Method to reduce pest infestation risk

Abstract

This disclosure relates to a method for reducing infestation by insect pests in an agricultural product, which comprises receiving temperature data indicative different temperatures for multiple periods of time; calculating, for the multiple periods of time, during which the agricultural product is subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures; and based on the mortality, generating a control signal to subject the agricultural product to a temperature treatment to achieve a desired mortality.

Description

"Method to reduce pest infestation risk"Cross-Reference to Related Applications

[0001] The present application claims priority from Australian Provisional Patent Application No 2024901278 filed on 3 May 2024, the contents of which are incorporated herein by reference in their entirety.Technical Field

[0002] This disclosure relates to reducing a pest infestation risk in an agricultural product by killing pest organisms through low temperature exposure.Background

[0003] Some pest organisms, such as insects, are sensitive to low temperature exposure. Therefore, in cases where an agricultural product should be essentially free of live insect pests, such as for biosecurity reasons, the agricultural products are stored at low temperature at a point during the supply chain, such as at the point of import. While this reliably kills most insect pests, it also introduces a delay in the supply chain that reduces shelf-life and increases costs. Therefore, it would be an advantage to provide an improved method for reducing the infestation of insect pests.

[0004] Any discussion of documents, acts, materials, devices, articles or the like which has been included in the present specification is not to be taken as an admission that any or all of these matters form part of the prior art base or were common general knowledge in the field relevant to the present disclosure as it existed before the priority date of each of the appended claims.

[0005] Throughout this specification the word "comprise", or variations such as "comprises" or "comprising", will be understood to imply the inclusion of a stated element, integer or step, or group of elements, integers or steps, but not the exclusion of any other element, integer or step, or group of elements, integers or steps.Summary

[0006] A method for reducing infestation by insect pests in an agricultural product comprises: receiving temperature data indicative different temperatures for multiple periods of time; calculating, for the multiple periods of time, during which the agricultural product is subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures; and based on the mortality, generating a control signal to subject the agricultural product to a temperature treatment to achieve a desired mortality.

[0007] In some embodiments, generating the control signal based on the mortality comprises, upon determining that the mortality is below a desired threshold, calculating a second period of time for temperature treatment to achieve the desired mortality.

[0008] In some embodiments, calculating the mortality comprises combining a mortality rate for each of the periods of time independently into the mortality.

[0009] In some embodiments, the multiple periods of time relate to multiple periods of time during transportation of the agricultural product.

[0010] In some embodiments, the method comprises measuring the different temperatures by temperature sensors during transportation of the agricultural product.

[0011] In some embodiments, multiple scenarios during transportation of the agricultural product are indicative of the different temperatures, respectively.

[0012] In some embodiments, calculating the mortality depends on one or more of: a pest species, a developmental stage of the insect pests, or a host species.

[0013] In some embodiments, the method further comprises calculating a duration of the temperature treatment to achieve the desired mortality.

[0014] In some embodiments, the method further comprises determining a transport plan to transport the agricultural product based on the duration of the temperature treatment.

[0015] In some embodiments, the method further comprises training a model for calculating the mortality based on historical data.

[0016] In some embodiments, the method further comprises: subjecting the agricultural product to temperature treatment according to a treatment protocol to achieve a desired mortality; detecting, during the temperature treatment, a deviation from the treatment protocol; and in response to detecting the deviation, calculate an updated treatment protocol.

[0017] In some embodiments, the method comprises adjusting a temperature of a transportation container, or cold storage based on a desired mortality at the end of transportation.

[0018] A computer system for reducing infestation by insect pests in an agricultural product comprises one or more processors configured to: receive measurements indicative of different temperatures for multiple periods of time; calculating, for the multiple periods of time, during which the agricultural product is subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures; and based on the mortality, determining a temperature treatment for the agricultural product to achieve a desired mortality.

[0019] A method for determining a prediction of or uncertainty in a mortality value of an insect pest comprises: creating a dataset that includes data related to a mortality of insect pests for different temperatures over periods of time; creating a mathematical model comprising model parameters, the model having temperature and time as input variables and mortality as an output variable; adjusting the model parameters to reduce an error between the output variable and the mortality of the insect pests in the dataset; and applying the model to determine the prediction of or uncertainty in temperature and time period conditions of an agricultural product to calculate the mortality value of the insect pest.

[0020] In some embodiments, the model is a multi-factor regression model.

[0021] In some embodiments, the model represents the mortality as a quantile of insect pests killed over each of the periods of time.Brief Description of Drawings

[0022] A non-limiting example will now be described with reference to the following drawings:

[0023] Figure 1 illustrates an example transport scenario.

[0024] Figure 2 illustrates a method 200 for reducing an infestation by a pest insect.

[0025] Figure 3 illustrates a predicted mortality from the final model versus observed mortality for the complete data. The bin colours indicate the frequency of observation pairs with the location values. The blue line is a loess smoother, and the red line is the equivalence line.

[0026] Figure 4 illustrates a trace of grape temperatures in a container travelling from Australia to China with the estimated mortality curve for B. tryoni at the pupae stage (most cold tolerant stage) and probit 9 mortality estimates from our model.

[0027] Figure 5 illustrates the mortality (probit scale) plotted against days for each trial shown as points and simple linear regression drawn for each. The line colours correspond to the temperature treatment (rounded to a degree) of each trial (number of trials = 320).

[0028] Figure 6 illustrates the daily mortality rate for each day within each experiment plotted against the day. There is reasonable consistency in mortality rate across the treatment period.

[0029] Figure 7 illustrates a computer system for reducing infestation by insect pests in an agricultural product.

[0030] Figure 8 illustrates a method for determining a prediction of or uncertainty in a mortality value of an insect pest.Description of Embodiments

[0031] This disclosure provides methods and systems for reducing infestation by insect pests in an agricultural product. Reducing infestation in this sense means reducing the number of live pests in the product, i.e. to kill the pests in the product. The agricultural product may be any type of product and would, in many examples, be fresh produce, such as fruit, nuts, vegetable, mushrooms, grains, etc. The product is provided fresh and not frozen, which means the pests should be killed by temperatures above 0 °C. It is further noted that the disclosed methods may be applied to other products or items, such as general cargo.

[0032] Most pests are not killed instantly by low (but not freezing) temperatures but die at a constant rate per day (probit mortality scale). The mortality rate (% mortalityper day) relates to the percentage of insects that die per day at a given temperature. In that sense, mortality accumulates over time as from a succession of mortality per day. In some scenarios, a minimum overall mortality is given as a desired value, such as 99.99% of all pests being killed. This value may also be converted to a probit value. The probit function is the quantile function associated with the standard normal distribution. Mathematically, the probit is the inverse of the cumulative distribution function of the standard normal distribution, which is denoted as O(z) , so the probit is defined as probitfor p G (0,1) . In some contexts, such as the biosecurity context, the probit value has a value of 5 added to it so as to keep the values positive.

