A comparison of obfuscation methods used for privacy protection: Exploring the challenges of polygon data in agricultural research

Abstract

A major challenge of sharing spatially explicit agricultural and agri‐environmental data is to identify the trade‐off between field parcel confidentiality and spatial pattern preservation. In this work, 27 point‐based obfuscation and evaluation methods were applied on agricultural data, collected by the Irish Nutrient Management Planning Online (NMP Online) platform, which is a high‐density polygon dataset developed to inform precision agriculture through nutrient management based on soil fertility and agronomic targets. Broad categorizations of methods—including N*Rand, Donut, Density, Pinwheel, AHilb, and k‐anonymity—were developed, combined, and modified to achieve the best trade‐off between security and accuracy. To improve geoprivacy and spatial pattern preservation of existing Donut and Density methods, qualitative approaches, including Donut‐k and Density‐k methods, were introduced which identify the optimal values of radii based on a combination of the Donut method and k‐anonymity satisfaction, and subsequently optimal k‐anonymity determination. Modified AHilb and Donut‐AHilb methods were also developed to generate smaller and arbitrary obfuscation areas to improve location security. The Donut‐AHilb method was found to be the best at spatial pattern preservation and satisfying larger k‐anonymity, but the risk of false identification and non‐unique obfuscation was high when considering the polygon nature of agricultural data such as field parcels. Further, we introduce the term “non‐unique obfuscated points,” which is important when obfuscating static objects as two or more points might have the same obfuscated location, which has relevance to the wider GIScience community.

Geoprivacy Protection of Agricultural Data

“…Several anonymization and randomization techniques on the centroid of field parcels were developed and tested in. 5 The performance of 27 point-based obfuscation methods were evaluated on the subset of the Irish Nutrient Management Planning Online (NMP Online) dataset with high point density and non-uniform distribution. Furthermore, several evaluation methods were tested to measure the ability of each method to satisfy both privacy concerns and spatial pattern preservation.…”

Implementation of novel polygon‐based obfuscation methods to improve privacy of agricultural data

“…Donut-k and Density-k are point-based qualitative techniques that were designed to maximize the geoprivacy protection and spatial pattern preservation by generating the smallest obfuscation area as far as possible (Nowbakht et al, 2022). It is important to bear in mind that when the spatial data are represented as polygons then the k-anonymity satisfaction means that the obfuscation area contains at least k − 1 other polygons such that the original polygon is unidentifiable among them.…”

“…The optimal value of k1 and k2 can be considered when the growth of radii are significantly large. This process was suggested by Nowbakht et al (2022) to provide higher levels of geoprivacy and spatial pattern preservation for point-based objects.…”

“…To explore the performance of point-based obfuscation methods on polygon nature data, the centroid of a polygon can be considered as point data. Recent research by Nowbakht et al (2022) developed and tested several anonymization and randomization techniques on the centroid of field parcels. Results highlighted a high percentage of false-identification and non-unique obfuscation, indicating the drawback of point-based obfuscation methods when applied to geographic features best represented as polygons.…”

“…Results highlighted a high percentage of false-identification and non-unique obfuscation, indicating the drawback of point-based obfuscation methods when applied to geographic features best represented as polygons. Seidl et al (2018) addressed the risk of false-identification for point features which occurs when the obfuscated point mistakenly links to another original point, while Nowbakht et al (2022) generalized false-identification for polygon features when the obfuscated point was inside the area of another original polygon, and more specifically when there was an intersection between the original and obfuscated polygons. While this study identified the importance of false-identification and non-unique obfuscation, as well as the main drawbacks of using point-based obfuscation methods on polygon nature objects (i.e., occurrence of non-unique obfuscation and false-identification) using polygon-based obfuscation methods in their own right still remains unexplored.…”