Theory and Practice of VR/AR in K-12 Science Education—A Systematic Review

: Effective teaching of science requires not only a broad spectrum of knowledge, but also the ability to attract students’ attention and stimulate their learning interest. Since the beginning of 21st century, VR/AR have been increasingly used in education to promote student learning and improve their motivation. This paper presents the results of a systematic review of 61 empirical studies that used VR/AR to improve K-12 science teaching or learning. Major ﬁndings included that there has been a growing number of research projects on VR/AR integration in K-12 science education, but studies pinpointed the technical affordances rather than the deep integration of AR/VR with science subject content. Also, while inquiry-based learning was most frequently adopted in reviewed studies, students were mainly guided to acquire scientiﬁc knowledge, instead of cultivating more advanced cognitive skills, such as critical thinking. Moreover, there were more low-end technologies used than high-end ones, demanding more affordable yet advanced solutions. Finally, the use of theoretical framework was not only diverse but also inconsistent, indicating a need to ground VR/AR-based science instruction upon solid theoretical paradigms that cater to this particular context.


Augmented Reality/Virtual Reality(AR/VR) Applications and Beliefs
Science education for primary and secondary school students are facing a variety of challenges nowadays. On the one hand, scientific knowledge often contains a large number of abstract and complex concepts [1], which is difficult for children and adolescents to internalize, even with the help of words and 2D images [2,3]. For example, food digestion has been documented as an essential topic in many countries' primary school science curriculums [4][5][6], but without vivid animation, it can be overwhelming for students to obtain accurate understanding with their pure imagination. On the other hand, implementing real scientific experiments is often bounded by reality conditions, such as a lack of materials, high cost for necessary equipment, safety risks, or difficulties in geographical distance [7].
To tackle the above challenges, researchers have resorted to computing technologies, which are suggested should play a crucial role in student learning [8], comprehension of science concepts, as well as scientific reasoning skill development [9,10]. This is especially true for Generation Z, who have been born in the digital era, and have technologies permeated into virtually every aspect of their lives [11]. The way Gen Z processes information requires educators to not only teach with basic technologies, but capitalize the full potential of e-learning 4.0 [12], which is more personalized, data-based, and gamified [13]. For instance, instead of viewing pictures of digesting organs, students may use Google Board to view food digestion in action, and see clearly how food is processed in each organ with the naked eye. Among all advanced computing technologies, virtual reality (VR) and augmented reality (AR) are increasingly capturing educators' and learners' attention. In particular, VR is defined as a real-time graphical simulation in which the user interacts with the system via Table 1. A comparative analysis of related reviews.

Reference
Covered Years Research Topics Technology Type Grade Level Analyzed Dimensions [28] 2004-2011 Science learning AR Not specified in subject terms) AND ("science education" or "science teaching" or "science learning" in abstract) AND ("primary school" or "elementary school" or "primary education" or "high school" or "k-12" in abstract). To ensure both quality and accuracy, only peer-reviewed journal papers with full text available have been included. This paper establishes the following inclusion and exclusion criteria (Table 2), and reviews each paper to determine whether it is eligible for analysis.

Inclusion Criteria Exclusion Criteria
Students used VR/AR devices to learn Not using VR/AR treatment as an independent variable The participants were primary or secondary school or high school students For preschool children, special education, college students, teachers and other adult learners Learning of science Non-science subjects Empirical studies Literature reviews, commentaries or meta-analysis Written in English Written in other languages On this basis, the researchers performed the PRISMA review process (Figure 1), including identification, screening, qualification and analysis. After several rounds of screening, 61 papers meeting the standard were eventually retained (listed in the Appendix A), labeling ID1-ID61 sequentially. establishes the following inclusion and exclusion criteria (Table 2), and reviews each paper to determine whether it is eligible for analysis.

Inclusion Criteria
Exclusion Criteria Students used VR/AR devices to learn Not using VR/AR treatment as an independent variable The participants were primary or secondary school or high school students For preschool children, special education, college students, teachers and other adult learners Learning of science Non-science subjects Empirical studies Literature reviews, commentaries or meta-analysis Written in English Written in other languages On this basis, the researchers performed the PRISMA review process (Figure 1), including identification, screening, qualification and analysis. After several rounds of screening, 61 papers meeting the standard were eventually retained (listed in the Appendix A), labeling ID1-ID61 sequentially.

Coding Scheme
To better understand these studies, seven types of coding scheme were either adapted or developed as follows: (1) Codes for bibliometric analysis. In reference to Zou et al.'s [34], the bibliometrics information may be categorized by published years, distributed journals, involved disciplines and grades. (2) Codes for theories. Zydney and Warner

Coding Scheme
To better understand these studies, seven types of coding scheme were either adapted or developed as follows: (1) Codes for bibliometric analysis. In reference to Zou et al.'s [34], the bibliometrics information may be categorized by published years, distributed journals, involved disciplines and grades. (2) Codes for theories. Zydney and Warner propose that there are three theoretical types, namely the grounded theoretical foundations, cited theoretical foundations and theoretical foundations not provided [35]. (3) Codes for learning activities. Based on Luo's approach, learning activities can be analyzed from the perspective of learning mode, such as collaborative learning, inquiry-based learning, receptive learning and so on [36]. (4) Codes for research design. Luo also categorizes research design in six aspects, including the type of research, research method, number of experiments, study length, data collection method, and data analysis methods [36]. ( [19,37], AR/VR technologies may be divided into four types, namely they are immersive VR, desktop VR, image-based (or tag based) AR and location-based AR. Meanwhile, Hwang et al. propose that devices of AR/VR refer to the hardware equipment they rely on, such as tablet computer, cameras, desktop computer, smart phone, etc. [38]. (6) Codes for content focus. In reference to Li and Tsai's classification of cognitive goals, we have coded the science learning content into six dimensions: scientific knowledge/concept, scientific reading, scientific process, problem solving, scientific thinking and scientific literacy [39]. It should be noted that scientific literacy is a comprehensive index, which includes the connotation of the first five indicators. (7) Codes for outcomes. Drawing upon Bloom's classification system [40], we have coded learning outcomes as one of the following: cognition, affection and behavior. Meanwhile, from the perspective of effectiveness, the papers were also classified as positive effect, negative effect or mixed effect. The positive effect means that the research results confirmed the research hypothesis; the negative effect means that the research hypothesis was refuted, and the mixed effect refers to having a positive effect in some of the variables and a positive effect in others.

