Digital elevation model uncertainty and hazard analysis using a geophysical flow model

Abstract

This paper describes a new methodology to quantify the variation in the output of a computational fluid dynamics model for block and ash flows, when the digital elevation model (DEM) of the terrain and other inputs are given as a range of possible values with a prescribed uncertainty. Integrating these variations in the possible flows as a function of input uncertainties provides well-defined hazard probabilities at specific locations, i.e. a hazard map. Earlier work provided a methodology for assessing hazards based on variations in flow initiation and friction parameters. This paper extends this approach to include the effect of terrain error and uncertainty. The results are based on potential flows at Mammoth Mountain, CA, and Galeras Volcano, Colombia. The analysis establishes the soundness of the approach and the effect of including the uncertainty in DEMs in the construction of probabilistic hazard maps.

Integrating data and compute-intensive workflows for uncertainty quantification in large-scale simulation: application to model-based hazard analysis

“…A number of fundamental changes in the design of hazard map generation were introduced and were largely possible through the use of the Netezza server. With Netezza's data mining capability, tessellation-based method [5,20] was completely replaced with a more reliable distance-based search for neighbours. This not only leads to a more precise control of the region Downloaded by [North West University] at 07:08 20 December 2014 of neighbourhood but also makes it more feasible to extend the problem to higher dimensions.…”

“…The localization and tessellation raises concern of overfitting which have been carefully addressed in earlier work [5] using a fit generated as above and tested with several hundred pieces of data outside those used in the fit. The cross-validations tests in [5,20] include comparing the BLM surface from 2048 simulations to an exhaustive evaluation of the surface with 10 4 simulations and prove that this is not the case here. While, computationally efficient, the tessellation approach is not directly related to the emulator quality, we propose here an alternate methodology based on a simple distance metric:…”

“…θ n is the vector of hyperparameters or roughness parameters and Ndim is the number of dimensions associated with the data set. The emulator incorporates an expensive nonlinear optimization technique to iteratively find the optimal values of the roughness parameters θ [5,20].…”

“…We focus in this work on the computational challenges arising from recent works [5,20] which provide a technique for generating hazard maps for regions with volcanic activity, based on uncertainty quantification methodology. In previous works [5,20], we proposed methodology based on localization and parallelization to make the problem tractable. However, the remaining challenges in data handling and computing are still too difficult for standard computational platforms like the large-scale cluster-based hardware used in most large computing work.…”

Large data and computation in a hazard map workflow using Hadoop and Neteeza architectures

“…We present here an approach designed to surmount issues arising from recent work [3,14] that integrates the outcomes of large full field flow simulations into a probabilistic hazard map -often required in near real time. Early versions of related work involving only the data warehouse appliance will appear in [12] A hazard map, as stated here, is a predictive map for a region which provides a probabilistic measure of a hazard (e.g.…”