Date of Award

Fall 1996

Project Type


Program or Major

Earth Sciences

Degree Name

Doctor of Philosophy

First Advisor

Robert C Harriss


The need for accurate estimates of forest cover and forest fragmentation is a critical issue for developing countries such as Costa Rica. Accurate estimates of forest cover can help in several sectors related to the environment and economic development. This dissertation focuses on providing an accurate and precise estimate of forest cover in Costa Rica. The year 1991 was used as a baseline. Landsat Thematic Mapper was the remote sensing sensor used in this analysis. This dissertation concludes that: (1) Twenty-nine percent ($\sim$1,400,000 ha) of the country was under primary forest (80% canopy closure) in 1991. Of the total forest cover, 71% is outside national parks and 29% is protected by the national parks. (2) Forest loss (for scene path 15/row 53) during five years period (1986-1991) was 224,970 ha, and it was estimated that the rate was $\sim$44,994 ha/yr. (3) Deforestation produced an increase in island fragments during the study period. Between 1986 and 1991, the total number of islands between three and 50 ha, and 50 and 100 ha increased by 524 and 45, respectively. Fifteen new islands with areas greater than 500 ha were created. (4) Results suggest that the extent of tropical deforestation go beyond estimations of total forest loss at the national level. The impacts at the national level have greater roots deeper roots when the data at the life zone level is considered. The results have important implication for biodiversity conservation and restoration, water resource management and climate change.

The issue of partial sampling of remote sensing data base was also explored through this dissertation. Partial sampling is important for the definition of sound deforestation monitoring systems in tropical environments. A data set from the Brazilian Amazon was analyzed in order to understand how stratified sampling, using persistence, would improve estimates of tropical deforestation over random sampling. Results show that stratification based on persistence contributes to the reduction of error, regarding estimates of total deforestation, when contrasted against random sampling without stratification (FAO methodology). Results are important to future monitoring programs in Costa Rica and the Central American region.