Geomembranes are generally welded in the field.The seam is considered a critical point for possible injuries and future leaks in the barrier system.Rollin et al.reported that the total number of faults in geomembranes occurred in landfills, ponds and basins, and 55% of them were found at the seams.Stark et al.evaluated seams made by industry and in the field of a polyolefin geomembrane for a water reservoir project.According to the authors, the seams made by industry can minimize dirt and moisture in the seam, ambient temperature changes and wind.After good statistical data work, the results of this research showed that the seams made by industry were 9% stronger than field seams for peel strength and 10% stronger for shear strength.Moreover, seams according to industry data showed less variability.Lavoie et al.evaluated two different HDPE geomembrane samples exhumed from a municipal landfill leachate pond and a sewage treatment aeration pond.The authors conducted thermal, physical and mechanical analyses.The thermogravimetric analyses showed an altered decomposition behavior for the landfill leachate pond sample, probably caused by the interaction with the leachate.The sewage pond sample presented low stress cracking resistance and low tensile elongation at break, compatible with the dynamic mechanic analysis , which showed an increase in the stiffness.Ewais and Rowe studied an HDPE geomembrane with 1.0 mm of the thickness of about five years immersed in synthetic leachate at different temperatures and in air and water at 55 ◦C.The authors observed changes in the sample’s stress crack resistance values before chemical degradation and before antioxidant depletion for temperatures lower than 70 ◦C.The major SCR value decrease was noted for the sample immersed in leachate at 55 ◦C, reaching 26% of the SCR value of the virgin sample.

This behavior was attributed to morphological changes during aging that affected the interlamellar connections due to the annealing that increased the strength of the inter-lamellar connections and the proposed chain disentanglement mechanism.HDPE geomembranes are often utilized as a flow barrier in landfills, mining facilities, canals, waste liquid ponds,mobile vertical farm and farm ponds.This research evaluated three exhumed high-density polyethylene geomembrane samples from two different shrimp farming ponds after 8.25 years of field exposure for the first pond and 3.0 years of field exposure for the second pond.Thermoanalytical and physical analyses were used to understand the final conditions of the HDPE geomembrane samples.Analyses such as thickness, density, carbon black content, melt flow index , tensile properties, stress crack resistance, oxidative-induction time , thermogravimetry,differential scanning calorimetry,and dynamic mechanical analysis were performed.Using an HDPE geomembrane as a liner of a shrimp farming pond is limited to some places globally, such as the northeast region of Brazil.This work aims to provide new data on the HDPE geomembrane performance in shrimp farming pond applications.This work evaluated three high-density polyethylene geomembrane samples exhumed from two different shrimp farming ponds in the northeast of Brazil with 0.8 mm of nominal thicknesses.The samples called “CAM” and “CAM1′′ represent the same geomembrane installed in a shrimp farming pond and were collected after 8.25 years of field exposure.The CAM sample was exhumed from the bottom liner and had been in contact with the salinized water.The CAM1 sample was exhumed from the same pond as the CAM sample, but it was taken from the slope liner and had been in contact for 8.25 years with environmental conditions.The third sample was exhumed from another shrimp farming pond after 3.0 years of service.This pond has the particularity of being covered with an agricultural plastic film.Figs.1 and 2 show, respectively, the shrimp farming pond representing the CAM and CAM1 samples and the shrimp farming pond representing the CAM2 sample.Table 1 presents the salinized water parameters used for shrimp cultivation.

