Cannabis Grow

The aim of this paper is to 1)conduct a comparative analysis of the SMILE Google Earth Engine machine learning classifiers and analyze their performance in crop type classification over a semi-arid irrigated heterogeneous landscape; 2) classify cannabis fields in the Bekaa region for the years 2016, 2017 and 2018 using multi-temporal multispectral imagery and GEE built-in classifiers; 3) evaluate the added value of the increase in temporal frequency due to the use of Landsat-Sentinel-2-Sentinel-1 in crop type classification. We use data from Sentinel 2, Sentinel 1, and Landsat 8. We evaluate the accuracy of four different SMILE classifiers: Random Forest, Support Vector Machine, Classification and Regression Trees, and Gradient Tree Boosting, which was recently introduced to GEE classifiers. We perform multi-temporal image classification using three different imagery combinations. Each combination consists of two or three sensors. The first combination consists of Sentinel 2 and Sentinel 1 imagery , the second combination consists of Landsat 8 and Sentinel 2 imagery , and the third combination consists of Landsat 8, Sentinel 2, and Sentinel 1 imagery . Sentinel-1 mission  is composed of two polar-orbiting sun-synchronized satellites, each carrying an imaging C-band synthetic aperture radar  instrument. The Sentinel-1 mission supports four imaging modes providing different resolutions: Interferometric Wide Swath , Extra Wide Swath , Strip Map , and Wave . We use the IW mode which supports operation in dual-polarization and has a ground range resolution of 5 × 20 m. Sentinel-2 mission  provides high spatial resolution multispectral satellite imagery with a high revisit frequency  and a wide field of view. The spatial resolution varies from 10 m to 60 m depending on the spectral band. Sentinel-2 carries the Multispectral Imager  sensor with 13 spectral bands in the visible, near-infrared, and shortwave infrared part of the electromagnetic spectrum.

Landsat 8 , a NASA and USGS collaboration, provides global moderate-resolution imagery  in the visible, near-infrared, shortwave infrared, and thermal infrared part of the electromagnetic spectrum. Fig. 2 represents the flow of work and the three main stages for crop classification. Sentinel 2 TOA reflectance dataset was obtained from GEE platform, it was then refined to remove cloudy pixels by cloud masking using QA band and pre-filtered at 5% or fewer clouds. We used the Normalized Difference Vegetation Index  and the Enhanced Difference Vegetation Index , the two most widely used vegetation indices as additional bands in the classification process. EVI provides improved sensitivity in high biomass regions while minimizing soil and atmosphere influences but requires a blue band. EVI2 performs similar to EVI but without the blue band . Other vegetation indices such as the soil adjusted vegetation index  were tested but did not offer an advantage. The images were filtered over six months, from March to August and one image from each month was chosen. This resulted in six images from each dataset each containing 13 bands . A median, maximum, and minimum composite were created from the image collection. All bands were then combined to form a single image containing median, maximum, and minimum values for each band . The calibrated and ortho-corrected Sentinel-1 product was accessed from within the GEE platform. It was then filtered to get images with VV and VH polarization. The images were filtered over the same period as Sentinel 2 images. A mean composite was obtained of all Sentinel 1 images which resulted in a total number of 12 bands . Three texture metrics  were also added for both bands . The images were converted to bands and added to the Sentinel 2 bands which resulted in a total of 282 raster features for the six months. Finally, cannabis grow system the GEE Landsat 8 Surface Reflectance dataset was filtered to mask clouds and cloud shadows over the study area over the same period as the previous two datasets. A median and variance composite of all Landsat 8 images was created. The result is six Landsat 8 images with 8 bands each . We combine Landsat 8 bands with Sentinel 2 bands for the L8S2 dataset, resulting in a total of 330 raster features. The third and final dataset combination contains a total of 378 raster features for S1, S2 and L8 combined over the full-time series. Table 1 summarizes the datasets used. The training data used includes ground-truth crop fields collected in 2016 & 2018 as polygons. After delineating surveyed crops, they were imported to GEE as assets and grouped into six crop groups. For the 2016 classification: wheat and barley fields were combined; beans, broccoli, cabbage, celery, coriander, corn, cucumber, lentil, lettuce, melon, mint, onions, parsley, pepper, radish, tomato, carrot, watermelon fields were combined into one crop group; and stone fruits and vineyards were combined.

For the 2018 classification: wheat, barley, alfalfa, and vetch fields were grouped into one crop group; onion, cucumber, eggplant, squash, pepper, tomato, lettuce, parsley, mint, cabbage, carrot, chickpea, peas, beans, corn, and melon fields were grouped into another; and apple, apricot, peach, plum trees, and vineyards were grouped. Cannabis fields were filtered into irrigated and rain-fed or not irrigated, and only the irrigated fields were used in this classification. We use irrigated cannabis fields for the training of the classifiers because tobacco and cannabis share a similar growing cycle . Since tobacco fields are rain-fed or non-irrigated, confusion between detection of tobacco and cannabis occurred. Thus, we decided to use only irrigated cannabis fields to avoid this type of confusion. The other two crop groups were potato and tobacco. Urban and fallow polygons were selected based on their NDVI being less than 0.18 over the whole year. Finally, each crop group and the urban/fallow group were given a label  starting from 0 until 6. Fig. 3 shows the spatial distribution of surveyed crops over the Bekaa plain in 2016. We use the GEE “sample Regions” as the sampling method to select training data. The method requires inputs that are a Feature Collection, a class property, and a scale. We provide the polygons in a form of a Feature Collection as an input. The class property in our classification is ‘Croplabel’. Hence, a ‘Croplabel’  is assigned to each class of the sampling data, and the scale size used is 50 m. The number of training pixels used for the 2016 classification and 2018 classification is shown in Table 3. The number of pixels used was obtained by taking the average number of pixels of 10 GEE permutations. The total number of pixels was divided such that and 70% of the pixels  were used for training and 30% of the pixels were used for validation. The number of training and testing pixels for each crop type  is also shown in Table 3. Classifying different crop types requires knowledge of their spectral characteristics. Spectral characteristics of different crops are an important factor that affects the accuracy of crop identification. Each crop type has a unique spectral signature that changes with time depending on crop phenology and crop growth stage . Thus, making the temporal feature essential in crop identification. Defining the best “identification time” provides opportunities to identify spectral differences during specific phenological stages. To enhance spectral discrimination among different crops, vegetation indices can be used to take advantage of the difference in reflectance between different wavelengths . In this study, we aimed to identify a period where the spectral differences are suitable to maximize classification accuracy. To achieve this, we analyzed the spectral characteristics of the crops in the study  during a whole year. We calculated the mean NDVI, using Sentinel 2 imagery, over 15 different surveyed fields from each crop group for the year 2016.

