Intercomparison and sensitivity assessment of lake primary production models for remote sensing

Lakes are highly biodiverse ecosystems and are providing a wide range of ecosystem services to human wellbeing such as drinking water, water for irrigation, access to fisheries and recreational areas. Anthropogenic activities threaten these services both through local impacts on water bodies (e.g. eutrophication) and globally (e.g. climate change). The trophic state and the aquatic carbon cycle are sensitive indicators to evaluate the state and health of lake ecosystems. Monitoring the spatial and temporal dynamics of primary production is therefore a high priority in lake research.
Primary production can be assessed in several ways. The most common approach involves the incubation and measurement of labelled carbon isotopes in lake water samples that are exposed to certain light conditions. Alternatively, primary production can be estimated using diel variations in oxygen concentration or fast repetition rate fluorometry. Both approaches are accurate but can hardly be used to cover large spatial heterogeneities. For global assessments, only bio-optical primary production models based on remote sensing data allow a consistent upscaling in a cost-efficient manner.
A wide range of bio-optical primary production models exist and have been applied to lakes. Generally, these models describe the availability of light in the water column and the efficiency of the algae particles to absorb photon energy and to use this energy for subsequent carbon assimilation. The main challenges related to such approaches are to accurately the retrieve required information from satellite data and to precisely estimate sensible model parameters. The upcoming hyperspectral mission Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) by NASA will help to improve the accuracy of primary productivity estimates.
This contribution aims to improve understanding of sensitivities and validity of available bio-optical primary production models to eventually maximise the benefits of improved information retrievals from PACE. We particularly reviewed state-of-the-art primary production models for remote sensing data of oceans and lakes, provided an overview of relevant model inputs and calculated Sobol sensitivity indices for a range of input parameters and models. Our results facilitate future applications of primary production models to hyperspectral PACE data and will particularly help to identify most sensitive input variables, to improve empirical model parameterizations and to guide the selection of suited models for freshwater systems.