Taiwan is rich in topography and landscape features. Therefore, in recent years, engineering projects, such as check dams and groundsills, have viewed landscape assessments and integration with the environment as being important, in addition to safety concerns. In mountainous areas, stream erosion control facilities should be considered to reduce the environmental impact, as well as related eco-friendly designs, to increase the harmony of the landscape and environmental sustainability.
Aimed at the design concept of check dam aesthetics, a previous study [1
] proposed that the three independent factors of form, color, and texture, as well as sense factors such as harmony, rhythm, and simplicity, could be used for the aesthetic assessment. Subsequently, a fuzzy logic system for landscape assessment was established using the analytic hierarchy process (AHP), and these principles were used to analyze the relationship of the various factors [2
Furthermore, one study [3
] used visual preference and four cognitive factors—naturalness, harmony, vividness, and closeness—to discuss the visual indicators, visual aesthetic experiences, and applications of the empirical relationship. Another study [4
] adopted a preference-based psychophysical landscape assessment method (scenic beauty estimation (SBE) method) to explore the aesthetic preferences and differences in various artificial structures in a catchment. The SBE method is a psychophysiological experimental method for assessing the beauty of forest landscapes. The concept derives from the stimulus and response models of behavioral psychology, and is based on the signal-detection method and the Thurstone measurement model, improved for assessing the beauty of landscapes [5
Participant-generated image methods, often used in social science research, present a long application history in social science [10
]. Research on participant-generated images (PGIs) in social science can be traced back to 1970. Volunteer participants were offered cameras to take pictures representing specific topics, which might be related to their specific life experiences or places visited. Several qualitative research analysis methods were then used to analyze the pictures taken by the subjects, who became the “participants” in the research.
As type of visual language translation method, the caption evaluation method (CEM), proposed by Koga et al. [11
], is a typical PGI method. Using this method, citizens in Minato, Tokyo, Japan, were requested to freely stroll in Minato and take various environmental pictures with a positive or negative impression. Meanwhile, they were invited to write down their feelings about each facility. The participants were further asked to record their reasons for being satisfied with or adverse to each facility.
Many Japanese scholars then began applying the CEM to evaluate various historical monuments, tourist attractions, and urban open spaces [12
]. For instance, Naoi et al. [18
] applied the CEM to evaluate visitors who selected historical areas as tourism destinations. A total of 30 Japanese college students and 27 Japanese adults interested in architecture and urban planning visited facilities in Japan’s historical and cultural areas and stated their perceptions of the selected facilities. The results revealed the relationship between the elements in the areas and the participants’ perceptions and desires. Furthermore, it also implied the participants’ positive assessment of old buildings and adverse perception of modern architecture.
Chen et al. [19
] collected comments regarding two hydropower dams in Canada from social media platforms such as Instagram. Each picture, and the title for each topic, was coded as a topic category for statistics. Each comment (e.g., dam construction would affect the perceived aesthetics and sense of home, or dam removal would cause lifestyle changes) might affect the relevant value and activity to predict the group items. For instance, the construction of a dam might affect the local citizens’ perceived aesthetics and sense of home, or the deconstruction of a dam might result in lifestyle changes. This type of research could be an extended research method of the CEM.
There are limited studies using visual language translation or the CEM in Taiwan; however, hydraulic engineers need a reference for aesthetic design to meet the landscape requirements. This study uses the CEM as a qualitative research analysis tool for the landscape evaluation of mountain stream facilities. Selecting award-winning projects as the research objects, case studies are conducted to evaluate their design styles and compare them with other types of mountain stream facilities. Meanwhile, the quantitative analysis, using methods such as SBE, is compared with the qualitative results. Finally, qualitative analysis is used to explore the correlations among the characterization factors, perception factors, and preferences. The research goal is to establish a foundation for subsequent aesthetic engineering design in the future. The entire research process is shown in Figure 1
2. Materials and Methods
2.1. Caption Evaluation Method (CEM)
This study applies the caption evaluation method (CEM) to evaluate the landscape assessment of mountain stream engineering facilities in Taiwan. The CEM, first developed from architectural psychology and proposed and practiced by Koga et al. [11
], is a participatory qualitative research method. The method attempts to investigate the reasons why the subjects take pictures at target locations, and the subjects are asked to state their opinions about the images. With such a method, the pictures of the targets taken by the participants are regarded as the trigger factors in their evaluation of the locations, and the acquired pictures are considered clues to their reactions.
