Intelligent Analysis of the Ecological State of Environment with Application of Distributed Expertise (on the Example of Bryansk Region)* Emilia Geger 1[0000-0003-0393-4274], Aleksandr Podvesovskii 2[0000-0002-1118-3266], Oksana Mikhaleva 2[0000-0001-6374-2827] and Anton Korsakov 2[0000-0002-4609-0246] 1 Bryansk Clinicodiagnostic Center, Bryansk, Russia emiliya_geger@mail.ru 2 Bryansk State Technical University, Bryansk, Russia apodv@tu-bryansk.ru, gordonmi@mail.ru, korsakov_anton@mail.ru Abstract. The paper considers the problem of assessing the ecological state of the environment in the region. An approach to the intelligent analysis and esti- mation of anthropo-technogenic pollution of a territory with the application of integral indicators which take into account environmental pollution is proposed. To estimate the integral indicator parameters, distributed group expertise tech- nology is used, supporting a mechanism for control of expert estimates con- sistency, taking into account experts’ competency in the relevant subject areas. Using the proposed approach, the problem of risk assessment of environmen- tal impact of chemical air pollutants has been solved. Methods for control of expert estimates consistency based on the procedure of feedback with experts made it possible to increase the reliability of evaluation results and also to de- crease the influence of a random expert error on the final assessment. The ob- tained aggregated risk estimates were used to construct, calculate and visualize the integral indicator of radioactive and chemical contamination of the districts of Bryansk region. Keywords: Environment, Anthropo-Technogenic Pollution, Integral Indicator of Pollution, Expert Estimates, Group Expertise, Consistency of Expert Estimates, Distributed Environment. 1 Introduction Today, scientific problems of monitoring and evaluating biological and medical con- sequences of anthropo-technogenic pollution of the environment are a priority for state policy in all economically developed countries [1, 2]. Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). * The reported study was funded by RFBR, project number 20-04-60185. 2 E. Geger, A. Podvesovskii, O. Mikhaleva, et. al. Analysis of environmental pollution using traditional statistical methods is not al- ways possible due to the lack of unity and required accuracy of measurement results of environmental pollution indicators used in monitoring and the absence of a unified structured system for assessing environmental pollution [3]. Parametric and nonpara- metric statistical methods have fairly strict assumptions – general homogeneity of observation conditions must be maintained, samples must be sufficiently representa- tive, with clear quantitative characteristics, etc. In cases where such assumptions can- not be made, it is proposed to move from statistical methods to intelligent decision- making support technologies. In particular, it is proposed to use methods of multi- criteria optimization in conjunction with the technology of distributed expertise to identify and estimate the parameters of this method. Multi-criteria optimization is a group of decision-making methods. These methods consist in finding an optimal solution that satisfies several criteria not reducible to each other, based on a certain optimality principle. An expert approach is based on the use of the collective opinion of experts (i.e., specialists in the relevant subject areas) in the preparation and decision-making pro- cess. Expert opinions are usually expressed partly in quantitative and partly in qualita- tive forms. Expert methods are used in situations where the choice, justification and assessment of the consequences of decisions cannot be performed on the basis of accurate calculations. The current level of information and communication technolo- gies makes it possible to organize distributed interaction of experts among them- selves, as well as with decision-makers and managers of expertise, using modern communication networks, primarily the Internet. Due to the above circumstances, a new phenomenon called networked expertise is emerging. Within its framework, expert networks and network expert communities are being created and are actively developing [4]. Due to the heterogeneity of environmental statistical data in estimating anthropo- technogenic pollution, it seems difficult to fully assess the risk of a specific impact of pollutants on the environment. Therefore, in this study, we have applied an approach to estimating the environmental situation on the example of Bryansk region using multi-criteria optimization methods together with the distributed expertise technology. Earlier in [5, 6] , integral criteria were proposed for each type of pollution based on the method of expert