A simplified material flow analysis employing local expert judgment and its impact on uncertainty

Mandalay, the capital of Mandalay region, is the second largest and most densely populated city in Myanmar. It is on the east bank of the Irrawaddy River on approximately 80 m above sea level and in the central dry zone of the country with hot semi-arid climate and around 800 mm of annual precipitation. The city occupies 912.7 km² with a population of 1.7 million [14]. This study selected five urban townships: Aung Myae Tharzan, Chan Aye Tharzan, Chan Mya Tharzi, Mahar Aung Myae, and Pyigyi Dagon, which cover 95% of the city’s urban areas (108 km²) and 1.2 million people with 0.8 km² of agricultural land [14].

Mandalay fully relies on on-site sanitation, which are mainly septic tanks receiving blackwater only. As of 2017, among emptying-experienced on-site sanitation in urban Mandalay, 32.8% were formally emptied and fecal sludge was transported to the designated disposal pond, while the remaining were emptied as informal business and majority of them were illegally dumped [15]. Fecal sludge is one of major issues in the city. No domestic wastewater treatment plant exists, and a combined open drainage system is used for greywater management in the city. Although there are three industrial zones inside the study area, there are no centralized industrial wastewater treatment plant apart from some stabilization ponds running by a few industries in the area. There are two landfill sites for solid waste management. As of 2016, the solid waste collection of the city development committee covered 80% of the city [16].

The sensitivity of each parameter to the total loading to the soil and water environment was analyzed following the methodology of a previous study [24]. Briefly, all the parameter values were fixed at the 50^th percentile (i.e., p50) of each PDF parameter, and the loading was estimated as p50. Keeping all the parameters at p50 except one parameter, the value of one parameter was changed to the 10^th percentile (p10) and the 90^th percentile (p90) of the PDF. The calculated total loadings were obtained as p10 and p90, respectively. The sensitivity of the parameter was expressed as the ratio of the total loading change as p10:p50 and p90:p50.

Table S5 summarizes the waste management practices in the study area, based on the intensive survey results. Briefly, 98% of excreta is discharged to on-site sanitation, with the remainder being directly discharged into the environment. This proportion is slightly higher than the proportion in urban areas of Mandalay region (92.3%) [25]. Of the greywater, 99% is discharged to open drainage. Of the kitchen waste, 80% is transported to disposal sites; the percentage is the same magnitude to the proportion in urban areas of Mandalay region (70.6%) [25]. Of the emptied fecal sludge from on-site sanitation, only 20% is formally disposed of at the designated pond, and the remaining is discharged to the environment. Similarly to the management of many cities in low- and middle-income countries [26], fecal sludge management is poor in the area. Of livestock manure, 57% is used in agricultural activities outside the study area, whereas the remaining is discharged to the environment. Of the agricultural residues, 84% of rice residues and 100% of vegetable residues are left in agricultural fields, whereas 16% of rice residues are used for livestock feeding. Of industrial organic solid waste, 72% and 11% are used for livestock feeding outside and inside the study areas, respectively; 11% is disposed at disposal sites; the remaining is discharged to the environment. All industrial wastewater is directly discharged into the environment.

The LEJ on waste management practices was compared with the intensive survey (IS), as shown in Fig. 2 and Table S5. Although the LEJs varied among ten local experts, the averages of LEJs and IS were of the same magnitude for on-site sanitation (81% and 82% for the septic system), greywater (97% and 99% to the environment), livestock manure (58% and 57% for agriculture), kitchen waste (65% and 80% to disposal sites), and industrial wastewater (100% and 100% to the environment, respectively).

In contrast, the averages of LEJs and IS were substantially different for agricultural residues (53% and 92% left on agricultural fields), fecal sludge (82% and 20% to disposal ponds), and industrial organic solid waste (73% and 11% to disposal sites, respectively). For agricultural residues, the selection of experts might create a gap between the LEJ and IS. According to a previous study [27], 8–15 local experts are necessary, and the selection of experts is still important. In the present study, although ten experts were used, seven of them were selected from the waste management sector, and they may not have had enough knowledge of local farmer practices regarding agricultural residues. For fecal sludge, the experts were selected from the sector responsible for fecal sludge emptying. However, the city has widespread illegal behavior concerning fecal sludge emptying; 65% of fecal sludge was emptied illegally in the city [15]. This intensive informal activity possibly resulted in a discrepancy between the LEJ and IS. Another possible reason for this discrepancy is response bias with social desirability [28], in which the responsible experts potentially desired formal emptying activities in the city.

Of the industrial organic solid waste, the large gap could be explained by two possible reasons. First, IS estimation was based on interviews with industrial managers. Potentially due to the response bias with dissimulation [28], they might have reported less disposal waste to reduce disposal fees, which was determined based on the disposal amount. Second, the local experts, responsible for waste management in the city, possibly missed the resource recovery of private sectors beyond the city area. No expert answered industrial solid waste burning on the site and the resource recovery actions outside of the boundary, although 3% and 72% of industrial solid waste were burned on-site and used for livestock outside the study area according to IS. In contrast, experts had a good estimate for industrial solid waste resource recovery inside the city area: 11% and 11% by LEJ and IS, respectively. These results partly imply the good applicability of LEJ to estimate waste and wastewater management practices, although a special attention is required for following: the proper selection of experts' expertise, response bias, illegal business, disposal fees, and recycling outside of the experts’ jurisdiction area.