Research Trends
The distribution of publication per year is shown in Figure 2. It can be seen that the number of papers published per year has maintained relatively stable from 2002 to 2018, with no more than four papers every year. However, it surged up to eight in 2019, and 16 in 2020, indicating that more scholars have been paying attention to this field.
propose that there are three theoretical types, namely the grounded theoretical foundations, cited theoretical foundations and theoretical foundations not provided [35]. (3) Codes for learning activities. Based on Luo's approach, learning activities can be analyzed from the perspective of learning mode, such as collaborative learning, inquiry-based learning, receptive learning and so on [36]. (4) Codes for research design. Luo also categorizes research design in six aspects, including the type of research, research method, number of experiments, study length, data collection method, and data analysis methods [36]. (5) Codes for VR/AR technologies and devices. According to Sun et al. and Chen [19,37], AR/VR technologies may be divided into four types, namely they are immersive VR, desktop VR, image-based (or tag based) AR and location-based AR. Meanwhile, Hwang et al. propose that devices of AR/VR refer to the hardware equipment they rely on, such as tablet computer, cameras, desktop computer, smart phone, etc. [38]. (6) Codes for content focus. In reference to Li and Tsai's classification of cognitive goals, we have coded the science learning content into six dimensions: scientific knowledge/concept, scientific reading, scientific process, problem solving, scientific thinking and scientific literacy [39]. It should be noted that scientific literacy is a comprehensive index, which includes the connotation of the first five indicators. (7) Codes for outcomes. Drawing upon Bloom's classification system [40], we have coded learning outcomes as one of the following: cognition, affection and behavior. Meanwhile, from the perspective of effectiveness, the papers were also classified as positive effect, negative effect or mixed effect. The positive effect means that the research results confirmed the research hypothesis; the negative effect means that the research hypothesis was refuted, and the mixed effect refers to having a positive effect in some of the variables and a positive effect in others.

Research Trends
The distribution of publication per year is shown in Figure 2. It can be seen that the number of papers published per year has maintained relatively stable from 2002 to 2018, with no more than four papers every year. However, it surged up to eight in 2019, and 16 in 2020, indicating that more scholars have been paying attention to this field.  The journal distribution is shown in Figure 3. The most published journals were Journal of Science Education and Technology (9), Computers and Education (7) and British Journal of Educational Technology (6). Other less frequently published journals include Journal of Educational Technology and Society, International Journal of Computer-Supported Collaborative Learning, Interactive Learning Environments, and so on. The journal distribution is shown in Figure 3. The most published journals were Journal of Science Education and Technology (9), Computers and Education (7) and British Journal of Educational Technology (6). Other less frequently published journals include Journal of Educational Technology and Society, International Journal of Computer-Supported Collaborative Learning, Interactive Learning Environments, and so on. The cross-distribution by scientific discipline and level of education is shown in Table  3. The subjects were unevenly distributed across disciplines, with most focusing on Physics (23) and Biology (13). As for the participants, 50% were primary school students, 30.6% were junior students, and 19.4% were high school students.  Astronomy  1  1  0  2  Biology  10  2  1  13  Chemistry  0  1  1  2  Environmental Science  1  1  2  4  Geography  6  3  2  11  Medical Science  0  2  3  5  Physics  13  7  3  23  Physiology  2  2  0  4  STEM  1  0  0  1  Science  2  3  2  7  Total  Note: Some studies involved multiple levels of education or disciplines, so the total number is more than 67.

Theories
With reference to Zydney and Warner, theories may be coded as one of three types: grounded theoretical foundations, cited theoretical foundations, and theoretical foundations not provided [35].

Grounded Theoretical Foundations
Grounded theoretical foundations refer to the explicit proposal to carry out research under the guidance of a certain theory. Among the 61 papers, 21 (34.4%) of them clearly indicated the theories they used, as shown in Appendix B. These theories cover a wide The cross-distribution by scientific discipline and level of education is shown in Table 3. The subjects were unevenly distributed across disciplines, with most focusing on Physics (23) and Biology (13). As for the participants, 50% were primary school students, 30.6% were junior students, and 19.4% were high school students. Note: a Some studies involved multiple levels of education or disciplines, so the total number is more than 67.

Theories
With reference to Zydney and Warner, theories may be coded as one of three types: grounded theoretical foundations, cited theoretical foundations, and theoretical foundations not provided [35].

Grounded Theoretical Foundations
Grounded theoretical foundations refer to the explicit proposal to carry out research under the guidance of a certain theory. Among the 61 papers, 21 (34.4%) of them clearly indicated the theories they used, as shown in Appendix B. These theories cover a wide range of fields, including pedagogy, psychology, and learning science. This demonstrates that VR/AR research has integrated the latest developments of contemporary pedagogy, psychology, and learning science research. Meanwhile, it also shows that solid understanding of theoretical paradigms are perceived as critical for effective VR/AR instructional design.