[0033] It is possible to apply a constant produce specific cooling regime to agricultural produce to achieve a desired mortality and then store all produce at a specific point in the supply chain (such as at the import customs clearance) at that temperature for a set time. However, there are multiple problems with this approach. In particular, some pest species are more susceptible to low temperature than others while mortality also depends on the host (i.e. which produce). Further, should the process of cold exposure be interrupted briefly, or if there are meaningful deviations from the set temperature, the regime may be forced to restart. Even further, the produce may have been subjected to low temperatures during the supply chain, which means a high percentage of insect pests have already been killed. Therefore, the full exposure is not necessary anymore. As a result, this approach of a set time and temperature leads to a delay (reduction of remaining shelf life) as well as costs for cold storage and handling. It is noted that low temperature in this context generally refers to temperatures that are below room temperature, so below 20 degrees. In other examples, low temperature can mean below 10 degrees, below 5 or 6 degrees, below 4 degrees or below 3 degrees. In many applications, the produce is sensitive to freezing and therefore, the temperature would be above 0 degrees.

[0034] This disclosure provides an improved method for reducing infestation by insect pests in an agricultural product that does not have a set time and temperature forkilling most pest. Instead of a set time and temperature, the disclosed method calculates a mortality for the different times and temperatures based on historical (past) data with different times, temperatures and mortality outcomes and then determines the time and temperature for achieving the desired mortality. In other words, the method may determine the time at a given temperature that leads to killing the pest organism to a desired mortality threshold. In further examples, the method predicts infestation risk in an agricultural product as a result of known but varying temperatures of the product. The advantage is that the storage time or end point treatment can be reduced.

[0035] The calculations use a model that has been built from a large number of experiments. Since those experiments have a range of parameters, the model can be used to interpolate and extrapolate to other parameters. This means that time values can be determined for a specific pest species or a host product including combinations not empirically tested in the studies. More specifically, a combination of multiple different pest species can be considered, and the model applied so that the mortality is calculated for all species together or for the species with the lowest mortality (e.g. the most robust). Further, the time values can be determined for a range of different temperatures. In particular, the time required for a temperature treatment to be effective can be calculated at any time during the cold chain.

[0036] While the proposed approach is robust, a regulator may still require specific experimentation before implementing a particular time / temperature routine. Such experiments can be cumbersome in general as a large number of pest organisms need to be available and then killed under different environments to find the most appropriate combination of parameters. In this scenario, the disclosed method is also useful in that it calculates a specific combination of time and temperature (potentially for different time periods) and then that specific combination can be tested in an experiment to validate the result for policy making. As a result, the experimental burden is greatly reduced. That is, the method can fast track experimentation by helping select the best combinations and identify gaps.

[0037] One feature of the disclosed method is the underlying insight that the probit mortality is constant over time. That is, the change in probit mortality in terms of a proportion of live pest organisms, on one day is the same as the mortality on a different day at the same temperature. In other words, the probit mortality change per day is stationary and only depends on the non-temporal parameters, like temperature, pest species and host. This insight supports the interpolation and extrapolation that the disclose method provides.

[0038] Figure 1 illustrates an example transport scenario 100 comprising an on-farm storage shed 101 (being located on a farm or orchard), a first leg 102 on a truck to a port, a second leg 103 on a container ship, a third leg 104 on a truck, a storage 105 and finally cold treatment 106 to kill a desired ratio of insect pests. It is noted that a large number of variations of different supply chains are possible including different legs on trucks, vessels, planes and in storage and so on. The figure also provides a temperature chart 110 to show the different periods with respective temperatures. In this example, during storage in shed 101 starting from time tl, the temperature is Tl, during first leg 102 starting from t2 the temperature is T2 and so on.

[0039] At the end of storage 105 at time t6, a large number of insects would have died as a result of the low temperatures during storage in farm storage shed 101, first leg 102, second leg 103, third leg 104 and storage 105. Nevertheless, cold treatment 106 is applied for a treatment time ttreat. The aim is to keep this treatment time as short as possible while achieving a desired mortality of the insect pest. As set out earlier, it is possible to assume that all insect pests survive until t6 and that the only the cold treatment 106 reduces the infestation to the desired ratio. However, this disregards the effect of the cooling across the supply chain and therefore, such a cold treatment would be unnecessarily long.

[0040] Therefore, this disclosure provides an improved method for reducing the infestation by the pest insect. In particular, Figure 2 illustrates a method 200 for reducing an infestation by a pest insect. The disclosed method comprises receiving 201, from temperature sensors, measurements of different temperatures for multiple periodsof time. This relates to the transport scenario in Figure 1 where temperature sensors measure temperatures T1 -T5 for periods tl to t6, respectively. The temperature sensors may comprise various different technologies, such as thermocouples, infrared sensors, thermistors and others. The temperature sensors may continuously monitor the temperature, such as every second, minute, hour, or day. Measuring the temperature may also be triggered by external events, such as by opening a container door, light in the container, etc. Receiving the measurements may comprise receiving digital or analog signals over a direct connection with the sensor or receiving data packets over a communication network, such as an loT network or the Internet. The measurements may not be received in real-time but may be delayed. This means the measurements may be stored for later processing. Such a delay is not detrimental because the mortality at point t6 can be calculated when the produce arrives at the cold treatment facility and there is no need for continuous calculation of the current, real-time mortality.

[0041] It is noted that the temperature sensors may measure the temperature of the air in the containers or other storage medium. In other examples, the temperature sensors measure the temperature of the actual agricultural product, which is also referred to as pulp temperature. In some cases, pulp temperature is more difficult to measure but provides more accurate outcomes for the final mortality calculation because it is more directly related to the temperature actually experienced by the insect pest. In some examples, there may be a mathematical relationship that calculates the pulp temperature from the air temperature over time dependent on a number of parameters, such as product type, storage type, air flow, etc. The measured temperature may be transmitted to a computing device, such as a monitoring server, regularly and potentially each time the temperature is measured (real-time). In other examples, the temperature measurements are recorded locally and then transmitted to the server in batch to reduce communication overhead.

[0042] In further examples, method 200 uses different scenarios for the different legs of the transportation. That is, method 200 receives an indication that the first leg 102 is by truck for time t3-t2. In response, method 200 retrieves the temperature T2 for thisscenario from a look-up memory. Similarly, the method retrieves temperatures Tl, T3, T4 and T5 from the look-up memory for respective scenarios. It is noted that there may be a combination in the sense that some legs provide measured temperatures while others are look-up values for a corresponding scenario. Further, it is possible that the scenario value is corrected or adjusted using the sensed value or vice versa.