However, nesting habitats and foraging resources for pollinating insects have decreased severely due to intensification in agricultural landscapes.Developing effective mitigation tools to counteract pollinator losses is essential for satisfactory pollinator diversity and sustainable agriculture.In Europe, agri-environment schemes are among the most valuable instruments to reduce the pressure of agriculture on the environment.The new Common Agricultural Policy of the EU also aims at increasing their share by targeting one-fourth of the direct payments to subsidise eco-schemes.AES offer two main tool types to support biodiversity and thus pollination, i.e.productive and non-productive schemes.Organic farming is one of the most favoured productive management systems where farming intensity is reduced with a preference for mechanical methods over agrochemicals.In their meta-analysis, Tuck et al.found that organic farming increased overall species richness by 30 %, with plants benefiting the most.Pollinator insects also profit from organic farming.Flower strips are flower-rich non-crop landscape features created at arable field edges.The annual or perennial strips are formed to enhance invertebrate pest control and pollination by providing forage, shelter, and nesting places.As AES schemes differ in concept, they should be evaluated against each other to prioritise them for benefits to biodiversity.Functional trait-based assessments offer an effective tool for studying community functionality, redundancy and stability in agroecosystems.Different management types can support various arthropod taxa by maintaining specific plant functional traits and high functional trait diversity.Plant traits associated with visibility, quantity, and quality of floral resources are the most important traits shaping pollinator communities.For example, plants with large, conspicuous inflorescences attract pollinators efficiently , while floral patterns resulting from UV reflection are also important signals to several major pollinator groups.Nectar is the primary reward for pollination; therefore, spatial and temporal variation in nectar volume influences pollinator behaviour and survival.As land-use significantly affects pollination-related flower traits , their diversity can indicate the quality and quantity of floral resources and can be used to evaluate AES schemes for pollinators.Different field parts with different management regimes can support biodiversity and pollination to varying degrees.The interior of arable fields covers the largest area of agricultural landscapes; therefore, its ecological condition may be decisive for biodiversity at the landscape scale.Intensive management within field interiors favours plants with ruderal strategies , and selects for species with early flowering and selfing.Crop edges are also shown to be keystone transitional zones for pollinators and support increased plant diversity compared to field interiors as they are expected to contain both typical arable weeds and species originating from grassy margins.

Linear non-crop landscape elements such as grassy margins along field edges are less intensively managed than field interiors and also provide important shelter and forage resources for various taxa in agricultural landscapes.The overall performance of AES schemes on pollination should be quantified by assessing all characteristic parts of the agricultural fields.Landscape-scale factors should also be considered beside local factors when studying the effects of AES schemes on pollinators and their services.The effectiveness of AES varies by landscape heterogeneity that has two main components.Compositional heterogeneity describes the relative amount of habitats in the landscape, whereas configurational heterogeneity refers to their spatial arrangement.Most studies assessed the effect of landscape composition on pollinator communities and found that pollinator abundance and richness were higher in landscapes comprising more high-quality habitats.The role of landscape configuration is less studied, though it may also be very important as landscapes with smaller fields have higher structural connectivity and more edges providing more floral resources for pollinators.Given this research gap, here we focus on the role of the less studied configurational heterogeneity by assessing landscape-scale mean arable field size.Our study presents a trait-based approach to estimate the potential of two agri-environmental schemes to support pollinators based on floristic composition.For this, we studied plant traits associated with pollinator attraction and reward accessibility.We calculated community-weighted trait means and functional trait diversity in organic and conventional winter wheat fields complemented with sown flower strips and compared them to conventional fields along a gradient of landscape configurational heterogeneity.We tested whether management intensity and landscape heterogeneity select for specific pollination-related plant traits, and result in differing trait diversity by using literature data on species traits.Our specific hypotheses were the following: Pollination-related plant trait diversity is reduced with increasing mean field size within the landscape.Flower strips and organic farming maintain functionally more diverse plant trait composition than control conventional fields.Pollination-related plant trait diversity decreases from grassy margins to field interiors in both conventional and organic fields.We considered only insect-pollinated plant species that may attract pollinators and provide nectar resources.We assessed the following traits: flower size, flower colour, UV pattern in the flower,vertical farming racks reward quantity, and flowering duration.Trait values were collected from the BiolFlor database and identification books.Flower size was expressed as flower diameter.For zygomorphic flowers, the mean of the longest and shortest diameter was calculated.In the case of compact inflorescences that can be regarded as one functional pollination unit, we used the size of the whole inflorescence.We expressed flower colours on a continuous wavelength scale.White-flowered species got the mean value of the visible light spectrum.The presence of UV pattern in the flower was treated as a binary variable.In the case of reward quantity, we considered only nectar resources because of no available information about pollen resources for the majority of the species.Nectar quantity was provided on an ordinal scale with four categories.

