Safe alternatives to pesticides that reduce pest damage are urgently needed for sustainable agriculture ambitions. Pests reduce global food production by 20–30%1 and rodents alone are responsible for an estimated loss of 70 million tonnes of cereals annually2. To combat this, reliance upon pesticides is increasing while growing concerns about environmental impacts limit their use3. Relatively few effective alternatives to pesticides are available for pests such as rodents, and alternatives like biocontrol agents often depend upon complex ecological interactions and carry their own side effects that limit uptake4.

A simpler approach to prevent pest damage is to manipulate decision making by problem animals5 and disrupt their ability to find at-risk foods. Foragers commonly rely on odours to locate food items6. Dispersing odour cues so they are not associated with food items is a form of camouflage that has the potential to make foragers ignore those food odours as hunger drives them to search for something easier to find7. Olfactory misinformation has recently reduced predation by introduced predators on nests of endangered birds in New Zealand8, where lethal pest control was ineffective.

Here we test whether olfactory misinformation can reduce damage to newly sown seeds caused by pest rodents. House mice (Mus musculus) are major agricultural pests that rely heavily on olfaction to find food6. In Australia, mice are managed via broadscale use of zinc phosphide at sowing to prevent millions of dollars in losses to valuable cereal crops, especially wheat9. However, stronger and more lethal doses of poison are now being required to reduce mouse impacts9. Mice dig up newly sown wheat seeds10, precisely targeting a seed’s location in the soil by the smell from the wheat germ (Fig. 1a), which is rich in nutritionally valuable oils11,12.

Fig. 1: How olfactory misinformation can undermine detection of seeds by mice.
figure 1

a, How mice use olfaction to locate newly sown wheat seeds, and an example of actual mouse damage on our study wheat crop. b, How odour pre-exposure and odour camouflage treatments can disrupt mouse foraging and reduce seed loss. Mouse silhouettes in a (middle) and b (right) created with

We tested two ways olfactory misinformation could protect newly sown wheat seeds from mice. First, dispersing wheat germ oil as a form of odour camouflage to decouple the otherwise tight association between seeds and seed odour cues, making it hard for mice to detect buried seeds6. Second, optimal foraging theory predicts that foragers rapidly learn the value of foraging cues13 and use Bayesian updating strategies to adapt foraging tactics14. Pre-exposing mice to unrewarding odour information before sowing (when no wheat seeds are available) should thus prompt mice to lose interest in using wheat germ odour to find seeds once sown7 (Fig. 1b).

Our study was conducted on a 27 ha wheat crop in southern Australia during a large-scale mouse plague (at least 300 mice per ha on our site). We dispersed wheat germ oil on spatially independent plots (10 × 10 m), beginning either 6 d before sowing of wheat seeds (pre-exposure treatment), or at the time of sowing (camouflage treatment) and then every 2–3 d until seedlings started to appear (after 8 d). We had two procedural controls (canola oil and trampling treatments) to account for the effects of oil or walking on the crops, as well as an untreated control. We quantified mouse damage after 7 and 14 d by counting conspicuous mouse digging holes along sowing lines in each plot. We also estimated seedling emergence after 2 weeks and used cameras to measure mouse visitations to plots. We predicted that (1) wheat germ odour dispersed as a camouflage would disrupt the ability of mice to find seeds by creating a uniform food odour distribution, reducing the amount of damage to treated plots; and (2) pre-exposure treatments applied before sowing would reduce damage immediately after sowing if mice had learned that the odour cue is unrewarding and visited pre-exposure plots less than other plots.

One week after sowing, there were 61% fewer mouse diggings on pre-exposure plots than on the untreated controls (incidence rate ratio (IRR) = 0.39, 95% confidence interval (CI) 0.19–0.80, P = 0.009) (Fig. 2a). The effect of the camouflage treatment was similar, but not significantly lower than the untreated control. Procedural controls were not different from untreated controls.

