Modelling approaches
River DOM processing is influenced by multiple dynamic drivers that
often respond non-linearly to hydro-climatological events across
catchments, such as floods, drought, and warming (Battin et al., 2023).
In the past, modelling of DOM, carbon and nutrient reaction and
transport through river networks was hindered by the lack of
high-resolution hydrology and hydrography data products at watershed,
national, and global scales. As a result, models were typically limited
to specific water body types (e.g. lakes only, river reaches/segments
only) or grouped catchments where output could not be discretized in
such a way to allow for spatiotemporal trends to be identified. The
intersection of sensor technology, river models and ML advances presents
new opportunities for aquatic scientists and managers to develop digital
representations of river systems (aka digital twins) to enhance aquatic
science and management.
Increasing volumes of sensor data have enabled the expansion of
metabolism estimation from the river reach scale to the network scale
(Figure 4) using a range of model methods, including process-based
(Segatto et al., 2020), empirical (RodrÃguez-Castillo et al., 2019), ML
(Segatto et al., 2021), or a combination (Pathak et al., 2022, Maavara
et al., 2023). As sensor networks can gather data on the physical and
chemical properties of rivers, such as temperature, light intensity,
dissolved oxygen and nutrient concentrations, these data are usually
used as input in process-based metabolism models to estimate reach-scale
processes (Demars et al., 2015, Appling et al., 2018). Local metabolism
rates can then be combined with information about the catchment
environment to upscale to the river network scale, and as inputs for ML
algorithms such as decision trees or neural networks.