Cited Theoretical Foundations
Among the 61 papers, 11 (18%) of them cited theories to analyze the research results. These theories were not directly applied to the design of VR/AR learning activities. Among the cited theories, constructivism was most frequently used (i.e., ID14, ID22, ID28, ID23, ID42, ID55), indicating that learners' active role and centrality were underlined in these studies. The second most cited theory was Mayer's cognitive theory of multimedia learning. Three papers (ID22, ID9, ID56) cited the continuous principle of the theory to demonstrate how learning materials designed according to the principle could effectively reduce the cognitive load of learners and improve learning performance. The other cited theories include cognitive load theory (ID22, ID56), cooperative learning theory (ID23, ID55), game-based learning theory (ID2), and so on.

Theoretical Foundations Not Provided
Thirty papers (49.2%) did not cite any theory to inform their learning or research design, but did mention certain terms closely related to particular theories. For example, Gnidovec et al. (ID36) studied 13-and 14-year-old students' technology acceptance of AR, which was a construct from the Technology Acceptance Model [41].

Learning Activities
The denotation of learning activities is shown in Appendix C. Among all the learning activities, inquiry-based learning was used the most (34 papers), followed by receptive learning (12 papers), problem-based learning (8 papers), game-based learning (6 papers), and collaborative learning (5 papers). It should be noted that in experimental research, only activities of the treatment group were accounted for, due to the inexplicit nature of the control group activity description.
The research using inquiry-based learning enabled learners to understand scientific concepts or phenomena through the operation and interaction of virtual things with the support of VR/AR. For example, Squire and Jan (ID2) required students to learn about polychlorinated biphenyls and mercury by exploring the cause of death of Ivan in VR games [42]. Sun et al. (ID6) built a VR model to simulate the movement of the Sun, the Moon, and the Earth [19]. Papers that adopted receptive learning used VR/AR to present virtual objects, so that learners could observe scientific things or phenomena in an intuitive way. For instance, Shim et al. (ID1) developed a VR system called VBRS simulating the iris and pupil of the human eye, through which students could see flowers of various shapes when they shifted between multiple viewpoints by pressing the number keys on the keyboard [43].
Three papers integrated collaborative learning, while they also adopted inquirybased learning at the same time. That is to say, learners inquired about certain objects or phenomenon in collaborative ways. For instance, Chiang et al. (ID10) used location-based AR to assist students' investigation of the ecological environment of the pond near the school [44]; Fidan and Tuncel (ID23) developed an AR-based application, which used sound and animation to create an inspiring atmosphere [1].
There was one paper on flipped learning, topic-based learning and design-based learning respectively, as shown in Appendix C.

Research Designs
The research methods were combed in terms of six aspects, and the statistical results are shown in Table 4. First of all, the number of experimental studies (47 papers) was far more than that of investigation studies (14 papers). Secondly, the majority of studies employed quantitative design (31 papers) and mixed-research design (26 papers). Thirdly, most studies (24 papers) used VR/AR for teaching within 0-3 h, as compared to over three hours, and 25 papers reported teaching with AR/VR for only one class session. Furthermore, questionnaires (44 papers) and knowledge tests (32 papers) were used as major data collection methods. Finally, a t-test was the most frequently adopted statistical measure (34 papers).

Technologies and Devices
In terms of technologies, four types of VR/AR were identified (see Figure 4), including immersive VR, desktop VR, image-or marker-based AR, and location-based AR [28]. The immersive VR system surrounds the user with a 360-degree virtual environment; the desktop VR system is displayed to the user on a conventional computer monitor, whereas a 3-D perspective displays technology projects 3-D objects onto the 2-D plane of the computer screen [19].
Specifically, seven (ID6, ID16, ID27,ID43, ID48, ID51, ID56) of the 61 papers used  immersive VR; 14 papers (ID3, ID20, ID4, ID1, ID5, ID24, ID41, ID42, ID49, ID52, ID55,  ID59, ID60, ID61) used desktop VR; seven papers (ID9, ID10, ID21, ID30, ID46, ID53 ID54) used location-based AR, 31 papers used image-or marker-based AR, and two papers (ID14, ID18) used two kinds of VR or AR at the same time.  As for the device or hardware equipment, it may be categorized as the following ( Figure 5). It can be seen that tablet PC (24 papers), desktop PC (18) and smart phone (14) were the most frequently used devices, whereas devices like the puzzle set were least employed. As for the device or hardware equipment, it may be categorized as the following ( Figure 5). It can be seen that tablet PC (24 papers), desktop PC (18) and smart phone (14) were the most frequently used devices, whereas devices like the puzzle set were least employed. As for the device or hardware equipment, it may be categorized as the following ( Figure 5). It can be seen that tablet PC (24 papers), desktop PC (18) and smart phone (14) were the most frequently used devices, whereas devices like the puzzle set were least employed.

Content Focus
The first type of content was scientific knowledge/concept, which was also the most targeted among other types. Specifically, in 47 out of 61 papers, researchers used VR/AR technology to help learners understand scientific knowledge and concepts. For example, Wrzesien (ID5) used an immersive interactive virtual water world software called E-Junior to let learners play the role of Mediterranean residents or fish in the sea, participating in daily activities of the Mediterranean, and learning the concept and knowledge of marine ecology through exploration in the virtual world [20].
The second type of content was science reading. There was one paper on AR technology that supported scientific reading. In the research of Lai et al.'s (ID25), students used mobile devices equipped with an AR science learning system to scan the textbooks, and the relevant pictures would immediately and dynamically appear above them [45]. The experimental results showed that, compared with the traditional multimedia science learning method, the treatment significantly improved the students' academic performance and learning motivation, and also significantly reduced their perception of the external cognitive load in the learning process.

