Advances in Smart Environment Monitoring Systems Using IoT and Sensors

Air quality, water pollution, and radiation pollution are major factors that pose genuine challenges in the environment. Suitable monitoring is necessary so that the world can achieve sustainable growth, by maintaining a healthy society. In recent years, the environment monitoring has turned into a smart environment monitoring (SEM) system, with the advances in the internet of things (IoT) and the development of modern sensors. Under this scenario, the present manuscript aims to accomplish a critical review of noteworthy contributions and research studies on SEM, that involve monitoring of air quality, water quality, radiation pollution, and agriculture systems. The review is divided on the basis of the purposes where SEM methods are applied, and then each purpose is further analyzed in terms of the sensors used, machine learning techniques involved, and classification methods used. The detailed analysis follows the extensive review which has suggested major recommendations and impacts of SEM research on the basis of discussion results and research trends analyzed. The authors have critically studied how the advances in sensor technology, IoT and machine learning methods make environment monitoring a truly smart monitoring system. Finally, the framework of robust methods of machine learning; denoising methods and development of suitable standards for wireless sensor networks (WSNs), has been suggested.


Introduction and Background
Sustainable growth of the whole world depends on several factors such as economy, quality education, agriculture, industries and many others, but environment is one of the factors that plays the most important role. Health and hygiene are key components of the sustainability of mankind and progress of any country, which comes from a clean, pollution free and hazardous free environment. Thus, its monitoring becomes essential so as to ensure that the citizens of any nation can lead a healthy life. Environment monitoring (EM) consists of proper planning and management of disasters, controlling different pollutions and effectively addressing the challenges that arise due to unhealthy external conditions. EM deals with water pollution, air pollution, hazardous radiation, weather changes, earthquake events, etc. The sources of pollution are contributed by several factors, some of which are man-made and others due to natural causes, and the role of EM is precisely to address the challenges so that the environment is protected for a healthy society and world [1]. With the more recent advances in science and technology, especially artificial intelligence (AI) and machine learning, EM has become a smart environment monitoring (SEM) system, because the technology has enabled EM methods to monitor the factors impacting the environment more precisely, with an optimal control of pollution and other undesirable effects. The design of smart cities is taking the place of old and traditional methods to create and plan urban environments. Smart cities are planned using wireless Figure 1. Smart environment monitoring (SEM) system highlighting water contamination and its monitoring using the cloud connecting internet of things (IoTs) and sensors.
One more example of a SEM system, highlighting a general purpose system with extended scope, is shown in Figure 2., which shows how the system is addressing various issues related to environment monitoring, such as humidity, temperature, radiation, dust, UV signal etc. The backbone of the system is a WSN that is establishing the actual interface between IoT devices and data captured through various types of smart sensors. This is a perfect example of a "smart city" [11,25,26], using a SEM system that ensures healthy environment for its citizen.   One more example of a SEM system, highlighting a general purpose system with extended scope, is shown in Figure 2., which shows how the system is addressing various issues related to environment monitoring, such as humidity, temperature, radiation, dust, UV signal etc. The backbone of the system is a WSN that is establishing the actual interface between IoT devices and data captured through various types of smart sensors. This is a perfect example of a "smart city" [11,25,26], using a SEM system that ensures healthy environment for its citizen.  By focusing on agriculture, as a relevant issue for the growth of any nation, it is easy to underline how SEM can play a significant role by providing a "smart or green agriculture" [14,20,27,28], that can deal with major challenges and factors involved in sustainable growth and enhancing productivity within the agriculture sector. One such smart agriculture scenario can be seen in Figure 3, where a SEM system is actually a smart agriculture monitoring system. In this case, the health of soil, moisture analysis, water contamination level, water quantity level and several other factors are very important in obtaining sustainable productivity in the agriculture sector. We can see in Figure 3 that the smart Sensors 2020, 20, 3113 4 of 18 agriculture monitoring system includes all such factors, controlled and monitored with the help of IoT  devices, suitable sensors capturing the agricultural data, then transmitted to the cloud through a WSN.   4 can deal with major challenges and factors involved in sustainable growth and enhancing productivity within the agriculture sector. One such smart agriculture scenario can be seen in Figure  3, where a SEM system is actually a smart agriculture monitoring system. In this case, the health of soil, moisture analysis, water contamination level, water quantity level and several other factors are very important in obtaining sustainable productivity in the agriculture sector. We can see in Figure 3 that the smart agriculture monitoring system includes all such factors, controlled and monitored with the help of IoT devices, suitable sensors capturing the agricultural data, then transmitted to the cloud through a WSN. We attempted to study the existing contributions by a critical survey on SEM methods; the literature suggests that the extensive reviews on SEM methods which have discussed significant findings are not found. We could not find much literature that reviews or surveys SEM techniques. A survey on smart agriculture systems [29], smart home technologies [30], smart health monitoring systems [31], environment monitoring [32], an IoT based ecological system [33], IoT for marine environment monitoring [34], and a survey on pollution monitoring system [35], are a few of the survey and review related articles highlighting different aspects of SEM. The environment is contaminated due to several factors, but water pollution, air pollution, radiation and sound pollution are mainly involved in most of the existing research. This motivates us to bring out an extensive review on SEM that covers all important factors affecting the health of the environment and predominant methods used to mitigate the challenges due to these factors, such as IoT and sensor technologies.
We have briefly discussed in this section the main issues related to environment monitoring, SEM, the role of IoT, AI and WSNs in implementing SEM. The next part of the paper is organized as follows: Section 2 discusses related research and study; Section 3 presents comparative analysis of advances in SEM systems; Section 4 highlights the significance of the study and recommendations).