The baseline nitrogen and phosphorus flows were estimated based on the medians of stochastic iMFA, and the details are shown in Figure S2(a) and (b). As summarized in Fig. 3, the total pollution loading to the soil and water environment (i.e., the environment) was estimated at 5,981 tons of nitrogen per year (tons/year) and 1,040 tons of phosphorus per year (tons/year), equivalent to 55 tons of nitrogen per year per square kilometer (tons/year/km²) and 9.6 tons of phosphorus per year per square kilometer (tons/year/km²). These loadings are of the same magnitude as those of other cities in Southeast Asia, such as Bangkok [7], Hue [29], and urban Hanoi [30]. Similarly to the above three cities of the total loading to the environment, the majority (85% of nitrogen and 81% of phosphorus) were derived from toilet effluent/leakage from on-site sanitation for nitrogen (3,670 tons/year) and phosphorus (560 tons/year) and greywater from households for nitrogen (1,369 tons/year) and phosphorus (277 tons/year). In addition, industrial wastewater from industries accounted for 7% (395 tons/year) of the total nitrogen loading and 6% (63 tons/year) of the total phosphorus loading to the environment. Agricultural runoff was limited, as the study sites were urban areas, and agricultural activity was not active. Of the total nitrogen and phosphorus input to the boundary, 40% was discharged into the environment. Thus, improvement of toilet/on-site sanitation management would be the most crucial challenge to establish sound nutrient cycle in urban Mandalay.

Focusing on resource recovery inside the study area, livestock inside the study area played a major role in recycling waste flows (Figure S2), similarly to sub-urban Hanoi [30]. As shown in Fig. 4a, livestock received 330 tons of nitrogen per year (tons/year) and 53 tons of phosphorus per year (tons/year) of organic solid wastes from industry for livestock feeding based on iMFA, accounting for 36% nitrogen and 27% phosphorus of the total input to livestock inside urban Mandalay and 519 tons of nitrogen per year (tons/year) and 128 tons of phosphorus per year (tons/year) of commercial feed from the market. Conversely, the waste resources generated inside the study area were actively used outside the boundary. Livestock outside the study area received 2,610 tons of nitrogen per year (tons/year) and 420 tons of phosphorus per year (tons/year) of organic solid wastes from the industry for feeding (Fig. 4b), corresponding to 19% nitrogen and 19% phosphorus of the total output from urban Mandalay. Along with livestock, agriculture outside the study area also contributed to the resource recovery of the wastes generated inside urban Mandalay: 270 tons of nitrogen per year (tons/year) and 88 tons of phosphorus per year (tons/year) of recovered manure from livestock inside the area (Fig. 4b), corresponding to 2% nitrogen and 4% phosphorus of the total output from urban Mandalay. Thus, the resource recovery of the waste generated inside urban Mandalay was mainly achieved by livestock and agricultural activities outside urban Mandalay rather than that inside.

As shown in Fig. 3, the medians of the total nitrogen and phosphorus pollution loading to the environment from the sMFA were 6,144 tons of nitrogen per year (tons/year) and 1,171 tons of phosphorus per year (tons/year), respectively; 3% and 11% higher than those of the iMFA. Corresponding to the baseline estimation by the iMFA, the four largest flows to the environment by the sMFA were effluent/leakage from on-site sanitation for nitrogen (4,390 tons/year) and phosphorus (534 tons/year), greywater from household for nitrogen (1,323 tons/year) and phosphorus (271 tons/year), wastewater from industry for nitrogen (671 tons/year) and phosphorus (217 tons/year), and manure from livestock inside the area for nitrogen (195 tons/year) and phosphorus (64 tons/year). Overall, the sMFA demonstrated the effectiveness of estimating pollution loading to the environment.

For these four major flows to the environment, the median estimates of the sMFA were similar to those of the iMFA within a 5% difference, except for industrial wastewater. The nitrogen and phosphorus loadings to the environment as industrial wastewater were 70% and 244% higher than those of the iMFA, respectively. This overestimation can be explained as follows: (1) as mentioned above, the estimation of industrial wastewater generation in the sMFA relied on the guideline values [21, 22] created outside Myanmar as the expert could not answer any LEJ during our interview; this might result in a significant error in the industrial wastewater flows estimated in the sMFA. (2) The gap in industrial wastewater flows between the iMFA and sMFA could be related to underestimating the iMFA caused by the response bias with dissimulation [28], similar to the industrial organic waste disposal in the previous section. Potentially, industrial managers provided information on lower water consumption and wastewater discharge than the actual values to avoid production capacity tax and waste disposal charges.

