Detection of potato storage disease by gas analysis.

The United Nations FAO (Food and Agriculture Organization) reports that a large quantity of global food production for human consumption is wasted every year (1.3 billion tons) with losses estimated between 40 and 50 % for all root and tuber crops, fruits and vegetables (FAO Global Initiative on Food Loss and Waste Reduction). Potatoes tubers are one the worldwide staple foods with an annual total production of circa 368 million tons. A major contributor to this loss is potato infection whilst in storage (in the UK circa 5% of the entire UK crop), with the main culprit being a disease of bacterial origin known as ‘soft rot’. This project attempts to address this post-harvest waste of potato tubers in storage through early detection and monitoring of the disease. The proposed approach for this research was to use gas phase biomarkers for the early detection of soft rot (in other words ‘smelling’ the disease). The first part of the project addressed past research on volatile detection and background on other sensing technologies not previously (or marginally) investigated in these studies. The section on past studies includes all the research carried out from the 1970s by Varns and Glyn to date. Most of the studies focused on Gas Chromatography or Gas Chromatography Mass Spectroscopy. The background section that follows addresses other technologies that could also be employed for volatile monitoring, namely Field Asymmetric Ion mobility Spectrometry, Photoionization Detection, Metal Oxide, Electrochemical, and Nondispersive Infrared gas sensors. Initial work focused on evaluating these gas sensing technologies for both symptomatic and pre-symptomatic progression of potato soft rot, under laboratory conditions. After preliminary investigation, the experimental method chosen consisted in assessing the sensors results at two time points, both for symptomatic and pre-symptomatic disease detection. A total of 80 potato samples (40 for each time point) were tested with 25 different gas sensors. To process the data the following techniques were employed: cumulative sensor responses, unsupervised PCA and k-means and machine learning models (LDA, MARS, SVM, random forest and the C5.0 algorithm). Results show that all these techniques yielded a very high discrimination rate between healthy control and diseased tubers (with 80 to 100% accuracy) for a number of sensors. PID, 3 metal oxide and 3 electrochemical, gas sensors were shortlisted for possible later work. The final part of the project focussed on deploying sensors identified in laboratory conditions in a real store. To this end, a bespoke instrument was developed solely for the monitoring of store environments. The instrument comprised the sensors tested in the previous part of this work that could be readily embedded into a research tool for in-situ experimentation (with the addition of few others for completeness). Three electrochemical, six metal oxide, one nondispersive infrared and the PID sensors were included in the unit. The experimental method employed was based on time course varying from few days to two weeks. Two types of experiments were carried out, namely laboratory work and store room monitoring. Time series results for four types of sample types (unwounded controls, wounded controls and two infected sets) show that some of the sensors (ethanol, ammonia, hydrocarbons, overall volatiles and carbon monoxide) deployed on the instrument could discriminate between the various sample batches and detect soft rot from a very early stage and throughout the experimental work. The bespoke instrument was then deployed in a research store setting for testing (at the AHDB Sutton Bridge Crop Storage Research Centre, UK). Four 1 ton wooden crates controls were placed inside a (56 m3) store room (at 95% RH and 15 ± 1 °C) followed after the fourth day by a batch of infected tubers (with a proportion of infected tubers to controls of 1%). Results over a period of circa three weeks show that some of the sensors (ethanol, ammonia and hydrocarbons) could detect soft rot from a very early stage and throughout the experimental work. In conclusion, the research reported here shows that gas analysis technology could be successfully applied for pre-symptomatic detection and monitoring of soft rot in a storage facility with readily available commercial sensors