estimates. Then they were summed up for the districts of Bry- ansk region, taking into account weight coefficients for the corresponding type of pollution in the district. Transferring the decision-making process to a distributed environment complicates the use of traditional methods of organizing expert activity [7]. Therefore, it is re- quired to develop new effective methods for supporting group expertise taking into account peculiarities of participants’ geographically-distributed interaction in this process. In this paper, it is proposed to consider implementation features of the technology for supporting group expertise in a distributed environment aimed at estimating the ecological state of the environment on the example of Bryansk region. Intelligent Analysis of the Ecological State of Environment with Application of… 3 2 Description of the group expertise methods in a distributed environment The purpose of applying networked expertise in this study is to assess the importance (priority, hazard) of chemical and radiation pollutants for each environmental media (air, water, soil) and food. One of the main tasks of expert estimates is to obtain weight coefficients of each of the specific environmental media pollutants. A separate subtask is to estimate the degree of influence of a particular pollutant on the environ- mental media based on the exposure hazard. To obtain an objective expert opinion, each of the conditions must be met: • availability of an expert group consisting of experts with subject matter expertise; • availability of an analytical group with knowledge of mathematical apparatus for obtaining and processing expert information. In general, preparation and conduct of the networked expertise includes such stages as formation of an expert group, choice of a type and method of obtaining expert esti- mates, assessment of expert estimates consistency, and determination of the final (aggregated) consistent expert estimate. In addition, an important task arises related to assessing experts’ competency in the relevant subject area and taking it into account in the estimation model, both at the stage of assessing expert estimates consistency and at the stage of forming the final assessment. Solution to the listed problems is based on the use of information technology for group expertise support in a distribut- ed environment [8], together with models and methods for control of expert estimates consistency [9]. The first step after obtaining a set of individual estimates of objects is to check the consistency of this set taking into account the experts’ competency. When working with numerical (cardinal) estimates, the consistency of the set of individual estimates W = {w1 , …, wm} can be evaluated using a spectral coefficient:  p p p        k k − k k −  k ln( k )   K S (W ) = 1 − k =1 k =1 k =1 p z , (1)  G k − ( p + 1) / 2 ln( p)      k =1  where k is the sum of the coefficients of experts’ relative competency ci; their esti- mates are represented by the k-th scale mark; G = m / ln(m) p ln(p) is a scale factor; 1 if z* = TRUE; z= (2) 0 if z* = FALSE, q −1 q −1 z* = [k (1) = 1]  [k (q ) = p]  [ k ( d ) =  k ( d +1) ]  [k (d ) − k (d + 1) = const]. (3) d =1 d =1 4 E. Geger, A. Podvesovskii, O. Mikhaleva, et. al. Expression (3) is a Boolean function that specifies the necessary and sufficient condi- tions for the equality of the consistency coefficient (1) to zero. In this case, q is the number of subgroups of experts who gave the same estimates, k(d) is the number of the scale mark corresponding to the estimate received from the experts of the d-th subgroup (d = 1, …, q), k(d) is the sum of the competency coefficients of the experts of the d-th subgroup (a more detailed description can be found in [10]). After calculating the consistency coefficient, it is necessary to compare its value with the threshold values TO (detection threshold) and TU (application threshold), and to assess whether the consistency degree of the set of expert estimates is sufficient to be used to calculate the aggregated estimate [10]. The comparison result can be used to decide on the possibility of further use of the set. If the consistency coefficient is higher than the application threshold, then the consistency of the set of expert esti- mates is sufficient. If the consistency coefficient is between the detection threshold and the application threshold, then the set of expert estimates contains information, but its consistency degree is insufficient to determine the aggregated estimate. If the consistency coefficient is below the detection threshold, then the set of expert esti- mates does not contain information, and it is necessary to suggest that all the experts should revise their estimates of the objects or to make a full or partial