Day 4

Detailed paper information

Back to list

Paper title Optical classification of inland waters using remote sensing: a synthesis of current research and future opportunities
Authors
  1. Emma Tebbs King's College London Speaker
  2. Aiyesha Capati King's College London
  3. Aidan Byrne King's College London
  4. Anujan Ganeshalingam
Form of presentation Poster
Topics
  • A7. Hydrology and Water Cycle
    • A7.06 EO for monitoring water quality and ecological status in inland waters
Abstract text Remote sensing can provide valuable information for monitoring the ecological status of inland waters. However, due to the optical complexity of lakes and rivers, quantifying water quality parameters is challenging. One approach is to use remotely-sensed reflectance to classify inland waters into discrete classes – or Optical Water Types - that correspond to different ecological states. These optical classes can then be used either to inform the selection of the most appropriate water quality retrieval algorithms or as valuable ecological indicators in their own right.

This review aimed to understand how remote sensing has been used to classify the ecological status of inland waters and which classification approaches are most effective, as well as identifying research gaps and future research opportunities. Using a systematic mapping methodology, a search of three large literature databases was conducted. The search identified an initial 174 articles, published between January 1976 and July 2021, which was reduced to 64 after screening for relevance.

Very few papers were published before 2008 but since then publications increased substantially. The number of waterbodies included in the studies ranged from one to more than 1000, with the vast majority of studies including five or fewer waterbodies. There was a geographical bias towards Europe, the US and China, with poor representation across Africa and the rest of Asia. The source of spectral data used for training the classifications was overwhelmingly from satellite data or in situ measurements, with relatively few using data from aircraft or UAVs. The most common satellite sensors used were the Landsat series, MERIS, MODIS, Sentinel-2 MSI and Sentinel-3 OLCI.

The classification frameworks used were primarily based on Optical Water Types or Trophic State Index, but many studies adopted their own bespoke classification schemes. The number of classes varied from 2 to 21, peaking at 3 classes. A variety of classification algorithms were utilised including unsupervised clustering, supervised (parametric and machine learning) methods, and thresholding of spectral indices. Most studies related the optical classes to in situ water parameters, particularly Chlorophyll-a, Total Suspended Solids and Coloured Dissolved Organic Matter. A variety of pre-processing steps were applied prior to classification including normalisation of spectral data and dimensionality reduction techniques such as Principal Component Analysis.

In this presentation, we summarise the strengths and limitations of different sensors, pre-processing methods and classification algorithms for optical classification of inland waters. Our results highlight important gaps, such as the geographical bias in studies and training data. We emphasize the need for greater transparency and sensitivity analysis to understand how decisions made about the choice of sensor, classification algorithm and pre-processing steps influence to resulting optical classes. Recommendations for future research are presented, including the need for standardized approaches to support transferability of methods and scaling up from local to global scales.