[0043] At time t6, method 200 calculates 202, for the multiple periods of time (between tl, t2, t3, t4, t5 and t6, respectfully), during which the agricultural product was subjected to the different temperatures Tl, T2, T3, T4, T5, a mortality of the insect pests as a result of the different temperatures. This means that method 200 calculates for each period separately the mortality and then combines the calculated mortalities (such as by multiplying them) the overall mortality. For example, 30% of insect pests may survive storage shed 101 to t2 due to the relatively low temperature Tl. 60% of the remaining insect pests at t2 survive the first leg 102 to t3 due to the relatively high temperature T3 and 50% of the surviving insect pests at t3 survive the second leg 103 to t4, of which 30% survive to t5 and 40% survive to t6. The total survival is 0.3*0.6*0.5*0.3*0.4=0.0108 which relates to a mortality of 98.92%. This is now the starting point of the cold treatment from t6. It is noted that the mortality can be expressed in a variety of different forms, such as percentage, probit, survival or other equivalent forms. In that sense, the term mortality should be understood as any value that is quantitatively indicative of the number of insect pests that have been killed or have survived up to this point. It is noted that the term “mortality” refers to a percentage of individuals that die for a given, arbitrary time period (potentially expressed as probit), such as during a transport leg or storage. “Mortality rate” refers to the percentage of pests that die each day or per any other constant time period.“Infestation rate” would be the number of individuals per quantity of the product / commodity / space.

[0044] It is noted that the mortality during the third leg 104 may not depend on the mortality of the first leg 102 and the second leg 103, which would allow a simplification of the mortality calculations in the sense that the individual mortalities can be simply multiplied. Further, the additional mortality that occurs on third leg 104may not depend on the mortality rate on the second or first leg but may depend on the cumulative mortality (or survival) at the start of the third leg and the temperature settings for that third leg. That is how the method deals with temperature varying mortality. In same cases this may involve moving between the probit and natural scales.

[0045] At point t6, the method 200 calculates a time period for temperature treatment that is suitable to reach a desired mortality. In this example, the desired mortality is 99.9% and therefore, the cold treatment should show a survival of less than 9%. The model disclosed herein can determine the time for this survival. In other examples, the method 200 calculates a temperature that is suitable to bring the mortality to the desired level for a given duration of cold treatment. In yet further examples, the method calculates both time and temperature to optimise a different cost function, which could involve a trade-off between shelf-life and energy expenditure or optimisation into the supply chain flow (e.g. to clear the treatment plant before the next consignment arrives). For example, for a shorter journey the temperature is lower to reach the threshold than for a longer journey. The focus is on the end state and then the method can optimise for the produce quality, energy consumption, time in storage and pest status simultaneously. These calculations may be specific for pest species, e.g. may be different for Mediterranean fruit fly than for Queensland fruit fly.

[0046] The disclosed method may further comprise determining a transport plan to transport the agricultural product based on the duration of the temperature treatment. In other words, the method uses the calculated duration of temperature treatment to decide when or where to transport the product. For example, if the product is to be shipped in a container with a known low temperature, and the calculated time to achieve the desired mortality rate is less than the expected travel time in that container, the method comprises deciding that the product remains in cold storage before shipment so that the desired mortality is achieved when the product arrives and no further treatment is necessary.

[0047] As a next step of method 200, a control signal is generated 203 based on the calculated mortality. That control signal controls how the agricultural product issubjected to temperature treatment to achieve a desired mortality based on the calculated mortality. The details of the model are provided below. The control signal can take a variety of different forms. The control signal may be a machine control signal that controls machinery that subjects the product to temperature treatment, such as movement machines that move the product into or out of or within the cold treatment facility. The movement machines may comprise conveyers, forklifts, or any one or more of the following warehouse robots:• Automated Guided Vehicles• Autonomous Mobile Robots• Palletizers• Depalletizing Robots• Conveyor Systems• Automated Storage and Retrieval Systems• Sorting Robots• Packing Robots• Pick and Place Robots• Inventory Robots

[0048] In other examples, the control signals are indicative of the desired mortality being achieved. In that case, the product is placed in cold treatment until the control signal indicates that the desired mortality is achieved. The product can then be removed from the facility. Further, the control signal can be provided visually or in other forms to a user who then operates the treatment facility to subject the produce to cold treatment. In yet another example, generating the control signals may comprise generating a report including the calculated mortality and / or the mortality after the calculated treatment time. The report may indicate whether or not the desired mortality has been achieved before or after cold treatment.

[0049] It is noted that the mortality can be re-calculated in response to measured temperature changes. For example, the cold storage 105 may have a set temperature, which is used in the model and the time the product remains in storage is a variable thatis measured through arrival and departure monitoring. However, the temperature in the cold storage may change unexpectedly, such as because of a power outage or arrival of further product. In that case, the time period between t5 and t6 can be split into two or more separate time periods with different temperature and the mortality is calculated separately for each period and then combined as explained above.

[0050] In a further example, the time period can be so short that the calculation essentially becomes in integral over a temperature curve. This is useful in cases where there is a ramp-up or ramp-down of the temperature over considerable period of time. In cases where the ramp time is short, the maximum temperature of the ramp can be used as a conservative approximation without an overly large error. However, if the ramp time is long (e.g. longer than 1 hour), the ramp may be split into multiple periods of time (such as one period every hour or every day). This way, the mortality for each period can be calculated assuming a constant temperature, which is the maximum temperature during that period.

[0051] In some cases, the mortality may already be at the desired level when the product arrives at t6. In that case, subjecting the product to temperature treatment based on the mortality means that the product is not subjected to temperature treatment. On the other hand, upon determining that the mortality is below a desired threshold, the method 200 may calculate a treatment time (period of time) for temperature treatment to achieve the desired mortality.

[0052] In yet a further example, the mortality rate is calculated for a time in the future. This calculation may be based on expected or predicted durations of transport legs and temperatures. For example, a product is to be shipped in a container that normally cools the product to a set temperature and the shipment normally takes a set time. In that case, the method can calculate a future mortality and determine whether that mortality is at least the desired mortality. The method can determine a transport plan, such as by modifying an existing transport plan, so that the desired mortality is achieved at the end of the journey.

[0053] The model used to calculate the mortality may comprise a range of different model parameters and structures. For example, the model may be a regression model, which may comprise a set of statistical processes for estimating the relationships between a dependent variable (often called the 'outcome' or 'response' variable, or a 'label' in machine learning parlance), such as the mortality, and one or more independent variables (often called 'predictors', 'covariates', 'explanatory variables' or 'features'), such as temperature, insect species, host, etc. In that sense, the model is a supervised machine-learning model and the model may involve the use of other machine learning methods, such as neural networks, decision models, etc.

[0054] In many cases, the model has a set of parameters and the values of those parameters are adjusted to optimise the prediction accuracy, i.e. to match the model output with the observations in the training data. For example, the method of ordinary least squares computes the unique line (or hyperplane) that minimizes the sum of squared differences between the true data and that line (or hyperplane). Other forms of regression use slightly different procedures to estimate alternative location parameters (e.g., quantile regression or Necessary Condition Analysis) or estimate the conditional expectation across a broader collection of non-linear models (e.g., nonparametric regression).