In the CEM, three points to caption the pictures are proposed:
To which elements do the subjects pay attention?
What are the features of such elements noticed by the subjects?
How do the subjects treat the noticed features?
Among the three questions, the first and the second questions are designed to gain an understanding of the features of the elements in the environment. The third question is aimed at inducing the subjects’ thoughts regarding the features of such elements. For instance, one subject provided the caption, “It was cool because there were few cars. It would be better when there was no height difference between sidewalks and traffic lanes” [18
]. In the example, “elements”, such as “cars”, “features” of the elements, such as “few”, and “perceptions” of the subject, such as “cool”, are mentioned. Such data could be used for various positive or negative qualitative analyses.
The CEM presents some advantages: (1) such a method could help the subjects focus on various facilities and elements, and state their opinions; (2) the pictures taken by the subjects could reveal the relationship between such elements and the observers’ perceptions, and the elements might be used to manage the destinations to satisfy the interviewers’ demands and provide ample opinions; and (3) the subjects can evaluate the facilities they experience.
On the other hand, the CEM also contains some disadvantages: (1) previous research has revealed that the sample size for such a method is relatively small, possibly because it takes a long time to complete taking pictures and recording, and it would be challenging to gather photographers at the same location for a long period; (2) in addition to the small sample size, the subjects’ attributes may need to be excluded from the investigation, such as demographic and psychological features, as they would affect the coverage of the results; and (3) in comparison with an in-depth individual interview, the CEM results may lack depth with regard to the topic [18
]. However, compared with the traditional questionnaire survey, more perspectives from different users’ experiences could be investigated through the CEM, allowing us to think about problems from various perspectives.
2.2. Analysis of Qualitative Data
Qualitative analysis, a modern method used in academic research, is more complicated, with multiple layers and distinct data analyses, than traditional questionnaires or data quantitative research. This method does not simply respond to a questionnaire or discuss the difference, correlations, or predictions among variables. Since the mid-1980s, the development of computer-assisted qualitative data analysis software (CAQDAS) has allowed researchers to properly record intuition, ideas, searches, and analyses to simplify qualitative analyses [20
]. The captions acquired through the CEM further preceded qualitative data analysis. Faced with former CEM survey results, many unstructured texts were generated that could not be quickly processed or perceived by humans or computers. For the next step, effective technology and algorithms are needed to mine and extract the most meaningful information [21
The captions obtained by the CEM need to be processed through natural language processing (NLP) before advanced analyses are performed. Natural language processing (NLP) is one of the essential applications of machine learning, for example, text-to-speech and sentence-to-sentence clauses, etc. With advancements in computing speed, the accuracy of natural language processing is gradually being accepted. Segmenting the sentences into a file is the first task in research on text analysis using a computer. “Words” are then used for analyzing and organizing the results; therefore, “word segmentation” can be regarded as the most basic word analysis task.
The present study uses the Jieba module as a Chinese word-segmentation tool. The Chinese meaning of “Jieba” is “stutter,” and the default word break principle of the Jieba module is simplified Chinese. Therefore, when using it in Taiwan, the traditional Chinese thesaurus was required to be downloaded in this study [23
]. After the word-segmentation process, there was still too much text information; furthermore, more important phrases needed to be chosen.
Among the distinct computer-assisted qualitative data analysis software, NVivo is currently the most popular qualitative research software with complete functions. Data sources of text records, relevant literature, records, videos, and social networking sites after interviews are often used in qualitative research. NVivo helps users systematically organize such information with context and in a mutual relationship. Such collected “Sources” are integrated contextually, and the contents are coded and defined for transformation into “Nodes.” When the “attribute value” is added into the coding process, it becomes “Cases.” In other words, “Cases” can be regarded as “Nodes with attribute data.” Set “Cases” with the same attribute value can be gathered together with the coding function in NVivo queries [24
]. The most meaningful and representative labels are obtained through cluster analysis. These labels are then used for encoding CAQDAS to establish the so-called “Case” and “Node” to clarify the research topic’s organizational and secondary levels. Through topic coding and classification, as well as comparison and analyses, an organizational hierarchy and sub-hierarchy were established to clarify the research structure and provide a reference for successive research.
