Inclusion of forest harvest legacies, forest type, and regeneration spatial patterns in updated forest maps: A comparison of mapping results

In Maine and other heavily forested states, existing land cover maps quickly become dated due to forest harvesting and land use conversion; therefore, these maps may not adequately reflect landscape properties and patterns relevant to current resource management and ecosystem studies. By updating an older land cover product (the 1993 Maine GAP map) using Landsat imagery and established forest change detection techniques, we demonstrate a practical and accurate means of providing contemporary, spatially explicit forest cover data needed to quantify landscape change. For a 1.8 million hectares study area in northern Maine, we quantify the accuracy of forest harvest classes and compare mapped harvest and regeneration area between the 2004 GAP update product and the 2004 Maine Landcover Dataset (MeLCD), a map recently developed in coordination with the 2001 National Land-Cover Database (NLCD). For the period 1995–2004, the overall harvest/non-harvest accuracy of the GAP update map is 87.5%, compared to 62.1% for the MeLCD. Producer and user accuracy for harvest detection is 92.4% and 89.7%, respectively for the GAP update, and 48.8% and 92.5% for the MeLCD. Mapped harvest area differs considerably, reflecting a systematic under-representation of recent harvest activity on the part of the MeLCD. By integrating older land cover data, the GAP update retains the forest disturbance legacies of the late 1970s through the early 1990s while simultaneously depicting 2004 forest composition for harvested and regenerating stands. In contrast, the MeLCD (and 2001 NLCD) over-represents the area and connectivity of older forest (undisturbed since the late 1970s), and provides no forest composition information for mapped forest regeneration. Systematic misclassification of forest age classes and harvest history has serious implications for studies focused on wildlife habitat modeling, forest inventory, and biomass or carbon stock estimation. We recommend the integration of older land cover data and time-series forest change detection for retention of harvest or disturbance classes when creating new forest and land cover maps.