Validation of the Terra-I Forest Loss Detection Products in Vietnam Using Landsat-8 Imagery: A Case Study of Dien Bien and Lam Dong Provinces

Date of Award


Degree Type


Degree Name

Master of Science in Geography



First Advisor

John Rogan

Second Advisor

Robert G. Pontius, Jr.

Third Advisor

Louis Reymondin


Over the past three decades, Vietnam has witnessed an estimated 2% of annual forest cover loss largely owning to timber harvest, cash crop production, and dam construction. These rapid deforestation and forest degradation processes necessitate frequent reporting of forest conditions to assist timely interventions. Terra-i is a near real-time monitoring system designed to detect tropical forest loss at 16-day intervals using MODIS and TRMM satellite imagery. Initially designed for the Colombian and Peruvian Amazon, Terra-i has not been evaluated in the tropical moist forested landscapes such as Vietnam. This study incorporated multitemporal 30 m Landsat-8 and the Carnegie Landsat Analysis System (CLASlite) to validate the 250 m Terra-i forest loss products in two provinces of Vietnam: Dien Bien (DB) in 2014, and Lam Dong (LD) in 2015. Results showed a greater amount of forest loss detected by CLASlite (5,961.9 ha in DB and 14,776.9 ha in LD) compared to Terra-i (1,864 ha in DB and 3,370.6 ha in LD). Accuracy assessment of Terra-i revealed high commission errors (94.7% in DB, 95.3% in LD) and high omission errors (98.3% in DB, 98.9% in LD) across all forest types and patch sizes. Bamboo forests and secondary successional broadleaf forests experienced the highest proportions of total forest loss, irrespective of the detection product used. Compared to CLASlite, Terra-i detected no large (>4.32 ha) forest loss patches and detected only 0.4% of small (≤0.72 ha) forest loss patches that accounted for 94% of total Landsat-detected forest loss events in both study areas. The Terra-i products did not provide reliable detection of forest loss events every 16 days in terms of quantity and location likely due to the prevalence of small-scale, human-induced forest disturbances in Vietnam. Additionally, the completeness of both forest loss products was greatly affected by the high cloud cover during the local wet seasons that limited capture of clear imagery for both Landsat and MODIS sensors. Forest change monitoring in Vietnam should be conducted using multiple satellite datasets and automated approached to overcome multiple issues regarding forest type, forest disturbances levels, and image acquisition conditions.