Content Focus
The first type of content was scientific knowledge/concept, which was also the most targeted among other types. Specifically, in 47 out of 61 papers, researchers used VR/AR technology to help learners understand scientific knowledge and concepts. For example, Wrzesien (ID5) used an immersive interactive virtual water world software called E-Junior to let learners play the role of Mediterranean residents or fish in the sea, participating in daily activities of the Mediterranean, and learning the concept and knowledge of marine ecology through exploration in the virtual world [20].
The second type of content was science reading. There was one paper on AR technology that supported scientific reading. In the research of Lai et al.'s (ID25), students used mobile devices equipped with an AR science learning system to scan the textbooks, and the relevant pictures would immediately and dynamically appear above them [45]. The experimental results showed that, compared with the traditional multimedia science learning method, the treatment significantly improved the students' academic performance and learning motivation, and also significantly reduced their perception of the external cognitive load in the learning process.
The third type of learning was scientific process, and two paper (ID15, ID61) focused on this. For example, Hsu et al. (ID15) used AR technology to build a surgical simulator to train students performing laparoscopic surgery and cardiac catheterization [46]. They found that students had positive cognition and high level of participation in AR courses and simulators, and their interest in learning greatly increased.
The fourth type was problem-solving (9 papers). For example, Kyza and Georgiou (ID21) used an AR application called TraceReaders, which allowed learners to write location-based AR applications for outdoor survey learning [47].
Three papers (ID2, ID37, ID44) embodied the fifth type of content, which was science thinking. For example, Chang et al. (ID37), with the support of mobile AR, aided students in contemplating about the dilemma of building nuclear power plants and using coal-fired power plants in virtual cities [48]. It is found that students' previous knowledge and beliefs had a certain impact on students' ability to participate in learning and reasoning.
Finally, there were also two papers (ID34, ID50) focusing on acquiring science literacy. Scientific literacy is the comprehensive embodiment of scientific knowledge, scientific thinking, and scientific ability [49]. Wahyu et al. (ID34) found that mobile AR assisted STEM learning could significantly improve students' scientific literacy than traditional learning methods [49].

Outcomes
The learning outcomes of 61 papers were classified according to the theory of Bloom's instructional objective classification [40]. As is shown in Figure 6, 46 papers set cognitive goals, and six of them reported mixed effects; 40 papers established affection goals, and five of them reported mixed effects. Three papers aimed to improve behaviors, all of which reported positive effects.
acy. Scientific literacy is the comprehensive embodiment of scientific knowledge, scientific thinking, and scientific ability [49]. Wahyu et al. (ID34) found that mobile AR assisted STEM learning could significantly improve students' scientific literacy than traditional learning methods [49].

Outcomes
The learning outcomes of 61 papers were classified according to the theory of Bloom's instructional objective classification [40]. As is shown in Figure 6, 46 papers set cognitive goals, and six of them reported mixed effects; 40 papers established affection goals, and five of them reported mixed effects. Three papers aimed to improve behaviors, all of which reported positive effects.
But there are still some studies (6 papers: ID3, ID5, ID33, ID38, ID49, ID52) concluding that VR/AR technology was no more effective than non-VR/AR technology in improving students' performance. For example, Chen et al. (ID3) developed an Earth VR motion system to help understand the changes of day and night and four seasons caused by the Earth's rotation. The researchers conducted a pre-test and a post-test on the students, and the scores of most post-test items were higher than those of pre-test. However, for the questions about the rotation of the Earth, the students' post-test score was significantly lower than the pre-test score due to the fact that the system did not provide sufficient information about the Earth's rotation [14]. Similarly, Wrzesien et al. (ID5) concluded that there was no significant difference in academic performance between the experimental group using VR technology and the control group using traditional learning methods [20]. A possible cause could be that the attraction of the virtual environment had diverted students' attention. They were more interested in operating virtual things than scientific concepts themselves. Wang (ID33) found that students who used e-Book learning materials had higher scores than students who used AR learning materials, although the difference was not statistically significant [51]. It may be inferred that well-designed AR content could limit students' thinking, because some students preferred studying directly based on the guidance of AR content as soon as they received the materials, and completed the tasks without thinking. E-books do not provide very detailed demonstration information, but the text information guided by graphics makes learners think first and then work. Chen (ID38) compared the differences between the game method and AR supported learning, and found that there was no significant difference between the two methods in improving academic performance [37]. It may be because several methods used in the experiment provided immediate reflection tips when students submitted wrong answers, while AR did not give full play to its advantages in multimedia learning.
However, there were still some studies that concluded with negative results in such dimensions as motivation (1 paper: ID32), technology acceptance (1 paper: ID46), selfefficacy (1 paper: ID48), satisfaction (1 paper: ID55), expectation (1 paper: ID56) and so on. It could be because that the use of VR/AR was too complex to operate appropriately and effectively, or that there was insufficient information provided, which could have resulted in learning difficulty. For example, Lu et al. (ID32) found that the experimental group using AR has lower learning motivation than the control group without AR. The author believed that the main reason was that learners were unfamiliar with materials and equipment, which posed certain learning challenges [55]. Lo et al. (ID46) found that the perceived usefulness of using AR was correlated with age. That is, older students tended to think that AR applications were not very useful. It was hypothesized by the authors that the older the students were, the harder it was for them to follow the teacher's instructions, or the more difficult for them to learn [56]. Shin (ID55) found that the learners did not enjoy the experience of desktop VR, because it did not generate a strong sense of immersion [57].

