Recent Advances in Reducing Food Losses in the Supply Chain of Fresh Agricultural Produce

: Fruits and vegetables are highly nutritious agricultural produce with tremendous human health beneﬁts. They are also highly perishable and as such are easily susceptible to spoilage, leading to a reduction in quality attributes and induced food loss. Cold chain technologies have over the years been employed to reduce the quality loss of fruits and vegetables from farm to fork. However, a high amount of losses ( ≈ 50%) still occur during the packaging, pre-cooling, transportation, and storage of these fresh agricultural produce. This study highlights the current state-of-the-art of various advanced tools employed to reducing the quality loss of fruits and vegetables during the packaging, storage, and transportation cold chain operations, including the application of imaging technology, spectroscopy, multi-sensors, electronic nose, radio frequency identiﬁcation, printed sensors, acoustic impulse response, and mathematical models. It is shown that computer vision, hyperspectral imaging, multispectral imaging, spectroscopy, X-ray imaging, and mathematical models are well established in monitoring and optimizing process parameters that a ﬀ ect food quality attributes during cold chain operations. We also identiﬁed the Internet of Things (IoT) and virtual representation models of a particular fresh produce (digital twins) as emerging technologies that can help monitor and control the uncharted quality evolution during its postharvest life. These advances can help diagnose and take measures against potential problems a ﬀ ecting the quality of fresh produce in the supply chains. Plausible future pathways to further develop these emerging technologies and help in the signiﬁcant reduction of food losses in the supply chain of fresh produce are discussed. Future research should be directed towards integrating IoT and digital twins for multiple shipments in order to intensify real-time monitoring of the cold chain environmental conditions, and the eventual optimization of the postharvest supply chains. This study gives promising insight towards the use of advanced technologies in reducing losses in the postharvest supply chain of fruits and vegetables. Sequential Forward Selection (SFS), Backpropagation Neural Network (BPNN), Extreme Learning Machine (ELM), Sparse Logistic Regression (SLR), Support Vector Machine(SVM), Radial Basis Function (RBF), Neural Networks (NN), Genetic Algorithm (GA), Support Vector Regression (SVR), Student–Newman–Keuls (SNK), Least Squares Support Vector Machines (LS-SVM), and Random Forest (RF).


Introduction
Food losses in the postharvest supply chain amount to a great loss of investments in the packaging, transportation, and storage operations. About 25-30% of global food produced is lost between on-farm food production and its storage at a retail facility, largely as a result of poor chain management and spoilage [1,2]. Food losses occur due to a reduction in quality and safety standards driven by consumer preferences, particularly in developed countries [3]. A high amount of losses (up to 30% per year) is often experienced during the postharvest handling of fresh agricultural produce, such as fruits and vegetables [4]. Advanced technologies are required to reduce the losses of fruits and vegetables in the postharvest supply chain. The reduction of these losses would increase the number of fresh produce available for consumption.
Fruits and vegetables are important sources of nutrients such as vitamins, minerals, and bioactive compounds, which provide many health benefits [5][6][7][8]. However, they are highly perishable goods that need to be appropriately preserved, to reduce the degradation of macro and micro-nutrients and extend shelf life [7,9,10]. As a result, fruits and vegetables are often packaged and kept in a desired low-temperature range using various refrigeration systems during the transportation and storage postharvest handling processes. This process delays or reduces microbial growth and enzymatic reaction, thereby improving overall quality, reducing mass loss, and extending shelf-life. The succession of refrigeration steps along these chains can be referred to as a postharvest cold chain of fruits and vegetables [11]. A description of the postharvest cold chain of fruits and vegetables is shown in Figure 1. Refrigeration is a key element in enhancing the quality of fresh produce and extending the shelflife, thereby enabling their adequate supply to an increasingly urbanized world [11][12][13]. However, more than 90% of perishable goods are still not refrigerated [1,14]. Inadequate refrigeration infrastructure or access to energy accounts for more than 20% losses of perishable goods [15]. These losses also encompass a huge amount of energy and water losses, but also carbon dioxide emissions [15,16]. Therefore, sustainable cold chain technologies in terms of being more resource-efficient, improving product quality retention, and reducing induced food losses become indispensable.
Several studies have been conducted on the postharvest cold chain of fruits and vegetables with a view to gain more insight on ways to address these technological and developmental challenges. The losses in the mass and nutritional qualities of strawberries, raspberries, red currants, drupes, cherries, and sour cherries were reduced using refrigerated containers at 4 °C when compared to storage at room temperature [17]. Packaging methods such as edible coating, active modified Refrigeration is a key element in enhancing the quality of fresh produce and extending the shelf-life, thereby enabling their adequate supply to an increasingly urbanized world [11][12][13]. However, more than 90% of perishable goods are still not refrigerated [1,14]. Inadequate refrigeration infrastructure or access to energy accounts for more than 20% losses of perishable goods [15]. These losses also encompass a huge amount of energy and water losses, but also carbon dioxide emissions [15,16]. Therefore, sustainable cold chain technologies in terms of being more resource-efficient, improving product quality retention, and reducing induced food losses become indispensable.
Several studies have been conducted on the postharvest cold chain of fruits and vegetables with a view to gain more insight on ways to address these technological and developmental challenges. The losses in the mass and nutritional qualities of strawberries, raspberries, red currants, drupes, cherries, and sour cherries were reduced using refrigerated containers at 4 • C when compared to storage During packaging, transportation, and storage of fresh agricultural produce, food losses are often induced as a result of a reduction in the quality (e.g., color, texture, mass) of the produce. These postharvest handling operations affect the nutritional and sensory quality of the agricultural produce, the mass of the fresh produce as well as the quantity of fresh produce available to the consumers. The quality of fresh agricultural produce can be referred to as the excellent characteristics of such products that are acceptable to a consumer [46]. Consumers typically purchase fresh agricultural produce based on their biochemical characteristics such as appearance, texture, flavor, and nutritive value [46,47]. Fresh agricultural produce such as fruits and vegetables provide an essential part of human nutrition, as they are important sources of vitamins, dietary fibers, minerals, and other biochemical (e.g., carbohydrate, protein, etc.) with tremendous health benefits [48]. Adequate in-transit monitoring of environmental conditions and changes in the quality attributes of fresh produce during transport and storage will help reduce food losses and ensure the availability and accessibility of fresh fruits and vegetables with high nutritional density to the consumers [49,50]. Therefore, emerging technologies are needed to help reduce the overall quality loss of fresh agricultural produce, thereby reducing food losses in the postharvest supply chain.