The average flowering duration was coded as a continuous variable.For each transect, we calculated the total number and average percentage cover of insect-pollinated plant species.We quantified transect-level community-weighted means for each trait by averaging trait values weighted by species abundance.We tested the effects of management type , transect position , mean field size on the species richness and cover of insect-pollinated plant species, Rao’s quadratic entropy and the CWM and FD values for each trait by using linear mixed-effects models.We used generalized mixed-effects models for species richness with Poisson error distribution and log link function.For each response variable, we started with a model with all two-way interactions of the fixed effects and then selected the minimum adequate model from all possible nested models with the minimum Akaike Information Criterion value.The effect of years was not assessed per se but was treated as a random effect in the analyses.The field, farmer, and landscape identities were also used as random effects nested within each other and crossed with year.We used log transformation for cover values and square root transformation for CWM and FD values if necessary to fulfil test assumptions.We restricted our assessment on those 151 transects that harboured at least one insect-pollinated plant species for flower colour analyses to have meaningful results.We derived marginal and conditional coefficients of determination of the models to estimate the variance explained by the fixed effects and the entire model including both fixed and random effects using the MuMIn.We conducted multiple GLMM tests, therefore, we controlled for the false discovery rate.Post hoc pairwise comparisons were made with the em means package , and were presented in the figures.All analyses were conducted in the R environment.We found that flower strips and organic farming outperformed conventionally farmed fields, as they had the highest insect pollinated plant species richness and supported higher trait diversity favourable for pollinators.Flower strips were superior with the highest cover of insect-pollinated plants and the highest ratio of flowers showing UV patterns.Although annual flower strips doubled the insect-pollinated plant species richness in the grassy margins, the benefits of flower strips completely diminished in neighbouring field interiors.In contrast, organic field interiors could be considered the most favourable pollinator habitat among the studied field interiors based on plant pollination traits.Thus organic farming can enhance pollinators at a larger spatial scale with more environmentally friendly crop production on the whole area of the arable land.We did not find any effect of configurational heterogeneity expressed as mean arable field size.We showed that organic farming outperformed conventional farming practices regarding insect-pollinated species richness, flower attractiveness, and food resources provided for pollinators.This is primarily due to lower overall farming intensity in organic fields.Even organic field interiors offer abundant and diverse flower resources to pollinators compared to conventional farms.Thus, we can consider organic farming the most beneficial method based on pollinators’ resources.In contrast, intensively managed conventional fields supported the lowest richness of insect-pollinated plant species showing that only a few plant species could establish here randomly.Annual flower strips sustained the most favourable vegetation for pollinators, considering the quantity and quality of flower resources.They had six times higher cover of insect-pollinated plants than the edges of the other two farming methods.

We assessed that the first 50 publications contain sufficient information for answering the research questions and thus were used as exclusion criteria. The primary studies with their authors and publication year are included in the appendix . To get insight into the relevance of the topic for this SLR, the results of the broad and narrow search results are grouped by year of publication and presented in Fig. 1 and Fig. 2, respectively. When looking at the number of primary studies published for the broad search query, it seems that there are downward trends in 2010 and between 2014 and 2018. This is however due to the high number of publications that were found and the exclusion criteria that affected publication in those years disproportionately. Many publications were excluded because they were outside of the 50 most relevant publications. However, there is a consistent and sharp increase in the number of publications since 2018 both for the narrow and broad search queries. To get insight into the focus of the primary studies, we have categorized the studies by topic . The categories clearly indicate that majority of the studies were on pig health. Most studies focussed on the presence, spread, or outbreak of diseases, and few studies focussed on the causes and prevention of boar taint. In the selected primary studies, pig health and the performance of pig production were researched often. The production performance was investigated by analysing various parameters, such as the DNA of pigs,rolling bench fat composition of pig meat, meat quality, and muscle thickness. This shows that the production performance of pig farming was mostly measured using the attributes of the meat measured after the animals are slaughtered. Besides pig health and pig production results, other topics were researched.