Behavioral Goals
There were three papers (ID11, ID54, ID57) that focused on realizing behavioral goals. These studies reached a consistent conclusion that the use of VR/AR could improve students' learning behavior. For example, Yoon and Wang (ID11) compared the time of interaction with devices and team cooperation between AR users and non-AR users. It was found that the former's time of interaction with devices was significantly higher than that of the latter, while the team cooperation was the opposite. This indicated that AR devices improved participation in learning, but also affected cooperation between teams to some degree [58].

Trends in the Integration of VR/AR in K-12 Science Education
First of all, there is a growing number of studies in VR/AR's integration in K-12 science education, indicating researchers' and practitioners' interest in using VR/AR to enhance learning science. For instance, 20 out of 60 papers were published in the last two years. Despite this, the majority of studies were published much more in generic educational technology journals, such as Computers and Education and Educational Technology and Society, which accounted for 85% of all. Contrarily, only few domain specific science education journals (i.e., Journal of Science Education and Technology) published such studies. This may be due to the fact that for most K-12 science teachers, VR/AR is an emerging technology that seems novel and inaccessible, and its effects on students is still ambiguous without conclusive findings or universal instructional design models [59,60]. Therefore, in future research more attention should be paid to the exemplary integration of VR/AR into teaching specific science topics, foster deep integration and enumerate the particular effectiveness of VR/AR application on students' learning outcomes, so that science teachers become more receptive of VR/AR uses.
Secondly, the theories involved appeared very diverse. On the one hand, this diversity demonstrates VR/AR's capacity of accommodating a multitude of theories; on the other hand, it also indicates the lack of an over-arching theoretical paradigm that could guide AR/VR-based science instructional design. Such a paradigm would not be possible without the collaborative effort from learning scientists, science teaching experts, instructional designers and VR/AR specialists. The absence of any of the stakeholders may lead to an ineffective design framework. It should also be noted that 45% of the reviewed papers did not cite any theory, which could lead to unsubstantiated interpretation of obtained results.
Thirdly, inquiry-based learning was the most adopted learning model (87.5%) among the reviewed studies, which is consistent with previous findings that inquiry-based learning was one of the most commonly used learning models [28][29][30][31]. Regardless, this learning model was not entirely gauged with the measured learning outcomes in the reviewed studies. That is, although students indeed used VR/AR devices, teachers did not necessarily capitalize on the benefits of inquiry-based learning, beside providing students with immersive or lifelike experiences. Previous studies have shown that inquiry-based learning without sufficient guidance is not significantly better than traditional textbook teaching [61]. Thus, it must be cautioned that there is a fine line between inquiry-based learning and simply asking students to explore or view an VR/AR object or environment. For example, Salmi et al. (ID19) developed a mobile AR application to enable students to explore the different reactions between a number of atoms and molecules, within which students only needed to interact with the AR system to view the structure of atoms and molecules; thus, it could be hardly deemed as inquiry-based learning [62].
Fourthly, in terms of the research methods, there were more quantitative studies (50.8%) than qualitative or mixed-method studies (42.6%), more experimental designs (77%) than investigation designs (23%). The emphasis on experimental studies could be because that those experimental studies were practically more welcomed than investigative studies in nearly all academic journals, owing to their more advanced statistical analysis measures and illustrations. Meanwhile, experimental studies help teachers make more instant and precise adjustment to their existing science teaching, such as integrating a certain VR/AR software, or a device. On the other hand, investigation studies are more suitable for understanding students' perceptions, attitudes or satisfaction toward the generic VR/AR technologies, the results of which may not be directly applied to specific instructional design or adaptation.
Last but not least, there were a variety of VR/AR technologies employed, such as location-based AR, image-or marker-based AR, immersive VR, and desktop VR, but the ratio of using advanced VR/AR technologies was very low. This is in direct contrast to Pellas, Dengel and Christopoulos's finding that 60% of the studies used high-end immersive devices, while nearly 30% used low-end solutions [36]. One major reason could be that school teachers were unlikely to purchase higher-end technologies, for experiment's sake without school's financial support. Moreover, considering K-12 students' cognitive ability and psycho-motor skills, it is not only appropriate but also safe for them to use lessadvanced and -expensive devices, so as to avoid the risks of under-utilization or damage. In other words, to increase the diffusion of AR/VR use in K-12 science education, there is a need to develop more affordable and portable devices that can be easily operated, so that both science teachers and students can utilize them effectively and efficiently. Also, given that there were only four papers (ID8, ID11, ID17, ID34, accounting for 10%) that focused on learning with AR/VR in informal environment, it may be suggested that VR/AR technologies that can be easily transported from one place to another be developed, so that students can learn with such technologies seamlessly in and out of class. For instance, students who were instructed to observe planets with VR/AR devices in class may continue to learn this topic at home by using both VR/AR technologies and their personal microscope.

