Using cell phone data to monitor population mobility and tourism recovery

Analytics on population movement has become increasingly important during the COVID-19 pandemic. In response, Data Ventures partnered with Vodafone and Spark to produce aggregated device counts at an hourly level for small geographies. We then developed a novel weighting methodology to create population estimates, broken down based on whether they are local to an area, domestic visitors, or international visitors. Our daily data provided insight into high mobility areas during lockdown. We now continue to use the data to inform the post COVID-19 recovery of international and domestic tourism.

For more information, see dataventures.nz and reports.dataventures.nz.

Te Matatini o te Horapa - a contagion network model for Aotearoa NZ

Contagion models for infectious disease can only be as good as the assumptions that they are built on. One common simplifying assumption is that the disease spreads through a well-mixed, homogeneous population (or populations). This has some obvious consequences for modeling the differential impact of disease on different groups. I will introduce an individual level, network based, contagion model, built from a range of data sources, including census data, and will show how it has been used for modelling the spread of CVOID-19 in Aotearoa NZ.

Constructing an individual level interaction network for modelling COVID-19 in Aotearoa NZ

Illnesses and deaths from the COVID-19 pandemic have not been evenly distributed across communities. Risk factors for serious disease (including age, ethnicity, & health conditions), and risk factors for infection (including dwelling size, inter-generational living, & type of work) are not independent. In order to capture the interactions between these factors we build on data from linked StatisticsNZ Integrated Data Infrastructure (IDI) to construct a network of the ~5million individuals and their interaction contexts, in particular homes, workplaces, and schools. I will highlight our progress so far, and priorities for future refinements.

Integrating Contact Patterns into Simple Models of Disease Spread

Simple models of disease spread often assume homogeneous mixing. While these are easy to construct and quick to solve, they ignore important detail. Contact matrices are an easy way of including heterogeneities in these simple models without introducing much complexity but require robust data. I will highlight why these contact matrices are important for our COVID-19 response, outline the best data we have, discuss what we are missing, and give examples of what other countries are doing in this space.

Participatory surveillance of influenza and COVID-19 symptoms in New Zealand

FluTracking is an online participatory surveillance system that has been used for collecting data on influenza seasons in Australia and New Zealand, wherein participants fill out a weekly survey on whether they have experienced symptoms. This has several advantages over sentinel-based data collection, allowing for near real-time reporting of symptoms and mitigating some bias. This study analyses New Zealand FluTracking data from April 2020 to April 2021, examining the impact of various factors on the incidence of flu- and COVID-like symptoms. I will discuss the various ways bias is introduced into these data, the methods employed to mitigate bias, and the potential of systems like FluTracking for public health surveillance.

Uncertainty Quantification for Complex Network Contagion Simulation Models

As the complexity of a model increases, so does the uncertainty inherent in the model’s output. I consider Uncertainty Quantification (UQ) techniques to assess the reliability of a complex network contagion model we have developed for informing Aotearoa New Zealand’s COVID-19 response. By fitting Gaussian process surrogate models to output from the network contagion model, we can quickly and efficiently apply UQ methods and perform inference on model parameters. Using New Zealand COVID-19 case number data, we can also condition these surrogates to make predictions without having to rerun the computationally expensive network contagion model.

Providing customised health intelligence in real-time – COVID-19 dashboards

Real-time health intelligence is essential to an effective public health response. The NZ COVID-19 dashboards were built with this need in mind drawing on ESR’s EpiSurv database for notifiable diseases. Once published, the dashboards and underlying data streams had to evolve as the response progressed e.g. to differentiate between cases occurring in the community and MIQ or to provide insight into identified outbreaks. Using a R Shiny framework allowed us to dynamically adjust the dashboards to response requirements, which was critical. The COVID-19 dashboards were built for the Ministry of Health in collaboration with ESR. Public dashboard available at: https://nzcoviddashboard.esr.cri.nz/.

Constructing a spatial index of public transport supply across Auckland, Canterbury and Greater Wellington.

In this study we constructed a low-level spatial index of public transport (PT) supply across the 25,000 meshblocks in the Auckland, Canterbury and Greater Wellington regions. Using Open Route planner and Google Transit Feed data, for each meshblock in the study area we calculated the walking distance to the nearest PT location, the number of accessible locations (per capita), the number of accessible routes and (peak-time) services and the transit time from place of residence to place of work via PT networks. These metrics were then combined into an overall PT Supply Score. The results of this study are visualised in an interactive Power BI dashboard.

Project Monty: an agent based model of the New Zealand transport system

The Ministry of Transport is developing an agent based model of the NZ transport system. The model builds a representation of the entire transport system as the sum of millions of different interactions as everyone seeks to achieve their own sets of goals and ambitions. We aim to offer better insights into complex systems by aggregating the interactions of individual agents (people) as they attempt to complete activities. Through their interactions with a transport network and each other, we get a more realistic representation of how individual travel choices impact the system.

NZ motor vehicle VKT (vehicle kilometres travelled) and fleet size estimation

Information on the number of motor vehicles in use in NZ and how much they are driven is of great interest to the public and stakeholders. The process for using Waka Kotahi NZTA’s MVR (motor vehicle register) data to estimate these. How, in R, the data is cleaned, processed and projected forward to allow for a real-time estimate of the current figures for both. More information available from Kain Glensor (k.glensor@transport.govt.nz)