Advanced Technologies for Quality Assessment in Postharvest Supply Chain: State-Of-The-Art
In recent decades, several modern food quality techniques have been applied to monitor, control, and predict the quality evolution of various fruits and vegetables in postharvest supply chains. These techniques include imaging systems, spectroscopy, multi-sensors, electronic nose (E-nose), acoustic impulse response (AIR), radio frequency identification (RFID), printed sensors (PTS), and mathematical modeling. In this section, we analyze the application of these techniques in advancing cold chain operations and process optimization during the packaging, storage, and transportation of fruits and vegetables within the past 10 years.

Application of Imaging Technology, Spectroscopy, Multi-Sensors, E-Nose, AIR, RFID and PTS in the Postharvest Supply Chain of Fruits and Vegetables
Imaging technology is an advanced method used by the food and agro-allied industries to monitor changes in food quality [51]. This technology includes computer vision (CV), hyperspectral imaging (HSI), multispectral imaging (MSI), thermography, and X-ray imaging. Image technology is particularly useful in detecting and evaluating the external quality attributes (color, geometrical, size, appearance, and surface structure) [52,53], and in some cases, the internal structures (X-rays and hyperspectral imaging) of fruits and vegetables. This technology involves collecting and analyzing spatial information gained from captured images of products, such as color, geometrical, size, appearance, and surface structure. The application of imaging technology in postharvest supply chains is mainly limited to surface detection. The surface properties of an object can be detected due to the interaction of light. A typical imaging system consists of a CCD camera, a light source, a computer, and related software ( Figure 2). The camera captures the images of the product based on the region of interest. The captured images are then processed to evaluate and quantify the quality changes that have occurred during a particular postharvest operation. The image processing steps often consist of image acquisition, segmentation, feature extraction and recognition, classification, and interpretation [54][55][56][57]. In order to adequately discriminate and analyze the captured numerous images, chemometrics and deep learning methods are often employed. These methods have already been found reliable in quantifying the accuracy of processed images and associated quality changes of fruits and vegetables in the postharvest supply chains (Table 1) The applications of different imaging and smart digital technologies in monitoring the quality of fruits and vegetables in postharvest cold chains are summarized in Table 1. These technologies include computer vision (CV), hyperspectral imaging (HSI), multispectral imaging (MSI), X-ray imaging, spectroscopy, multi-sensors, E-nose, and acoustic impulse response.
From Table 1, the majority of the study was done using CV [58][59][60][61][62][63][64][65][66]. This involves the capturing of images of a product using a digital camera, and the ability of computers to understand the processed image data using computational intelligence tools (e.g., chemometrics, deep learning) [54,55]. This imaging method is rapid, reliable, and consistent. However, this technique has some limitations such as the use of artificial lighting during image capturing and the inability to detect internal attributes. Table 1 further shows that the bulk of the studies using CV was on cold storage, mostly to monitor and detect spoilage, chilling injury, and shelf-life of grapes, lettuce, tomato, zucchini, banana, strawberry, oranges, and mango [58][59][60][61][62][63][64]66]. CV with several chemometric and statistical analytic approaches was able to quantify quality losses with 75-92% accuracy (Table 1). Only a single study explored the application of CV in quantifying the quality losses of lettuce as a result of packaging material used, with 86% accuracy (Table 1) [65].
Hyperspectral and multispectral imaging (HSI and MSI) are advancements of computer vision, which involves the capturing of image data at a different wavelength (e.g., continuous 400-1700 nm in steps of 1 nm for HSI and targeted 400-1100 nm in steps of 20 nm for MSI) across the electromagnetic spectrum [67][68][69][70]. Hyperspectral imaging particularly integrates both imaging and spectroscopy features to simultaneously gather spectral and spatial information from a product, thus making it a more powerful imaging technology compare to CV and multispectral imaging [71]. Both In order to adequately discriminate and analyze the captured numerous images, chemometrics and deep learning methods are often employed. These methods have already been found reliable in quantifying the accuracy of processed images and associated quality changes of fruits and vegetables in the postharvest supply chains (Table 1) The applications of different imaging and smart digital technologies in monitoring the quality of fruits and vegetables in postharvest cold chains are summarized in Table 1. These technologies include computer vision (CV), hyperspectral imaging (HSI), multispectral imaging (MSI), X-ray imaging, spectroscopy, multi-sensors, E-nose, and acoustic impulse response.
From Table 1, the majority of the study was done using CV [58][59][60][61][62][63][64][65][66]. This involves the capturing of images of a product using a digital camera, and the ability of computers to understand the processed image data using computational intelligence tools (e.g., chemometrics, deep learning) [54,55]. This imaging method is rapid, reliable, and consistent. However, this technique has some limitations such as the use of artificial lighting during image capturing and the inability to detect internal attributes. Table 1 further shows that the bulk of the studies using CV was on cold storage, mostly to monitor and detect spoilage, chilling injury, and shelf-life of grapes, lettuce, tomato, zucchini, banana, strawberry, oranges, and mango [58][59][60][61][62][63][64]66]. CV with several chemometric and statistical analytic approaches was able to quantify quality losses with 75-92% accuracy (Table 1). Only a single study explored the application of CV in quantifying the quality losses of lettuce as a result of packaging material used, with 86% accuracy (Table 1) [65].
Hyperspectral and multispectral imaging (HSI and MSI) are advancements of computer vision, which involves the capturing of image data at a different wavelength (e.g., continuous 400-1700 nm in steps of 1 nm for HSI and targeted 400-1100 nm in steps of 20 nm for MSI) across the electromagnetic spectrum [67][68][69][70]. Hyperspectral imaging particularly integrates both imaging and spectroscopy features to simultaneously gather spectral and spatial information from a product, thus making it a more powerful imaging technology compare to CV and multispectral imaging [71]. Both hyperspectral and multispectral imaging technologies can detect internal and external quality attributes of fresh produce. However, they also require artificial lightning, are very sensitive to environmental conditions, have limited penetration depth, and are very expensive to use. Closely following CV, several studies have been conducted on the application of hyperspectral imaging in monitoring the quality of spinach, mushrooms, cucumber, mango, apples, and citrus fruit during cold storage and packaging (Table 1) [72][73][74][75]. This imaging system coupled with PCA, CFS, GIA, SFS, SLR, LDA, kNN, and NN was able to give 89-98% accuracy (Table 1). However, not so much for multispectral imaging, as only two studies quantified the quality losses of mangoes during the cold storage, with a classification rate of ≈92% using PLS-DA (Table 1) [70,76].