The most spectral differences among these crop groups were observed between April to August in both years. However, we applied the classification on different periods and found that the highest classification accuracy was between March to September, which was adopted in this analysis. Random Forest classifier is an ensemble-based tree-structured classification method that grows several classifiers instead of one classifier to perform better classification . It creates several decision trees  and aggregates their results, thus predicting a response from several decision trees. The variation between individual trees results in more diversification and better accuracy. The classifier requires two user-defined parameters: the number of decision trees to be grown and the number of features used at each node. In our classification, the number of trees is chosen to be 20 trees based on parameter tuning to minimize computational timeouts, reduce run times, and avoid over-fitting . We performed hyper-parameter tuning by training the classifier using a different number of trees  and plotting the overall accuracy results for each number of trees. The number of splits at each node is set by default to the square root of several variables. Gradient tree boosting , similar to the random forest, is a tree-based ensemble learning algorithm that uses gradient descent as its optimization algorithm to minimize the loss function. It works by sequentially training several weak learners, which are shallow decision trees. During the training process, the mis-classified pixels – due to the weak learners that classified them earlier – are assigned to stronger weights and thus classified correctly . A loss function is associated with the model such that, each time a new tree is added it minimizes the loss function. With every addition of a new tree, the overall prediction error will decrease . Classification performance of gradient boosting is highly influenced by parameter tuning. Parameters include the number of trees, shrinkage, learning rate, loss function, sampling rate, and maximum node size. In this classification, we use 15 trees and the default parameters set by GEE: “Least Absolute Deviation” as loss function, shrinkage of 0.005, and sampling rate of 0.007. Classification and regression tree  is a tree-based machine learning algorithm . It can be used to classify numerical and categorical  data. The algorithm behind CART classifier builds a decision tree starting from the roots, which will split at each node. The training data will pass down the tree through the splits and a decision is made at each node to decide the next direction of the data. The decision is made to reduce the impurity at each node, which depends on the splitting rule. The splitting rule metrics for classification trees include, but are not limited to, misclassification error, Gini index, Entropy index, and Twoing . The splitting will continue until there is only one sample left, and a final decision is made at the terminal node . The inputs for CART classifier are a feature collection which is the 70% training data, a class property which is ‘Croplabel’, and input properties which are all the bands being used for classification. We classify crops in 2017 using the previously created models from 2016. We prepare a composite, similar to Section 2.2, for the year 2017. We extract the spectral, spatial, and textural characteristics of the 2016 training polygons using different dataset combinations. We train the classifiers using the 2016 training data and we classify the 2017 composite using the trained classifiers in 2016. We choose the year 2017 because we have obtained marijuana grow system fields’ survey data during 2017. We perform an accuracy assessment as described earlier. The surveyed fields were all used as a testing data set since this data was not used for training. We perform several permutations, using the most robust classifiers, in order to decide which model classifies cannabis in 2017 better than the other.

The most accurate model in classifying cannabis is presented below. The application of machine learning algorithms in crop-type and land cover classification has gained popularity in the past decade. However, the identification of cannabis plantation areas has not received enough attention. Our study provides a crop classification method to identify cannabis fields using freely available medium resolution imagery  and machine learning algorithms. We perform several permutations using different dataset combinations and different metrics for each satellite data. The highest accuracy was achieved using the mean composite of Sentinel-1 and median and variance composite of Landsat 8, however, for Sentinel-2, using min, max and median composite achieved better results. Our results are in agreement with Carrasco et al.  who evaluated the use of different metrics  derived from Sentinel-1, Sentinel-2, and Landsat 8 as a time series for land cover mapping, and classification was achieved using the mean composite of Sentinel-1, the median composite of Sentinel-2, and the median and variance composite of Landsat 8. Previous studies have reported the improvement in classification accuracy after combining radar sensors such as Sentinel-1 to optical sensors . Our classification results show a slight improvement in classification accuracy when Sentinel-1 is added using RF and SVM classifiers.This marginal improvement in accuracy was also observed earlier by Boryan et al.  when identifying winter wheat in two different locations using optical and SAR data. The authors suggest that texture features from SAR are not necessarily important when cloudless optical data is available.

Multivariate modeling indicated that adolescents with a history of cannabis use had lower perceived risk of harm compared with those who had a history of using other substances. This is an important finding as studies have found that cannabis consumption decreases perceived risk of harm from using cannabis . Limitations to consider when interpreting results of this study include the inability to conclude causal relation, limited generalizability, and response biases. Although complex sampling was used to have the most generalizable adolescent population to the US, some differences in perceived risk of harm from cannabis use and history of substance use were detected between the full sample of US adolescents from the NSDUH dataset and the analytic subsample that had complete responses to the pertinent survey questions, thereby limiting generalizability. Furthermore, there is potential for response biases with retrospective, self-report questions that may result in under reporting or recall bias. For example, reported perception of peer use has been linked to a respondent’s own substance use . Recommendations for future research are to conduct prospective studies to confirm the findings of the current study. Furthermore, research analyzing longitudinal data to monitor trends in risk perceptions and consumption, differentiating by state is essential as cannabis laws differ by state. Additionally, research examining the effects of interactions between age or sex and peer influence on cannabis risk perception will be useful for adapting prevention services tailored for age and sex. Understanding the effects of legalization of cannabis on adolescent use can better inform state officials on ways to implement programs to educate adolescents about the risk of harms associated with its use. Adolescent cannabis use prevention programs might include age-appropriate messaging about the risk of harm from using cannabis and elements that bolster the protective influences of peers and schools.