2.3. Scenic Beauty Estimation (SBE)
Scenic beauty, an abstract concept, was previously studied using qualitative methods. Prior to the psychophysical paradigm phase, Daniel et al. [25
] developed scenic beauty estimation (SBE) to quantitatively analyze scenic beauty. The SBE method, a psychophysiological experiment to evaluate the scenic beauty of forests, originated from the stimulus and respondence model in psychology and was improved according to the signal detections method and the Thurston scaling model. In addition, the viewers’ preference for landscape or beauty was represented by the perceived scenic beauty distribution of the evaluators.
For the experimental procedure, the subjects viewed representative color slides (stimulus) and gave responses according to the preference scale of 1 to 5, revealing low-quality to high-quality scenic beauty. Statistics was used to standardize the evaluators’ values using distinct evaluation criteria to solve the possible differences caused by different evaluation scale baselines. This was expected to accurately measure the public perceived preference for various landscapes. When the sample size was large enough, the randomly sampled perception value would become a normal distribution.
In summary, the SBE method presents the following advantages [26
]: (1) It can include the intangible value of resources into the quantitative evaluation, and combine psychology and statistics to exclude individual subjective judgments of managers and planning designers, but adopt the public “perception preference” for different landscapes to respond to the viewers’ perceived preference for landscapes. (2) In terms of the validity test, the SBE method provides similar results for the randomly shot slides or photos of the subjects being evaluated by the on-site evaluation. It could therefore save human resources and time. (3) Regarding the reliability test, the SBE method proves the consistency of photos taken at the same site but at different times, excluding special event factors. Therefore, the method shows high reliability. (4) In addition to forest landscape assessments, the SBE method can be applied to various landscapes. The results prove that it is a beauty-estimation method with high reliability and validity. (5) The SBE method eliminates evaluation errors caused by differences in individual aesthetic concepts. (6) The SBE index can be applied to managing large-area landscapes. (7) With regard to evaluation, the listed landscape factors appear to have positive and negative effects on recreation users that could be a reference for future management decision-making.
Nonetheless, the SBE method also presents the following disadvantages [26
]: (1) It is arguable whether the selection of landscape samples could represent general landscape groups. (2) Regarding framing, a photographer’s techniques and angles affect the evaluation results. (3) It is not easy to view the exact evaluated areas and special regions from photos. (4) The SBE method cannot distinguish whether a viewer’s evaluation preference is landscape perception or cognition. (5) The calculated coefficients in the analysis lack definite explanations, would change with different people, and are comparatively subjective.
Traditional landscape qualitative research is usually based on narrative analysis methods that are too subjective, although discussed in depth. However, the commonly used quantitative analysis method often fails to deeply explore the reasons behind the data. Therefore, this study proposes a new method, namely, visual language translation, or the CEM, which combines the features of qualitative research and quantitative analysis. This has also rarely been mentioned in previous studies, and this is considered a new attempt.
Different from the traditional research method, visual language translation, or the CEM used in this study, clearly understands the subjects’ perception of the images and sample areas. Factors are accurately acquired through the subjects’ descriptions, and such factors, through qualitative analysis, are coded with NVivo to find the best ones. However, too much open information can easily result in information dispersion and difficulties with induction and analysis. For this reason, both the CEM and SBE are selected to acquire complete information for analysis and comparison. In this study, these methods were established and performed well with good results.
Using SBE, we can acquire the subjects’ preferences for images and sample areas, which, corresponding with the CEM, accurately comprehends the subjects’ factor weights. In this way, in addition to obtaining a quantitative landscape preference, it is also possible to further understand the possible reasons for the preference. For instance, wooden lattice-framed embankment-matching stairs reduce the visual impact on the entire environment and enhance the subjects’ preference. A stone embankment could reduce the sense of artificiality to make the subjects feel safe and stable. Different from individual analysis in the past, two methods are utilized in this study, and an on-site investigation is integrated to understand the sample areas more deeply and to accurately acquire factors for subsequent research and design.
However, this study has limitations. It is mainly based on case studies, of which we still need more varied research objects and a wider survey sample. It is suggested that the CEM could be continuously used for discussing the correlations among representation, perception, and preference. Furthermore, experts’ interviews and questionnaires could be combined to build a list of the possible factors in landscape evaluation and to facilitate discussion of the basic concept of aesthetic design in mountain stream engineering.