[0055] The method of calculating parameter values may follow a number of different approaches, such as ordinary least squares, weighted least squares, generalized least squares, Linear Template Fit and others. Yet further examples include maximum likelihood estimate, ridge regression, least absolute deviation, and adaptive estimation. Another example is quantile regression, which focuses on the conditional quantiles of y given X rather than the conditional mean of y given X. Linear quantile regression models a particular conditional quantile, for example the conditional median, as a linear function pTx of the predictors.

[0056] As noted before, the model used to calculate the mortality is a multi-variate model with a number of different variables. In example, the model is a linear, multivariate model, such as a multivariate regression model. The multivariate regressionmodel is a compact way of simultaneously writing several multiple linear regression models. The multiple linear regression model may be compactly written as Y = XB+U where Y is a matrix with series of multivariate measurements (each column being a set of measurements on one of the dependent variables), X is a matrix of observations on independent variables that might be a design matrix (each column being a set of observations on one of the independent variables), B is a matrix containing parameters that are usually to be estimated and U is a matrix containing errors (noise). The errors are usually assumed to be uncorrelated across measurements and follow a multivariate normal distribution. If the errors do not follow a multivariate normal distribution, generalized linear models may be used to relax assumptions about Y and U. The more detailed formulation below may use different notation since various different forms of expression are possible.

[0057] The general linear model may incorporate a number of different statistical models: analysis of variance (ANOVA), analysis of covariance (ANCOVA), Multivariate analysis of variance (MANOVA), Multivariate analysis of covariance (MANCOVA), ordinary linear regression, t-test and F-test. The general linear model is a generalization of multiple linear regression to the case of more than one dependent variable. If Y, B, and U were column vectors, the matrix equation above would represent multiple linear regression.

[0058] Hypothesis tests with the general linear model can be made in two ways: multivariate or as several independent univariate tests. In multivariate tests the columns of Y are tested together, whereas in univariate tests the columns of Y are tested independently, i.e., as multiple univariate tests with the same design matrix.

[0059] The parameters of the model can be estimated using computational techniques, such as those implemented in the lme4 package of R, which is a programming language for statistical computing. The package includes linear (Imer), generalized linear (glmer) and nonlinear (nlmer) functions.

[0060] In the present case, the model is fitted to data from different pest species and host species (the type of product). Therefore, calculating the mortality depends on a pest species or a host species or both. In that sense, the model is trained on historical data. This historical data captures or is indicative of the at which the insect pest has died under different temperature conditions in the past. More particularly, the parameter values of the model are adjusted to reduce the error between a prediction of the model and measured outcomes in the historical data. Further below, this disclosure provides a detailed description of the datasets that can be used to train the model.

[0061] In particular, the model was trained on a dataset to construct a model to calculate the mortality of several species of pest fly infesting different host fruit kept at different set temperatures, ranging from 0 °C to 7 °C, over a period up to 20 days. The best fitting model was then used to predict the mortality across the wider design space as a scenario tool that can predict how long it would take for a particular fruit fly species to die within a given temperature range.

[0062] The model estimates mortality of pest species in consignments of fruit in refrigerated containers using real-time collected temperature data carrying grapes from Australia to China. The results revealed that the mortality for the protocol is reached even when there are slight fluctuations in temperature, and also the desired mortality is reached 7 days earlier than the end of the regulated quarantine period. Since the quarantine period would not need to restart, the saved time may avoid costs associated to maintaining a refrigerated storage space. This can make trade faster, increase shelflife of fruits, and provide opportunities for export countries to access different markets that couldn’t be reached because of time and money constraints. Therefore, there is potential to shift the process of treating pests in export fruits from a strictly somewhat arbitrary rules-based method to a more flexible, efficient, risk-based method, which would be beneficial to all stakeholders.Computer system

[0063] Figure 7 illustrates a computer system 700 for reducing infestation by insect pests in an agricultural product. The computer system 700 comprises a processor 701 and memory 702. Processor 701 is configured to perform the steps of method 200. The processor may be implemented with multiple processors or in a cloud computing environment or by an internet of things (IOT) device on the edge. Processor 701 accesses program code stored on a non-transitory computer-readable medium, such as memory 702, that causes processor 701 to perform method 200. That is, processor 701 is configured to receive measurements from temperature sensors 703 of different temperatures for multiple periods of time. Receiving measurements may involve sensors 703 being directly connected to processor 701 or processor 701 receiving the measurements remotely, such as via the Internet or through an Internet of Things (loT) network. The measurements may be stored remote from processor 701 for some time (e.g., on the edge) and then sent to the processor 701 at a later time.

[0064] The processor 701 then calculates, for the multiple periods of time, during which the agricultural product was subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures. Finally, processor 701 determines, based on the mortality, a temperature treatment for the agricultural product to achieve a desired mortality. The processor 701 may then send data or control signals to actuators or temperature treatment devices to cause those devices to subject the agricultural produce to the calculate temperature for the calculated time.Container control

[0065] In one example, the agricultural product is transported in one or more cooled shipping containers. This includes 20 foot or 40-foot containers for sea freight and also cooled trucks, air cargo containers, warehouses etc. In that case, processor 701 calculates a temperature for one or more of the shipping containers to achieve a desired mortality. This calculation may be based on a predicted time the product will be in the shipping container. For example, for a shipment on an ocean vessel, processor 701 may receive an estimated travel time and then calculate the temperature for that travel time to achieve the desired mortality. Then, the processor 701 may send the calculatedtemperature to the controller of the shipping container to set the calculate temperature. It is noted that the method of calculating the temperature or calculating the mortality for a given temperature may be performed by a stationary server or may be performed on the edge by a processor integrated with the shipping container.MethodsModel Species

[0066] In one example, the method uses 10 different species of pest flies from the Family Tephritidae and Family Drosophilidae to build a statistical model - Mediterranean fruit fly (Ceratitis capitata), Queensland fruit fly (Bactrocera tryoni), Oriental fruit fly (Bactrocera dorsalis), Melon fly (Zeugodacus c lieu rb it ae). Peach fruit fly (Bactrocera zonata), Bactrocera invadens, Carambola fruit fly (Bactrocera carambolae), Guava fruit fly (Bactrocera correcta), Pumpkin fruit fly (Zeugodacus tau) and Spotted winged drosophila (Drosophila suzukii). Some datasets used to train the model include mortality data on the Mediterranean fruit fly (Ceratitis capitata) and the Queensland fruit fly (Bactrocera tryoni), which are the two main fruit fly species that threaten Australia’s $13 billion horticultural industry.Resource data set

[0067] The model is trained on data on time dependent pest fly mortality rates at cold temperatures on different host fruit. The data describes 13 different fruit types (Appendix 1), up to 4 developmental stages (egg, larval stages 1, 2, and 3), and exposed temperatures ranging from 0°C to 7 °C. Exact zero and one values for mortality rate were removed from the data before fitting the models.

