Detecting change in landscape greenness over large areas: An example for New Mexico, USA

作者：Nash Maliha S^*; Bradford David F; Wickham James D; Wade Timothy G

来源：Remote Sensing of Environment, 2014, 150: 152-162.

DOI：10.1016/j.rse.2014.04.023

• Landscape
• NDVI change
• New Mexico
• Local factors
• Climate factors
• Autoregression model

Monitoring and quantifying changes in vegetation cover over large areas using remote sensing can potentially detect broad-scale, slow changes (e.g., climate change over decades), as well as more local and rapid changes (e.g., fire, land development over weeks and years). A widely used indicator for detecting change in land cover is a measure of greenness, the Normalized Difference Vegetation Index (NDVI), derived from the Advanced Very High Resolution Radiometer (AVHRR). Detecting change in the NDVI, however, can be confounded by time-dependent patterns (e.g., seasonal effects) and variation associated with climate factors. In the present study we provide a method to address these confounding factors by evaluating the NDVI change using autoregression techniques that compare results from univariate (i.e., the NDVI vs. time) and multivariate analyses (the NDVI vs. time and climate variables) for similar to 314,000 1-km(2) pixels comprising the state of New Mexico over an 18-year period (1989-2006).
The ability to detect NDVI trend was greatly improved by including climate variables in the multivariate analyses of the NDVI over time. Specifically, the fraction of pixels with a significant NDVI trend (mostly increasing) doubled from 5.2% of the pixels for the univariate autoregression analyses to 11.9% for the multivariate autoregression analyses. The comparisons of univariate and multivariate analyses also revealed that for most of the pixels with a significant NDVI trend in either analysis, the trend was consistent with changes in local factors rather than to broad-scale, slow changes (e.g., climate change); only 0.8% of the pixels had a significant NDVI trend associated with change in the climate variables. This latter finding is somewhat surprising given that several climate variables changed significantly over much of the state during the 18-year period, and the NDVI was significantly related to these variables in the multivariate autoregressions for much of the area. Close examination of several areas suggested that NDVI change in these areas was attributable to wildfires, agriculture, habitat restoration, and tree mortality associated with insect infestation. Published by Elsevier Inc.

• 出版日期2014-7