Contemporary Trends in Land Surface Phenologies from the Aral Basin Suggest Weather- or Irrigation-Driven Changes in GPP not Land Cover Change

The transformation of the Aral Sea into the Aral Lakes is a startling example of anthropogenic environmental degradation. However, ongoing impacts of hydrologic modification and climatic change are occurring throughout the nearly 2 million km2 Aral Basin (AB), thus requiring synoptic, basin-wide monitoring of environmental trends. The MODIS 500m global land cover (LC) product is a potential source of such information in a data-sparse region like the AB. However, we find that it is particularly unstable in this arid/semi-arid region. For example, 25% of AB pixels change LC class between 2001 and 2005 versions of the global LC using the IGBP LC scheme. Here we present an alternate analysis of environmental trends in the AB using MODIS data at the same spatial resolution, but incorporating temporal information embedded within contemporary land surface phenologies (LSPs), in contrast to temporally delimited LC. NDVI time series from 2001-2008 were derived from 500m MODIS NBAR data (16-d composites every 8 d; MCD43A4). Focusing on the period from 1 March to 1 October, each pixel time series was comprised of 27 NDVI values per year (or 216 composites in total). Trend analysis using the nonparametric Seasonal Kendall test revealed a number of spatially coherent hotspots of highly significant (p<0.01) positive and negative trends in NDVI dynamics over the 2001-2008 interval. Relative to the 2001 and 2005 LC classifications, positive trends tended to coincide with the following transitions between IGBP classes: open shrub → closed shrub; grassland → {open shrub, cropland}; {open shrub, cropland/natural vegetation mix} → cropland. These findings suggest that apparent LC change events might be false-positives caused by weather-driven increases in GPP. In turn, negative trends were largely coincident with putative LC change to grassland from mostly woody classes: {closed shrub, open shrub, woody savanna, savanna, snow/ice, barren} → grassland. Here, potential false-positives may reflect effects of regional drought on woody classes, declining mountain snowpack, or original misclassification of grassland as barren, i.e., we would expect declining NDVI in drought-affected grassland, with limited sensitivity to NDVI trends in the barren class. In the intensively irrigated subregion of Karakalpakstan, the MODIS LC product indicates an elevated rate of change from natural vegetation to cropland and vice versa. These transitions tend to coincide with highly significant (p<0.01) negative or positive NDVI trends. However, visual inspection of 2001 and 2005 Landsat imagery from Karakalpakstan did not indicate extensive LC change. Here we interpret apparent false-positives in the global LC product as arising from localized changes in irrigation or cropping practices. Overall, this work demonstrates the utility of temporally dense LSPs for large-area monitoring of environmental change, as opposed to occasional LC classifications which, in arid regions like the AB, may exhibit instability and sensitivity with respect to a region’s normal range of interannual spectral variability.