Issues in the Integration of VR/AR in K-12 Science Education
Despite its apparent advantages, VR/AR also has its limitations or issues. The first type of issues reflected in previous studies are technical issues, which refer to either the inherent limitations of VR/AR technologies, or the associated technological glitches, such as lack of mobility and inconvenience of using, especially for immersive VR. For example, HMD, trackers and other VR-related utilities like the Cave Automatic Virtual Environment could often cause such difficulties [14].
The second type of issues are pedagogical issues. Teachers who use VR/AR to teach science may have problems in using it effectively and efficiently, including identifying the most suitable resources, designing the most appropriate activities, or conducting the most precise assessments. For instance, VR/AR has been reported as distracting and visually overloading. Wrzesien and Raya (ID55) found that there was no significant difference between the results of the experimental group using virtual devices and the control group without virtual devices. Learners were easy to get lost in the virtual environment, and a lack of sufficient learning information was the main reason for this phenomenon [20]. Teachers thus are obligated to sift through various VR/AR resources, and identify those that are age-appropriate, visually comfortable, and mentally congruent. Also, as Charsky and Ressler (2011) point out, the lack of teaching methods and objectives can make students confused and depressed, and even increase their knowledge overload and reduce their learning motivation [63]. Some studies noted the limitations of VR/AR technology and sought to overcome them with supplementary activities. For example, Yoon et al. (ID8) used knowledge prompts, a bank of peer ideas, working in collaborative groups, instructions for generating consensus, and student response forms for recording shared understanding [64]. These scaffolds could promote collaboration within the peer groups by encouraging students to discuss their observations and reflections of their experience. Another pedagogical issue lies in the comprehensive and accurate evaluation of student learning outcomes. For instance, students' cognitive and affective outcomes were mainly measured, whereas behavioral change was less emphasized.
The third type of issues can be categorized as social issues. For instance, the price of VR/AR devices is considered a social issue, rather than a technical issue, because the price is not solely determined by the technical complexity or sophistication, but its relative novelty among other technologies as well as the income level of its targeted consumers. Meanwhile, whether teachers can integrate VR/AR into science teaching is greatly dependent upon the social perceptions of such technologies, as well as their school support, both of which constitute the context for our topic. For instance, according to Chih et al., not all schools were willing to pay a high price for virtual display devices and real-world devices [14].
There are also several research issues. In terms of the research length, about 62% of the studies observed usage for less than 10 h. Under such circumstances, probability factors like the novelty effect could hardly be eliminated. Also, while a multitude of variables were examined, including scientific reading, scientific process, scientific problem solving, scientific literacy and so on, most studies still focused on low-level cognition through knowledge tests; high-level thinking ability has not received adequate attention. According to Bloom's goal classification, memory, understanding, and application correspond with low-order thinking abilities, whereas analysis, evaluation and creation belong to high-order thinking abilities [65]. Academic research shows that "injection" mode is usually used to cultivate high-level thinking ability in science learning; that is, the learning of thinking skills is integrated with the learning of the science curriculum. In this mode, students are fully involved in thinking practice, focusing on the learning process and understanding of meaning. After solving certain challenging problems, high-level thinking skills can be developed [65]. However, the emphasis on higher-order thinking has been absent in most reviewed studies in this paper. This is consistent with previous research that the application of VR/AR in science education mainly focuses on the understanding of scientific concepts and phenomena [28,29]. For example, 85% of the studies focused on students' mastery of scientific knowledge or concepts, without mentioning critical thinking, social reasoning ability, innovation tendency and other high-level thinking ability. Moreover, the data analysis methods relied mostly on t-test (55.7%), which would be insufficient to analyze more complex relationships or phenomenon.

Implications and Recommendations
Based on the issues identified above, we offer the following suggestions in both theoretical advancement and practical improvement of the VR/AR's integration in K-12 science education. As for science teachers, it is paramount to be familiar with both psychological and pedagogical theories, so that VR/AR-based activities can effectively and efficiently promote student learning interest as well as achievement. They should also be very selective in choosing the most appropriate and authoritative VR/AR apps or resources, so as to not only meet the learning demand of students, but also avoid foreseeable technical glitches. When designing learning activities, it is essential for teachers to target more advanced skills, such as critical thinking, in order to cultivate students' inquiry-based mindset. What's more, with the knowledge of trending VR/AR practices, teachers may embrace more learning models like collaborative learning and project-based learning into their science instruction. Researchers, on the other hand, are suggested to conduct more mixed-method studies, which offer a comprehensive and profound understanding of students' experiences and changes in cognition, affection and behavioral skills. They may as well include teachers as research participants, instead of focusing on students only, so that barriers in teachers' intention or proficiency of VR/AR integration could be identified and addressed at an early stage. When possible, studies that last longer and have repeated trials are strongly recommended. Longer interventions with repeated evaluation could help solidify the benefits of VR/AR-based science instruction, and boost teachers' confidence with its exemplary uses. Finally, technical experts or software engineers may be prompted to develop more affordable, portable and personalized subject-specific VR/AR technologies, and program more science-related immersive VR/AR environment that cater to different grade levels' needs. For instance, lower-grade level students may use AR/VR to gain new experience and direct observation, while higher-grade levels may use it to foster the ability to analyze, evaluate and even create.

Limitations
The current research also has its limitations. For example, the review was very selective, meaning that we intentionally chose journal articles from renowned databases only to ensure quality, rather than including also conference papers or theses. Another limitation was that the citation or reference network analysis were not included, in order to keep the paper more focused and tightened. Future research that aims to conduct a more comprehensive review could enlarge the scope and utilize knowledge mapping software to illustrate the trend and research hot spots with sophisticated displays.

Conclusions
VR/AR is advantageous in K-12 science education [18]. The purpose of this paper was to examine the theoretical and practical trends and issues in existing research on VR/AR's application in K-12 science education between 2000 and 2021, including the publication data, adopted theories, research methods, and technical infrastructure, etc. It was found that there has been a growing number of research projects on VR/AR integration in K-12 science education, but studies pinpointed the technical issues rather than the deep integration of AR/VR with science subject content. Also, while inquiry-based learning was most frequently adopted in reviewed studies, students were mainly guided to acquire scientific knowledge, instead of cultivating more advanced cognitive skills, such as critical thinking. Moreover, there were more low-end technologies used than high-end ones, demanding more affordable yet advanced solutions. In terms of research methods, quantitative studies with students as the sole subjects were mainly conducted, calling for more mixed-method studies targeting both teachers and students. Finally, the use of theoretical frameworks was not only diverse but also inconsistent, indicating a need to ground VR/AR-based science instruction upon solid theoretical paradigms that cater to this particular domain.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the fact that it was written in mostly Chinese.

