A Bayesian model for quantifying errors in citizen science data: application to rainfall observations from Nepal

Abstract

High-quality citizen science data can be instrumental in advancing science toward new discoveries and a deeper understanding of under-observed phenomena. However, the error structure of citizen scientist (CS) data must be well-defined. Within a citizen science program, the errors in submitted observations vary, and their occurrence may depend on CS-specific characteristics. This study develops a graphical Bayesian inference model of error types in CS data. The model assumes that (1) each CS observation is subject to a specific error type, each with its own bias and noise, and (2) an observation's error type depends on the static error community of the CS, which in turn relates to characteristics of the CS submitting the observation. Given a set of CS observations and corresponding ground-truth values, the model can be calibrated for a specific application, yielding (i) number of error types and error communities, (ii) bias and noise for each error type, (iii) error distribution of each error community, and (iv) the single error community to which each CS belongs. The model, applied to Nepal CS rainfall observations, identifies five error types and sorts CSs into four static, model-inferred communities. In the case study, 73 % of CSs submitted data with errors in fewer than 5 % of their observations. The remaining CSs submitted data with unit, meniscus, unknown, and outlier errors. A CS's assigned community, coupled with model-inferred error probabilities, can identify observations that require verification and provides an opportunity for targeted re-training of CSs based on mistake tendencies.