Spectroscopy, which is the study of the interaction of electromagnetic waves, including ultraviolet, visible, and infrared spectra, has been applied to monitor and optimize the cold storage process of peach and mango (Table 1). Although only a few studies have been carried out (Table 1) [77,78], this spectral approach can however give an accurate prediction (≈96%) of the total soluble solids, and phenolic content of peaches during cold storage. This technique gives the advantage of repeatable spectral data and provides high resolution of spectra. Additionally, this method is toxic-free. Nevertheless, the spectra data often contains redundant information due to hundreds of spectral variables, limited sensitivity to minor components, and complicated analysis.
X-ray imaging was applied to detect both the internal disorder and external changes (firmness) of pears and kiwifruit during cold storage, respectively (Table 1) [11,79]. This technology involves the production of electromagnetic radiation by an X-ray tube when passed through a product to absorb part of X-ray beam energy [80]. Using SVM, FEA, NN, and LDA, X-ray adequately quantified the columella firmness of kiwifruit and discriminated healthy pear from defective ones, with accuracy ranging from 90% to 95% (Table 1).
Other techniques used to monitor the quality of fruits and vegetables in postharvest cold chains are multi-sensors, electronic nose (E-nose), acoustic impulse response, radio frequency identification (RFID), and printed sensors (PTS). In this study, multi-sensors involves the use of numerous sensors placed at different locations on the produce and in the cold chain equipment (storage container or transport vehicle) to capture important quality attributes (e.g., color, firmness) and food losses (weight loss, temperature, time) metrics [81]. Data from the sensors are processed using sensor fusion (soft sensors). Soft sensors are virtual software code to process multiple sensor information for identified quality classifiers and for the development of warning systems (e.g., quality decline in fruits) [82]. They can be developed using different methods including mechanistic modeling based on physics of specific measured quality and food loss metrics, statistical modeling based on low-level representations in the feature space, and chemometrics or deep learning-based sparse representation techniques for multi-modal event modeling. Figure 3 depicts the use of multiple sensors (e.g., temperature, humidity) coupled with imaging technology during the cold storage of fruits and vegetables. From Table 1, only two studies applied multi-sensors to improve the accuracy of continuous sensor data acquisition in order to enhance transparency and traceability of the cold storage and transportation logistics of pear [83,84]. The multi-sensors monitored critical parameters that affect the quality attributes of fresh produce, including temperature and relative humidity (using portable low-energy-demanding temperature and humidity sensors). The detection of these parameters during shipment allows for effective control of safety and quality changes of the pear. Similarly, a study was reported on the use of an electronic nose (an instrument used to detect volatile organic compounds) [85] and acoustic impulse response to evaluate the quality of tomato and apple during cold storage, respectively ( Table 1). The accuracy of the measured quality attribute (firmness) and mass loss data using ANN was ≈85%. This value is lower than those obtained using the aforementioned imaging technologies. This could be because of the complex nature of the method, which is based on the measurement of the sound emitted by fruit as it vibrates in response to a gentle tap with a small pendulum [86]. compared to RFID [91]. However, their application as tools in monitoring and optimizing cold chain processes is scarce.
Furthermore, there is no study on the application of imaging technology and smart digital technologies to monitor food quality losses during the transportation of fruits and vegetables. This is surprising considering that food losses in the transportation stage of the food supply chain can be as high as 30% as in the case of Poland, for example [92]. For this reason, future studies on the application of imaging technology, spectroscopy, multi-sensors, electronic nose, acoustic impulse response, RFID, PTS in the postharvest cold chain of fruits and vegetables should focus on the transportation chain.  RFID has also been applied as an advanced tool for identifying internal and external changes in the physical, biochemical, and physiological processes of packaged food [87][88][89]. This non-contact identification communication technology can automatically identify multiple objects moving at high-speed, and therefore can be applied in the transport cold chain, specifically as an IoT enabler [90]. Similarly, PTS which uses the printing process, such as inkjet printing, nanoimprinting, screen printing, etc., to prepare electronic circuits on a flexible substrate enables the monitoring of temperature, moisture, pressure, and motion of fresh produce [91]. This technology has the advantage of flexibility when printed on substrates, ease of distribution, and low cost especially when compared to RFID [91]. However, their application as tools in monitoring and optimizing cold chain processes is scarce.
Furthermore, there is no study on the application of imaging technology and smart digital technologies to monitor food quality losses during the transportation of fruits and vegetables. This is surprising considering that food losses in the transportation stage of the food supply chain can be as high as 30% as in the case of Poland, for example [92]. For this reason, future studies on the application of imaging technology, spectroscopy, multi-sensors, electronic nose, acoustic impulse response, RFID, PTS in the postharvest cold chain of fruits and vegetables should focus on the transportation chain.

Imaging technology computer vision Storage
To provide an application designed for embedded devices such as mobile Android smartphones to objectivize the measurements using machine vision Tomato and zucchini HA The proposed method successfully achieved by predicting the quality characteristics of the products during cold storage [62] 12. Imaging technology computer vision system Storage To develop a computer vision system to predict the quality levels of vegetables during storage Lettuce SNK The color information measured by the computer vision system achieved nondestructively to evaluate the quality level of iceberg lettuce with R 2 of 0.77 [63] 13

Imaging technology computer vision Storage
To demonstrate the applicability of Random Forests (RF) for estimating the internal qualities of fruits based on peel color

Mango RF
The relationship between peel color and fruit quality was strongly found in different storage temperatures with a correlation coefficient up to 0.98 [64] 14.
Imaging technology X-ray Storage To design a methodology for sorting fruits with X-ray image processing and pattern recognition techniques Kiwifruits NN, LDA The model built with LDA predicted the columella firmness in kiwifruit with a 94.6% [11] 15. Imaging technology thermography Packaging To study the temperature distribution on a pallet of fruits during plastic boxes and cardboard packaging Apples ANN Thermal imaging showed the cardboard boxes to be a better packaging material for apples compared to plastic boxes [93]  19.