Fig. 2: Olfactory misinformation reduces mouse damage and wheat seed loss.
figure 2

a, The number of mouse diggings on control, camouflage and pre-exposure plots (100 m2; n = 12 per treatment) 1 (red) and 2 (blue) weeks after sowing. Negative binomial regression showed a significant treatment effect after 1 (X2(4) = 16.6, P = 0.002) and 2 (X2(4) = 27.6, P = 1.52 × 10−5) weeks, and P values were interpreted with Holm-Bonferroni correction for multiple comparisons. Numbers on graph indicate week and different letters indicate significant differences between treatments. An outlier of 572 diggings from a control plot in week 2 is not shown for graphical clarity. b, The number of seedlings estimated to have been lost on control, camouflage and pre-exposure plots (100 m2) (n = 12 per treatment) 2 weeks after sowing. Negative binomial regression showed a significant treatment effect (X2(4) = 22, P = 0.0002), and P values were interpreted with Holm-Bonferroni correction for multiple comparisons. Different letters indicate significant differences between treatments. An outlier of 3,628 seedlings lost from a control plot is not shown for graphical clarity. Boxplots show median (centre lines), first and third quartiles (box limits), data range (whiskers), outliers (dots) and means (black triangles).

Two weeks after sowing, once most seeds had germinated, there were 74% fewer diggings on pre-exposure plots than on untreated controls (IRR = 0.26, 95% CI 0.15–0.48, P = 1.05 × 10−5), and 63% fewer diggings on camouflage plots (IRR = 0.37, 95% CI 0.21–0.68, P = 0.001) (Fig. 2a). Procedural controls were not different from the untreated control.

Pre-exposure plots had 72% fewer seedlings lost (that is, not emerged) than the untreated controls (IRR = 0.28, 95% CI 0.15–0.50, P = 1.5 × 10−5), while camouflage plots had 53% fewer seedlings lost than the untreated control (IRR = 0.47, 95% CI 0.26–0.84, P = 0.01) (Fig. 2b). Procedural controls were not different from the untreated control.

Our results support our hypothesis that uniform food odour distribution (odour camouflage) decouples food items from their associated odour cues to hamper foraging success and reduce damage, even during a population outbreak6. Mice leave distinct, well-defined holes after removing seeds, indicating their search for seeds is targeted and precise. Such diggings on camouflage treated plots indicate that mice could still differentiate between wheat germ odour and buried seeds. However, the reduction in such digging suggests that finding seeds on treated plots was cognitively taxing and other food was easier to find. Switching to easier options due to cognitive load is common in many species including mice and humans15,16. In contrast, seeds on untreated plots would have been strongly associated with the odours they produced, allowing mice to locate them beneath the soil with precision6. Decoupling this association between food and its odour cue using olfactory misinformation is a surprisingly simple but highly effective way of reducing damage to cereal crops. Moreover, pests are unlikely to overcome its effects because it employs the same information they rely on to find food.

It is likely that mice also rapidly learned that foraging was difficult on our odour-treated plots. Mice seemed to learn to dig for the newly sown seeds, as digging activity during the second week after sowing was on average 2.5 times higher than during the first week. However, only the pre-exposure treatment resulted in fewer diggings compared with control plots after 7 d, indicating that mice had developed a learned response to this odour treatment. To forage efficiently in noisy information-rich environments, foragers must prioritize rewarding information and filter out information that is not rewarding17 as it wastes time and energy if pursued. Decreased interest in unrewarding foraging information is a widespread response of animals to variable food availability, and such habituation underpinned the efficacy of odour misinformation to protect endangered birds from alien predators18. Camera traps and chew cards set out the night before sowing showed that mouse activity on pre-exposure plots at the time of sowing and immediately after were not different from that on other treatments, indicating that mice were still present but reduced their digging activity, providing seeds with some early protection via habituation effects (Supplementary Figs. 1 and 2).

Overall, however, the reduction in seed loss at germination did not differ between the pre-exposure and camouflage treatments. Thus, in contrast to past work on olfactory misinformation8, most of the misinformation effect we detected was due to camouflage rather than habituation. Past use of the approach has involved uncommon and patchily dispersed food items8,19, whereas wheat seeds are sown along drill lines with even spacing, meaning that mice could encounter food rewards often enough to extinguish habituation20. The camouflage effect was nonetheless strong despite high mouse density at its seasonal peak when food is limiting and mice are most hungry. This is a major finding given that seasonal peaks in mouse populations in south-eastern Australia generally coincide with the sowing time for wheat crops21.