Related Research and Study
The current research suggests that environment monitoring systems are implemented smartly as SEM for various purposes and using different methods. A huge number of contributions on SEM, We attempted to study the existing contributions by a critical survey on SEM methods; the literature suggests that the extensive reviews on SEM methods which have discussed significant findings are not found. We could not find much literature that reviews or surveys SEM techniques. A survey on smart agriculture systems [29], smart home technologies [30], smart health monitoring systems [31], environment monitoring [32], an IoT based ecological system [33], IoT for marine environment monitoring [34], and a survey on pollution monitoring system [35], are a few of the survey and review related articles highlighting different aspects of SEM. The environment is contaminated due to several factors, but water pollution, air pollution, radiation and sound pollution are mainly involved in most of the existing research. This motivates us to bring out an extensive review on SEM that covers all important factors affecting the health of the environment and predominant methods used to mitigate the challenges due to these factors, such as IoT and sensor technologies.
We have briefly discussed in this section the main issues related to environment monitoring, SEM, the role of IoT, AI and WSNs in implementing SEM. The next part of the paper is organized as follows: Section 2 discusses related research and study; Section 3 presents comparative analysis of advances in SEM systems; Section 4 highlights the significance of the study and recommendations).

Related Research and Study
The current research suggests that environment monitoring systems are implemented smartly as SEM for various purposes and using different methods. A huge number of contributions on SEM, both based on purposes and types of methods, have been studied and therefore the related research has been discussed in three main subsections, namely the study based on smart agriculture monitoring systems (SAMs), smart water pollution monitoring systems (SWPMs), and smart air quality monitoring systems (SAQMs). In this manuscript the authors have attempted to critically report the major findings and limitations of the current research on SEM. Soil monitoring (SM) [14,15,36], ocean environment monitoring (OEM), marine environment monitoring (MEM), air quality monitoring (AQM) [10,11,37,38], water quality monitoring (WQM) [14,39], and radiation monitoring (RM) [1,36] have been covered, by offering a wide analysis of different application fields of SEM.
While studying the existing literature on SEM methods, especially on advancements in IoT and sensor technologies for SEM systems, we found that an extensive review on this topic has not been much reported. We found some interesting literature on specific areas of research addressing some challenges of environmental factors such as water pollution, air quality, radiation, and smart agriculture. We aimed at bringing out major advances in IoT and sensor technologies used for addressing the challenges in SEM and thus we included some significant research studies and contributions of various sources highlighting specific classic work on SEM methods. The current study on advances in IoT and sensor technologies used for SEM provides insight to the scientists, policymakers, and researchers in developing a framework of appropriate methods for monitoring the environment that faces challenges mainly due to poor air quality, water pollution and radiation. These factors also affect agriculture which is backbone of any developed and developing economy and thus smart agriculture monitoring (SAM) has also been studied in this section. Table 1 shows major research studies and contributions on the above SM, OEM, MEM, AQM, WQM and RM areas of interest. Soil monitoring methods were reported to have been affected by greenhouse effects. Ocean and marine SEM systems have been implemented using sensors, WSN and IoT and these methods have mainly suffered with cost, coverage and installation issues [40][41][42]. Air pollution control and AQM [1,10,11,16,[43][44][45] have been suggested using a mobile sensor network, wireless sensors and IoT devices that operate on AI and machine learning. In a similar manner, we can see in Table 1 that the different types of SEM systems are designed and implemented for various purposes and there is no robust method that can address any of the challenges of environment. Heterogeneous sensors Air quality [12] Air quality monitoring Large area monitoring; noisy data; accuracy and cost issues

