TRANSLATING BIG DATA MODEL OUTPUT TO INFORM POLICY AND DECISION MAKING
Tools and techniques to turn strings of numbers into logical information for technical experts and the community
M van der Sterren, M Griffith, S Manning, P Tate, J Dixon
Publication Date (Web): 2 May 2017
Increased computing power and improvements in technology and innovation have allowed the development of large complex numerical models, but how do we make sense of the terabytes of output generated?
Sydney Water was recently faced with this challenge following the completion of the Hawkesbury-Nepean River and South Creek model. With over 130 scenarios generating approximately 100 GB of output each, the sheer volume of output was quite overwhelming. However it is essential that the outcomes of the modelling are understood and clearly communicated to management, stakeholders and the community, so they can inform policy and management decisions. This led to the development of innovative statistical and visual analytical techniques.
Figure 1: Animation comparing two scenarios generated from the HN model - click here to view animation
One of the techniques used to analyse the model output is based on the Healthy Rivers Commission’s objective values. Performance is indicated by calculating the proportion of results for a particular constituent that fall within the objective value, expressed as a percentage. A second technique is to simply graph the average. When combined, these two analysis techniques provide an effective first pass big picture view of the output. This level of analysis may be adequate to address the objective, may trigger the development of new scenarios or it may prompt the need for more detailed analysis.
Analysing the model output in flow classes provides greater insight into the scenario results. The six flow categories used (very low, low, moderate, fresh, flood and extreme) were defined using the baseline scenario for a site or stretch of river. The modelled water quality was then analysed to determine the flow weighted total loads (presented as bar plots) or flow weighted concentrations (presented as box-whisker plots). Using flow divisions enables the identification of changes in flow regime and water quality as a result of the different management options.
In addition to separating the output using flow categories, the above techniques can be applied to ‘dry weather’, ‘wet weather’ or ‘all weather’ model conditions. Dry weather analysis is particularly important to Sydney Water so we understand potential impacts on water quality during extended dry weather, while land managers may be more concerned with wet weather when diffuse runoff dominates the receiving water quality.
While complex statistical analysis is essential to understand the outcomes from the modelling and to develop solutions, it is often unnecessarily technical for management or the wider community. Animations were found to be a useful technique to present scenario results to these audiences.
All the techniques described in this paper to analyse the Hawkesbury-Nepean model output were programmed in MATLAB.
A number of key learnings arose from the scenario development and analysis process:
This paper presents different techniques used to analyse the water quality model outputs from the Hawkesbury-Nepean River and South Creek model.
- Ensure the objective is clearly defined and all assumptions are understood BEFORE running the model
- Keep the initial analysis simple. Presenting the average constituent concentration in conjunction with a comparison to relevant guidelines or objectives is often sufficient
- Depending on the level of detail required, separate the load or concentration by flow classes. This provides an improved understanding of how increased flows with lower concentrations or decreased flows with higher concentrations, influence the water quality within the river
- Animations are an effective technique to communicate the overall findings to management and the community, however, they are not suitable for detailed analysis.
Click here to read the full paper
Please note: You need to login to access this member-only content.
Not a member? You can purchase the paper in our Online Document Library.