Modeling Remote Sensing Data Uncertainty for Decision Analytics: A Case Study for Flood Management

Remote sensing (RS) data, increasingly used for information about floods, wildfires, and other severe environmental conditions, embodies uncertainties that can impact decisions that rely on that data. Understanding the uncertainties in RS data and its implications for decisions is particularly salient for local (municipality or county-level) decision-makers who rely on information generated by models that may lack accuracy at smaller spatial scales. In the environmental monitoring and decision-making cases, there are two key types of uncertainties: epistemic uncertainties that exist due to imperfect information about prevailing environmental conditions, and aleatoric uncertainties in future environmental conditions and events. The epistemic uncertainties are linked to technical sensing capabilities of RS systems, data acquisition coverage and frequency, and quality of data processing, and thereby can be reduced with targeted technology improvements and strategic investments. Furthermore, these epistemic uncertainties can be modeled, and should be rigorously included in decision analytics that rely on RS data. However, in many cases, the uncertainties in RS data products are either not fully known, or disclosed, and often not incorporated in evaluating decisions and their economic implications. Here, we develop an approach to show how uncertainties in RS data can be first modeled and then fused within decision evaluations. We employ a framework, grounded in the value of sample information concept, that explicitly links RS data quality to actionable decisions. We investigate the case of flood management in regions near wetlands and examine how radiometric and altimetric uncertainties in RS data may affect decisions that rely on accurate understanding of wetland water storage capacity. We evaluate the combined effects of uncertainties in digital elevation models (DEM) and satellite imagery on both short-term decisions (e.g., emergency evacuations due to predicted inundation) and long-term planning (e.g., infrastructure development). The framework is demonstrated with a case study of wetland regions and shows how local decision makers can evaluate what value they may obtain from using imperfect RS data for informing decisions that affect public safety and community resilience.