Estimation of suspended sediment concentration using satellite remote sensing data in Lake Tana, Ethiopia

Soil erosion by water from upland areas results in on-site and offsite problems in the catchment and aquatic ecosystems such as lakes and rivers. As such, up-to-date information on erosion rates and suspended sediment fluxes in rivers is indispensable to manage the impact of soil erosion, sediment transport, and sedimentation. However, detailed spatial and temporal information on erosion and sediment dynamics is rare, as it is labor-intensive and costly to obtain, particularly in developing countries. The advent of satellite remote sensing applications has provided the opportunity to monitor sediment fluxes by assessing the suspended sediment concentration (SSC) of rivers and lakes. This approach may provide a cost-effective alternative to ground-based sampling schemes. However, satellite-based approaches to monitor sediment fluxes require calibration and validation as the relation between SSC and optical properties of the water recorded by satellite sensors may vary with changing sediment properties. 

Here, we used empirical models to estimate SSC values from optical sentinel 2 data for Lake Tana in Ethiopia using in situ collected water samples. Moreover, in situ reflectance data, which were measured using an ASD Field Spec 4 spectroradiometer instrument, from water samples collected at Lake Tana are used as well. SSC and in situ reflectance measurements were conducted for 546 water samples collected from the lake, particularly from the river plumes of the two most important rivers draining to Lake Tana, i.e. Gumara and Giligel Abay.  The sample SSC values ranged from 1.50 mg/L to 4,146 mg/L. The samples were classified into two categories:  low SSC (≤ 250 mg/L) and high SSC (> 250 mg/L), as the optical properties of water are significantly influenced by its constituents. The individual bands in the NIR and visible spectrum exhibited a good correlation (R² = 0.73, RMSE = 30.69 mg/L) for low SSC-values over Lake Tana. Moreover, the multilinear regression (MLR) analysis using both the visible and NIR bands of low SSC-conditions improved results compared to using individual bands (R² = 0.84, RMSE = 23.37 mg/L). In contrast, high SSC water samples from Lake Tana did not correlate well with individual bands. However, combining the NIR and red bands generally improved the estimation of SSC for high SSC values (R² = 0.9, RMSE = 0.26 mg/L).  

The established relations between optical properties and field-based SSC values will be applied to long-term timeseries of optical data to assess the temporal variations in sediment concentration in rivers draining to Lake Tana, and in Lake Tana itself. These timeseries will be compared to optical data on vegetation changes in the catchment to identify hot spots in both space and time that are responsible for elevated fluxes of sediment to Lake Tana.