Incorporating contextual factors in agri-environmental research – options for obfuscation to support wider data application

Geoprivacy protection is a significant concern when sharing agricultural data with geographical information. Open and transparent access to agricultural data is essential to optimise its use, as well as integration with other data sources, such as environmental and climate data to support agri-environmental research. However, the choice of obfuscation method is crucial for achieving maximum statistical accuracy and geoprivacy protection in shared agricultural data. The primary objective of this research is to evaluate the effectiveness of obfuscated data in solving real-world problems. Random forest and multilinear regression models were used to predict soil organic carbon (SOC) based on internal (soil-related) and external (climate) features under three scenarios. The model estimations of SOC using original data were compared to model estimations of SOC using two datasets derived from obfuscation techniques, the environmental similarity obfuscation method (ESOM) and Rand. ESOM is a novel cluster-oriented obfuscation method that considers the importance of the contribution of contextual factors (external/climate conditions) in agri-environmental research, which guarantees that obfuscated data are relocated to areas of the same external conditions (in this instance climate) as the original data. Rand is a simple, yet widely used, obfuscation method that transfers the original location to a random location inside an obfuscation area. The results demonstrate that ESOM SOC estimation is more accurate than that obtained using the Rand method, despite the low modelled impact of climate conditions on the SOC estimation. In addition to the importance of feature selection and prediction models in achieving high accuracy and enhancing model performance, the choice of obfuscation method is another crucial factor when the use of original data is restricted. Therefore, while integrating internal and external features can improve results, it is essential to carefully assess how these features impact accuracy when selecting an obfuscation method.