The current study adds further evidence to justify capitalizing on the potency of peer and social influences in substance use prevention interventions. Adolescents in this study who perceived risk of harm from monthly cannabis use had lower odds of believing their peers used outdoor cannabis grow, higher odds of perceiving their peers disapproved of using cannabis, higher odds of their parents limiting their time out with friends during school nights, higher odds of regarding school as important, and higher odds of reporting participation in extracurricular activities. This study further emphasizes the need for effective, multidimensional programs that target adolescent attitudes and beliefs about cannabis use through peer education, enhancing school engagement, and promoting youth clubs, athletics and other school-based or community social events. Tobacco companies have long employed numerous tactics to advertise their products to youth and young adults , and young people who report viewing tobacco advertisements are at greater risk for tobacco use initiation, progression to regular use, and development of nicotine dependence.As a result, the 1998 Tobacco Master Settlement Agreement  limited the marketing of tobacco products in ways that might entice under aged youth to use them and movies, use of cartoon characters such as “Joe Camel”. However, following passage of the MSA, more subtle product placement strategies continued to be used in TV and movie productions with tobacco products featured as a part of the plot or character development. Similar to direct tobacco advertising, viewing tobacco products on TV/movies is also positively associated with youth tobacco use . Several U.S. states, including California, legalized the sale, possession, and use of cannabis recreationally among adults, beginning in 2012.However, unlike with tobacco, there are relatively few restrictions on cannabis marketing, in part because cannabis is classified federally as a schedule I drug despite legal sales of recreational cannabis in 11 states and the District of Columbia . Consistent with studies that show viewing tobacco marketing increases risk for tobacco use, a small body of cross-sectional work has also shown that viewing cannabis advertisements is associated with higher odds of cannabis use , as is seeing cannabis use in TV/movies . The tobacco and emerging cannabis marketplaces have changed considerably over the past decade, resulting in a proliferation of new tobacco and cannabis products , which have become increasingly popular among YA . There is also evidence that tobacco and cannabis companies are marketing these products in new  ways – through online platforms such as social media , which may disproportionately impact YA who rely on the Internet more heavily than older adults .

For example, a recent study examining JUUL’s  marketing campaign revealed that thousands of Instagram posts, emails, and other advertisements were targeted to youth, and non-smoking populations.Similarly,Medmen recently initiated a well-funded national advertising campaign, including advertisements on the Howard Stern and Adam Carolla shows, YouTube videos, billboards, and social media advertisements . Given the increasing array of tobacco and cannabis products and methods for marketing them , it is important to identify the extent to which YA recall seeing marketing, for which products, and through which channels . Marginalized populations, including sexual and gender minorities, racial/ethnic minorities, and populations of lower socioeconomic status, use tobacco and cannabis products at higher rates, relative to the national average . Historically, these groups have also been disproportionately exposed to cigarette advertisements targeted specifically to minority populations . However, little is known about whether, or how viewing of marketing for new and emerging tobacco and cannabis products differs across sociodemographic characteristics, such as race/ ethnicity, gender identity, sexual identity, and socioeconomic status in YA. This study had two aims. First, we assessed prevalence of recalling online advertisements, as well as seeing product use in TV/movies, for a wide range of tobacco and cannabis products among a diverse sample of YA  cannabis products in California from Southern California. Second, we assessed sociodemographic differences in recalling online advertisements and seeing product use in TV/movies for any tobacco products and any cannabis products. All analyses were limited to never-users of tobacco and cannabis products, respectively. Sociodemographic characteristics were first calculated, separately among never users of tobacco and cannabis products . Then, prevalence estimates were calculated for recalling seeing tobacco and cannabis marketing. Unadjusted estimates are reported for both Internet- and TV/movie-based marketing, and F-tests assessed whether prevalence differed significantly by marketing source for each product. Finally, sociodemographic differences were assessed for recall of marketing for any tobacco  or cannabis  products, separately for Internet- and TV/movie-based marketing channels.Multivariable logistic regressions also assessed associations between all sociodemographic characteristics, in combination, on recalling any tobacco and any cannabis marketing. All analyses were limited to never-users of tobacco  and cannabis  and were conducted in 2020 using Stata SE version 15. Sample characteristics of users vs. never-users of tobacco and cannabis were compared in Supplemental Table 1. This study assessed prevalence of, and sociodemographic differences in recalling tobacco and cannabis grow equipment product marketing among a sample of Southern California YA reporting no history of tobacco and cannabis use, respectively. For the majority of products – all combustible tobacco products, combustible cannabis, and edible cannabis – respondents had higher odds of seeing use of those products on TV/movies than they did seeing online advertisements. Portraying tobacco use in TV/movies has been an effective – and profitable – way to advertise tobacco products , increasing risk for tobacco use initiation among youth .

While considerably less research has assessed the role of seeing cannabis products in TV/movies on initiation of cannabis use among young people, cannabis brands have been successful in negotiating product placements with entertainment studios, and with almost no regulation . While it is impossible to discern the degree to which respondents saw tobacco and cannabis products in TV/movies through intentional product placements and/or through the artistic discretion of the TV/filmmakers, our results highlight that shows and movies reaching young people include a considerable amount of tobacco and cannabis product use. Effective measures to reduce exposure to this form of marketing may include giving programs that display tobacco or cannabis use R  and TV-MA  ratings and prohibiting the display of recognizable brand names, among other actions. Consult the Truth Initiative  for a full list of measures endorsed by the organization . While there is ample evidence that JUUL and other e-cigarette brands are indeed promoted on TV/movies , respondents in this sample had higher odds of recalling seeing online advertisements  for these products. Given that youth and YA remain the largest demographic group of Internet users , and that the proportion of young people using e-cigarettes has risen , online advertisements for e-cigarettes may disproportionately influence underaged youth and YA to experiment with, and become regular users of e-cigarettes. A future direction for effective tobacco regulation might include limiting online marketing for e-cigarette products. While logistically challenging, online marketing should ideally be regulated in such a way that ensures first amendment protections to e-cigarette companies, while also limiting exposure among YA never users  . A number of sociodemographic differences were also found with regard to viewing tobacco and cannabis marketing. For example, women reported seeing online cannabis advertisements at higher rates than men. Compared to men, greater percentages of women also reported seeing tobacco and cannabis products on TV/movies. It is possible these findings stem from documented gender differences in processing and recall of advertising details, with women recalling details of advertisements more clearly than men . However, it is also plausible that young women  who recall seeing tobacco and cannabis products in TV/movies may be at especially high risk for using those products themselves. In prior longitudinal work among non-smokers, young women who watched a favorite actor smoke on screen had a nearly twofold increase in risk for smoking themselves. This association was not significant among young men . In multivariable analysis, LGB-identified YA also recalled seeing online cannabis advertisements at a higher rate than straight respondents, and prior research has shown that LGB youth have a greater willingness to use cannabis products than their straight peers . Together, these findings highlight that young women and LGB people may be priority populations for public health efforts to prevent tobacco and cannabis use. Several racial/ethnic differences were found. Interestingly, racial/ ethnic minority respondents had lower odds of recalling marketing, across a number of comparisons. For example, Asian YA had about half the odds of recalling seeing online cannabis advertisements and seeing use of cannabis products in TV/movies, compared to White respondents. Further, compared to White respondents, Black respondents had about 40% odds of recalling both tobacco and cannabis use in TV/movies, and respondents reporting an “other” race/ethnicity had about 50% odds of recalling cannabis use in TV/movies. While cigarette companies have a long history of targeting advertisements to Black populations , our results suggest that among never-users, White YA are more likely to see online advertisements for tobacco and cannabis, or to see those products used in TV/movies.

