Automated Phytosensing: Ozone Exposure Classification Based on Plant Electrical Signals

Till Aust, Eduard Buss, Felix Mohr, Heiko Hamann · 2024

Read Automated Phytosensing: Ozone Exposure Classification Based on Plant Electrical Signals by Till Aust, Eduard Buss, Felix Mohr, Heiko Hamann (2024) 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-2412-13312.txt.

Introduction

Plants respond to air pollution with faint electric signals. Till Aust, Eduard Buss, Felix Mohr, and Heiko Hamann want to read those signals automatically. Their WatchPlant project imagines living plants as a sensor network for city air. In the lab they exposed ivy plants to ozone and recorded the signals. A software toolchain extracts generic features from each signal trace. Automated machine learning then picks and tunes the classifier. A forward selection step prunes the feature set.

The pipeline classified ozone exposure with up to 94.6 percent accuracy on unseen data. It also worked for other species and stimuli. The authors aim at cheap, dense urban air monitoring built on plants.

Catalog record

CREATORTill Aust, Eduard Buss, Felix Mohr, Heiko Hamann
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-13312

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Automated Phytosensing: Ozone Exposure Classification Based on Plant Electrical Signals Till Aust, Eduard Buss, Felix Mohr, Heiko Hamann Abstract In our project WatchPlant , we propose to use a decentralized network of living plants as air-quality sensors by measuring their electrophysiology to infer the environmental state, also called phytosensing. We conducted in-lab experiments exposing ivy ( Hedera helix ) plants to ozone, an important pollutant to monitor, and measured their electrophysiological response. However, there is no well established automated way of detecting ozone exposure in plants. We propose a generic automatic toolchain to select a high-performance subset of features and highly accurate models for plant electrophysiology. Our approach derives plant- and stimulus-generic features from the electrophysiological signal using the tsfresh library. Based on these features, we automatically select and optimize machine learning models using AutoML . We use forward feature selection to increase model performance. We show that our approach successfully classifies plant ozone exposure with accuracies of up to 94.6% on unseen data. We also show that our approach can be used for other plant species and stimuli. Our toolchain automates the development of monitoring algorithms for plants as pollutant monitors. Our results help implement significant advancements for phytosensing devices contributing to the development of cost-effective, high-density urban air monitoring systems in the future. I Introduction Since decades city populations are increasing at a fast pace, a phenomenon called urbanization. 1 1 1 https://unhabitat.org/wcr/ One effect is increased traffic volume, which negatively impacts air quality. For example, in Europe just under two thirds of air pollution is accounted to traffic 2 2 2 https://www.eea.europa.eu/highlights/emissions-from-road-traffic-and and up to a third of European citizens are exposed to concentrations of air pollutants that exceed the European Union’s air quality standards, especially due to particulate matter and ozone. 3 3 3 https://www.eea.europa.eu/soer/2015/europe/urban-systems Hence, in addition to ongoing efforts to reduce pollution, also suitable air pollution monitoring techniques are needed to identify potential hazards early and to enable swift action [ 1 ] . Usually, such techniques incorporate expensive, fixed monitoring stations that provide highly accurate measurements with the drawback of low spatial resolution [ 2 ] . Hence, there is a paradigm shift towards more decentralized measurement approaches but a fundamental challenge remains in finding cost-effective methods to monitor all pollutants accurately [ 3 ] . Another approach addressing low spatial resolutions is to use mobile pollution sensing systems. However, the increased spatial resolution comes with a tradeoff of reduced temporal resolution. In our EU-funded project WatchPlant [ 4 , 5 ] (2021-2024) we address both the challenge of measuring many pollutants in an inexpensive and effective way and the tradeoff of temporal and spatial resolution by proposing a network of low-cost biohybrid air-quality sensor nodes. Our vision is to deploy these air-quality sensor nodes densely throughout the city (possible due to their low-cost) and connecting