Agricultural Land Cover Dynamics on the Ganges-Brahmaputra Delta: 1988-2014

Abstract:

We seek to understand spatiotemporal (ST) patterns of agricultural land cover dynamics on the lower Ganges-Brahmaputra Delta (GBD). Recent availability of accurately coregistered, radiometrically intercalibrated Landsat TM, ETM+ and OLI imagery collected since 1988 allows for synoptic scale ST analyses of vegetation phenology. We use multitemporal spectral mixture analysis of exoatmospheric reflectance to represent land cover and water bodies as continuous fields of soil and sediment substrates (S), vegetation (V), and dark surfaces (D; water & shadow). This study analyses 61 cloud-free Landsat acquisitions across two geographic scenes to identify ST patterns of winter cropping and interconversion between agricultural fields and ponds used for aquaculture. We also use MODIS 16-day EVI composite time series post-2000 and high spatial resolution imagery to extend and vicariously validate the Landsat-derived observations. We use temporal moment spaces (derived from temporal mean, standard deviation, and skewness) and temporal feature spaces (derived from spatial Principal Components) to characterize the full range of phenological patterns observed at 30 m scales throughout the lower delta. For each year with sufficient cloud-free coverage, we distinguish between areas with a high likelihood of use for aquaculture versus areas with a high likelihood of use for agriculture based on a combination of reflectance and phenology. From changes in these patterns we infer changes in land use on seasonal to interannual timescales. Many of the phenological patterns we observe occur on the scale of individual polders, suggesting decision making at community scales. While there appears to be considerable loss of agricultural land to aquaculture in many areas of the lower delta, we also observe intensification of dry season cropping in other areas. MODIS reveals frequent instances of both gradual and abrupt decreases in seasonal peak EVI as well as many localized instances of abrupt cessation of agriculture. In terms of area and magnitude, the single largest disturbance we observe immediately follows Cyclone Aila in 2009. We observe a band of phenological disruption on the polders along almost 400 km of the periphery of the Sundarbans.