Geomatics sensors and IoT
Pollution monitoring [16] Air pollution monitoring System Real time monitoring; accuracy issues Sensors with MQ3 Model, Raspberry Pi and IoT Sensor based AQM [37] Air pollution monitoring system Efficient for low coverage area; low cost; easy to install; less number of pollutants are covered Gas sensor and LASER sensor SEM [48] Dust and humidity monitoring Wide coverage and efficiency; low cost and small size IoT Radiation [36] Radiation monitoring High cost and low stability against temperature variation HPXe chamber Aqua farming and energy conservation [49] Aqua Farming Water quality and quantity control; higher carbon emission and energy requirement Odor, pH, conductance and temperature sensor Table 1. Cont.

Research Purpose Findings and Challenges Method/Device Used
Multi-agent supervising system [50] e-health monitoring system due to temperature and radiation changes around the surroundings Detection of emergency situations Supervising system and AI SEM in winter season [51] Effect of surroundings during winter season only Effect of batteries and other radiation Wireless sensor network LoRa technology for climate monitoring [32] Climate and ecology monitoring Study of emissions in the environment LoRa technology and sensor network Smart city and SEM [52] Monitoring of data center radiation Temperature, humidity and energy consumption in data centers monitored for smart city and SEM

IoT
ZigBee based environment monitoring [53] Smart industry environment To study hazardous effects in industries ZigBee and WSN LoRa: Long Range

Study based on Smart Agriculture (SAM)
This section presents studies and research on smart agricultural monitoring (SAM) systems covering the measures for crop monitoring, pest control, fertilizer control etc. The research study summary for a few important works can be seen in Table 2. Plant growth monitoring [54] was implemented and named as "gCrop", using IoT, machine learning and WSN. The work uses a regression model of the 3 rd degree and provides a prediction accuracy of 98% but the computational complexity was high. The analysis of crop quality [14,46] assessment was made using SAR data for monitoring the quality of paddy rice. Support vector machines (SVMs) with back-scattering features were used in this assessment of the rice quality, with a limited sample size. Leaf area and dimension also play an important role in the assessment of various types of crops, as means to determine if the growth is satisfactory or not. One such work was reported in [55], that was used to measure the leaf area index using SVM as the machine learning technique, with a Gaussian process model [56] and the accuracy of measurement found as 89% with a limited sample size also in this case. An expert system using AI has been implemented in [57] using the Naive Bayes [58] method and machine learning which operates on sensor data captured in agriculture. This work was useful in monitoring the quality of fertilizer, pesticides and the amount of water to be irrigated in the crops. Some other works studied crop quality assessment [21,59,60] and [61] used for monitoring of the soil health, suitable for soya bean crop on the basis of phenological data and unmanned aerial vehicle (UAV) real-time images. There are a few other important studies on various application of SEM systems for different applications, such as smart farming [62], pest monitoring [63], and crop area monitoring [61].
The environment conditions affect the health of crops and consequently the agriculture growth. Therefore, we aimed at studying the status of research on SEM using IoT, sensors and AI techniques. The factors involved in agriculture such as soil condition, moisture condition, water pollution, air quality, temperature etc. have been taken into consideration while reviewing the advances in SEM methods. The focus is given to studies on water pollution monitoring and air quality monitoring methods also, which are discussed in next sub-sections.

