In order to identify crucial factors which are working as obstacles in the field of implementing ML models in environmental sectors, we conducted a reconnaissance survey with some of the stakeholders in Bangladesh. During this reconnaissance survey, 5 experts and 5 non-experts on ML were participated from government, non-government, and academic sectors. Besides, an extensive literature review was done to find out the important issues to incorporate as the existing obstacles in the questionnaire.

This study was designed based on the perception of stakeholder those are working on different aspects of environmental management and monitoring but have not expert level experience regarding machine learning in Bangladesh. Among the different type’s institutes in Bangladesh, institutes were selected purposively. After identifying 20 institutes including Government Organizations (GOs), university, and Non-Government Organizations (NGOs) related to environmental subject, list of stakeholders was prepared. Then, an online questionnaire form was sent to all of the stakeholders with a consent form including in details of ethical consideration of this study. Among them, a total number of 20 stakeholders and practitioners in the field of environmental (monitoring, management, science, engineering and policy-making) participated in this survey (Table 1). Here, a structured questionnaire was used to conduct this survey. Also, to understand the fact in details following the objectives of this study, 5 In Depth Interviews (IDIs) were conducted using zoom platform. Besides, the resource persons were selected based on their experiences with ML models and practitioners who have never used ML models and have little knowledge about ML models, and facing obstacle to be an expert on this. Background information of the respondents is given in Table 2.

A total of 20 KIIs were conducted with key informant as presented in Table 1 and Table 2. All of the respondents answered the questions from 1 to 10 value, where 1 = less agree and 10 = fully agree, and 2 to 9 represents the level of statement on this scale. After that, all of the data was checked twice to ensure the

Here, AUST = Ahsanullah University of Science and Technology, BUET = Bangladesh University of Engineering and Technology, CC = City Corporation, CUET = Chittagong University of Engineering and Technology, ICCCAD = International Center for Climate Change and Development, KUET = Khulna University of Engineering and Technology, LGED = Local Government Engineering Department, MIST = Military Institute of Science and Technology.

correct form of entry and Statistical Package for the Social Sciences (SPSS) 2020 software was used to analyze the data to fulfill the objectives. The list of questions asked to the participants (a scale of 10 was selected) and answered as numeric entries between 1 and 10.

The value of mean and the peak of the curve shows that most people are facing obstacles due to lack of appropriate resources (Figure 1(a)). In case of absence of reliable resources on an average intensity of participants is 9.15, when the number of observation is 20. Also, the mean value is 9.15, and the standard deviation is also significantly less, which is. 933. Besides, working experience as a factor of determining the connection with lack of resources, where 75th percentile value or the upper quartile of all the boxes are above 9 and the maximum value is 10. The majority of the participants are facing challenges due to lack of

proper resources and from the simple box plot diagram it can be clearly estimated and assumed that all the participating institutions depicts the upper quartile or the 75th quartile is above 8.00 out of scale 10.00. More specifically the participants from BUET and MIST strongly mentioned lack of machine learning relevant resources is a prime fact behind halting researchers and practitioners to implement machine learning in the environmental management field (Figure 2(a)).

Furthermore, hence it can be said that the result of our assumption is positive. Besides, from the qualitative findings, it is observed that, the practitioners who participated expressed their consensus that ML models are definitely excellent sources of process-based modelling and can act as a great tool for aiding in decision-making, but the resources available are quite vague and inscrutable. As a result, it has become increasingly difficult to use them in the practical decision-making process for people working on environmental sectors.

Researchers and other practitioners often consider ML models as “black boxes” as their patterns and learned relationships are not similar and not interpretable as classical statistical models and differential equations [25]. Intensity of participants in favor of absence of lack of information exchange is on an average 9 when the number of observations is 20. It can be increased by increasing information exchange and reducing the standard deviation 1.124 (Figure 1(b)). There is a positive relationship between working experience and stakeholders facing challenges due to lack exchange of information. So higher experience can relatively reduce the challenges.

