Solutions for Improving Model Simulation in the Virtual Watershed Platform

Loading...
Thumbnail Image

Authors

Painumkal, Jose T.

Issue Date

2017

Type

Thesis

Language

Keywords

accuracy , elastic server , machine learning , simulation , watershed , web

Research Projects

Organizational Units

Journal Issue

Alternative Title

Abstract

This thesis is a collage of the works implemented to enhance the modeling capabilities of the NSF EPSCoR-supported Western Consortium for Water Analysis, Visualization and Exploration (WC-WAVE) Virtual Watershed Project. The core components of this work are a web-based tool to conduct scenario-based studies with watershed models, a proposed server-usage optimization strategy to enable cost-effective deployment of model containers to reduce the waiting time of jobs in a cloud environment, and a web tool to minimize the prediction errors of computer simulated models using a generic machine learning based approach. The developed prototype application includes an elastic hybrid server cluster comprising owned and rented servers that can facilitate on-demand provisioning of the computing resources based on job arrivals and ensures reduced waiting time for the modeling jobs within the allocated budget amount. The prototype contains a dashboard to track the progress of model run jobs and a user feedback monitoring module to generate auto alerts during severe performance issues. The model-scenarios component in the application could help hydrologists in simulating user-defined model scenarios using the PRMS (Precipitation Runoff Management System) model. The tool facilitates the download of watershed datasets available in the Geographic Storage, Transformation and Retrieval Engine (GSToRE) and enables the insertion of model simulations to GSToRE. The proposed model accuracy component uses a generic machine learning approach to process the predictions made by a computer-simulated model and helps in improving the accuracy of the model by minimizing the prediction errors. The prototype system supports the processing of the model data set using four machine learning regression techniques and enables the fine tuning of the model predictions.

Description

Citation

Publisher

License

In Copyright(All Rights Reserved)

Journal

Volume

Issue

PubMed ID

DOI

ISSN

EISSN