However, it should be noted that while this analysis was focused on identifying sociodemographic disparities in seeing marketing, all groups did recall seeing some degree of marketing . Several steps have been taken at the state and federal levels to regulate underage exposure to tobacco marketing . However, while many U.S. states have legalized the sale of cannabis products, they remain illegal federally. This limits the ability to effectively regulate accessibility to cannabis products for adults and those who may benefit from them , while also limiting exposure to those most vulnerable to misuse . Respondents in this sample were below the legal purchasing age for cannabis products in California, yet a large proportion of them – all of whom reported never using cannabis in the past – reported seeing online advertisements and use of these products in TV/movies. These results highlight a need for more research related to cannabis marketing exposure and subsequent use among YA, and the role of regulations to limit exposure. While individual states may be unable regulate online advertisements, they would be able to regulate local advertisement , should they be shown to deliberately and effectively target underage youth. More work is still needed to understand how to effectively regulate pro-use messages online and in TV/movies. First, our main outcome measure was self-reported recall of tobacco and cannabis marketing, which may not reflect actual marketing efforts to place ads where YA will see them. Instead, our measure signifies where YA were most likely to notice tobacco and cannabis advertisements. Second, these analyses were cross-sectional; we were unable to assess whether recalling marketing was associated with tobacco and cannabis use initiation. Third, this study assessed whether respondents recalled seeing marketing both online and in TV/movies, though there exist a host of other ways in which tobacco and cannabis products are marketed .

Approximately one-third of people with epilepsy will experience treatment-resistance which is defined as failure of adequatetrials of two tolerated, appropriately chosen antiepileptic drugs to achieve seizure-freedom.Children unresponsive to conventional treatments face an increased risk of cognitive, behavioral,and psychosocial dysfunction that can have a negative impact on their health and development . This prognosis has led to strong consumer interest in and uptake of alternative treatments such as artisanal ‘cannabidiol -rich’ products as a way to manage seizures in children with epilepsy . However, such products are typically of unknown quality, composition, and safety,and their use may conceivably pose unpredictable health risks to these children.Despite increasing access to legal pharmaceutical-grade cannabis grow facility products globally, many consumers continue to use artisanal cannabis preparations.

This may be done for various reasons including lower cost relative to the prescribed product, lack of awareness or knowledge of the patient access pathways, bias against pharmaceutical products, or perceived superior effectiveness and/or tolerability of artisanal products relative to the This current analysis of ‘artisanal’ cannabis samples administered to children with epilepsy in the Australian community found potentially unsafe levels of residual solvents, mainly ethanol, in approximately one quarter of the cannabis samples tested. In the manufacture of artisanal cannabis preparations, the incomplete evaporation of ethanol and other solvents prior to reconstitution with an oil-based diluent can lead to consumers ingesting higher amounts of residual solvents than anticipated, particularly if products are taken at high doses and/or for prolonged periods of time.There are legitimate concerns around the potential harmful effects of ethanol on the developing brain, as well as the fact that alcohol consumption, particularly chronic and/or acute use of considerably large amounts of alcohol , and sudden alcohol withdrawal, can increase the risk of seizures .

Other alcohol-related factors for increased seizure risk include impaired sleep quality and interactions with antiepileptic drugs. Despite these concerns, ethanol is commonly used as a solventin many oral liquid preparations for pediatric populations to improve drug solubility and/or as a diluent . According to ICH guidelines, ethanol and isopropanol are ‘Class 3 solvents’ which are regarded as less toxic and of lower risk to human health.Such solvents may be administered in concentrations higher than the toxicity limit  provided this is underpinned by good manufacturing practice or other quality-based requirements.The exact implications of this observation are unclear, but it suggests that pesticide contamination is a legitimate concern which requires further investigation across a larger set of samples.At the time the ‘PELICAN’ study was collecting samples from participants , legal pathways to accessing medical cannabis grow system in Australia were still evolving and highly bureaucratic,time-consuming, and expensive for patients . This represents a time in history when consumers had few alternatives to accessing medicinal cannabis and, artisanal ‘black market’ cannabis products,by comparison, were cheaper and easier to access. There are now better legal options available for accessing medicinal cannabis that avoid the concerns identified with unregulated products.

In Australia, Epidiolex is now a registered and government-subsidized medicine for the treatment of Dravet syndrome and Lennox–Gastaut syndrome  and an array of other CBD-containing products are available on prescription via schemes overseen by the Therapeutic Goods Administration . Despite this, the use of artisanal cannabis products will undoubtedly continue because of the perception that artisanal products are more effective and/or better tolerated than pharmaceutical-grade cannabis products, and that the addition of D9-THC and minor-cannabinoids may harness a supposed ‘entourage effect’ that enhances overall efficacy . To-date, no randomized, controlled studies have compared pharmaceutical-grade CBD against artisanal cannabis preparations in apopulation with epilepsy, although preclinical studies are starting to shed light on the pharmacological interactions between cannabis constituents . Meanwhile, in North America, concerns continue around an overall lack of mandatory testing of cannabis products to ensure patients are obtaining safe, quality-controlled product from licensed producers. Several recent reports have described cannabis-derived products contaminated with microbes, heavy metals, pesticides,and other toxins .

