Datasets produced during the EOAfrica "Sentinel 2 user-relevant water quality monitoring in small southern African water bodies" project

This repository contains the datasets produced during the "Sentinel 2 user-relevant water quality monitoring in small southern African water bodies (SWAM)" project. The project focused on water quality parameters (WQPs), specifically Chlorophyll a (Chl-a) concentration and Total Suspended Matter (TSM) concentration.  

File naming convention: {WQP}_{DatasetType}_{StartYear}_{EndYear}_{Lake}.{FileType}

• {WQP}:
  • CHL: Chlorophyll-a concentration in units mg m$^{-3}$
  • TSM: Total Suspended Matter concentration in units g m$^{-3}$
• {DatasetType}:
  • cube: netcdf cube in format (time, lat, lon)
  • timeseries: csv file containing columns for water body name (Lake_name), satellite overpass date and time (DateTime_UTC), date (Date), 10th percentile of WQP over the lake (e.g. CHL_10P), mean WQP concentration (e.g. TSM_mean), median WQP concentration (e.g. CHL_median), 90th percentile of WQP concentration over the lake (e.g. TSM_90P), number of valid satellite pixels remaining after application of atmospheric correction and WQP algorithm (N_valid). Data are only included where >5% of the total water body area contained valid pixels after atmospheric correction and WQP algorithm application. 
• {StartYear}:
  • 2015: the year of the first timestamp 
•  {EndYear}:
  • 2025: the year of the last timestamp
• {Lake}:
  • TW: Theewaterskloof Dam
  • VV: Voelvlei Dam
  • CW: Clanwilliam Dam
  • MV: Misverstand Dam
  • ZV: Zeekoevlei
  • RV: Rietvlei
• {FileType}:
  • csv: csv file
  • nc: netcdf4 file

All satellite data access and processing was performed on the EO Africa Innovation Lab, a cloud-based virtual research environment with co-located data and computing services. Access to Copernicus data is enabled through Data and Information Access Services (DIAS) technology, with direct data access to CreoDIAS. All workflows were produced in JupyterLab using python. The Earth Observation Data Access Gateway (EODAG) python library was used to search for the file location of all Sentinel 2 L1C data within CreoDIAS during the period from 1 August 2015 to 30 August 2025 for each water body. Each Sentinel 2 file was read, resampled to 20 m spatial resolution, spatially subset to the smallest region of interest encompassing the entire water body, atmospherically corrected and written to NetCDF. In the case of the C2-Nets, the IdePix S2-MSI processor (v9.0.2) was applied to the L1C data to provide additional data flagging and masking options. 

Total Suspended Matter (TSM) concentration products were produced using the Case 2 Regional CoastColour (C2RCC) atmopsheric correction processor (Brockmann et al., 2016) with the C2X-Complex neural network to produce the remote sensing reflectance product, and then deriving TSM using the Jiang et al (2023) algorithm. 

Chlorophyll-a (Chl-a) concentration products were produced using the ACOLITE-RAdCor atmospheric correction (Vanhellemont & Ruddick 2016; Castagna & Vanhellemont, 2025) to produce the remote sensing reflectance product, and then deriving Chl-a using the Mixture Density Network (MDN) from Pahlevan et al (2020). 

This work was supported by the EO Africa R&D Facility, a joint initiative of ESA and the African Union Commission, in the context of the bilateral research project 03_SWAM_ZA_GB_EOAC3 between the Council for Scientific and Industrial Research (South Africa) and the University of Stirling (United Kingdom of Great Britain and Northern Ireland) under ESA Contract No. 4000133905/21/I-EF