Several feeding strategies have been developed and evaluated. Further, research was done to test if the origin of meat could be traced and on-farm environmental influencers, such as disease spread in manure and temperature effects on pigs. The narrow search query focused more on the application of ML and the topics of research are summarised in Table 7. According to the search results, most of the research focussed on performance related issues, such as pig performance, feeding strategy, and genetics. Pig health is the second most researched area, focusing mainly on diseases. The trend of using ML for disease detection through image and video analysis is reflected in the number of primary studies found on this topic. Pig detection and behaviour analysis were also performed using ML. Feeding strategy was analysed using basic statistical analysis and using ML. There were also primary studies that focus on detecting the origin of meat, detecting stress in pigs. A detailed description of the topics of research per primary study is shown in Appendix B. As can be seen in Table 8, a variety of data were used for research in the pork industry. Data were largely collected from body and tissue . The rest of the data reported refers largely to measurements of the external characteristics of the animals and their living environment or feed . There were also primary studies that used data on manure and micro-arrays. The type of data used per primary study is shown in Appendix E. The results of the analysis of data from the second search query focused on the application of ML to pig farming are shown in Table 9. The table shows the widely used features used in research. The features used can be grouped into features on pigs and features on climatic conditions.Pork meat information, images and recordings, growth, occupancy and reproduction, genetics, and pig movement have been used to build and validate models. The features on climatic data consist of data about indoor or outdoor climatic conditions, or both. Data used in those primary studies consist of outside temperature, indoor temperature, and relative humidity. The features used per primary study are shown in Appendix H.

The location of data gathering provides insight into the main sources of research data. We have categorized the locations where data are gathered into groups , which shows that most of the data is gathered at farms and slaughterhouses. The farms were sometimes conventional farms where a specific group of animals was monitored closely; other times, they were experimental farms used for precisely monitoring growing conditions and pig behaviour. Slaughterhouses serve as one of the main data gathering locations. The quality and composition of meat are important indicators of production performance; therefore, data from slaughterhouses where these indicators are routinely measured, are important sources of data. Other locations of data gathering for research include laboratories , externally collected data, previous research, or large database where data are shared. Further, data was collected at distribution centres, urban abattoirs, and there was a case of data from a virtual simulation. The data gathering location per primary study is shown in Appendix C. To get an overview of data analysis methods used in the primary studies, the algorithms or principles used were calculated as shown in Table 11. Least squares regression was used most often, followed by logistic regression. The primary studies found by the broad search query did not often use complex algorithms; instead, they applied variations of the linear regression model. All data analysis methods used are shown in Appendix F. All primary studies found by the second search query used ML algorithms. Random forest algorithm was used most, followed by neural networks and support vector machine, as is shown in Table 12. Compared to the primary studies found by the broad search query, those found by the narrow search query used more complex analysis models.

The algorithms used per publication are shown in Appendix G. Eight studies using models based on random forest and/or neural networks recorded an R-squared accuracy scoring higher than 0.7. The challenges stated include difficulty in scaling into large-scale farms due to differences in the environment between the experimental setup and large scale farm, and other features that influence the performance. Table 13 shows what has been studied, and if the research subjects were pigs, it shows the size of the group of pigs studied. A total of 26 primary studies considered live pigs, from which measurements were made. The number of pigs used for research varies significantly, from 5 to 908,582 pigs. Three primary studies considered farms as a whole, focusing on the growing environment of the pigs. Further, data on meat quality and faecal samples were used by some studies. There were also studies that used externally collected data, virtual experimental data, RNA pools, and liver pasts. Appendix D shows the research group types and sizes per primary study. Much research has been using data on genetics, pig breeds, meat quality after processing, disease recognition, and factors influencing diseases. Research that focused on the prevention of negative aspects of the final product is, for example, research done on boar taint and the attempt to reduce it. Boar taint in meat results in a lower price for the carcass. A variety of challenges have been stated in the studies regarding ML. The largest challenge is making the algorithms perform consistently high for all situations. Many studies reported that it was feasible to define a well performing model for the specific research setup and condition the model is trained and tested on but it was in general not possible to guarantee the same level of performance in practice. Mostly, more data and more features are required in order to improve the overall performance of the models developed.The original aim of this study was to review the literature on the use of advanced data analytics techniques in real-life business scenarios. We tried diverse search strings in order to find studies that are not based on experimental set-up be based on the use of data sources from real-life production chains. But the number of results we obtained was low.

