Embracing uncertainty for better science and better decision making
by Fridolin Haag
ZMT
As humans we dislike uncertainty in many situations. Not knowing can make us vulnerable. The tendency to avoid uncertainty also appears in human practices such as science or decision making. However, as I will illustrate, knowing the extent of our ignorance can help us take better action. Science, specifically predictions made by science, can be viewed as one important input to decision making, for example, around conservation or management of coastal and marine areas. These predictions come with large uncertainties – though this might be invisible when we focus on point estimates. As I show with a simulation study, large prediction uncertainties do not necessarily prevent the identification of good management options, but their consideration can increase our trust in the robustness of the decision. When uncertainties in our models are indeed overwhelming, addressing specifically those uncertainties that impact a decision allows us to focus costly data collection efforts. A systematic approach that I present facilitates prioritizing the relevant sources of uncertainty. This value of information analysis can be understood as a form of probabilistic sensitivity analysis. Such analysis requires an explicit formulation of the parametric uncertainties in our prediction models with probability distributions that essentially quantify our ignorance. Obtaining such distributions can be a challenge. However, in a Bayesian framework for parameter inference they are a direct result. Thus, I briefly discuss Bayesian approaches of fitting models to data and their (dis-)advantages. By providing examples of how it can be tackled, I hope to spark discussion on how we can find comfort with uncertainty in scientific work. With the perspective on decision support as the intended endpoint of scientific inquiry, I would also like to contribute to the discourse on”user-centered” science, as discussed in transdisciplinarity research.
Drones and SfM-MVS techniques applied to coastal environments: What are the scales of observable processes and errors associated with reconstructed 3d models?
by Elisa Casella (Co-authors are listed in the presentation since two different works are presented)
ZMT | University of Bremen
Drones, coupled with Structure-from-Motion and Multi-View-Stereo methods (computer-vision science) have enhanced the capability of observing earth processes. Consumer-grade drones have been on the market for less than eight years. The large availability of this technology led, in relatively few years, to an exponential growth of the number of scientific papers where drones have been employed to study the environment. Coastal areas present challenging conditions for the use of drones, due to the nature of the environment. At the same time, coastal environments are prone to rapid changes due to the interaction between land and marine environments (e.g. beaches) and can be difficult to survey (e.g. shallow-water reefs) with traditional techniques. Drones and SfM-MVS methods proved to be essential in providing new insights into the study of the coastal environment, allowing for more accurate and faster monitoring. This work is divided into two parts. First, we present a review of about 50 papers using drones-SfM-MVS for beach surveys. We show that thanks to drones-SfM-MVS methods, it is possible to map beach environments with a resolution of few centimeters, and with a vertical error of 19 to 0.5 cm. Multi-temporal studies show the ability of the method in catching seasonal volumetric changes. The second part of this work presents one of the first 3d-reconstructions of shallow coral-reefs environment using drones-SfM-MVS. The refraction of light passing between the two different media (air,water) represents the main source of error in the reconstruction of the 3d of the seafloor using aerial images. The results of these studies show that: i) for land coastal environments, drones-SfM-MVS can provide higher-resolution information, with an accuracy comparable with that of the more precise traditional survey techniques (e.g. DGPS); ii) for shallow underwater environments (e.g. coral reefs), it is possible to gather high-resolution maps, but the vertical accuracy of the reconstructed environment presents limitations to study small-scale changes.
Music in your data: Hydroacoustic observations and the art of noise removal
by Tim Dudeck
ZMT
Signal-to-noise ratio is a common term in data analysis. But what is noise? For example, hydroacoustic observation systems record a lot of real and “unreal” noise in their raw data output. These can be of biological, physical or electrical origin. Just like in any field data analysis, removing this noise is a critical part of hydroacoustic data analysis. In this study, I show different types of noise and how to deal with them. How do we define noise, what is variance and when does noise become a signal? What are the risks of excessive noise removal or too less noise removal? Moreover, there is often a thin line between unbiased noise removal and data manipulation. Examples will be shown from the modern EK80 hydroacoustic system and how it was possible to observe the deep-scattering layer in the Benguela Upwelling System through noise removal. The ideas and methods behind the examples can be applied to other data and in regards to ever growing datasets, are meant to raise awareness of the risks and benefits of noise.
Poster
Tackling spatial heterogeneity of groundwater discharge between scales
by Nils Moosdorf and Till Oehler
ZMT | ZMT
Submarine groundwater discharge (SGD) occurs along most coastlines of the world. It can be a major source of nutrients to coastal environments. In order to estimate the impact of those nutrient fluxes, its important to budget them at the regional scale (e.g. for a single bay). Yet, observations of nutrient fluxes are local. Since spatial and temporal variability of SGD is huge, we have not yet managed to meaningfully extrapolate local results to the regional scale.
At the point scale (meters), direct observations of SGD from seepage meters are available. At the local scale, detailed tracer studies representing subsurface heterogeneity based on geophysics are possible. But at the regional scale, the resolution of observation methods becomes too small to reasonably depict the exact heterogeneity.
SGD is extremely sensitive to heterogeneity, and we do not have the capabilities to sufficiently document it at the regional scale. Thus, a method is needed to represent this uncertain heterogeneity at the regional scale based on local scale samples. Here, we will present the problem in order to initiate discussion with ZMT colleagues about possible solutions and ways forward toward regional scale estimates of SGD and its associated solute fluxes and ecosystem impacts.