[0068] Table 3 shows the marginal counts of observations in the categorical factors and means and ranges for continuous variables.Statistical Model

[0069] In one example, a statistical model uses data that reflects the relationship between mortality and temperature with time, considering different host commodities, pest species and their life stages. The data is shown in Figure 5.

[0070] The statistical model used to build the calculator is a linear mixed model of the mortality measured as a probit (normal quantile). The features included in the models are pest species, host commodity, life cycle stage, temperature, and day (i.e. time). The model fitted included all factor interactions with day as well as a random intercept and slope term for each trial. Each estimate measures how much that term modifies the mortality rate on the underlying probit scale. The full data set consisted of a total of 320 trials with varying number of data points, varying from 2 up to 13. Each trial tested a pest species infesting a host commodity cultivar at a fixed temperature for a period of time. Of those with more than 3 data points (260 of the 320), the R2values for the linear models of probit mortality on days had a median value of 0.965 and a lower 5% percentile of 0.752 (Figure 6).

[0071] The model formula is defined according to the syntax of the Tmer’ function provided by R’s ‘lme4’ package and is shown in Table 1 with the model Akaike information criterion (AIC) and Bayesian information criterion (BIC). Given a collection of models for the data, AIC estimates the quality of each model, relative to each of the other models. Thus, AIC provides a means for model selection. These models are available to predict a mortality curve for similar pest species, host commodities, stages and temperatures.

[0072] All the models tested fit the intercept term for each trial dataset as a random variable with the initial model (Equation 1 in Table 1) including a random term for the mortality rate for each experiment. Adding higher order terms did not improve the model fit. This suggests the parameter space has increased beyond the ability of the data to support the larger models. We choose the model with the smallest AIC.

[0073] In some examples, the model includes weightings according to the precision of validity for each trial dataset.

[0074] In one example, the dataset includes the average number of puparia in the controls. These range between 652 to 780. The model may treat the average mortalities as individual data points in a mixed model formulation.

[0075] In the dataset, 4560 fruit were used (9120 for lemons) for C. capitata, however the numbers for B. tryoni are not available. This prevents a logistic model as it is not clear from this trial and several others the exact number of trials (or fruit flies) that were exposed in the various trials, i.e., the denominator of a binomial response.Combining datasets becomes even more challenging where the data is not clear about the number of fruit flies used in each trial. To try to incorporate some measure of sample size into the model, the method may comprise sample size as a covariate. In cases where not all trials provide details about puparia tested, the combined data set can be reduced. Therefore, comparison of models using statistics such as AIC may be difficult.

[0076] To test and apply the proposed model to a real-world scenario, we used realtime collected air temperature data from a refrigerated container carrying table grapes from Australia to China. The data is generated by calibrated air temperature trackers in the crates of shipments going through the supply chain. We applied the data to our model to understand how the mortality rate of B. tryoni, if present, would be affected and the effect of this on the number of days required in the refrigerated container to reach a mortality of Probit 9.ResultsModel testing and building

[0077] The analysis used is a linear mixed model of the mortality on the probit scales. In some examples, there may be a caveat to prediction of mortality curves for combination of factor levels that are not included in the datasets used to train the model. Specifically, the model may, in some cases, not guarantee mortality curves to increase with time. This may occur for some combinations of species, hosts, stages, andtemperatures, particularly when temperatures are high, and the estimated mortality rate falls below zero.

[0078] In the combined analyses, uncertainty associated with under-represented factor levels such as pest and host species may result in lower precision for the effects and consequently lower precision in predictions involving those factors and more so for interactions between those factors.

[0079] Some datasets may comprise the nominal number of insects and some the number of fruit. Models using the number of insects can be tried. However, for reasons stated above, comparison of models can be problematic because of the reduced datasets with complete sample size numbers.Table 1: AIC and BIC values for pest mortality models represented by Equations 1 and 2 as well as 3rdand 4thorder interaction models and a model similar to Equation 2 but with an estimated (non-zero) intercept.Model Formula AIC BICEquation 1 probits ~ Species + Host + Stage + 2,098 2,771 Temperature + Days + Species:Host + Species:Stage + Species:Temperature + Species:Days + Host:Stage + Host:Temperature + Host:Days +Stage:Temperature + Stage:Days + Temperature:Days +(1 + Days | Study / experiment)3-way interactions probits ~ (Species + Host + Stage + 2,574 3,876Temperature + Days)A3 + (1 | Study / experiment)4-way interactions probits ~ (Species + Host + Species + 2,670 4,319Temperature + Days)A4 + (1 |Study / experiment)Estimated but single probits ~ Days + Days:(Species + Host + 2,095 2,278 intercept (Equation 2 Species + Temperature) + (1 + Days | with estimated Study / experiment) intercept)Equation 2 probits + 3 ~ Days + Days:(Species + Host + 2,147 2,326Stage + Temperature) + (1 + Days |Study / experiment) - 1Equation 2 with Days as probits + 3 ~ fDays + fDays:(Species + Host + 2,268 3,640 a factor (fDays) Stage + Temperature) + (1 + fDays |Study / experiment) - 1

[0080] The model given the current dataset with the smallest AIC value (Equation 1) included main effects and 1storder interactions between species, host, stage and temperature and an interaction term for temperature and days which models the change in mortality rate with change in temperature. The model also includes random terms for the trial intercept (starting probit) and the daily mortality rate or mortality increase across the studies. This random rate term may shrink extreme slopes to the average of all trials, the amount of shrinkage according to the number of observations in each of the trial. Larger trials may exhibit less shrinkage than smaller trials since they provide more information than smaller trials.

[0081] Figure 3 compares the observed probit values across all datasets with those estimated by the chosen model. At lower mortalities, corresponding to observations in the first one or two days of the experiments, the model tended to predict higher mortalities than observed, however from 50% mortality (probit of zero) the bias was negligible.

[0082] The estimate of prediction uncertainty for linear mixed models may involve a simulated distribution for all these parameters. In one example, the R function ‘predictlnterval’ from the TmerTools’ package can be used.Applying the model as a supply chain logging tool

[0083] Temperature and other physical conditions such as humidity and wind speed can be recorded in refrigerated containers used to transport of fresh fruit or other food commodities from source (usually a farm) to its final destination (usually a distributor) often located in a different country or a different region within the same country. In either situation, quarantine regulations stipulate that the fruit to be treated to prevent the importing of pest species like fruit fly from the source location. Cold storage can be used. Some quarantining regulations require the fruit to be held below a specified temperature for a specified number of days. Should the temperature of the fruit rise above this temperature the consignment is deemed to have failed the requirements. The fruit need to be chilled below the required temperature to start a new quarantine period.