We were unable to find any study that fused data from the whole production chain except one by Ma et al. . Ma et al. developed an intelligent feeding equipment and network service platform,roll bench using sensors, software, and monitoring units. This system helped in optimizing feeding methods and improving pig management. We considered, therefore, two search strings, a broad and a narrow search string: the former focussing on general data analytics and the second on machine learning. In both cases, we used more generic terms and relied on exclusion criteria to select the relevant studies. We consulted five scientific databases in this SLR, namely: Science Direct, Scopus, Web of Science, Springer Link, and Wiley. We did not consult Google Scholar initially which returned 16,300 publications, a vast majority of which were irrelevant, and thus filtering out the relevant publications was infeasible. The exclusion criteria we selected ensured that only publications that are accessible and of high quality were considered. The exclusion criterion on year of publication was based on our initial test search results. While research performed 20 years back is likely outdated, the set year allowed us to get a clear view of the trend in the studies published. This made it clear, for instance, that the number of primary studies increased substantially from 2018 onwards, particularly on the application of machine learning in the pork sector. When analysing the results of the SLR, we grouped the attributes such as features, data types, or algorithms we found. Grouping the results allowed us to derive clear conclusions. The detailed results are however included as appendices to this study. As can be seen in the increasing number of primary studies since 2018, the application of ML will be important in the pork sector in the future. When looking at the algorithms used in this context, studies that used features and data on images and recordings often used neural networks, specifically convolutional neural networks algorithm. For example, Cowton et al. , Mukherjee et al. , and Fernandes et al. focused on pig recognition and deviations in behaviour. Studies that used pork meat information often had random forest as best performing algorithm. The focus of the research using pork meat information was to investigate growth performance or the state of health .

What is striking in our review is that most of the research conducted use only groups of pigs selected for an experiment. Most of the studies are conducted in large pig farms; however, the numbers of animals studied within the farms were often small, indicating the studies were conducted on an experimental basis and not on the entire herd. While the focus on a selected group of pigs helps to accurately track and monitor pigs, it does not reflect the real-life situation. Therefore, it remains unclear whether the solutions found in research could be applied in real-life business cases. Out of the 41 primary studies we considered through the broad search query, only 10 involve more than 500 pigs, and of which only two, performed by Naatjes et al. and Haraldsen et al. , used real-time data from farm management systems. Research based on real-life data is scarce and most studies focussed mainly on isolated and one-off problems. We suggest that future research should consider the complexity encountered in real-life circumstances and the integration of data analytics within the management systems so that the analytical results can be used within routine business processes. The Republic of Botswana has an installed capacity of 550 MW with an estimated population of 2.3 million of inhabitants. It is reported that the country has one of the highest national electrification rates in Africa determined as 60 % with 77 % of the population connected to electricity in urban areas and 37 % connected in the rural areas. According to Statistics Botswana, it is stipulated that the Government has set national electricity access target of 82% by 2016 and 100 % by 2030. With an average energy demand of more than 400 MW, Botswana relies mainly on fossil fuels energy sources such as coal, fuel wood, and petroleum and an imported power from South Africa to meet needs of the population. Additionally, the country is also known for its particularity in renewable energy sources such solar, wind and various forms of bio-energy such as bio-fuels and biomass wastes.