Imaging technology computer vision systems Storage
To evaluate fruit quality nondestructively by computer vision Grapes RF The system achieved a cross-validation classification accuracy up to 92% which support its capability of powerfully, flexibly, and continuously monitoring the quality of the complete production along the whole supply chain [66] 20.

Visible/shortwave near-infrared spectroscopy Storage
To establish optimal spectral models for the assessment of fruits in actual production and therefore obtain early warning information of mechanical injuries during storage

Peach GA
The optimal spectral model through the GA-PLS method found prediction accuracy ranges from 0. 89

Application of Mathematical Modeling Techniques in the Postharvest Supply Chain of Fruits and Vegetables
Modeling is the act of representing phenomena or processes in such a way as to explicitly describe an observed system and to predict or optimize different behaviors, parameters, and conditions [94]. Mathematical modeling is essential for efficient engineering design and optimization. With adequate mathematical models, undesirable effects that significantly causes food losses such as weight loss, or quality changes can be predicted, thus cold chain logistics can be optimized or controlled.
Mathematical modeling techniques are becoming increasingly popular as an alternative to expensive and difficult experiments of postharvest cold chain operations as a result of the sophistication and reliability of computers as well as the affordability and availability of modeling software [95][96][97][98][99]. Agricultural and food engineers, and other researchers have over the years developed different mathematical models for postharvest supply chains. Depending on the complexity, these different modeling techniques have been developed to predict heat and mass transfer, fluid flow, and quality changes in and around fresh produce. Gas exchange and in-depth understanding of migration from packaging material to fresh produce have been described using mathematical models. Additionally, several deterministic, stochastic and kinetic models have also been developed to predict the overall quality of fresh produce, mass loss, fluid flow, and heat and mass transfer during the transportation and storage of fruits and vegetables [98,[100][101][102][103][104][105][106].
In this study, mathematical models used to enhance cold chain operations chain can be separated into six different types based on their specific process application, namely: migration models (MM), membrane gas separation (MGS) model, heat and mass transfer (HMT) model, structural behavior models (SBM), stochastic models (SM), and kinetics rate models (KRM) ( Table 2). MM are often used to study the migration of organic compounds such as Benzophenone, Diisobutyl phthalate, and Phenanthrene from packaging material to fresh produce. On the other hand, MGS models are often used in a modified atmosphere (e.g., modified atmosphere storage or modified atmosphere packaging) to study the lifespan of fresh produce by reducing the respiration rate through the adequate regulation of atmospheric conditions (e.g., CO 2 , O 2 ). Additionally, to abstract the packaging of different fruits and vegetables [107]. This type of model works on four ideal flow patterns for a mixed gas module including co-current flow, cross flow, counter-current flow, and perfect mixing [108][109][110].
HMT modeling also called hygrothermal modeling involves using a numerical physics-based method such as computational fluid dynamic (CFD) to solve the governing partial differential equations of heat and mass transport phenomena in a system, often using finite element analysis (FEA) [100,105,111,112]. HMT models describe the underlining physics inside fresh produce and how they are affected by the surrounding conditions ( Figure 4). They generally are independent of experimental calibration and validation. HMT models can also be used to investigate the impact of different packaging designs on the convective heat transfer rate of fruits and the surrounding [105]. By reducing heat transfer from the outside environment, effective packaging can help to shield the product from temperature variation in the storage and transport process. In addition, increasing the packaging heat transfer resistance can also ensure temperature stability of the fresh agricultural produce [113,114]. There is therefore a need for simulation tools and numerical models that analyze all factors affecting optimum packaging design.  SBM involves the study of structural and mechanical properties of packaging materials for fresh produce using FEA (Table 2). This modeling approach encompasses a geometric representation, material representation, and boundary conditions (loading and restraints).
SM involves predicting the variability of certain generated data following their probability distribution, and then evaluating of results statistics until the minimum error becomes constant [115]. These models are often used to analyze the effect of biological variability on food quality and losses during the supply chain, and also to quantify the efficiency of the cold chain technology [102,103,116]. Although they do not provide a fundamental understanding of the underlying physics, they are however very reliable and flexible.
KRM are temperature-dependent and are frequently used to study the combination of the rate of reaction with the material balance to predict the behavior of a particular system. Nutritional and sensory qualities of fresh agricultural produce in the postharvest supply chain can be quantified based on kinetics, such as zero-order, first-order, second-order, mixed order, or higher-order reactions [117].
SM involves predicting the variability of certain generated data following their probability distribution, and then evaluating of results statistics until the minimum error becomes constant [115]. These models are often used to analyze the effect of biological variability on food quality and losses during the supply chain, and also to quantify the efficiency of the cold chain technology [102,103,116]. Although they do not provide a fundamental understanding of the underlying physics, they are however very reliable and flexible.
KRM are temperature-dependent and are frequently used to study the combination of the rate of reaction with the material balance to predict the behavior of a particular system. Nutritional and sensory qualities of fresh agricultural produce in the postharvest supply chain can be quantified based on kinetics, such as zero-order, first-order, second-order, mixed order, or higher-order reactions [117].

Heat and mass transfer modeling Packaging
To develop a 3D heat and mass transfer model of a fresh food produce packaging system to predict airflow and heat transfer characteristics Citrus Airflow, temperature, and turbulence property distribution inside a single product are nonuniform; good correlation between air velocity and temperature; the accuracy of the model depends on geometry, thermal, physical, and chemical properties of the package, the cooling air, and the produce [120] 13. Membrane gas separation modeling Packaging; storage; transportation To develop a mathematical model to investigate the effect of stage cut on the gas separation performance of hollow fiber membrane modules

Avocado
The model utilizes pure gas permeances of membrane material to predict the mixed gas separation performance; the performance of gas separation declines over time as they age with exposure to pressure, temperature, and contaminant loading [121]  To study the influence of box material and plastic cover on the distribution of 1-MCP (methyl cyclopropane) in cold storage to delay ripening of fruit Apple Diffusion, convection, and adsorption were modeled to simulate the temporal distributions of 1-MCP inside a storage container, boxes, and fruits; the plastic cover does not affect the adsorption of 1-MCP; wooden boxes notably adsorbed 1-MCP from the treatment atmosphere and may reduce the efficacy [122] 16.