From the box plot diagrams 50th and 75th percent quartile values it can be said that in this case almost all the participants are also very much positive that information exchange is also a crucial factor that is to be dealt aptly in order to integrate the current technologies in monitoring and management of environment. The participants from AUST, BUET, CUET and LGED very strongly recommended this factor as a great challenge (Figure 2(b)). The available information cannot be applied to the practical work as smoothly as the practitioners want. From the histogram and the box plot, it is evident that the majority of the participants are supporting that lack of appropriate and relevant information is another key obstacle against using ML models as decision aided tools.

Here the Intensity of participants in favor of absence of less exposure of research upshots is on an average is 8.50, when the number of observations is 20 (Figure 1(c)). So, less exposure of research upshots reduce the intensity. There is a positive relationship between working experience and stakeholders facing challenges due to lack of exposure of research upshots. More experienced person gets more facilities to expose research upshots.

The researches which are conducted on the implementation of ML models are relatively opaque to the majority of the participants who are working at environmental sector, and it is quite evident from the histogram and the box plot diagram (Figure 2(c)). The mean value on the histogram is 8, and the 75th percentile value of the box plot diagram is above 8 in maximum cases; hence we can consider it as another key factor. It is crucially important to depict the research upshots explicitly and implicitly both. Because research upshot has got a very positive correlation with funding as well. However, to encapsulate it can be easily said that the assumptions that were made are positively appropriate.

The practitioners emphasized difficulties in funding as the 4th key component as per our study which is another negative catalyst against the implementation of ML models for environmental monitoring and management. In case of lower funding standard deviation is higher than other case which is 2.412 (Figure 1(d)). Intensity of participants facing problems due to difficulties in funding is on an average is 7.65, so many practitioners fail to take forward steps towards experimentation with ML models. There is a negative relationship between working experience and stakeholders facing challenges due to difficulties in funding. The mean value obtained from the histogram and the upper quartile of the boxplot depicts a value in between 7 to 8 so it was considered the 4th key obstacle. The lowest whisker value is obtained from BUET and the quartile values obtained from BUET illustrates that compared to other participating organizations BUET has got less difficulties is managing funding (Figure 2(d)). But the other organizations are facing more problems regarding funding issues as an obstacle.

In order to inculcate the urgency and importance of using up to dated technologies like Machine learning in environmental sector it is very important that the practitioners bear strong interest in that field. While the interview sessions were done along with literature reviews it is evident that interest also plays a crucial role here. The average frequency of the participants in support of augmenting interest in the relevant field is approximately 5.90 and the value of standard deviation from the histogram shows the value as 2.614 (Figure 3(a)). Again from the results obtained from the box plot diagram it can be observed that internal quartiles of each of the participating organizations are at different ranges and so their perception is not fully common. MIST, ICCCAD, CC considers it as a challenge but BUET is comparatively facing less challenge in this field. And the remaining organizations are moderate about this factor (Figure 4(a)).

Traditional engineering or science is also known as sequential engineering or science. In order to complete tasks faster with high accuracy integration of technology is crucial. Sticking to conventional techniques may hinder development process. In this study very renowned organizations and institutes participated

and also addressed the old conventional techniques as a major challenge as well (Figure 4(b)). The histogram depicts that more than the average number of participants highlighted that giving priority to old conventional techniques may lead to halting of pace and accuracy of analysis (Figure 3(b)). The internal quartiles range depicts that 75th and 50th percentile values in maximum cases are within range 6 to 8. Based on the analysis it is evident that old conventional techniques are to be upgraded.

In generic language adaptation is a process of adapting oneself or thing with a new or uncommon work or action. Many find it difficult initially to adapt with a new system. After the literature review was conducted, problems in adaptation with machine learning was identified as a factor that is not allowing practitioners and researchers to indulge themselves in this field effectively. From the histogram diagram the obtained mean value is 6.65 which illustrates that more than 50% of the participants are supporting and coinciding to this point that fear of adaptation with machine learning is one of the many challenges (Figure 3(c)). Again, the internal quartile ranges of the maximum number of participating organizations and institutes are above 75th percentile with a bit of exception in case of BUET, MIST and LGED (Figure 4(c)).