replacement of the expert group. Considering this fact, we can say that one of the important aspects of organizing group expertise in a distributed environment is the control of expert estimates con- sistency based on conducting a feedback procedure with experts [9]. This procedure consists in contacting selected experts with a request for the possibility of changing their estimates and recommendations aimed at increasing the consistency of the set of individual estimates. If the expert agrees to change his estimate, then, based on the results of the change, the consistency coefficient is recalculated. When working with numerical (cardinal) estimates, at each step, an expert is selected for whom the largest value is | wi − w0 | i = , (4) ci characterizing the deviation of the estimate given by him from the average estimate in the group. In this case, expert’s competency is taken into account – with a decrease in the expert's competency, the degree of trust in his opinion decreases when it differs from the mean group one. Here, w0 is the average estimate calculated taking into ac- count the experts’ competency: m w0 = c w . i =1 i i (5) Intelligent Analysis of the Ecological State of Environment with Application of… 5 3 Solving the problem of expert estimation of the degree of influence of a pollutant on a pollution object Let us consider, for example, the problem of expert estimation of the risk of the im- pact of chemical air pollutants on the environment. The expertise objects are 15 chemicals (CO, NO, NO2, ammonia, ethanol, suspended solids, formaldehyde, acetic acid, hydrogen fluoride, manganese, iron oxide, hydrogen sulfide, petroleum hydro- carbons, xylene, toluene). To analyze and estimate the objects, an expert group was formed, which included six independent experts specializing in various environmental aspects. The experts’ competency was assessed on the basis of their level of knowledge in the expertise subject area. The corresponding competency coefficients are presented in Table 1. Table 1. Experts’ competency coefficients. Expert’s number, i 1 2 3 4 5 6 Competency 0,286 0,238 0,190 0,143 0,095 0,048 coefficient, ci A numerical rating scale with values from 1 (least influence) to 15 (greatest influence) was chosen as the estimation scale. To process the results of expert estimation, the initial spectral consistency coeffi- cients of each obtained expert estimate set were calculated, the detection and applica- tion thresholds were determined. In cases where the consistency coefficient value appeared to be less than the value of the application threshold, the method of increas- ing the consistency was used based on the procedure of feedback with experts. Ta- ble 2 presents the results of increasing the consistency, as well as the final aggregated estimates and the corresponding ranks of the expertise objects. Analysis of the results of solving the estimation problem showed the following. As a result of increased consistency, it was possible to ensure complete consistency of expert estimates for objects such as manganese, xylene, and toluene – the final consistency coefficient exceeds the application threshold (the corresponding cells in the table are highlighted in green). For formaldehyde, the consistency coefficient of estimates exceeded the application threshold from the beginning, which allows us to speak about the consistency of the initial set of estimates for this object. The initial coefficient value of the expert estimate consistency for four objects (CO, ammonia, acetic acid, petroleum hydrocarbons) did not exceed the detection threshold. This means that the consistency of the corresponding estimates is below the acceptable threshold (the corresponding cells in the table are highlighted in red). Con- sistency increasing procedure led to an increase in the consistency coefficient, but the application threshold was not exceeded. Also, it was not possible to ensure high con- sistency of expert estimates for the suspended solid object, although the initial value of the consistency coefficient exceeded the detection threshold. The cells in the table corresponding to the situation when the estimates consistency coefficient is between the detection and application thresholds are highlighted in yellow. 