A developing brain subject to chronic cannabis-exposure may to a greater extent affect these underlying mechanisms, and hence result in increased severity of diminished expression. Apathy, on the other hand, is linked to reward expectancy and cost-benefit-computation . It is possible that these mechanisms have other determinants, and therefore present as associations to male sex and depressive symptoms. A previous study by Strauss et al.  also found male sex to be linked to the apathy-dimension. And depression shares many features with the apathy-dimension of negative symptoms, such as anhedonia. Especially anticipatory anhedonia is affected in schizophrenia , but is also found as a feature in depression,vertical grow system and may therefore contribute to drive this association. The frequency of cannabis use at baseline was also associated with the severity of diminished expression after 12 months.

Our interpretation is that higher frequency of cannabis use before baseline predicts less improvement in symptom severity over the first year of treatment. In contrast to Sabe et al.’s findings of less severe negative symptoms in recent cannabis abstainers, we found no difference in symptom severity in either dimension when comparing abstainers to continued-users and non-users. And continued use did not contribute to symptom severity at 12-month follow-up. A possible explanation for this is the abstaining groups’ heterogeneity with regards to amount of intake, i.e. that both heavy and more recreational users are included in the abstainer group, with consequences for the effect of abstaining. In our sample, the “abstainer”-group was too small to do further sub-categorization. We could, however, speculate that abstaining from heavy continued use  would have beneficial effects on the development of negative symptoms.However, both the effect of different classes of drugs, and the severity of substance misuse will vary significantly. We consider it unlikely that the intake of drugs of abuse protect against or reduce negative symptoms.

Rather, it may suggest that individuals with a heterogenous intake of illicit drugs constitute a subgroup with lower levels of primary negative symptoms. The main strength of this study is the use of a validated two-dimensional model of negative symptoms in a large sample of FEP participants. This enabled us to counteract some of the limitations found in previous studies, and provides a more differentiated investigation of negative symptoms in line with the current theoretical understanding of its phenomenology. We also used a more differentiated measure of cannabis use, encompassing the frequency and recency of use and thus enabling study of potential dose-response effects. The sample size and the inclusion of relevant clinical and sociodemographic characteristics enabled statistical control for potential confounding group differences associated with both cannabis grow equipment use and negative symptoms. There are also important limitations. First, the chemical composition of cannabis may vary significantly, especially with regards to the THC content. We could not correct for this in the analyses. From police confiscate in Norway, THC content has been estimated to vary from 30 to 45% .

In line with this, the assessment of “instances of use” as a proxy for the amount of cannabis used is no measure of the actual amount of cannabis consumption, or the effect of other illicit drugs that may have been used simultaneously. Second, it is widely accepted that side-effects of antipsychotics, such as sedation and extrapyramidal symptoms, may constitute sources of secondary negative symptoms . Clinical measures of these were not included in the analyses. Different antipsychotics display different side-effect profiles , and this variation is not fully captured by the measure of DDD. It may be that the dose dependent effects are less relevant than the receptor profile of the different antipsychotics. Since this is a naturalistic study, the treating clinicians may also have adjusted the dose or changed medication to reduce side-effects. The absence of an association in our data does not contradict antipsychotics’ potential to cause secondary negative symptoms. Finally, there was a substantial loss to follow-up. However, there were no significant differences between the drop-outs and those who completed follow-up.  The use of cannabis for medical purposes is increasing worldwide . With the changing public and political opinion, more countries are implementing medical cannabis legalization. Although approved in many regions, safety data from clinical trials are not as robust for medical cannabis as for other pharmacotherapies.

Brief individual interventions addressing substance use motivations and expectancies have been successful in reducing adolescent cannabis use ; however, research on preventing initiation through brief intervention and among JIY is nascent. Extension of expectancies research with JIY samples is necessary, particularly using prospective data and examining the role of positive expectancies and cannabis use outside detention when there is greater opportunity for use. Studies of school-based and general adolescent samples have also demonstrated the importance of understanding reasons for and protective factors against cannabis grow lights use. Data from the Monitoring the Future Survey examining past 10-year trends demonstrates adolescents cite more coping-related reasons than any other motivations for use . Individual factors that positively influence social cognition and behaviors  appear to buffer against substance use among early adolescents in public school , and higher self-esteem is associated with less substance use  among Black adolescents exposed to community violence and with high family stress.

Enhanced emotion regulation skills, which are influenced by social cognitive factors , are also protective against cannabis use initiation among Black adolescents . Justice-involved youth, who experience high rates of trauma, poverty, stigma and discrimination, may cite multiple reasons to use cannabis as a coping strategy, however, research in this area is lacking. Understanding how individual level, substance-related attitudes, beliefs and social cognitions influence JIY’s cannabis use, while accounting for known factors associated with increased likelihood of use, such as psychiatric symptoms , other substance use , and externalizing behaviors, is key to shaping the development of feasible systems-embedded brief substance use prevention interventions. Identifying individual social cognitive factors that might protect against cannabis use initiation in first-time JIY allows incorporation of a strengths versus deficit framework; a theoretical approach still largely lacking in the study of cannabis use and juvenile justice. In this prospective cohort study of first-time JIY, we aimed to understand rates of early onset cannabis use  and individual level factors associated with early onset use and new initiation in the 12 months after first court contact.

We hypothesized more psychiatric symptoms, other substance use, pro-cannabis use beliefs, attitudes and intentions, and lower self-concept and less self-control would be associated with early onset use and new initiation over follow-up. Three variables were derived from youth baseline self-report  and 3 follow-up assessments  over a 12-month period . Youth who had no lifetime use at baseline but reported cannabis use during the 12-month follow-up period were coded as new initiation. Descriptive statistics were examined at baseline. Next, we determined factors associated with lifetime cannabis use reported at baseline using bivariable measures of association . Third, among youth who reported lifetime cannabis grow tent use at baseline, we compared those who did and did not report early onset use at baseline. Fourth, we conducted modified Poisson regression to determine the independent associations between baseline factors and two primary outcomes:  lifetime cannabis use  reported in the entire sample, and  early onset use  in the subset of participants who reported baseline lifetime cannabis use.  Covariates were selected for inclusion in the multivariable models based on the standard cut-off rule of p < 0.05 in bivariable analyses except age, gender, and race/ethnicity, which were included in all models.