Conflicts of Interest:
The authors declare no conflict of interest. An AR-based mobile learning system to improve students' learning achievements and motivations in natural science inquiry activities Cited Meyer's multimedia design theory and used the language related to inquiry learning theory but did not cite it

Appendix A
Inquirybased learning ID10 [44] Students' online interactive patterns in AR-based inquiry activities Grounded on knowledge construction theory Inquirybased learning and collaborative learning ID11 [58] Making the invisible visible in science museums through AR devices Not provided Inquirybased learning ID12 [68] Employing Augmented-Reality-Embedded instruction to disperse the imparities of individual differences in earth science learning Used the language related to learning style theory but did not cite it Inquirybased learning ID13 [69] Constructing liminal blends in a collaborative augmented-reality learning environment Grounded on distributed cognitive theory Inquirybased learning and collaborative learning ID14 [70] Enhancing learning and engagement through embodied interaction within a mixed reality simulation Grounded on embodied learning theory; cited constructivism theory; used the language related to learning attitude theory, self-efficacy theory, learning participation theory but did not cite them Inquirybased learning ID15 [46] Impact of AR lessons on students' stem interest Used the language related to learning motivation theory but did not cite it Inquirybased learning ID16 [71] An augmented-reality-based concept map to support mobile learning for science Used the language related to learning motivation theory, learning attitude theory but did not cite it Inquirybased learning and receptive learning ID17 [72] How AR enables conceptual understanding of challenging science content Not provided Receptive learning ID18 [45] The influences of the 2-D image-based AR and VR on student learning Grounded on cognitive load theory; Used the language related to technology acceptance but did not cite it Inquirybased learning Impacts of an AR-based flipped learning guiding approach on students' scientific project performance and perceptions Used the language related to critical thinking theory, group self-efficacy theory, learning motivation theory, and psychological load theory but did not cite it Flipped learning ID21 [47] Scaffolding AR inquiry learning: The design and investigation of the Tracereaders location-based, AR platform Grounded on the theory of experiential learning and used the language related to the theory of inquiry learning but did not cite it Inquirybased learning ID22 [73] Impacts of integrating the repertory grid into an AR-based learning design on students' learning achievements, cognitive load and degree of satisfaction Grounded on situated learning theory and cited constructivism theory, cognitive load theory, and cognitive theory of multimedia learning Receptive learning ID23 [1] Integrating AR into problem based learning: The effects on learning achievement and attitude in physics education Grounded on Situational learning theory; Cited constructivism theory, cooperative learning theory, Self-guidance theory, situational learning theory; Used the language related to learning attitude theory but did not cite it Problem-based learning ID24 [74] Applying VR technology to geoscience classrooms Not provided Problem-based learning ID25 [45] An AR-based learning approach to enhancing students' science reading performances from the perspective of the cognitive load theory Grounded on cognitive theory of multimedia learning and cognitive load theory; used the language related to learning motivation theory but did not cite it Problem-based learning ID26 [21] A usability and acceptance evaluation of the use of AR for learning atoms and molecules reaction by primary school female students in Palestine Not provided Receptive learning ID27 [75] The effect of the AR applications in science class on students' cognitive and affective learning Used the language related to learning motivation theory, learning interest theory, and meaningful learning theory but did not cite them Receptive learning The effect of using VR in 6th grade science course the cell topic on students' academic achievements and attitudes towards the course Cited Piaget's learning theory and used the language related to learning motivation theory but did not cite it Receptive learning ID29 [76] The effect of AR Technology on middle school students' achievements and attitudes towards science education Used the language related to learning motivation theory but did not cite it Topic-based learning ID30 [54] Integration of the peer assessment approach with a VR design system for learning earth science Used the language related to learning motivation theory, critical thinking theory, creative ability theory, and cognitive load theory but did not cite it Design-based learning ID31 [77] Students' motivational beliefs and strategies, perceived immersion and attitudes towards science learning with immersive VR: A partial least squares analysis Used the language related to motivation theory, self-regulation theory, learning attitude theory but did not cite it Inquirybased learning ID32 [55] Evaluation of AR embedded physical puzzle game on students' learning achievement and motivation on elementary natural science.
Used the language related to learning motivation theory but did not cite it Game-based inquiry learning ID33 [51] Integrating games, e-Books and AR techniques to support project-based science learning Used the language related to learning motivation theory but did not cite it Inquirybased learning ID34 [49] The effectiveness of mobile AR assisted stem-based learning on scientific literacy and students' achievement Not provided Inquirybased learning ID35 [78] Using AR to teach fifth grade students about electrical circuits Used the language related to learning attitude theory but did not cite it Receptive learning ID36 [41] Using AR and the Structure-Behavior-Function Model to teach lower secondary school students about the human circulatory system Used the language related to technology acceptance but did not cite it Receptive learning ID37 [48] Students' context-specific epistemic justifications, prior knowledge, engagement, and socioscientific reasoning in a mobile AR learning environment Used the language related to situational cognitive theory, learning engagement theory but did not cite them Inquirybased learning ID38 [37] Impacts of AR and a digital game on students' science learning with reflection prompts in multimedia learning Used the language related to situational learning theory but did not cite it Inquirybased learning ID39 [79] Use of mixed reality applications in teaching of science Used the language related to learning motivation theory, learning attitude theory but did not cite them Receptive learning ID40 [50] Perceived learning in VR and animation-based learning environments: A case of the understanding our body topic Used the language related to constructing knowledge and so on but did not cite it Receptive learning  The impact of internet virtual physics laboratory instruction on the achievement in physics, science process skills and computer attitudes of 10th-grade students Grounded on cognitive and social constructivism theory Problem-based learning Appendix B Table A2. List of grounded theoretical foundations cited in reviewed studies.