We estimated seed loss to mice in our study to be >6% on untreated plots when mouse densities were at least 300 ha−1. Our misinformation treatments reduced this loss by more than 60%. Damage reduction is likely to be achieved with much less odour application than we used; delivering ca. 50 times the oil in the seeds estimated to be on a plot every 2–3 d. Past work protecting patchy prey with olfactory misinformation has been successful with <10 times as many odour points as prey locations8. To develop cost-effective delivery of odour camouflage, it will be necessary to determine how long wheat germ odours persist on a crop and how regularly treatments need to be reapplied, if at all.

Dispersing food odours to hide food from foragers that search by smell represents a new approach to pest control that is non-lethal and works when foragers are at high densities. To protect wheat seeds, we used wheat germ oil, an inexpensive by-product of the wheat-milling process12, and farm machinery commonly used to deliver other products such as fertilizer when sowing may enable efficient delivery of odour camouflage. We suggest that olfactory camouflage techniques have the potential to protect many other cereal seeds that are vulnerable to pre-germination loss due to pest animals, providing a simple and ethical solution to managing pest impacts.


This study was conducted on a farm approximately 10 km north-west of Pleasant Hills, NSW, Australia (35° 27′ S, 146° 47′ E). Daily temperature over the study period ranged from 13.4–26.3 °C (average of 19.5 °C) and average daily rainfall was 1 mm. Our study field grows a rotation of winter wheat (Triticum aestivum) and canola (Brassica napus) and had wheat stubble from the previous crop, which is retained to preserve moisture, but also provides habitat for mice22. We established 60 plots (each ca. 33 sowing rows) across three blocks to account for spatial variation in mouse activity. Plots were set >12 m from fence lines to avoid edge effects and >20 m apart to ensure independence given mouse movements23 (Supplementary Fig 3). Wheat was sown in May at 80 kg ha−1 (800 g per 100 m2 plot = ~16,000 seeds).

We used 5 treatments, which were randomly allocated to 4 plots within each of the 3 blocks (n = 12 plots per treatment). ‘Odour camouflage’ was a wheat germ oil solution applied immediately after sowing (D0), and then at days 1, 3, 5 and 8. Note that seedlings had begun to appear on day 8. ‘Pre-exposure’ was the same solution applied at 6, 4 and 2 d before sowing, and then at days 0, 1, 3, 5 and 8 after sowing. A ‘canola oil control’ accounted for oil effects and was applied as in the camouflage treatment. A ‘trampling control’ involved walking on plots as on oil-treated plots. The ‘untreated control’ had no treatment applied and was not walked on.

Each application was applied at 50 times the oil in seeds sown on a plot to substantially override odours from buried seeds. Each 1 ml of wheat germ oil (organic cold pressed, Leonardi Laboratories) requires approximately 5,000 seeds12 and 160 ml of oil equated to 50 times the 16,000 buried seeds on a plot. This oil was diluted into 2.5–3 l of water and applied using a pressure sprayer to disperse a fine mist along drill lines and adjacent stubble. Consumer-grade canola oil (diluted as in wheat germ oil) was applied using different sprayers to avoid cross-contamination.

Background mouse density was estimated via live trapping, mouse activity on individual plots was indexed using chew cards, and cameras recorded mouse visitations to control, pre-exposure and camouflage plots in the 12 d after sowing (Supplementary Note 1).

On day 7, mouse damage was assessed by two observers counting the digging holes (~1 cm deep) left by mice extracting seeds (Fig. 1a) along all drill lines on a plot. By day 14, increased and patchy mouse damage required an adaptive sampling method to efficiently count diggings and estimate seedling damage24 (Supplementary Note 2). The number of seedlings taken by mice was estimated from the expected number of seedlings (sowing at 80 kg ha−1) minus the actual number of seedlings counted.

We used negative binomial regression to model digging count data at days 7 and 14, and seed loss data at day 14, comparing the effects of treatments to the control using IRRs and Holm-Bonferroni correction for multiple comparisons. Observer and block were included as random factors but removed if P > 0.05. Visitation counts were compared among treatments also using a negative binomial regression model as above, including treatment, time and interactions. The relationship between digging counts and mouse activity estimated from chew cards on individual plots was also tested using a linear regression and showed no significant relationship. There was also no treatment effect on mouse activity. All analysis was conducted using R v.4.1.025 and the MASS package26.

Animal ethics statement

Animal ethics approval was granted by the University of Sydney’s Animal Ethics Committee (protocol number 2021/1988).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.