Land-cover change in the Caucasus Mountains since 1987 based on the topographic correction of multi-temporal Landsat composites

Download Buchner_etal_RSE_2020_LC_change_Caucasus

Mountainous regions are changing rapidly across the world due to both land-use change and climate change.
Given the importance of mountainous regions for ecosystem services and endemic biodiversity, monitoring these
changes is essential. Satellite data provide a great resource to map land-cover change in mountainous regions,
however mapping is especially challenging there because topographic complexity affects reflectance. The socalled
‘topographic effect’ has been successfully corrected for in case studies of small areas, but a comparison of
large-area classifications and land-cover change analyses with and without topographic correction is missing.
Here, we performed a long-term land-cover change assessment for a large mountainous region, i.e., the Caucasus
Mountains with topographic correction. Our two goals were 1) to examine the effect of topographic correction
on land-cover classification for a large mountainous region, and 2) to assess land-cover changes since 1987
across the Caucasus based on the full Landsat archive. Both the complex topography and the history of land-use
changes, especially after the collapse of the Soviet Union in 1991, make the Caucasus Mountains an ideal study
area to understand topographic effects on large-area land-cover mapping for the last three decades. First, we
compared a non-topographically-corrected Landsat classification for 2015 with a classification that was topographically-
corrected with an enhanced C-correction for the same year and assessed the accuracy of both.
Second, we derived topographically-corrected Landsat classifications for six dates to assess changes in cropland
and forest from 1987 to 2015, based on class probabilities and post-classification comparisons. In regard to our
first goal, topographic correction improved the overall accuracy of the classification only by 2% (from 79 to
81%), but disagreement rates were as high as 100% in mountainous regions, especially among forest types. In
regard to our second goal, we found that cropland loss was the most prevalent change process since 1987.
Cropland loss was particularly widespread in Georgia and Armenia until 2000, and in Azerbaijan until 2005. The
North Caucasus (the Russian Federation) had more stable cropland over time, most likely due to different land
reforms after the collapse of the Soviet Union, and the prevalence of flat landscapes and very fertile soils, which
make cultivation easier than in the South Caucasus. Rates of forest change throughout the Caucasus Mountains
were surprisingly low, with forest loss and forest gain being roughly equal. Forest loss was most likely related to
both illegal logging and natural disturbance, whereas forest gain was most likely due to cropland abandonment.