Kinetic modeling Packaging
To investigate the impact of temperature and relative humidity on fruit transpiration rate (TR); to develop a prediction model for quantifying TR; to integrate TR model into engineering packaging design and quantify

Strawberries
Temperature and relative humidity have a significant impact on the transpiration rate of strawberries; increase in relative humidity increases TR; decreasing temperature decreases TR; the model predicted the water vapor barrier properties required for maintaining optimal relative humidity inside package [123] 17.

Heat and mass transfer modeling Storage
To model the airflow and temperature distribution in a natural connection thermal energy storage refrigerator; to determine the performance of the refrigerator with different phase change material (PCM) vertical-horizontal -Horizontal PCM configuration produces lower compartment temperatures than a vertical configuration; combining horizontal and vertical configuration gives better design performance [140] 18.

Migration modeling Packaging; storage
To study the diffusion, convection and adsorption of 1-MCP gas in cold stores; to understand the mechanism of 3D distribution of 1-MCP

Apple
The model demonstrated the absence of significant spatial variation of 1-MCP gas in a container; diffusion-convection in air and diffusion-adsorption in the product [124] 19.

Heat and mass transfer modeling Packaging
To develop a porous medium model was develop on volume averaging of transport equations of momentum and 1-MCP in air and product

Apple
The velocity field in and around the stack was well reproduced by the porous medium model; the porosity, skin mass transfer coefficient, and specific surface area strongly affected the simulation process [100] 20.
Heat and mass transfer modeling Packaging (bunch carry bag and plastic liners) To determine the effect of the packaging component and box stacking on airflow, heat, and mass transfer rate Grape The use of carry bag resulted in an increase in the cooling time; the addition of plastic liner over the bunch carry bag increased cooling time; moisture loss was most prevented using nonperforated liners; CFD simulation determined optimum

Kinetic model Storage
To describe the product time-temperature history along the cold chain; the model considered front and rear air circulation in the cold room

Apples
The model adequately predicted cooling rate, the temperature at different positions, and weight loss; the model has a short CPU computational time (<1 s) when compared to CFD models. This enabled a rapid evaluation of input parameters such as air temperature [106] 24.

Stochastic modeling Storage
To evaluate the quality of perishable foods using a generic algorithm + center of gravity model; to estimate the environmental level

Fruits
The algorithm adequately predicted temperature and humidity levels; the algorithm was integrated to gauge the use of RFID (radio frequency identification) and sensors for real-time information gathering [102] 25. Stochastic modeling Transport; storage To estimate the heat generation and also the cooling efficiency during cold transport and storage chain Banana 10% accuracy for the heat of respiration and cooling efficiency was reached after 4-7.5 days of transport [103] 26.
Membrane gas separation modeling Packaging; storage To describe product respiration and gas exchange through package using Michaelis-Menten kinetics + Fick's equation; taking into account diffusive gas permeation through packaging film and perforation respiration rate and storage temperature

Tomato
The model adequately predicted the required package surface area and perforation diameter to achieve a specific O 2 concentration in the headspace; the model can be used to set a specific equilibrium concentration of O 2 and CO 2 by modifying the configuration of the package [126]  The model was able to describe the cooling behavior and uniformity of fruits in fiberboard boxes; there was large variability in convective heat transfer coefficients from the apples and pear filling; the fruit shape affects the model accuracy [130] 33.

Heat and mass transfer modeling Packaging
To develop a 3D HMT model to quantify cooling behavior of 10 different carton designs based on cooling rate, energy consumption, uniformity, weight loss, and chilling injury of apples

Apples
The model adequately quantified the effect of airflow, and packaging design on the product quality; vent area, shape and number of vent have less impact on the fruit cooling; homogeneity and symmetry of packaging vent positions have more impact on the fruit cooling rate; the model proposed airflow velocity between 0.4 and 1.0 m/s [131] 34.

Structural behavior modeling Packaging
To develop a validated structural behavior model to study the structural behavior of VCP packages by considering the geometrical nonlinearities of the packages

Fruits and vegetables
The model accurately predicted the compression strength of the corrugated paperboard, control package, and standard vent package; compression strength of the standard vent packages was found to be linearly affected by paperboard liner thickness; increasing and decreasing the baseline liner thickness of the standard vent package by 80% resulted in an increase and decrease in compression strength by about 15% and 19%, respectively; from the contact FEA model, maximum Von Mises stress was produced at the corners of the package; Von Mises stress was significantly affected by the coefficient of friction [132] 35.
Heat and mass transfer modeling Storage; transport To investigate the airflow distribution inside two types of refrigerated shipping containers (T-bar floor and flat floor) used for transporting fresh fruit handling The airflow distribution in the two container designs was markedly different. Good agreement was found between measured and predicted values of air velocities. The reefer with T-bar floor design exhibited a noticeable reduction of air recirculation zone and enhanced uniform vertical air movement compared to the reefer with flat floor design [138]

Kinetic modeling Storage
To examine the effect of relative humidity (RH) conditions on the shelf life of strawberries including both the sensory and nutritional quality; to study the kinetics of sensory and chemical changes occurring in strawberries during storage by comparing three kinetic models; to examine and predict through modeling the waste that would occur depending on the storage conditions Strawberries Weight loss significantly increased when storage RH decreased; the weight loss was correlated with the changes that occurred in visual appearance and chemical properties. Overall appearance (i.e., the average score of color and shriveling), was modeled with a zero-order kinetics model for the various RH conditions; lower RH increased the rate of appearance deterioration thereby reducing the remaining shelf life of strawberries. The Weibull model adequately fitted the chemical properties data and it was found to be an important tool in describing the changes that occur with varying storage conditions [98] 37.

