Limits to detectability of land degradation by trend analysis of vegetation index data

This paper demonstrates a simulation approach for testing the sensitivity of linear and non-parametric trend analysis methods applied to remotely sensed vegetation index data for the detection of land degradation. The intensity, rate and timing of reductions in seasonally-summed NDVI are systematically varied on sample data to simulate land degradation, after which the trend analysis was applied and its sensitivity evaluated. The study was based on a widely-used, 1km 2 AVHRR data set for a test area in southern Africa. The trends were the most negative and significant when the degradation was introduced rapidly (over a period of 2-5years) and in the middle of a 16-year time series. The seasonally-summed NDVI needs to be reduced by 30-40% before a significant negative linear slope or Kendall's correlation coefficient was apparent, given an underlying positive trend caused by rainfall. The seasonally-summed data were reordered to remove this underlying positive trend, before simulating degradation again. With no underlying positive trend present, degradation of 20% resulted in significant negative trends. Since areas widely agreed to be degraded show only 10-20% reductions compared to non-degraded areas, this raises doubts over the ability of trend analyses to detect degradation in a timely way in the presence of underling environmental trends. Residual Trends Analysis (RESTREND) was applied in an attempt to correct for variability and trends in rainfall. However, a simulated degradation intensity ≥20% caused the otherwise strong relationship between NDVI and rainfall to break down, making the RESTREND an unreliable indicator of land degradation. The results of such analyses will vary between different environments and need to be tested for sample areas across regions. Although the paper does not claim to solve the challenge of detecting land degradation amidst rainfall variability, it introduces a method of assessing the sensitivity of land degradation monitoring using remote sensing data. © 2012 Elsevier Inc.