6 E. Geger, A. Podvesovskii, O. Mikhaleva, et. al. Table 2. Results of solving the estimation problem. Chemical Initial consistency Final consistency Aggregated Rank No. substances (objects) coefficient, KS(W) coefficient, KS´(W) estimate (weight), w0 1 CO 0,290 0,470 8,569 9 2 NO 0,650 0,790 9,334 8 3 NO2 0,670 0,770 10,286 5 4 Ammonia 0,350 0,507 6,142 12 5 Ethanol 0,562 0,796 3,620 14 6 Suspended solids 0,469 0,570 6,382 10 7 Formaldehyde 0,864 13,713 1 8 Acetic acid 0,404 0,529 6,331 11 9 Hydrogen fluoride 0,517 0,733 12,237 2 10 Manganese 0,563 0,818 4,095 13 11 Iron oxide 0,554 0,702 3,428 15 12 Hydrogen sulfide 0,593 0,752 11,236 3 Petroleum 13 0,426 0,617 11,093 4 hydrocarbons 14 Xylene 0,602 0,876 9,662 6 15 Toluene 0,549 0,840 9,571 7 Detection threshold TO = 0,429 Application threshold TU = 0,822 For all the other objects, the initial value of the consistency coefficient exceeded the detection threshold. As a result of the consistency increase, it was possible to make it close to the application threshold though not exceeding it. This also allows us to con- clude that the final expert estimates for these objects are sufficiently consistent. In the course of increasing the consistency, reasons for the low consistency of es- timates of some objects were identified. It was found that, in most cases, the opinions of experts with numbers 5 and 6, who had the least competence in the subject area, differed significantly from the other experts’ opinions. In this regard, it was decided to replace these two experts with one new expert whose competence roughly coin- cides with the competence of expert 4 and is estimated by the coefficient с5´ = 0.143 (exactly equal to the sum of the values c5 and c6). The estimation process was repeat- ed with the renewed expert group. The results are shown in Table 3. The results of the repeated expertise lead to the following conclusions. For seven objects (CO, NO, ethanol, formaldehyde, manganese, xylene, toluene), the final value of the consistency coefficient of expert estimates exceeds the applica- tion threshold, which means that the estimates are completely consistent. For the rest of the objects (NO2, ammonia, suspended solids, acetic acid, hydrogen fluoride, iron oxide, hydrogen sulfide, petroleum hydrocarbons), the application threshold could not be exceeded, but it was possible to approach it to an adequate degree This allows to calculate the final aggregated estimate with acceptable accura- cy. Intelligent Analysis of the Ecological State of Environment with Application of… 7 Table 3. Results of the repeated solving the estimation problem. Chemical Initial consistency Final consistency Aggregated Rank No. substances (objects) coefficient, KS(W) coefficient, KS´(W) estimate (weight), w0 1 CO 0,580 0,840 8,132 9 2 NO 0,670 0,825 9,460 8 3 NO2 0,660 0,774 10,686 5 4 Ammonia 0,570 0,780 6,192 12 5 Ethanol 0,562 0,847 3,400 15 6 Suspended solids 0,470 0,570 6,382 10 7 Formaldehyde 0,880 13,813 1 8 Acetic acid 0,480 0,746 6,245 11 9 Hydrogen fluoride 0,534 0,735 12,879 2 10 Manganese 0,563 0,818 4,095 13 11 Iron oxide 0,569 0,710 3,528 14 12 Hydrogen sulfide 0,613 0,778 11,431 4 Petroleum 13 0,456 0,678 11,568 3 hydrocarbons 14 Xylene 0,720 0,879 9,653 6 15 Toluene 0,549 0,840 9,469 7 Detection threshold TO = 0,429 Application threshold TU = 0,822 4 Application of expertise results to estimate environmental pollution in the districts of Bryansk region The resulting aggregated risk estimates of the impact of chemical air pollutants on the environment were used to construct an integral indicators of environmental pollu- tion in accordance with the methodology described in [5]. Table 4 and Fig. 1 show the results of calculating the integral indicator of environmental pollution for various districts of Bryansk region. Along with the indicator of chemical pollution, the results of calculating the indicator of radioactive pollution are presented. The calculation was carried out on the basis of a similar methodology with the expert estimation of the risk of exposure to radioactive substances. Table 4. Integral indicators of environmental pollution in the districts of Bryansk region. Integral indicator of pollution No. Districts radioactive rank chemical rank 1 Brasovsky 0,0225 9 0,00045 12 2 Bryansky 0,0171 15 0,00223 3 3 Dubrovsky 0,0158 20 0,00037 15 4 Dyatkovsky 0,0242 8 0,04321 1 5 Gordeyevsky 0,2031 2 0,00025 17 6 Karachevsky 0,0202 12 0,00059 11 7 Kletnyansky 0,01357 28 0,00013 23 8 Klimovsky 0,07294 6 0,00001 28 8 E. Geger, A. Podvesovskii, O. Mikhaleva, et. al. Table 4. (Continued). Integral indicator of pollution No. Districts radioactive rank chemical rank 9 Klintsovsky 0,1135 5 0,00018 19 10 Komarichsky 0,0216 10 0,0034 4 11 Krasnogorsky 0,20101 3 0,00011 25 12 Mglinsky 0,0044 27 0,0002 26 13 Navlinsky 0,0213 11 0,00016 20 14 Novozybkovsky 0,2800 1 0,00076 7 15 Pogarsky 0,01845 14 0,0016 5 16 Pochepsky 0,00544 24 0,00043 13 17 Rognedinsky 0,015995 18 0,00016 21 18 Sevsky 0,01491 22 0,00003 27 19 Starodubsky 0,03275 7 0,0004 14 20 Surazhsky 0,01496 21 0,00014 22 21 Suzemsky 0,0171 16 0,00061 10 22 Trubchevsky 0,01862 13 0,00064 9 23 Unechsky 0,0159 19 0,00076 6 24 Vygonichsky 0,0162 17 0,00032 16 25 Zhiryatinsky 0,0054 25 0,00013 24 26 Zhukovsky 0,0065 23 0,00074 8 27 Zlynkovsky 0,199 4 0,00025 18 28 Bryansk City 0,00491 26 0,037 2 As can be seen from the presented data, the most chemically contaminated territories are Dyatkovsky district and the city of Bryansk; in terms of the density of radioactive contamination, the following districts are the most contaminated: Novozybkovsky, Gordeevsky, Krasnogorsky, Zlynkovsky, Klintsovsky, Klimovsky, Starodubsky and Dyatkovsky districts. 