Study based on Smart Water Pollution Monitoring (SWPM) Systems
Different literature has been studied on smart water pollution monitoring (SWPM) methods and systems using machine learning methods, IoT and wireless sensors. Table 3 depicts a few major contributions in the area of SWPM. Remotely sensed images were analyzed and machine learning was applied for prediction of the pollution level in the lagoon water, useful for agriculture [64]. This work used ordinary neural network based machine learning and the prediction results were not very satisfactory. Classification of water contamination [65] has been studied and water was classified as clean or polluted water, using machine learning methods and IoT devices. The paper presented a realtime contamination monitoring system, though the data captured were in a limited area only. The assessment of various pollutants mixed in water has been implemented in [66] and the pollutants were classified using a DSA-ELM model [66] with the evaluation of the model itself. AI and neural network based prediction of water quality parameters was studied in [67], and alkalinity, chloride and sulphate contents were estimated. The work mainly focused on prediction of water quality parameters and values of sulphate or chloride present in the water. Big data analysis and issues in classification of water contamination were discussed for classification of the contamination using SVM in [68]. Quality assessment of drinking water and its classification into drinkable and non-drinkable water were presented in [69,70], as a real time monitoring system and AI-SVM based classification technique respectively. A video based surveillance of water quality and pollutants, was studied in [71], and the surveillance helped to stop the man-man sources of pollutants. The work employed IoT tools for video-surveillance and machine learning for classification of water as polluted and clean water. One more work on a drinking water prediction model was suggested in [61], and a feature based model helped in analysis of drinking water to further predict its quality before usage. In another work, Sensors 2020, 20, 3113 8 of 18 chlorophyll-A concentration in lake water was assessed using different machine learning models [72], and the work was recommended as method for a realtime lake water cleaning management system. Chlorophyll-a in lake water [72] Chlorophyll-A concentration in lake water machine learning based classification of water BPNN, SVM with 78% accuracy Water quality monitoring [74] Water quality monitoring IoT for surface water quality assessment IoT with smart sensors

Study based on Smart Air Quality Monitoring (SAQM)
Research on SAQM methods and systems have also been studied, and Table 4 presents a summary of different SAQM approaches used in recent literature on air quality monitoring systems. Air quality characterization [58] has been implemented using heterogenous sensors and machine learning methods. The monitoring as well as characterization of water quality was achieved but interoperability issues were reported in this work due to use of heterogenous sensors. Air quality evaluation using fixed as well as mobile nodes of sensors [75] was implemented, capable to check the air quality in stationary as well as mobile ways. In this latter case, the compatible sensors were deployed as mobile nodes which can work satisfactorily in a moving environment. Data captured through smart sensor nodes were processed and analyzed with the help of machine learning techniques. Another air quality control process was studied using IoT and machine learning techniques in [76], with a focus on assessment of air pollution, deploying gas sensors which help in capturing air particles and analyzing the pollutants mixed in the air. Sensor networks have been established in moving vehicles for monitoring air quality with the help of machine learning; in [77], mobile sensor nodes and WSN were deployed. Infrared sensors were deployed to evaluate the air quality, especially analyzing volatile organic compounds (VOCs) in [25], with the help of machine learning methods. The elements of VOCs were detected and analyzed using spectroscopic observations. There are a few components present in the air that help assessing the quality of the air; one such component, called PM2.5, was predicted in [78], using extreme machine learning techniques tested upon spatio-temporal data collected in a certain duration of time over a range of distances covered by the sensors. Different forecasting models were suggested in [44] for quality evaluation of urban air and the components like O3, SO2 and NO2 were determined and a comparison was made for the models used in the work. RFID and a gas sensor based air quality control mechanism were implemented in [79], to determine the level of pollution in the air by predicting the pollution value; IoT was employed to analyze sensory data captured through gas sensors. RFID was primarily used in this work for detection of pollutants and communicating to WSNs with the help of IoT devices connected across a WSN architecture. An SAQM system has been studied in [80], using a LoRaWAN (long range WAN) [80][81][82], and this work has been very useful for detecting temperature, dust, humidity and carbon dioxide components in the air. An intelligent air quality system was presented for detection of CO2, NOx, temperature and humidity in [83] using AI and machine learning techniques for developing expert systems for air quality assessment. Furthermore, PM10, PM2.5, SO2, oxides of nitrogen (NOx), O3, lead, CO and benzene components were detected, on the basis of machine learning methods trained by spatio-temporal data, in [84]. This was extended using deep learning for detection and detailed analysis of O3 components only. Another work employing heterogenous sensors was studied in [85]. SVM was used for analyzing the sensor data, captured through heterogenous sensors, and air quality was estimated. Table 4. Research on SAQM systems using machine learning and IoT.