Stochastic modeling Whole chain
To develop an Agro-Chain Greenhouse gas emissions calculator (ACGE calculator) to calculate the percentage of food losses per chain stage

Cut vegetables
The ACGE calculator can be applied for understanding the impacts of different operations along a postharvest cold chain and for analyzing chain configuration scenarios: such as weighing impacts of the operations/impacts along the chain, comparison between various options for supplying a specific food component, and comparing a reference scenario with an 'improved scenario'; low temperature in the cold chain may result in extended retail shelf life and a lower percentage of losses, but will cost more energy; application of protective packaging leads to a reduction of losses, but at the cost of the packaging [141] 38.
Concerning the transportation supply chain, only a few authors have applied SM, HMT models, and KRM to estimate heat generation, cooling efficiency, temperature distribution, and the expected fraction of perishable products (Table 2). These products include spinach, peaches, and banana [103,104,143].
The above analysis shows that mathematical models have been widely developed and applied in the packaging and storage of fruits and vegetables with the view to improve quality and reduce food losses. However, not so much modeling study for the transportation supply chain of fruits and vegetables. Future mathematical modeling studies should focus on the transportation supply chain taking into account the shipment time, the varying environmental conditions (e.g., temperature, humidity, and airflow), packaging, and vehicular movement. In addition, the KRM could be integrated with HMT models to give more insight into what extent the quality attributes of fruits and vegetables are preserved better and also to quantify the effect of other drivers for decay processes (e.g., relative humidity, light) in the fresh produce supply chain. This can be achieved by developing a digital twin of the product (see Section 4.2).

Emerging Opportunities in Reducing Food Losses in the Postharvest Supply Chain
The application of IoT and digital twins in optimizing shelf-life and reducing food losses during an entire shipment has gained significant interest in recent years. This section analyzes the potential of applying the Internet of Things (IoT) and digital twins in reducing food losses in the postharvest supply chains of fruits and vegetables.

Application of IoT in the Postharvest Supply Chain of Fruits and Vegetables
IoT has emerged in different fields such as e-commerce [145,146], manufacturing [147,148], education [149][150][151], medicine and healthcare [152][153][154][155][156][157][158], and agriculture [159,160]. This is because of the enormous number of devices connected to the Internet, as well as the widely available internet and data storage service providers [148,161,162]. Basically, IoT allows humans, objects, and things to connect and communicate at any time and anywhere. The European Commission Information Society defined IoT as different things exhibiting identical and virtual personalities, connecting and communicating in a smart space using intelligent interfaces within social, economical, and user contexts [163].
The IoT system consists of networks of physical objects that contain embedded technology to sense, communicate, and interact with their internal states or the external environment [164]. The key enablers for a typical IoT system include RFID, printed sensors, web service, machine-to-machine communication (M2M), WSN, imaging system, multi-sensors, cloud, blockchain, among others, but not necessarily altogether [87,91,147,161,165,166].
The application of IoT is well-established in various agricultural production sectors such as controlled environment agriculture, open-field agriculture, and livestock applications [167,168]. In recent years, the use of IoT has gained significant interest in the food industry for product tracking, traceability and the monitoring of environmental conditions (e.g., temperature, humidity), weight loss, and the overall quality loss in the postharvest supply chain [87,161,165,[169][170][171]. This technology has also received significant attention in developing intelligent packaging in the food sector [172][173][174].
Intelligent packaging involves the use of sensors (biosensor, printed, chemical, and gas sensor) and indicators (time-temperature indicators, freshness indicators, gas indicators, and integrity indicators) to detect biological, chemical, or gaseous changes from packaged fresh produce [24,25,27,91,170,[174][175][176]. The sensor-based RFID tags as an example can detect hygrothermal and chemical changes (e.g., temperature, CO 2 , light exposure, pH, etc.) of the fresh produce in the post-harvest supply chain [25,87,175]. The timely information obtained within the package system can be used to inform stakeholders in the supply chain of an event that may damage the packaging material or the fresh produce itself.
Generally, the application of IoT in different cold chain processes results in a large amount of real-time data which can pave the way for new computational approaches such as artificial intelligence and big data analytics [161,177]. This data will help various stakeholders in the supply chain control and optimize the cold chain technology to reduce quality loss and also make informed decisions regarding food safety. However, the application of IoT in controlling cold chain technologies in order to reduce food losses in the supply chain of fruits and vegetables is still inadequate. Table 3 shows that several studies applied IoT in tracking and tracing temperature changes and food quality during the shipment of fruits and vegetables in the past decade [20,159,165,178,179]. Two studies applied IoT on the packaging of fruits and vegetables, as well as during cold storage [159,178]. The application of IoT in the shipment of fruits and vegetables is accompanied by multi-sensors such as temperature and humidity sensors, light exposure sensors, and global positioning system (GPS) sensors ( Figure 5) during shipment of fruits and vegetables (Table 3). These sensors are installed in the food containers to monitor the changes in the environmental cold chain conditions such as air temperature, airspeed, light exposure, and relative humidity using a sensor data fusion (soft sensors). They are connected to a wireless network and computers to communicate with control stations, producers, or other stakeholders in the supply chain. The collected data can then serve as input data in analyzing the changes in the food attributes (e.g., weight loss, shelf life, nutritional, or sensory qualities), using a mechanistic physics-based model or a digital twin. It is worth mentioning that multi-sensors (e.g., chemical sensors, biosensors, etc.), imaging systems (see Section 3.1), E-nose (see Section 3.1), spectroscopy (see Section 3.1), and AIR (see Section 3.1) can also be used to directly measure changes in some quality attributes of fresh produce in the postharvest supply chain. IoT has become a very important tool in monitoring and controlling the process conditions of food, allowing the controllers to implement proper decisions. All of these can help to significantly reduce food losses. More so, the reduced cost of software and hardware wireless devices [180], digital sensors, accompanied by IoT technology in food transportation, packaging, and/or storage already increases the potential of IoT as a veritable and sustainable tool for reducing food losses. producers, or other stakeholders in the supply chain. The collected data can then serve as input data in analyzing the changes in the food attributes (e.g., weight loss, shelf life, nutritional, or sensory qualities), using a mechanistic physics-based model or a digital twin. It is worth mentioning that multi-sensors (e.g., chemical sensors, biosensors, etc.), imaging systems (see Section 3.1), E-nose (see Section 3.1), spectroscopy (see Section 3.1), and AIR (see Section 3.1) can also be used to directly measure changes in some quality attributes of fresh produce in the postharvest supply chain. IoT has become a very important tool in monitoring and controlling the process conditions of food, allowing the controllers to implement proper decisions. All of these can help to significantly reduce food losses. More so, the reduced cost of software and hardware wireless devices [180], digital sensors, accompanied by IoT technology in food transportation, packaging, and/or storage already increases the potential of IoT as a veritable and sustainable tool for reducing food losses.