5 Conclusion The paper considers an approach to the intelligent analysis of ecological situation, based on the application of multi-criteria optimization and technology for supporting group expertise in a distributed environment. An informative and reliable method for estimating anthropo-technogenic pollution of a territory is proposed. It uses integral indicators taking into account environmental pollution. This method was used to as- sess radioactive and chemical contamination of the districts of Bryansk region. Intelligent Analysis of the Ecological State of Environment with Application of… 9 Fig. 1. Displaying indicator values on the map of the Bryansk region. Expert technologies make it possible to overcome the limitations imposed by methods of statistical data processing as well as those associated with the impossibility of pro- cessing heterogeneous environmental and statistical data by traditional methods. Meanwhile, the transition from traditional methods of expert estimation to network ones provides an effective form of distributed interaction among the participants in this process and helps to reduce its total duration. In turn, the use of models and methods for control of expert estimates consistency contributes to an increase in the reliability of estimation results and a decrease in the influence of a random expert error on the final assessment. 10 E. Geger, A. Podvesovskii, O. Mikhaleva, et. al. References 1. Onishchenko, G.G.: O sanitarno-jepidemiologicheskom sostojanii okruzhajushhej sredy [On sanitary and epidemiological state of the environment]. Gigiena i sanitarija [Hygiene and sanitation] 92(2), 4–10 (2013). [in Russian]. 2. Popova, A.Yu., Gurvich, V.B., Kuzmin, S.V., Mishina ,A.L., Yarushin, S.V.: Sovremen- nye voprosy ocenki i upravlenija riskom dlja zdorov'ja [Modern issues of the health risk assessment and management]. Gigiena i sanitarija [Hygiene and sanitation] 96(12), 1125– 1129 (2017). [in Russian]. doi: 10.18821/0016-9900-2017-96-12-1125-1129 3. Geger, E.V.: Jekologicheskaja ocenka vlijanija tehnogennyh nagruzok okruzhajushhej sredy na formirovanie biologicheskih i medicinskih pokazatelej zdorov'ja naselenija (na primere Brjanskoj oblasti) [Environmental assessment of the impact of man-made envi- ronmental loads on the formation of biological and medical indicators of public health (on the example of the Bryansk region)] (Sc.D. in Biology Thesis), Bryansk: Bryansk State University (2014). [in Russian]. 4. Gubanov, D., Korgin, N., Novikov, D., Raikov, A.: E-Expertise: Modern Collective Intel- ligence. Springer International Publisher (2014). doi: 10.1007/978-3-319-06770-4 5. Geger, E.V.: Metodologija razrabotki integral'nogo pokazatelja zagrjaznenija dlja ocenki jekologicheskogo sostojanija okruzhajushhej sredy [The Methodology of Development of Integral Indicator to Assess the Environmental Pollution]. Perspektivy nauki [Science Per- spectives] 5(32), 190-193 (2012). [in Russian]. 6. Geger, E.V., Zolotnikova, G.P.: Vlijanie radiacionnyh i himicheskih nagruzok na izmenen- ija biohimicheskih pokazatelej jendokrinnogo gomeostaza u zhitelej iz jekologicheski razlichnyh rajonov Brjanskoj oblasti [Influence of Radiating and Chemical Loads on Changes of Biochemical Indices of the Endocrine Homeostasis in Inhabitants from Eco- logically Various Areas of the Bryansk Region]. Gigiena i sanitarija [Hygiene and sanita- tion] 95(5), 428–432 (2016). [in Russian]. doi: 10.18821/0016-9900-2016-95-5-422-428 7. Cai, J.: A social interaction analysis methodology for improving E-collaboration over the Internet. Electronic Commerce Research and Applications, 4(2), 85-99 (2005). doi: 10.1016/j.elerap.2004.10.007 8. Podvesovskii, A.G., Mikhaleva O.A.: Tehnologija podderzhki gruppovoj jekspertizy v raspredelennoj srede [Technology for Group Expertise Support in Distributed Environ- ment]. XXI vek: itogi proshlogo i problemy nastojashhego pljus [XXI Century: Resumes of the Past and Challenges of the Present plus] 8(4), 20–25 (2019). [in Russian]. 9. Podvesovskii, A., Mikhaleva O., [et. al.]: A Model of Control of Expert Estimates Con- sistency in Distributed Group Expertise. In: Kravets, A. et. al. (eds.) CIT&DS 2017, CCIS, vol. 754, pp. 361-374. Springer International Publishing AG (2017). doi: 10.1007/978-3- 319-65551-2_26 10. Totsenko, V.G.: Spectral Method for Determination of Consistency of Expert Estimate Sets. Eng Simulat 17, 715-727 (2000).