Research Purpose Data and Technique
Air quality characterization [58] Air quality monitoring Heterogeneous sensors; machine learning based predictive model Air quality modeling [75] Air quality monitoring Mobile nodes Air pollution [76] Air quality monitoring Gas sensors from mobile vehicle data, IoT and machine learning Air quality in vehicular sensor network [77] Air quality monitoring Sensors in mobile nodes [25] Organic compound detection Infrared sensors, spectroscopy and machine learning PM2.5 estimation [78] Air quality in terms of PM2.5 concentration levels

Detection of VOC in air
Spatio-temporal geographic data, Extreme machine learning technique Urban air [44] Urban air pollution in terms of O3, NO2 and SO2 concentrations Forecasting models Air pollution prediction [79] Air pollution control RFID, Gas sensors and IoT Smart air quality [80] Air quality Temperature, humidity, dust and carbon dioxide sensor; LoRaWAN Intelligent air quality system [83] Air quality for detection of CO2, NOx, temperature and humidity UV light, AI and sensors
The extensive study on SEM methods brings out the following major observations for the discussion: • The research on SEM includes various purposes, mainly on SAM, SWPM and SAQM. The study of water pollution, air quality, soil moisture and humidity can help in modeling and design of healthy environment systems that would also help smart agriculture for sustainable growth of the economy.

•
The methods under each of the purposes are divided in terms of sensory data used, machine learning methods used, IoT devices used, and types of sensors involved. The current study made by us mainly focused on impact of existing research on water quality monitoring, air quality assessment, applications of SEM and smart agriculture systems.

•
In most of the SEM methods, especially SAM and SWPM, CNN based deep learning methods are used by the researchers and other deep learning models are not very frequently used.

•
The sensory data vary in most of the applications of SEM and there is no robust data over which a maximum number of methods are operating. The data type and regions of interest are not the same for various research work.

•
The methods have been used for either classification or prediction; for example, water is classified as polluted or clean water; similarly, the water and air quality can be predicted (e.g., level of degradation).
The studies reported for all purposes of SEM systems do not have any common challenges and vary from application to application, but the major challenges observed are as follows: • Wherever heterogeneous sensors are used, there is problem of interoperability in the analysis of the data captured through different types of sensors.

•
Sample size is limited in many of the contributions.

•
Noisy data poses a challenge in analysis. Noise is present in the data captured through sensors used for various purposes. The noise may be contributed by several internal and external factors.

•
The machine learning methods which have been employed for training the data and for classification are mostly traditional methods of machine learning, such as SVM, neural network, etc. • Fuzzy based methods and deep learning approaches are used in a few research studies and implementations, but the research suffers with either big data issues or huge computational complexity.

