggregating the 30-m map to 250- and 500-m MODIS cell sizes, the training errors are more than halved. The final 250-m map was assessed using a structural vegetation map of the area and an overall rmse of 8.5% is estimated. The 250-m map was sampled and used to derive a tree cover continuous field product using 3 Level 113 MODIS time slices, approximating the acquisitions of the ETM+ data. The results are promising as an overall root mean square error between the ETM+ derived tree crown cover map and an aggregated MODIS 500-m map was 5.2%. Results from this test in Zambia show that the MODIS 250-m bands should allow for improved depictions of percent tree cover. (C) 2002 Elsevier Science Inc. All rights reserved.