A Nonparametric Bayesian Model to Adjust for Monitoring Bias with an Application to Identifying Environments Stressed by Climate Change
Jonathan Auerbach, Theresa M. Crimmins, David Kepplinger, Ruishan Lin, E.M. Wolkovich · 2025
Read A Nonparametric Bayesian Model to Adjust for Monitoring Bias with an Application to Identifying Environments Stressed by Climate Change by Jonathan Auerbach, Theresa M. Crimmins, David Kepplinger, Ruishan Lin, E.M. Wolkovich (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-04924.txt.
Introduction
Jonathan Auerbach and co-authors fix a bias in field timing data in a 2025 paper. Volunteers who watch plants report bloom or leaf states on visit days, but the true event may fall between visits or be misread. Plain methods then place the event date off target, which warps studies of plant timing and climate.
Their Bayesian model uses smooth rising curves to infer the true event time and correct both error types. Tests on fake data show the fix works. Applied to lilac blooms watched in the northeast United States, it moved peak bloom far later than plain methods said. Some sites also stood out as truly late once the bias was stripped away.
Catalog record
| CREATOR | Jonathan Auerbach, Theresa M. Crimmins, David Kepplinger, Ruishan Lin, E.M. Wolkovich |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-04924
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Nonparametric Bayesian Model to Adjust for Monitoring Bias with an Application to Identifying Environments Stressed by Climate Change Jonathan Auerbach, Theresa M. Crimmins, David Kepplinger, Ruishan Lin, E.M. Wolkovich Abstract We propose a new method to adjust for the bias that occurs when an individual monitors a location and reports the status of an event. For example, a monitor may visit a plant each week and report whether the plant is in flower or not. The goal is to estimate the time the event occurred at that location. The problem is that popular estimators often incur bias both because the event may not coincide with the arrival of the monitor and because the monitor may report the status in error. To correct for this bias, we propose a nonparametric Bayesian model that uses monotonic splines to estimate the event time. We first demonstrate the problem and our proposed solution using simulated data. We then apply our method to a real-world example from phenology in which lilac are monitored by citizen scientists in the northeastern United States, and the timing of the flowering is used to study anthropogenic warming. Our analysis suggests that common methods fail to account for monitoring bias and underestimate the peak bloom date of the lilac by 48 days on average. In addition, after adjusting for monitoring bias, several locations had anomalously late bloom dates that did not appear anomalous before adjustment. Our findings underscore the importance of accounting for monitoring bias in event-time estimation. By applying our nonparametric Bayesian model with monotonic splines, we provide a more accurate approach to estimating bloom dates, revealing previously undetected anomalies and improving the reliability of citizen science data for environmental monitoring. Keywords nonparametric Bayes \cdot monotonic splines \cdot monitoring bias \cdot bias correction \cdot crowdsourcing, phenology \cdot climate change \cdot citizen science 1 Introduction We consider the problem of estimating the time of an event which cannot be observed directly. Instead, individual observers make multiple visits to the location where the event is to take place and record whether the event has already occurred or not at each time of their arrival. The records are then used to estimate the event time. We are motivated by a real-world problem in phenology, the field which studies the timing of seasonal life cycle events in plants and animals ( Forrest and Miller-Rushing, 2010 ) . For recent applications, see ( Dennis et al., 2024 ) and ( Schwob et al., 2023 ) . Citizen scientists monitor the status of flowering (flowers open or not open) of lilac in the northeastern United States, and anomalously early or late bloom dates are used as evidence of environmental stress caused by climate change. Typically, bloom dates are considered early or late relative to historic records before major anthropogenic warming began in 1980 ( Stoker et al., 2013 ) . The problem with these comparisons is that changes in how citizen scientists monitor a site can make the site appear more anomalous than it actually is. We focus on the peak bloom date, which is the first date when more than 50% of the flowers have bloomed for a plant in a location. We consider two sources of variation. The first is that the citizen scientists may not arrive on the day that the lilac bloom. The second is that the citizen scientists may report the flowering status of the lilac in error. We show that when these sources of variation are not properly accounted for, substantial bias can result. We present our work in five sections. In Section 2 , we review how the field of phenology relies on "status monitoring" to identify environments stressed by climate change. The purpose of this section is to further motivate the monitoring bias problem and our proposed solution. However, the problem is not unique to phenology. Correcting for monitoring bias is a general problem in the agricultural, biological, and environmental sciences. In Section 3 , we state our assumptions and provide a demonstration of monitoring bias. In particular, we show that the size of the bias depends on the frequency with which the monitor visits the site. In Section 4 , we describe our solution. We propose a nonparametric Bayesian model that jointly models citizen science data across multiple sites. In Section 5 , we apply the proposed model to a real-world monitoring problem in which citizen scientists monitored the status of common lilacs ( Syringa vulgaris ) in 2022 at 25 sites throughout the northeast U.S. and reported the observed status of open flowers to the USA National Phenology Network. We find that failing to correct for monitoring bias underestimates the peak bloom date of the lilac by 48 days on average. We also find that, after adjusting for monitoring bias, 2 of the 25 sites had anomalously late bloom dates. These sites did not look atypical before adjusting for monitoring bias, and the proposed model allows us to quantify how certain we are that these sites are in fact anomalous. We conclude in Section 6 with a discussion. 2 Background Climate change impacts environments in complicated ways that are difficult to observe directly. Scientists have increasingly turned to data collected by citizen scientists to better understand these impacts. A key source of data comes from the volunteers who monitor the seasonal activity of plants and animals around the world ( Battle et al., 2022 ; Menzel et al., 2020 ) . A major strength of citizen science data is the quantity of observations that can be amassed. Citizen science data provide considerably more information than what can be collected by paid scientists alone. In the United States, seasonal activity in plants and animals is documented by volunteers and professional scientists at thousands of locations through the USA National Phenology Network Nature’s Notebook platform ( Rosemartin et al., 2014 ) . Since the program’s launch in 2009, over 40 million records have been contributed, and these data have been used in hundreds of scientific analyses and to support decision-making in a wide range of applications ( Crimmins and Crimmins, 2022 ) . As statistical methods for analyzing phenological data become more critical to science and policy, understanding and correcting their biases is increasingly important. We show that current methods used to determine the timing of seasonal onsets are sensitive to a bias that we refer to as monitoring bias. The bias occurs because observers use “status monitoring” to document what they see: each time they visit a site, they report whether they see plants or animals expressing a series of life cycle stages called phenophases ( Denny et al., 2014 ) . For example, a monitor might record whether they observe a particular plant has open flowers on the date they make the observation. They do not record the date that the plant bloomed. Common estimates of the bloom date, such as the first day the monitor observes a bloom, are biased both because the monitor rarely arrives on the actual bloom date but also because the monitor may report the bloom date in error. Indeed, it is not uncommon for a monitor to report the status of a plant as bloomed on one visit