[0084] For the predictors particular for a chosen container journey, the method predicts the mortality rate (per day) from the disclosed model and the uncertainty around that prediction. This may include specifying the pest species, the fruit type being transported, and the temperatures recorded for a particular consignment, sometimes referred to as a track.

[0085] Let mt= 1 — strepresent the probability of an insect dying in one day at temperature t , and stthe corresponding survival probability (also referred to as constant period mortality rate). So, the probability of an insect dying after D days exposed to a sequence of temperatures t1(t2, ... , ti, ... , tDis P (survival for D days) = n?=i$t i {o}

[0086] The assumption here is that survival for a given species, host and stage is dependent on the current (daily) temperature and not on previous conditions hence the probability at day i is the product of the series of probabilities up to and including that day. The mortality rate mtis estimated from our statistical model by back transforming the predicted probits from the model to the probability scale.

[0087] The method may simulate a track survival by drawing a probit value from N ~ (mi, for the corresponding average daily temperature tj recorded from the track dataset, matching that to the temperature for each day and generating a survival / mortality curve for the track. The method may generate 1000 simulated tracks and extract the mean and 97.5thpercentile for each day.

[0088] Figure 4 shows both the daily temperature as piecewise constant curve (left axis) for a consignment of grapes throughout the journey as well as the predicted mortality (right axis) from the model for B. tryoni infesting grapes at the temperatures logged. In this case, the daily temperature drops below the quarantine value of 2 °C on the second day and then stays below 2 °C until close to the end of the journey. So, the quarantine period starts on day 2 and finishes on day 18. However, the model estimates a probit 9 before day 10, 8 days before the quarantine endpoint.Conclusion

[0089] This disclosure provides a method of cold treatment of export fruit to prevent pest outbreaks. It quantifies the relationship between cold temperature and mortality of fruit flies with time and discusses how it can leverage that relationship to calculate a remaining quarantine period (that is supposed to re-start if the ambient temperature goes above the required set temperature). This is based on the observation that the mortality rate continues to be affected by the cold temperatures, even though the rate might alter, death of all potentially present fruit flies can occur after a period of time.Application of model through real-time temperature data in supply chain

[0090] The results obtained through running the model using the real-time collected temperature data reiterated the usefulness of the model, where we could show that even though the temperatures at times went above the required set temperature of 2 °C for 16 days, the desired mortality of Probit 9 was reached before the end of the quarantine period as mortality occurred even when the temperature was above 2 °C even though the rate was lesser. The time that can be potentially saved through this modification,can reduce costs associated with the requirement for additional cold storage in case the temperature was breached in-transit. As the desired mortality is reached quicker, the fruit can also be taken to market faster and increase the shelf-life and reduce food waste as well. This may also lead to an increase in the diversification of export markets that are available to a country as there are less time and money constraints.

[0091] In one example, the method uses air temperature measured by a temperature logger (e.g., from Escavox Pty Ltd, Australia) as a proxy for pulp temperature (on which the protocols are based). In other examples, there is a quantitative relationship between fruit pulp temperature and air temperature, in order to accurately predict the mortality of fruit flies using the real-time collected air temperature data.Summary statistics of datasets used to build models

[0093] Table 3: Summary statistics of datasets used to build models. Counts (% out of 1539) for categorical variables. The mean (min, max) for days and mortality were 5.5 (0.3, 16.0) and 0.76 (0.02, 1.00), respectively.Resource data set

[0094] The used dataset comprises 28 trials including data on 13 different fruit types, 10 species of fruit flies with up to 4 developmental stages (egg, larval stages 1, 2, and 3), and exposed temperatures ranging from 0 °C to 7 °C. The species Bactrocera invadens was reclassified as B. dorsalis.

[0095] Data that provided mortality-time pairs was transcribed to a single datafile. We extracted data about the host commodity (fruit type, including cultivar), pest species, life (or developmental) stage, exposed temperature, exposed time duration at each temperature, and the mortality (given as either a Probit 8.7 (99.99% mortality) or 9 (99.9968% mortality) value or a lethal dose (LD) LD90 or LD99 value) from each trial.

[0096] Where the developmental stage was given as “larvae” we assigned it as “larval stage 2”, young larvae as “larval stage 1” and mature larvae as “larval stage 3”. We grouped the different cultivars of the same fruit type together, e.g. Thompson’s seedless grapes, Ruby Seedless grapes, and Flame Seedless grapes were all grouped together as “grapes”. Although the resource dataset (Table 1) has a large number of data points, the model parameter space is also large and highly unbalanced, resulting in a sparse data array with some parameters not estimable. Exact zero and one values for mortality rate were removed from the data before fitting the models as probits of 0 and 1 are negative and positive infinity respectively. Also, mortality values of 1 subsequent to the first value of 1 in a trial is not admissible.

[0097] The dataset may comprise data related to temperature control of pest species during an “exploratory trial” to target the most tolerant life cycle stage, followed by a much larger “confirmatory trial” in which many (often tens of thousands of individuals) individuals at this developmental stage are subjected to cold temperatures (or a single temperature) until some endpoint such as the time to reach the lethal dose to kill 99% of individuals (LD99) or Probit 9. When building the model, the data is used from the exploratory trials to obtain mortality rates at different times and temperatures tocalculate the relationship between these variables. However, data from the “confirmatory trials” of the trials was used to build the model to validate the model for different pest species and host commodities.Statistical Model

[0098] Referring back to Figure 5, the relationship between mortality on the probit scale over time is largely linear indicating a constant mortality rate throughout the treatment period. To explore this further, for each experiment in each of the studies, we calculated the change in mortality between each observation in time to calculate a daily mortality change. Figure 6 shows these daily rate changes grouped by day of the experiments.

[0099] In one example, the model may comprise a linear mixed model of the mortality measured as a probit (normal quantile). The features included in the model are pest species, host commodity, life cycle stage, temperature, and day. The model fitted includes all factor interactions with day as well as a random intercept and slope term for each trial. Each estimate measures how much that term modifies the mortality rate on the underlying probit scale. The full data set used in this example consists of a total of 289 trials with varying number of data points, varying from 2 up to 13. Each trial tested a pest species infesting a host commodity at a fixed temperature for a period of time. The R2values for the linear models of probit mortality on days for each of the 289 experiments had a median value of 0.979 and a lower 5% percentile of 0.787.

[0100] The model is able to predict a mortality curve for similar pest species, host commodities, stages and temperatures. We treated the average mortalities as individual data points in a mixed model formulation.Model testing and building

[0101] Not all models tried are discussed here but the model development involved statistical arguments using forward and backward elimination of terms and comparingmeasures such as AIC and Bayesian information criterion (BIC) as well as physiological arguments (Table 1)

[0102] Equation (1) below (and in Table 1) represents a main effects and first order interaction mixed effects model with random intercept terms to explain the variation in the average trial mean for trial and experiment in mortality fitted as probits.

