Multi-Decade Vegetation Productivity Decline in Northern Wisconsin and Minnesota Forests Driven by Increased Rates of Disturbance.

摘要

Disturbance events that alter forest carbon (C) stock are complex and challenging to capture in large areas over long periods of time. Historical remote sensing data coupled with available ancillary data may provide critical spatial and temporal information to quantify C implications of these events in ecosystem modeling. Methods to identify, extract, and extrapolate the nature, duration, and spatial extents of disturbance events that alter the C cycle are needed. Advanced Very High Resolution Radiometer (AVHRR) normalized difference vegetation index (NDVI) version 'g' declines from the Global Inventory Modeling and Mapping Studies (GIMMS) group have been observed from 1982 - 2007 in the northern mixed Laurentian forest of Wisconsin, and Minnesota. We set out to determine their cause and quantify the impact to C by identify trends in net primary productivity. We developed an algorithm to extract different long-term disturbance dynamics, from Landsat using a combination of proven spectral indices including NDVI, tasseled cap, and disturbance index, and evaluated the results. Multiple insect outbreaks have occurred over the study period, and were dominated by forest tent caterpillar defoliation. Forest harvest is a predominant disturbance agent in the region. Both have reduced forest net primary productivity (NPP) by > 20 gCm2/mo in disturbed sites. We found spectral indices used in our classification showed sensitivity to different disturbances through evaluation of reported events from forest inventory assessments, and aerial photography with an overall accuracy ranging from 56 - 82% for insects and logging respectfully. We found the successive increase in forest disturbance rates that have reduced AVHRR NDVI at the coarse scale impacted 21.8% of the forested area (7,349 sq. km) over our study period. Disturbance rates have exceeded any AVHRR recovery signal that is observable at Landsat resolution. We provide a semi-automated approach to quantify 30 m disturbance dynamics into regional C models driven by AVHRR that would otherwise be ignored.