Digital Twin as an Advanced Tool in Reducing Food Losses in the Postharvest Supply Chain of Fruits and Vegetables
Digital twins have recently gained significant interest in postharvest engineering, as a way of expanding mathematical models and computer simulations by linking input data to the solutions implemented after the simulation study [181]. Simply, a digital twin of a product can be defined as a virtual model of the product's real-life representation containing all realistic characteristics. The virtual model contains all essential elements, including geometrical components and material properties, and accurately and realistically simulates all relevant physics and their kinetics throughout the product's life-cycle. Digital twins can be mechanistic (physics-based), statistical (empirical-based), and intelligent (e.g., machine learning, deep learning) in nature. However, only the physics-based mechanistic digital twins can adequately evaluate the processes that cause quality loss in fresh produce. This involves linking measured sensor data of the environmental conditions

Digital Twin as an Advanced Tool in Reducing Food Losses in the Postharvest Supply Chain of Fruits and Vegetables
Digital twins have recently gained significant interest in postharvest engineering, as a way of expanding mathematical models and computer simulations by linking input data to the solutions implemented after the simulation study [181]. Simply, a digital twin of a product can be defined as a virtual model of the product's real-life representation containing all realistic characteristics.
The virtual model contains all essential elements, including geometrical components and material properties, and accurately and realistically simulates all relevant physics and their kinetics throughout the product's life-cycle. Digital twins can be mechanistic (physics-based), statistical (empirical-based), and intelligent (e.g., machine learning, deep learning) in nature. However, only the physics-based mechanistic digital twins can adequately evaluate the processes that cause quality loss in fresh produce. This involves linking measured sensor data of the environmental conditions (e.g., the air temperature around the fruit), as input data to the currently still uncharted product's quality evolution of fresh produce (Figure 6), preferable in a real-time update, using a physics-based model. In this way, the digital replica reacts hygrothermally and biologically in the same way as its physical counterpart (a real fresh fruit or vegetable). By enriching current real-time monitoring capabilities using sensors, digital twins can be used to diagnose and predict potential problems in the supply chain that will increase food losses. These problems can be caused by physiological (e.g., chilling injury), thermal/hygrothermal (e.g., mass loss), biotic (e.g., phytosanitary pests, pathogens), and mechanical (e.g., puncture injury, bruising) effects. This unique attribute shows that the digital twin has the potential of incorporating several physics-based thermal, physiological, mechanical, biological, and decay models for corresponding quality and shelf-life metrics. This insight can then help remotely analyze the quality performance of the fresh produce in each shipment and also predict the remaining shelf life days. Based on the analysis, a proactive preventive measure can be taken early to reduce quality losses throughout the cold chain. Such measures can also help predict and optimize future product quality and process design.
As a next step, digital twins can be implemented in real-time with actual multiple shipments. This is expedited with the integration of the already available big data technologies (e.g., IoT devices, blockchain devices, soft sensors, cloud systems, etc.) [182,183]. However, such a system is not yet in place, to the best of our knowledge (Table 3). From Table 3, only two studies developed a digital twin for mango. The mechanistic models developed for these studies included HMT models, as well as KMR for various quality attributes such as firmness, soluble solids content, and vitamin content [184] [185]. The air temperature data of the actual mango cold chain, collected as input from a temperature sensor was linked to these models to create a digital twin of a virtual mango fruit [184,185]. With the digital twins, the fruit quality evolution was quantified for multiple overseas shipments. However, the twin did not use other significant input data history such as the humidity of the products at the different supply chains. The temperature data collected was not in real-time, but offline, so aposteriori. In addition, the digital twin did not integrate models to estimate the mass loss, chilling injury, and other biochemical models which are important for quantifying food losses in the entire cold chain. The use of packaging material was also not considered (Table 2). By enriching current real-time monitoring capabilities using sensors, digital twins can be used to diagnose and predict potential problems in the supply chain that will increase food losses. These problems can be caused by physiological (e.g., chilling injury), hygrothermal (e.g., mass loss), biotic (e.g., phytosanitary pests, pathogens), and mechanical (e.g., puncture injury, bruising) effects. This unique attribute shows that the digital twin has the potential of incorporating several physics-based thermal, physiological, mechanical, biological, and decay models for corresponding quality and shelf-life metrics. This insight can then help remotely analyze the quality performance of the fresh produce in each shipment and also predict the remaining shelf life days. Based on the analysis, a proactive preventive measure can be taken early to reduce quality losses throughout the cold chain. Such measures can also help predict and optimize future product quality and process design.
As a next step, digital twins can be implemented in real-time with actual multiple shipments. This is expedited with the integration of the already available big data technologies (e.g., IoT devices, blockchain devices, soft sensors, cloud systems, etc.) [182,183]. However, such a system is not yet in place, to the best of our knowledge (Table 3). From Table 3, only two studies developed a digital twin for mango. The mechanistic models developed for these studies included HMT models, as well as KMR for various quality attributes such as firmness, soluble solids content, and vitamin content [184] [185]. The air temperature data of the actual mango cold chain, collected as input from a temperature sensor was linked to these models to create a digital twin of a virtual mango fruit [184,185]. With the digital twins, the fruit quality evolution was quantified for multiple overseas shipments. However, the twin did not use other significant input data history such as the humidity of the products at the different supply chains. The temperature data collected was not in real-time, but offline, so a-posteriori. In addition, the digital twin did not integrate models to estimate the mass loss, chilling injury, and other biochemical models which are important for quantifying food losses in the entire cold chain (Table 2).

Internet of Things Transportation
To build an intelligent model for food quality monitoring and control in a multi-temperature distribution center

Food
The food spoilage rate of food reduced during transportation due to the real-time food quality monitoring and control using an intelligent model [165] 2.