all of them to aggregate pollution information for real-time monitoring to help policy makers to decide and inform citizens about potential hazards, for example, via their smart phones. By engaging citizens and avoiding private data collection (e.g., connection history) we mitigate ethical concerns. We anticipate a positive ecological impact as our biohybrid system promotes the expansion of urban green spaces. One biohybrid air-quality sensor node consists of an electronic device, our PhytoNode [ 6 , 7 ] , and a natural plant (in our case Hedera helix ). The PhytoNode is a light-weight and inexpensive sensor used for phytosensing. In phytosensing the electric physiology of living plants is measured to sense, for example, the environmental conditions [ 8 ] . One advantage of phytosensing is that we can potentially infer multiple pollutants, which usually requires multiple different sensing devices, by only measuring one natural plant. A cost-effective and robust phytosensing approach is to measure electric differential potentials holistically across the entire plant. This can be done, for example, by inserting two needles as electrodes, one close to a leaf and one close to the bottom of the plant’s stem. A remaining challenge is to infer the environmental stimulus based on the measured holistic electric differential potential because this macroscopic level of electric physiology in plants is neither well studied nor thoroughly understood as of now [ 9 ] . Due to the recent successes of machine learning (ML) across many domains [ 10 , 11 ] , also several studies applied ML to analyze holistic plant electrophysiology. Most studies utilize phytosensing in the context of agriculture, for example, to detect drought stress using support vector machines [ 12 ] , ensemble boosted tree classifiers [ 13 ] , or (extreme) gradient boosting [ 14 , 15 ] . Other studies utilize deep neural networks to detect salt stress [ 16 , 17 ] , nitrogen deficit [ 18 ] , soil moisture [ 19 ] or other abiotic stressors [ 20 ] (classical machine learning approaches for abiotic stressors [ 21 , 22 ] ). Fewer studies consider phytosensing in the context of air pollution monitoring, by experimenting with gas stimuli [ 23 ] . Chatterjee et al. [ 24 ] experimented with tomato plants and exposed them, among other stimuli, to ozone (16 ppm, that is, 266 times the threshold for preserving the citizen health stated by the EU 4 4 4 threshold value for ozone of 60 ppb (parts per billion), e.g. German Federal Environment Agency: https://www.umweltbundesamt.de/daten/luft/ozon-belastung#zielwerte-und-langfristige-ziele-fur-ozon ) for one minute every 2 hours. Chatterjee et al. propose to use discriminant analysis classifiers to classify ozone pollution based on 11 features, with background subtraction, derived from the plant electrical signal. Depending on the selected features and classifiers, they can distinguish between ozone and other stimuli (different concentrations of sodium chloride and sulfuric acid) with an accuracy of up to 95 %. This work has been extended by using more statistical features [ 25 ] , curve fitting coefficients as features [ 26 ] , or different classifiers [ 27 ] . However, they did not study whether a stimulus can be distinguished from the resting state (i.e., no stimulus). Dolfi et al. [ 28 ] propose a two step detection algorithm to assess ozone air pollution based on plant electrical signals. In a first step, they use a derivative-based algorithm for change detection in the signal. If a change is detected, they employ correlation waveform analysis to assess if the change is due to ozone pollution. They exposed Ligustrum texanum and Buxus macrophilla to either ozone concentrations of 200 ppb or gradually increasing ozone levels from 50 to 200 ppb. They achieve an overall detection accuracy of 87 %. However, their method requires to manually select three plant specific thresholds for the detection algorithm and plant specific waveform responses for the correlation measure. They also did not study if this methods generalizes to other pollutants or stimuli. All the above studies have in common that they choose the ML models ad hoc without providing a rationale for choosing one over another. They appear to have selected a functioning model through manual parameter adjustments or by comparing a restricted set of seemingly randomly chosen models. As a possible solution we propose to use automated machine learning (AutoML). Methods of AutoML automatically compose and parameterize ML algorithms

Read the complete text