We created final multivariable models using a sequential backwards selection approach, in which variables with the largest p-values were removed sequentially, with the final model having the lowest AIC. Next, among youth who did not report lifetime cannabis use at baseline , we compared baseline factors associated with cannabis use initiation over follow-up using the same methods as described above. We then conducted a survival analysis using Cox proportional hazards regression to determine baseline factors associated with time to cannabis use initiation among youth who reported no lifetime cannabis use at baseline. We estimated the length of follow-up by calculating the difference between the interview date during which the first instance of cannabis use was reported and the interview date of the baseline survey.All variables significant at p < 0.05 in bivariable survival analyses were included in the multivariable Cox proportional hazards regression model; we also included age, gender, and race/ethnicity and obtained a final model using a sequential backwards selection procedure, as above.

This study underscores the importance of studying in-home cannabis smoking, which occurs at a higher rate than in-home tobacco smoking; however, more research is needed. Cannabis users comprise a sizeable population, both in absolute and relative terms, and with cannabis laws becoming more liberal, the number of users is rising. A major public health goal for future studies is to identify how often in-home cannabis smoking occurs in the general public and in high-use populations. Knowing the rates in the overall population will reveal the full scope of the behavior while understanding rates in high-use groups will quantify the behavior among those who have the most to gain from intervention. Obtaining these prevalence data will inform interventions to eliminate in-home smoking and will encourage further research, including the identification of public health messages designed to prevent in-home smoking  and research on the health consequences of first hand, secondhand, and third hand cannabis grow tray smoke exposure.

As with all studies, results of the present analysis should be viewed considering the strengths and limitations of the methods used. While we included data from 107,274 men and women from 17 countries, the data were collected using anonymized web-based surveys specifically targeted toward those who use drugs through advertisements on social media and drug-scene-related media outlets. While non-probability-based sampling and self-selection bias prevent generalizable estimates of prevalence , the inferences that were presented comparing in-home tobacco and cannabis smoking by use-status are expected to have significantly less bias since it is unlikely that the primary prerequisite for selection bias  was present. In the Introduction, we referenced online information on tobacco and cannabis control policies that highlight significant variance in the type and implementation of policies across countries.

To our knowledge, no study has attempted to quantify the impact of such regulatory differences on the in-home use of cannabis and tobacco, an investigation that was beyond the scope of the present study. However, the presence and diversity of public policies toward both substances should be considered when interpreting our findings. To reduce bias related to confounding by participant-level characteristics and by country-level characteristics such as regulatory policies, our models statistically controlled for several potential confounders,vertical grow system for sale and included country as a fixed and random effect. Future investigations of how tobacco and cannabis control policies impact in-home smoking are warranted, especially studies that explore how legalized cannabis use, with and without legalized outdoor use, impacts in-home use and subsequent SHS and THS exposure. The self-report origin of our cannabis and tobacco use status and in-home smoking status is also a limitation, although the GDS is a well-respected entity among the drug-using community and is known for preserving anonymity, making it more likely to elicit accurate responses.

We were unable to study usual in-home smoking patterns  including the frequency of in-home smoking or household composition because we did not collect usable data on these important factors—future studies should capture more detailed data. As a result, our estimates likely overestimate problematic in-home smoking, although this is not expected to greatly bias results since the overestimation is likely non-differential—i.e., similar for both in-home cannabis and in-home tobacco smoking.  Transmission of severe acute respiratory syndrome coronavirus 2 , the virus responsible for causing coronavirus disease 2019 , has led to unprecedented morbidity and mortality across the U.S. . Risk factors for COVID-19-related severe illness resulting in possible hospitalization include: active or former smoking status and/or having pre-existing comorbidities or an immunocompromised status . Combustible and non-combustible tobacco users are vulnerable to clinical morbidities, including impaired pulmonary function and respiratory illnesses.Additionally, vaping nicotine, flavorings, and/or tetrahydrocannabinol  products may place individuals at increased risk for COVID-19- related symptomatology and illness due to impairment of normal pulmonary defenses to inhaled viral pathogens . Smoking, and possibly e-cigarette use, can upregulate the angiotensin-converting enzyme-2  receptor, which is the receptor for SARS-CoV-2 .

The non-linear  effect of age of first tobacco use on average HONC score can be described as a slight decrease until age 8, followed by a slight increase between 8 and 12 years, and then a sharp decrease with age. Hazardous substance use in adolescence can increase risk for substance use disorders in adulthood, which remain a major public health issue in the United States. So, it is essential to develop new tools for identifying adolescents at risk of developing hazardous substance use among co-users in order to possibly increase the impact of prevention and intervention programs. This study took a preliminary step by determining risk factors associated with hazardous use of alcohol, cannabis, and tobacco — the three of the most commonly used substances by adolescents — and employing them to build joint models for predicting risk of hazardous substance use. To our knowledge,mobile vertical rack this is the first study that attempted to jointly model hazardous use of multiple substances by adolescents specifically for the purpose of quantitative risk prediction.

With information about salient combinations of individual, family and socio-demographic factors, the proposed model can be used to predict scores on quantitative measures of hazardous use for an adolescent user of alcohol, cannabis, and tobacco. A predicted score can be compared with its distribution  to assess the relative level of risk for the user. An important novelty of this work is the application of relatively new statistical and machine learning methods for analyzing multiple outcomes. To the best of our knowledge, these methods have not been used so far in the substance use literature. They are better suited for risk prediction than the classical multivariate regression because one  allows outcome-specific predictors, thereby providing flexibility for joint modeling that may lead to higher predictive accuracy; and the other  allows regularization of regression coefficients to protect against over fitting, thereby helping the model to generalize well on future unseen data. For comparison, we also fitted the classical multivariate regression model, whose results are in Supplementary Table 3.