Multiple intelligences theory
Students were asked to explore in the virtual environment, so their multiple senses were stimulated, and their ability to establish intellectual and emotional connections with their own world was enhanced. (ID4) The researchers attempted to stimulate primary school students' musical intelligence, bodily-kinesthetic intelligence, spatial intelligence, interpersonal intelligence, and intrapersonal intelligence in VR environment. (ID5) The theory of leisure A Serious Virtual World was constructed with VR, which enabled primary school students to find their potential and skills in the leisure environment, compare with other players in the game, and learn in the cooperative game. (ID5)

Knowledge construction theory
Learning scaffoldings, such as knowledge prompt and peer thinking database, were designed to support 6-8 grade students' knowledge construction in AR environment. (ID8) A location-based mobile device AR system was developed to help learners construct knowledge through discussing problems and sharing knowledge.

Theories Application Scenarios
Multimedia learning theory In this study, an AR-based science learning system was developed based on the contiguity principle of multimedia learning theory, which was used by students to interact with textbooks. (ID25) Based on the interactivity principle, students set up the AR experiment and observed the results. (ID35) According to the multimedia learning theory, an AR game was designed to test its learning efficiency. (ID38) The theory of experiential learning Students collected virtual elements to mimic reality experience. (ID5) The principle of experience continuum and interaction of experiential learning theory were used to design primary school student's learning activities of visiting outdoor space, motivate them to learn, and exert a positive impact on their cognitive and emotional outcomes. (ID21) The researchers developed a system with AR technology that allowed the learner's body to move freely in a multimodal learning environment to enhance embodied learning. (ID58) Situated learning theory An AR-based learning system called Mindtool was designed, which enabled students from fourth graders to explore concepts or solve problems. (ID22) AR environment was used to create heuristic problem situation, so that students aged from 12 to 14 could learn through PBL. (ID23) A health education board game applying AR was developed. This game included eight topics, such as health check, hospital, ambulance and so on, helping students learn health knowledge in a realistic situational environment. (ID47)

Theory of immersion
A research model to understand the learning perception of immersion was proposed, which tested the learning characteristics and evaluated the immersion variables through the individual's motivational beliefs and strategies. (ID31) Technology acceptance theory TAM theory was used to study users' adoption patterns from the perspective of perceived usefulness and perceived ease of use, and a blueprint for the research to be explored was constructed. (ID46) Theory of inquiry learning A VR system named Multi-User Virtual Environments was developed to enable multiple simultaneous participants to enact collaborative learning activities of various types. (ID49) Piaget's cognitive theory Research questions were put forward according to Piaget's cognitive theory and the Inventory of Piaget's Developmental Tasks was used in the study for learners to complete. (ID51) Theory of collaborative learning The learning activity was designed according to theory of collaborative learning, including three parts: (1) a new mixed-reality learning scenario, (2) a student participation framework, and (3) a curriculum. (ID57) Other theories Lin et al. (ID47) used five theories to design their AR health education board game. In addition to the situational learning theory mentioned above, other four theories are scaffolding theory, dual-coding theory, over-learning theory, and competition-based learning theory. In their AR health education board game, users needed to use the developed App to scan question card on the inspection report. Guidance and correct answers were provided at the back of the question card (scaffolding theory). pictures and text were added to the question card as study aids (dual-coding theory). To answer the question rightly, the users needed to repeat practicing again and again (over-learning theory), and the competition mechanism was used by the game to enhance the learning motivation of learners. (ID47) Appendix C Table A3. List of learning activities in reviewed studies. Inquirybased learning   ID2, ID3, ID4, ID5, ID6, ID7, D8, ID9, ID10, ID11,  ID12, ID13, ID14, ID15, ID16, ID18, ID21, ID31,  ID32, ID33, ID34, ID37, ID38, ID41, ID42, ID43,  ID46, ID48, ID49, ID51, ID52, ID56, ID57, ID58 Learners interacted with virtual environment or virtual objects created by VR/AR, and learned scientific knowledge and scientific concepts or phenomena by exploring.  ID1, ID16, ID17, D19, ID22,  ID26, ID27, ID28, ID35, ID36,  ID39, ID40 VR/AR could help learners better understand scientific concepts and phenomena by visualizing invisible things, simplifying complex things, concretizing abstract things, and combining real-world learning objectives with digital content.
Problem-based learning ID23, ID24, ID25, ID41, ID44, ID50, ID60, ID61 Researchers used the environment created by VR/AR as the basis for raising problems and the source of materials for solving problems.
Game-based learning ID45, ID47, ID48, ID53, ID54, ID59 Learning activities were carried out in the form of games. Learners used scientific knowledge to solve problems through interaction with the environment or other learners. The main types of games are story game (ID45), health education board game (ID47), role playing games (ID48 and ID59), and collaborative role playing game (ID53 and ID54).

Flipped learning ID20
Learners used AR-based flipped learning system, to watch videos in advance, finishing homework, and discussing in class.
Topic-based learning ID29, ID55 The researchers developed an AR-based activity manual with 32 learning activities. In the experimental group, teachers used these activity manuals for theme teaching, and students completed learning activities according to the content of the manual. (ID29) The learning content was organized according to different topics, which indicated learning subjects of earth science education. (ID55)

Design based learning ID30
Researchers developed a peer assessment approach and incorporated it into VR design activities, in which students designed their own VR projects to raise environmental awareness and cultivate earth science knowledge.