Internet of Things Transportation
To improve transport efficiency in order to save the fruits from spoilage

Fruits
IoT system in a truck refrigerator adequately used to monitor the quality condition of fruits during transport [20] 3. Internet of Things Packaging, storage, and transportation To monitor food quality and safety Fruits and vegetables IoT obtain real-time food traceability and monitoring data to control the logistic and process parameters causing quality loss [178] 4. Internet of Things Packaging, storage, and transportation To reduce the food losses during the food chains since 50% is lost Agro-food IoT used to automate the packaging system with proper tracking and monitor the temperature of the produce at cold storage and during transportation [159] 5.

Internet of Things Transportation
To monitor the temperature changes and inefficient management during transportation as the insufficient temperature can pose a high risk to food quality Fruit and vegetables Application of IoT in food chains facilitated safety, intelligence, and deliver quick decisions [179] 6. Digital twin Storage; transport To develop a digital twin for the cold chain shipment of fruits Mango Based on measured environmental conditions, the impact of shipment duration, heat of respiration, airspeed and delivery air temperature history on quality of mango for different transport pathways was easily quantified using a digital twin [184] 7.
Digital twin All chain To gain a better insight into how fruits behave under convective cooling Mango At low speeds, a more uniform cooling can be achieved and thereby a more homogeneous quality decay within the mango; digital twin was able to evaluate the heterogeneity of the temperature field and identified the zones with the highest temperature inside the product, which can be valuable information for the placement of temperature probes [185]

Future Opportunities to Reduce Food Losses in the Postharvest Supply Chain of Fruits and Vegetables
With the gradual depletion of resources, there is a need to look at sustainable ways of achieving food security by reducing food losses in the postharvest supply chain. Looking ahead, the major challenges that cause food losses in the postharvest supply chain have to be addressed.
One emerging field is the development of intelligent packaging systems to reduce food losses, especially fresh agricultural produce. Intelligent packaging systems through the use of internal and external monitors (sensors, nanosensors, and indicators) provide valuable information on the interaction of food with the packaging material and the environment at different phases of processing, transportation, and storage. It also takes into consideration the ergonomic features of the packaging to reduce inconvenience in the transportation, storage, use, and eventual disposal of the packaging material [174,186]. With the recent interest and development in intelligent packaging, there is a need to integrate the sensors, indicators, and data carriers technologically to provide real-time information about fresh food in different cold chain logistics through the use of IoT based technologies and digital twin.
Although the potential of a digital twin in minimizing quality losses and increasing the shelf-life of fresh produce has been demonstrated (Table 3), the holistic implementation of digital twins in the entire value chain (from planting-fork) and for a wide range of fresh produce is yet to be demonstrated. The existing digital twins (Table 3) should be improved upon to include other relevant models that simulate thermal, physiological, mechanical, and biological damages that cause food losses in the postharvest supply chain. For example, a mass loss model can be included to quantify the salable weight at the end of the chain. This can help quantify the market value of fresh produce due to the subjective acceptable consumer product appearance. Additionally, tropical fruits such as banana or papaya experience chilling injury due to low-temperature storage and long cold chain process (Table 1). Therefore, thermal damage models predicting chilling injury during cold chain processes should be included. The potential of linking pathogens with decay severity should also be a future focus. Future digital twins should also capture the biological variability of fresh produce in order to give more realistic actionable metrics as multiple fresh produce have different individual pre-harvest and postharvest history. This can be achieved by integrating stochastic simulations (e.g., Monte Carlo simulations) with the existing digital twin physics-based models.
An additional future focus is to integrate IoT systems (including soft sensors) in real-time with digital twins. This real-time coupling will enable stakeholders to monitor and control each supply chain shipment at all times and take dynamically corrective measures to reduce quality loss and increase the remaining shelf-life days. Furthermore, by adding more "intelligence" to the coupled IoT and digital twins system, the cold chain technology (e.g., a refrigerated container) can independently optimize its process parameters to increase the shelf life of fresh produce and reduce food losses of the entire shipment. This added value can be easily quantified especially in this current time, considering the COVID-19 situation. Due to COVID-19, food producers have seen a decrease in the timely distribution of fresh produce to supply chain retailers. This is attributed to the decrease in transport labor, and longer shipment time because of shipment re-routing. This development has led to increased food losses. As a consequence, for example, about 5 billion US dollar worth of fresh fruits and vegetables were lost in the USA alone during the COVID-19 peak period from March 2020 to June 2020 [187]. With intelligent coupled IoT and digital twins, different processes and cold chain technology can be optimized to reduce the dependency on human labor, faster shipment duration, and possible damages caused by physiological, hygrothermal, mechanical, and biotic factors. A reduction in damages on the fresh produce implies a reduction in food losses.

Conclusions
Fruits and vegetables are important sources of nutrients such as vitamins, minerals, and bioactive compounds, which provide many health benefits. However, due to poor postharvest management processes, large quantities of fruits and vegetables perish before they reach the consumer. Of all the techniques for extending the shelf life of perishable produce, storage at low temperatures is by far the most effective.
This study, therefore, provides in-depth insight on the application of advanced technology in improving food security, by reducing food losses during postharvest cold chain processes for fruits and vegetables. It has been found that:

•
Computer vision, hyperspectral imaging, multispectral imaging, spectroscopy, and X-ray imaging are already widely used in monitoring and optimizing the cold chain processes of fresh agricultural produce.

•
The application of MM, MGS models, HMT models, SBM, SM, and KRM in improving the cold chain processes and evaluating the quality losses of fresh produce is well established. These models can help control and optimize the packaging and storage operations of fruits and vegetables in order to reduce quality losses.

•
IoT is widely applied in monitoring and controlling the process conditions of fresh produce, enabling various stakeholders to implement proper decisions.

•
Digital twins are significant in quantifying the quality evolution of fresh produce in each shipment and also predict the remaining shelf life days.

•
There is a very huge potential for coupling digital twins with big data technologies (IoT devices, printed sensors, RFID, multi-sensors, soft sensors) to monitor, optimize, and make significant changes that will reduce food losses in the postharvest supply chain of fresh produce. However, such a system does not exist.
The augmented insight on the application of emerging technologies can serve as a roadmap for future cold chain studies on fresh agricultural produce.