This model identified a total of ten risk factors, all of which were also identified by MCGLM, but the latter additionally identified family history of hazardous alcohol use and had higher predictive accuracy. None of the individual models identified age of first cannabis use, age of first use of other substances, and family history of hazardous alcohol use, which were identified by MCGLM. Of these, the age of first cannabis use, an important risk factor as per the literature, was identified by MCGLM to be associated with both CPQ-A and HONC scores. Although the individual models identified guardian, parental monitoring, and parental past substance use, which were not identified by MCGLM, the effects of these variables could have been captured partially by similar variables included in the joint model, e.g., vertical grow rack parental attachment and family history variables. One of our key findings is that early age of onset of tobacco use is associated with hazardous use of all three substances. In line with previous literature, one study found that, among adolescent drinkers, past-year smokers were at a higher risk for alcohol use disorders than non-smokers .

Another study evaluating the association between e-cigarette use and the use of other substances showed that early onset of e-cigarette use was associated with increased use of alcohol and cannabis.Another key finding is that lower levels of parental attachment are associated with higher scores of hazardous cannabis and tobacco use. Lower levels of parental monitoring and attachment are also known to be associated with adolescent substance use disorders . Other findings in this study are also generally consistent with the literature. For example, increased age is associated with higher RAPI and HONC scores. Indeed, there is a higher percentage of young adults with alcohol use disorders compared to adolescents, and cigarette use tends to increase with age . The observed reduction in CPQ-A scores with an increase in maternal education is supported by literature showing lower socio-economic status is associated with hazardous substance use . The finding that increased CPQ-A and HONC scores are associated with greater early life stress is also consistent with literature demonstrating that various forms of early life stress such as physical and/or sexual abuse, negligence, parental divorce, and domestic violence tend to increase hazardous use of alcohol, cannabis, and tobacco . Our result indicating that family cigarette use is associated with higher HONC scores is consistent with other studies reporting on the association between family smoking habits and increased adolescent smoking .

More recently, a simple cumulative risk index was developed to classify which adolescents are at risk for developing persistent substance disorders in adulthood using risk factors from childhood and adolescence . Another recent study built a model for predicting quantitative risk of developing cannabis use disorder in adults based on personal risk factors using statistical and machine learning approaches . Yet another recent study  built models for predicting risk of developing substance use disorder by thirty years of age using separate sets of predictors from late childhood to 22 years of age. There is a substantial literature on exploring factors that lead to co-use of multiple substances . However, to our knowledge, modeling of hazardous use of multiple substances jointly has not been considered especially in the context of risk prediction modeling. More specifically, the key differences between these previous studies and our present work are that they  do not provide a measure of quantitative risk or score,model concurrent use rather than hazardous use of multiple substances, and  focus on a selected number of risk factors  rather than a comprehensive set of potential risk factors.

As such, there is a need to develop risk prediction models for hazardous use of multiple substances based on personal risk factors of adolescent users. This study aims to fulfill this need by developing preliminary models for jointly predicting hazardous use of alcohol, cannabis grow set up, and tobacco for adolescents who have used all three substances in their lifetime. Joint statistical modeling of multiple outcomes utilizes the correlation between them, which can lead to higher power for detecting association between risk factors and outcomes and can additionally provide insight into the shared underlying mechanisms. As our goal is statistical risk prediction rather than hypothesis testing, we consider a set of potential risk factors as suggested by the literature.However, it assumes a common set of predictors for all outcomes, which limits its applicability in our context of risk prediction. This is because if a variable is predictive of one outcome but not another, model parsimony dictates that the variable should be included only in the model for the former but not the latter. Adding unimportant variables to a model adversely affects its ability to predict accurately for new  participants that are not included in building the model. Moreover, regularization of regression coefficients in the model can protect against over fitting of the model especially when sample sizes are not large.

An over fitted model is sub-optimal for the purpose of predicting for new participants . However, regularization is not available in the classical approach. Therefore, we apply two relatively new statistical and machine learning methods, each of which addresses one of these limitations. Specifically, we utilize multivariate covariance generalized linear models  and penalized multivariate regression with a lasso penalty . These methods have not been used to model multiple outcomes in the substance use literature perhaps because the development of joint risk prediction models has not yet been considered in a formal way. Supplementary Table 1 summarizes the 18 risk factors. They include participant demographics ; general environmental factors ; their own substance use ; and family substance use . Not all participants used substances other than alcohol, outdoor cannabis grow, and tobacco, and hence for them there was no corresponding age of first use of other substances. Therefore, to include this age variable in the model, a binary indicator of lifetime use of other substances was added to the model together with its interaction with the age of first use. This way, the interaction term had a non-zero value only for the users of other We used two multivariate statistical modeling frameworks for joint modeling of the three outcome variables: MCGLM and multivariate lasso. MCGLM: It is a novel generalization of the classical multivariate regression  allowing modeling of the mean structure, variance function, and within-response covariance structure. MCGLM allows outcome-specific predictors, i.e., the predictors need not be shared across all outcomes. Its predictive accuracy was measured by root mean square error , computed using leave-one-out cross-validation .This measure differs from the ordinary RMSE and allows assessment of model performance on future unseen data more accurately by protecting against over fitting.

In the past year, 25% of EA ages 18–24 have initiated/increased cannabis use in response to negative stress and emotions – the steepest rise in use across all queried age groups; further, 43% of EA reported past year cannabis use, the highest rate observed in 35 years . While the literature supports important links between negative affect and social in- equality , as well as between negative affect and cannabis use , there is a gap in our understanding of how these processes interact with each other to contribute to EA cannabis use, as well as in our understanding of the cognitive and developmental neuroscience behind these relationships.For example, during some phases of development, social inequality may directly alter emotion and regulatory networks to increase risk. At other stages, objective and subjective social inequality may increase cannabis grow system use risk either via direct influences  or as contextual moderators of risk .

