F1000Research article template


Points to address here:

  • Use of EO data

  • Challenges in using EO data

  • What we try to achieve in here

In recent years, there has been a heightened interest in monitoring inland water quality, as rapid changes in human activity have been linked to rising eutrophication levels and increased silting. Decision makers have also noted the importance of monitoring networks, and water monitoring networks are starting to be deployed by governments, their growth occurring in parallel to legislation on water quality. While these networks are a positive development, they are not ubiquitous, limited both in space and in time. Earth Observation (EO) data in the optical domain has proven useful (e.g. (Carpenter 1983)) in this endeavour, as surface reflectance varies as a function of water composition. In principle, EO data has the ability to frequently monitor inland waters even in situations where no in situ observational networks are in place. Moreover, since sensor such as Landsat have been in operation for over three decades, it opens the possibility of assessing long term dynamics in water composition.

A number of complications need to be addressed in order to use EO data effectively to monitor lakes, however. For one, data acquisition is contingent on orbital selection, as well as cloudiness. In regions where clouds are prevalent, our ability to monitor lakes is thus severely reduced. Additionally, the sensors capture information on the state of the land surface, but are also affected by the optical properties of the atmosphere. Atmospheric properties are due to scattering and absorption by gases and particles, effects such as aerosol concentration and type, ozone concentration and water vapour. Atmospheric correction techniques are therefore required to compensate this important contribution. Finally, the interpretation of the data is complicated by the limited spectral sampling, and thus requires of models. Some of these models are empirical, and thus rely on local calibration, which might be sparse or directly unavailable. Mechanistic models are complex to develop, but some exist. If they are simple enough to require few data inputs to model most of the signal, they can be inverted to infer the state of the water.

(Wang 2007)