Spatio-temporal inventory of floodplain wetlands at basin scale using wetness index time series derived from Landsat datasets

Abstract:

Floodplain wetlands are known to perform various hydrological functions such as groundwater recharge, flood attenuation, baseflow sustenance. However, these wetlands are under extreme anthropogenic stress and are rapidly being converted into agricultural and residential areas. Wetlands are in many ways a proxy for hydrological wellbeing of the basins for which a comprehensive spatio-temporal inventory of wetlands is a prerequisite but labor and computationally intensive. Here, we present an algorithm using MNDWI (Modified Normalized Difference Water Index) derived from multi-temporal Landsat series satellite dataset to map the surface waterbodies in floodplains and to evaluate their temporal stability (e.g. persistent, lost, newly formed) at basin scale. The algorithm uses basic arithmetic in a GIS framework and the entire workflow can be automated in any GIS environment. The algorithm has been implemented on the floodplains of the Ramganga Basin (25,000 km²) in the West Ganga Plains, India for the post-monsoon data of the period 1994-2019 at annual scale.

Our results show that there were 158 km² of area covered by persistent surface waterbodies in 1990s, which has reduced to 130 km² in recent years (2014-2019). Further, a total of 104 km² of area covered by such waterbodies have been lost during this period, and new waterbodies covering 76 km² of area have formed. There are 62 km² of area which remained wet for 90-10% of the observation period, 75 km² remained wet for 75-90% time and 322 km² remained wet for 50-75% of time. Therefore, a total of 459 km² of the area in the basin remained wet for at least 50% of the period of observation which spanned nearly 3 decades. However, most of these contributions come from 1990s and early 2000s, and in recent times, the basin is exhibiting a drying trend losing several water bodies. The algorithm developed in this work is intuitive, yet robust and easily implementable at any spatio-temporal scale using freely available datasets such as the Landsat series.