Further, effects of objective and subjective inequality may be additive or themselves interact. Future research would benefit from incorporating metrics of subjective experiences of social inequality  in addition to the widely used “objective ”metrics such as zip code, income, parent education, employment, and environmental instability  when studying relationships to EA hazardous cannabis use. In addition, differentiating the multi-component processes of emotion regulation and studying the constituent parts  may elucidate behavioral, functional, and neural correlates  that have implications for EA decision-making risk in “high-stakes ”emotional contexts like cannabis use . We are extremely far from understanding the potential neurodevelopmental and socio-emotional resilience factors in these equations . As clinicians and clinical scientists, operating without these data directly obstructs our capacity to identify essential risk/protective factors of EA socio-emotional health in a way that meaningfully helps providers prevent and intervene with EA cannabis use .

Of even greater relevance and salience, data to help elucidate meaningful neuro-developmental mechanisms of experienced subjective and objective social inequality are essential to informing and advancing better prevention and intervention approaches for underserved EA . In terms of potential future directions, noting of course that these relationships may not only be due to environment, but can be exacerbated in diathesis:stress contexts, recent studies have highlighted the importance of examining day-to-day shifts in mood and anxiety symptoms, both in terms of cannabis use risk and resilience in this age group , and also in terms of social interactions and prosocial behavior during experiences of subjective and objective social inequality and subsequent heightened stress for youth . Most important within these studies is the indication of the rapid and dynamic shifts that happen in mood and marijuana grow system use, on a day-to-day level in this age group, and the critical importance of capturing that variability within a platform such as ecological momentary assessment  and/or daily diaries, which can accurately and flexibly capture these highly dynamic emotional and subjective and objective social inequality experiences .

These approaches facilitate the capture of dynamic variance across multiple levels of the human experience,as well as the cumulative consequences of persistent and/or ongoing  experienced subjective and objective social inequality . Equally relevant to future evaluations, the study of experienced social inequality necessitates designs that combine idiographic methods to examine individual-level patterns of variability and change with more traditional nomothetic methods to examine patterns of differences be- tween groups . In this regard, Ong advocates for use of “daily process designs ”, which evaluate across a continuum of:  experienced social inequality in near real-time;  the tracking of changes in emotions, proximate to their incidents and moments of change;  preservation of temporal sequences; and  minimization of retrospection bias, in order to better capture the interactions between stressors , adaptive resources , and outcomes. Finally, in terms of future work, it is important to note that most EA do not move into full sustained cannabis use disorders. Evaluations that may be able to engage a longitudinal prospective research protocol to identify and uncover which EA are most protected from risk trajectories would be highly valuable for youth healthcare providers. Work that can help determine which neuro-developmental factors protect EA will be crucial to guide more impactful prevention and intervention programming for this age group –and would be invaluable to helping support young people thrive even amidst current and future emotional risk challenges.

Even EA who share experiences of inequality or similar challenges in emotion regulation may not experience similar outcomes some will experience stable or worsening problems while others will transition towards resilience and recovery. In some contexts, such as for individuals in high subjective and objective inequality contexts, differences in emotion regulation and the associated underlying neural response may be more salient predictors of outcome than in other con- texts.In the past year, 25% of EA ages 18–24 have initiated/increased cannabis use in response to negative stress and emotions – the steepest rise in use across all queried age groups; further, 43% of EA reported past year cannabis use, the highest rate observed in 35 years .While the literature supports important links between negative affect and social in- equality , as well as between negative affect and indoor cannabis grow system use , there is a gap in our understanding of how these processes interact with each other to contribute to EA cannabis use, as well as in our understanding of the cognitive and developmental neuroscience behind these relationships.

For example, during some phases of development, social inequality may directly alter emotion and regulatory networks to increase risk. At other stages, objective and subjective social inequality may increase cannabis use risk either via direct influences  or as contextual moderators of risk . Further, effects of objective and subjective inequality may be additive or themselves interact. Future research would benefit from incorporating metrics of subjective experiences of social inequality  in addition to the widely used “objective ”metrics such as zip code, income, parent education, employment, and environmental instability  when studying relationships to EA hazardous cannabis use. In addition, differentiating the multi-component processes of emotion regulation and studying the constituent parts  may elucidate behavioral, functional, and neural correlates  that have implications for EA decision-making risk in “high-stakes ”emotional contexts like cannabis use . We are extremely far from understanding the potential neurodevelopmental and socio-emotional resilience factors in these equations.

As clinicians and clinical scientists, operating without these data directly obstructs our capacity to identify essential risk/protective factors of EA socio-emotional health in a way that meaningfully helps providers prevent and intervene with EA cannabis use . Of even greater relevance and salience, data to help elucidate meaningful neurodevelopmental mechanisms of experienced subjective and objective social inequality are essential to informing and advancing better prevention and intervention approaches for underserved EA . In terms of potential future directions, noting of course that these relationships may not only be due to environment, but can be exacer- bated in diathesis:stress contexts, recent studies have highlighted the importance of examining day-to-day shifts in mood and anxiety symptoms, both in terms of cannabis grow set up use risk and resilience in this age group , and also in terms of social interactions and prosocial behavior during experiences of subjective and objective social inequality and subsequent heightened stress for youth . Most important within these studies is the indication of the rapid and dynamic shifts that happen in mood and cannabis use, on a day-to-daylevel in this age group, and the critical importance of capturing that variability within a platform such as ecological momentary assessment and/or daily diaries, which can accurately and flexibly capture these highly dynamic emotional and subjective and objective social inequality experiences .

These approaches facilitate the capture of dynamic variance across multiple levels of the human experience , as well as the cumulative consequences of persistent and/or ongoing  experienced subjective and objective social inequality . Equally relevant to future evaluations, the study of experienced social inequality necessitates designs that combine idiographic methods to examine individual-level patterns of variability and change with more traditional nomothetic methods to examine patterns of differences between groups . In this regard, Ong advocates for use of “daily process designs ”, which evaluate across a continuum of:  experienced social inequality in near real-time;  the tracking of changes in emotions, proximate to their incidents and moments of change;  preservation of temporal sequences; and  minimization of retrospection bias, in order to better capture the interactions between stressors , adaptive resources , and outcomes. Finally, in terms of future work, it is important to note that most EA do not move into full sustained cannabis use disorders. Evaluations that may be able to engage a longitudinal prospective research protocol to identify and uncover which EA are most protected from risk trajectories would be highly valuable for youth healthcare providers.