Analysis of geospatial behaviour of visitors of urban gardens: is positioning via smartphones a valid solution?

TL;DR

This study evaluates the accuracy of smartphone GNSS data for analyzing visitor behavior in urban gardens, demonstrating that it can be a reliable and automated method for spatial behavior analysis.

Contribution

The paper introduces a framework for automating the collection and analysis of smartphone GNSS data to study spatial behavior in urban gardens, validating its accuracy and consistency.

Findings

Horizontal accuracy below ~2.3 m at 95% confidence

No significant difference between device types or survey methods

Strong autocorrelation of residuals confirms data reliability

Abstract

Tracking locations is practical and speditive with smartphones, as they are omnipresent devices, relatively cheap, and have the necessary sensors for positioning and networking integrated in the same box. Nowadays recent models have GNSS antennas capable of receiving multiple constellations. In the proposed work we test the hypothesis that GNSS positions directly recorded by smartphones can be a valid solution for spatial analysis of people's behaviour in an urban garden. Particular behaviours can be linked to therapeutic spots that promote health and well-being of visitors. Three parts are reported: (i) assessment of the accuracy of the positions relative to a reference track, (ii) implementation of a framework for automating transmission and processing of the location information, (iii) analysis of preferred spots via spatial analytics. Different devices were used to survey at different times and with different methods, i.e. in the pocket of the owner or on a rigid frame. Accuracy was estimated using distance of each located point to the reference track, and precision was estimated with static multiple measures. A chat-bot through the Telegram application was implemented to allow users to send their data to a centralized computing environment thus automating the spatial analysis. Results report a horizontal accuracy below ~2.3 m at 95% confidence level, without significant difference between surveys, and very little differences between devices. GNSS-only and assisted navigation with telephone cells also did not show significant difference. Autocorrelation of the residuals over time and space showed strong consistency of the residuals, thus proving a valid solution for spatial analysis of walking behaviour.