[0103] The term P here denotes pest species, H is host, S is stage, T is the temperature in degrees Celsius and D denotes the number of days exposed to the fixed temperature. The PQ term represents the common probit value at day zero and the remaining P terms (Pi to Pis) represent vectors of scale terms that define the effect sizes of each of the levels for each of the factors. Interactions between these terms are represented as concatenations of these terms, e.g. the TD term represents the multiplicative effect of temperature and day and ?15represents the size of that relationship. The indices p and e indicate the study and trial (experiment), respectively. The intercept of the model was initially estimated and then set to a near zero mortality term.The terms in the last row of Equation 1 represent random intercepts for studyand trial (ye(p) ) as well as random slopes (Sp, ve(p)J- These random effects are assumed to have Gaussian distributions. The suffixes e(p) signify that the experiments e are nested within studies p. The residual error is denoted in Equation 1 as s.

[0104] Several models similar to Equation (1) that included higher order interaction terms were tested and rejected as they generated higher AIC and BIC values. We modified Equation (1) to a simpler model with a common intercept atday zero for all pest species, hosts, developmental stages and temperatures. Since the probit of zero is not defined we set the common intercept to a probit value of - 3. This supports the postulate that the mortality value at the beginning of the treatment period should be zero regardless of species, hosts, stages and temperatures. We chose the mortality probit value of -3 (equivalent to a mortality rate of 0.13%) to represent a very small mortality occurring at time zero. This new model defined by Equation (2) above and in Table 1 had a slightly higher A1C but lower BIC value.

[0105] The absolute mortality varied with trials (and to a lesser extent experiments within trials) but the variation in mortality rates among trials and experiments were small. Although the two random slope (mortality rate) variances are small compared to the random intercept variances, removing these from the model increased the AIC. This random rate terms will shrink extreme slopes towards the average of all studies, the amount of shrinkage according to the number of observations in each of the studies.

[0106] Models allowing for mortality rate to vary with both species and hosts differently and species and stages gave higher AIC and BIC values and therefore these terms were considered unnecessary.

[0107] The ANOVA table below for the final model shows the importance of the fixed effects.Table 4 The ANOVA table for final model.Term Equation Sum Mean Numerator Denominator F Pr(>F)2 terms Sq Sq DF DF valueDays D 7.02 7.02 1 19.94 65.18 <1 e-04PD 12.0Days:Species 1.34 9 43.16 12.43 <1 e-044Days:Host HD 3.77 0.31 12 24.82 2.92 0.012Term Equation Sum Mean Numerator Denominator F Pr(>F)2 terms Sq Sq DF DF valueDays: Develop SD6.99 2.33 3 101.65 21.64 <1 e-04 mental stageDays:Temperat TD0.43 0.43 1 80.13 3.96 0.050 ure

[0108] The importance of the regression of probit mortality on day is shown by its large F value with interaction with species and between day and each of the covariates, species, host, stage, and temperature with days which allows variation in mortality rate with these factors. The model also includes random terms for the trial and experiment intercept (starting probit) and the daily mortality rate or mortality increase across the studies and experiments.

[0109] The parameter estimates provided by the data are shown in Table 5 below. The estimates represent changes in probit units per day. The reference ‘Day’ term (0.7001) is for the mortality of D. suzukii eggs in avocado held at 0 °C. This equates to 75% / day. The C. capitata fly has the lowest mortality rate among all species. Stages 1, 2, and 3 show increasingly lower mortality rates than eggs. Mortality rate decreases as temperature increases. Some hosts show higher mortality rates compared to that for avocados. Fruit flies have higher mortality rates in papaya.

[0110] Table 5: Summary of fixed effect estimates for the final model with a common intercept defined at minus three probit units (approx. 0.13%) mortalityTime independence assumption

[0111] The assumption of a constant change in probit mortality through the treatment period was investigated. It appears to have remained reasonably consistent across the treatment periods (Figure 6), with the exception of day 2, which has a low sample size (n = 5). Along with support for this assumption provided by the plots in Figure 5 the assumption of constant (& linear) change in probit mortality across the treatment period is reasonable.Effects of different covariates

[0112] To help illustrate the differences in mortality rates among the different covariates, we fitted linear regression models (with no intercept) to experiments that had more than three data points. This amounted to 193 experiments.Effect of temperature as a covariate

[0113] Treatment temperature had a significant effect on mortality rate, where an increase in mortality with decreasing temperatures is observed in the well supported temperature treatments. The data related to the 5, 6, and 7 °C temperature treatments in which mangosteens were used as a host. The decrease in mortality rate for each degree increase in temperature is estimated as (negative) 0.0172 (s.e. 00.92) (probits / day / degree), negative but small. Moreover, removing the data associated with the higher temperatures from the model, i.e. 5, 6, and 7 °C, made little difference to this rate change estimate or to the fit of the model.Effect of host as a covariate

[0114] We explored the mortality rates for each experiment categorised by host fruit. The mortality rate did vary with host (F = 2.92). Generally, mortality rates were higher for berries (grape and blueberries) than for citrus. Within the types of citrus, the mortality ranked highest for mandarin, then lemon and orange; although the lowest was for grapefruit it had only a few experiments. The mortality rate for fruit flies was least in mangos and guavas.Effect of developmental stage as a covariate

[0115] The F statistic for the stage interaction with day is significant (F = 21.6). The mortality rates for the different stages are shown below. The mortality rates for eggs are higher on average than those for stage 1, 2 and 3.Model performance on end-point treatment

[0116] Since the model disclosed herein is built on the data from the exploratory trial, where the mortalities are provided during the treatment period until no survivors remained, we can validate the performance of the model’s predictive ability against theconfirmatory data provided by some of the trials. We validate our model using data from two such trials below.

[0117] When all flies have died, the probit value is infinity and confidence, or predictive intervals cannot be calculated using the Gaussian approximation to the binomial. We have chosen the likelihood method to estimate the confidence intervals which are 100being the required confidence level and n represents the number of individual flies tested and we show the lower limit for mortality when necessary.Summary

[0118] The disclosed model demonstrates the mortality of fruit fly species given the temperatures exposed to in a cold treatment for biosecurity purposes. It includes the number of days exposed to the temperature, host commodity, developmental stage of pest species, and interactions as variables in the model to represent their potential effects on the mortality rates. Overall, we found that the fruit fly species and development stage have large effects on the mortality rate, however, the host commodity had little influence on the mortality rate. We discuss these effects and possible reasons for these below.Effect of different covariates: Temperature

[0119] Most of the data used to build the model consisted of temperatures ranging from 0 to 4 °C, with some data above 4 °C. When fruit is chilled in commercial conditions, the time to obtain the target temperatures of around 2C may take 24 hours or more. During this time, it is likely that significant mortalities of fruit flies occur.Effect of hosts

[0120] Our model revealed no distinct effect by the host commodity on the mortality rate of the pest species. This aligned with the United States Department of Agriculture,Animal and Plant Health Inspection Service, Plant Protection and Quarantine (USDA- APHIS-PPQ) Treatment Manual, which has separate cold treatment protocols for different pest fruit fly species, however, the host fruits are not differentiated to have different cold treatment protocols but was dependent on the country of origin.Effect of developmental stage of pest

[0121] In the exploratory trials in cold treatment studies, the most cold tolerant developmental stage is found in order to proceed with the large confirmatory trials.