•
There is no robust approach of machine learning reported, that can be employed in addressing the challenges of the environment irrespective of the purpose of the monitoring and control, types of data, and types of sensors used.
Research trends were also analyzed to assess the quantum of research carried out in the area of SEM [14,21,31,[103][104][105][106][107][108][109][110][111][112][113] and Table 5 shows a summary of quantity of research in this case. The study of trends was made by using a publication search in the Science Direct databases in year-wise manner. In this analysis, the duration has been chosen from year 1995 to year 2020. It can be clearly seen in Table 5 that the quantum of research on SEM has been increasing with the time in both the case, namely the research employing IoT and WSN, as well as the research using IoT and machine learning. An interesting fact is an outcome of the table: the research using modern machine learning methods is still lagging behind those which do not use any machine learning. However, if IoT devices are used, and deployed in a WSN, then the role of AI cannot be overlooked. The analysis of research trends are shown in Figure 4, highlighting the research trends in two main categories, namely SEM using IoT and WSN, and IoT and machine learning, respectively. The trends suggest that the SEM has yet to be implemented and studied widely on machine learning based training and subsequent classification or prediction. The research is reported to have increased every year but more impact of IoT and WSN can be seen in Figure 4.

11
In this analysis, the duration has been chosen from year 1995 to year 2020. It can be clearly seen in Table 5 that the quantum of research on SEM has been increasing with the time in both the case, namely the research employing IoT and WSN, as well as the research using IoT and machine learning. An interesting fact is an outcome of the table: the research using modern machine learning methods is still lagging behind those which do not use any machine learning. However, if IoT devices are used, and deployed in a WSN, then the role of AI cannot be overlooked. The analysis of research trends are shown in Figure 4, highlighting the research trends in two main categories, namely SEM using IoT and WSN, and IoT and machine learning, respectively. The trends suggest that the SEM has yet to be implemented and studied widely on machine learning based training and subsequent classification or prediction. The research is reported to have increased every year but more impact of IoT and WSN can be seen in Figure 4. The above discussion and analysis helps us recommending the following for better, robust and smarter environment monitoring systems: • A framework of machine learning methods needs to be developed.
• A robust set of classification, prediction and forecasting models has to be designed that can operate on any data, irrespective of the purpose of using the SEM. The above discussion and analysis helps us recommending the following for better, robust and smarter environment monitoring systems: • A framework of machine learning methods needs to be developed. • A robust set of classification, prediction and forecasting models has to be designed that can operate on any data, irrespective of the purpose of using the SEM.

•
Suitable denoising methods are required to be implemented as pre-processing to the SEM major stages, since most of the research has failed using de-noising the data and its appropriate pre-processing.

•
Data deduplication approaches and other methods are needed to deal with big data issues involved in a few significant studies. • SEM aims at sustainable development of any nation and the smart agriculture and smart environment play a most important role in achieving the sustainable goals, but in rural areas, in most of the developing and underdeveloped nations, the necessary infrastructure for setting up IoT, WSN and other sensors is still a challenging task. This requires governmental level involvement both at local as well as global perspectives.

•
Interoperability issues in implementing various types of sensors, can be addressed by developing suitable standards and protocols that can make the data compatible for all acquisition and analysis systems.
An attempt was made to include major observations of a few significant review articles on SEM but it was very difficult to report any such extensive review on the SEM in particular. This motivated us to study the most important contributions on research addressing environmental challenges due to main factors. This review helped us to reach to some conclusions and make recommendations for designing a robust SEM systems that can handle all possible challenges using a framework of AI and sensor technologies.

Conclusions and Future Scope of Work
This paper has presented an extensive and critical review of research studies on various environment monitoring systems used for different purposes. The analysis and discussion of the review suggested major recommendations. The need of extensive research on deep learning, handling big data and noisy data issues, and a framework of robust classification approaches has been realized. We have focused mainly on water quality, and air quality monitoring as smart agriculture systems that can deal with environmental challenges. The major challenges in implementation of smart sensors, AI and WSN need to be addressed for sustainable growth through SEM. The participation of environmental organizations, regulator bodies and general awareness would strengthen SEM efforts. The poor quality of sensory data can be preprocessed using appropriate filters and signal processing methods to make the data more suitable for all subsequent tasks associated in SEM. The future scope of the work aims at studying other factors of environment such as sound pollution and disasters etc.
Author Contributions: All authors contributed equally to the work. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.