[0122] However, our model showed that the developmental stage had little effect on the overall mortality rate of the pest. Therefore, although knowing the most cold tolerant developmental stage may give an accurate estimation of the minimum number of days required to kill the most hardy developmental stage, the mortality rate for eggs and all larval stages would be similar.Usefulness of the model

[0123] The disclosed model can provide an initial estimate of time to mortality based on collated data and the underlying relationships from the different studies, which have been integrated into a generic predictive model. This allows the model to handle more temperatures, fruit fly species, larval stages, and host commodities than a single trial alone. There is also potential for this model to make predictions on mortality rates (and time to achieve Probit 9 mortality) for other host commodities that have not been empirically tested. This could be used to set interim protocols or determine the specific parameters to be examined in further dedicated experimental testing.

[0124] If temperature values are available, the model can also contribute towards calculating the biosecurity risk of export produce, from the start of cold storage at the export country, throughout the sea or air freight till the commodity reaches the destination country. If an initial infestation rate is known, the probability of infestationat the destination country can be provided since our model can calculate the mortality rate of the pest species.Method for predicting mortality

[0125] Figure 8 illustrates a method 800 for determining a prediction of or uncertainty in a mortality value of an insect pest. The method 800 comprises creating 801 a dataset that includes data related to a mortality of insect pests for different temperatures over periods of time. The dataset may contain data as described above. Creating this dataset may comprise collating data records from multiple sources or re-formatting the data to make it suitable for efficient processing. For example, the data may be inserted into a database, such as SQL, or into a datafile with a specified format.

[0126] The method 800 then comprises creating 802 a mathematical model, such as a statistical model that models uncertainty, comprising model parameters, such as a supervised machine-learning model that may be a multi-linear regression model. Creating the model may involve calling functions of a software package, such as lme4 of R to specify the model. Accordingly, a computer processor may allocate memory to store the model characterisation, including the model parameter values. As discussed above, the model has temperature and time as input variables and mortality as an output variable. These variables are available in the dataset created in step 801. The method 800 then estimates 803 the model parameters to reduce an error between the output variable and the mortality of the insect pests in the dataset. As set out above, this may involve a range of different optimisation strategies. There may be an additional step of selecting one of multiple models. This selection may be based on a selection criterion, such as Akaike information criterion (AIC), Bayesian information criterion (BIC), or other criteria.

[0127] Finally, the method 800 comprises applying 804 the model to temperature and time period conditions of an agricultural product (the product to be tested) to determine a prediction of or uncertainty in the mortality value of the insect pest. Applying the model to these input variables means that the processor calculates the uncertainty or themortality value of the insect pest by executing the calculations that are defined by the model. The output mortality can then be used to adjust the time period for a final step in the transport chain to achieve a desired mortality.

[0128] The transport chain may also be automated in the sense that products that meet the desired mortality at a point in the transport chain are automatically diverted to further processing (green lane) whereas products that have an insufficient mortality (i.e. below the desired level) are automatically diverted to cold storage. The products may also be automatically labelled to indicate meeting the desired mortality. In yet another example, the computer processor updates an entry in a logistics database that a particular consignment or container or package has met the desired mortality to indicate that this item can proceed without further cold treatment.

[0129] It will be appreciated by persons skilled in the art that numerous variations and / or modifications may be made to the above-described embodiments, without departing from the broad general scope of the present disclosure. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive.

Claims

CLAIMS:

1. A method for reducing infestation by insect pests in an agricultural product, the method comprising: receiving temperature data indicative different temperatures for multiple periods of time; calculating, for the multiple periods of time, during which the agricultural product is subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures; and based on the mortality, generating a control signal to subject the agricultural product to a temperature treatment to achieve a desired mortality.

2. The method of claim 1, wherein generating the control signal based on the mortality comprises, upon determining that the mortality is below a desired threshold, calculating a second period of time for temperature treatment to achieve the desired mortality.

3. The method of claim 1 or 2, wherein calculating the mortality comprises combining a mortality rate for each of the periods of time independently into the mortality.

4. The method of any one of the preceding claims, wherein the multiple periods of time relate to multiple periods of time during transportation of the agricultural product.

5. The method of any one of the preceding claims, wherein the method comprises measuring the different temperatures by temperature sensors during transportation of the agricultural product.

6. The method of any one of the preceding claims, wherein multiple scenarios during transportation of the agricultural product are indicative of the different temperatures, respectively.

7. The method of any one of the preceding claims, wherein calculating the mortality depends on one or more of: a pest species, a developmental stage of the insect pests, or a host species.

8. The method of any one of the preceding claims, wherein the method further comprises calculating a duration of the temperature treatment to achieve the desired mortality.

9. The method of claim 8, wherein the method further comprises determining a transport plan to transport the agricultural product based on the duration of the temperature treatment.

10. The method of any one of the preceding claims, wherein the method further comprises training a model for calculating the mortality based on historical data.

11. The method of any one of the preceding claims, wherein the method further comprises: subjecting the agricultural product to temperature treatment according to a treatment protocol to achieve a desired mortality; detecting, during the temperature treatment, a deviation from the treatment protocol; and in response to detecting the deviation, calculate an updated treatment protocol.

12. The method of any one of the preceding claims, wherein the method comprises adjusting a temperature of a transportation container, or cold storage based on a desired mortality at the end of transportation.

13. A computer system for reducing infestation by insect pests in an agricultural product, the computer system comprising one or more processors configured to:receive measurements indicative of different temperatures for multiple periods of time; calculating, for the multiple periods of time, during which the agricultural product is subjected to the different temperatures, a mortality of the insect pests as a result of the different temperatures; and based on the mortality, determining a temperature treatment for the agricultural product to achieve a desired mortality.

14. A method for determining a prediction of or uncertainty in a mortality value of an insect pest, the method comprising: creating a dataset that includes data related to a mortality of insect pests for different temperatures over periods of time; creating a mathematical model comprising model parameters, the model having temperature and time as input variables and mortality as an output variable; adjusting the model parameters to reduce an error between the output variable and the mortality of the insect pests in the dataset; and applying the model to temperature and time period conditions of an agricultural product to determine the prediction of or uncertainty in the mortality value of the insect pest.

15. The method of claim 14, wherein the model is a multi-factor regression model.

16. The method of claim 14 or 15, wherein the model represents the mortality as a quantile of insect pests killed over each of the periods of time.

Images