As shown in Fig. 4a, for resource recovery, industrial solid waste recycling by livestock inside the area by the sMFA was 40% lower than that of the iMFA. Kitchen waste recycling as animal feeding in livestock by the sMFA (46 tons/year for nitrogen and 15 tons/year for phosphorus) was higher than that by the iMFA (3 tons/year for nitrogen and 1 ton/year for phosphorus). The remaining resource recovery options within the system boundary were similar to those of the iMFA. As seen in Fig. 4b, no waste was used on livestock outside the area by the sMFA, despite being one of the major recycling flows by the iMFA. As mentioned in the previous section, experts could not recognize the resource recovery action of industrial waste by the livestock outside. In contrast, the resource recovery of livestock manure to agriculture outside the area showed similar results for the iMFA and sMFA. As many low- and middle-income countries still face challenges in establishing good industrial statistics [31], the lack of good local industrial statistics could be a challenge to increase the accuracy of the estimation by both the iMFA and sMFA.

The variability of total and individual nitrogen loadings to the environment by the iMFA and sMFA was examined by the normalized median difference and the normalized 80% CI of the estimated loadings of the sMFA compared with the iMFA (Eqs. 3 and 4). Overall, the total loadings of the iMFA and sMFA were 5,214; 6,419; and 8,059 tons of nitrogen per year (tons/year) and 5,595; 6,805; and 8,446 tons of nitrogen per year (tons/year), and 913; 1,117; and 1,380 tons of phosphorus (tons/year) and 1,078; 1,283; and 1,549 tons of phosphorus (tons/year) (1st; 2nd; and 3rd quartiles, respectively). The normalized median differences of total loading were, + 0.03 and 0.11 for nitrogen and phosphorus, respectively (Figs. 5 and S4). The normalized width of 80% CI of the total loading was − 0.05 and − 0.11 for nitrogen and phosphorus, respectively. These indicate the limited impact of the simplification of MFA and the validity of the sMFA to estimate the total loading to the environment.

The distribution of the individual nitrogen loadings from each component to the environment are shown as histograms in Fig. 5 (refer to Figure S4 for phosphorus). Overall, the distribution patterns of individual nutrient flows were similar between the iMFA and sMFA, except for fecal sludge and industrial wastewater. The normalized median differences of on-site sanitation effluent/leakage, greywater, and livestock manure for nitrogen, which were three of the four major flows, were − 0.05, − 0.03, and − 0.03, respectively; those of the same three flows for phosphorus were − 0.05, − 0.02, and − 0.03, respectively. The normalized widths of 80% CI of these three major flows for nitrogen were + 0.02, 0.00, and + 0.17, respectively, and those for phosphorus were + 0.01, 0.00, and + 0.17, respectively. Such results indicate the limited impact of MFA simplification and the good performance of the sMFA in terms of the loading estimation and the estimates variability of these three major flows.

In contrast, a large gap was observed between the iMFA and sMFA for industrial wastewater, which is the remaining of the four major flows. The normalized median differences of the loading of industrial wastewater were + 0.70 and + 2.44 for nitrogen and phosphorus, respectively; the normalized widths of 80% CI of the loading were − 0.75 and − 0.70 for nitrogen and phosphorus, respectively. The large median difference in the industrial wastewater loading could be explained by the possible underestimation by the iMFA, as mentioned in the previous section. The normalized median difference was particularly large for phosphorus (+ 2.44). The livestock feeding industry dominantly contributed to the phosphorus loading of industrial wastewater (Table S4), derived from the difference in wastewater generation for the livestock feeding industry between the iMFA and sMFA, as both models used the same phosphorus concentration data of the manure (Table S2). Because of the large variability, no expert could provide estimation on industrial wastewater generation. The PDFs (triangular) of industrial wastewater amount for each industry in the sMFA were based on two references [21, 22], whereas the PDFs (normal) in the iMFA were based on wide-ranging results of 65 industry surveys conducted in the present study (Table S4). These differences in the data source and created PDFs resulted in a smaller variability in industrial wastewater flow estimates in the sMFA than in the iMFA. As this flow accounted for 11% and 19%, respectively, for the total loading of nitrogen and phosphorus to the environment, the negative values [− 0.75 (Fig. 5) and − 0.70 (Figure S4)] of the normalized width of 80% CI of the industrial wastewater loading adversely affected the normalized width of 80% CI for the total loading to the environment (− 0.05 and − 0.11, respectively), indicating smaller variability in the estimates of total loading in the sMFA.

Apart from the four major flows, the large difference between the iMFA and sMFA was also observed in human excreta and fecal sludge, showing the limited performance of the sMFA. The normalized median difference for human excreta and fecal sludge was, respectively, + 1.00 and − 0.88 for nitrogen (Fig. 5) and + 1.00 and − 0.82 for phosphorus (Figure S4); the normalized widths of 80% CI were + 1.11 and − 0.88 for nitrogen and + 1.04 and − 1.03 for phosphorus. The median differences in human excreta corresponded to the proportions of the population without toilet facilities estimated by IS (median: 0.02) and LEJ (0.065), as shown in Fig. 2. This overestimation of LEJ could be explained by less-updated information by local experts under the rapidly increasing coverage of the septic tanks in urban areas of Mandalay region: 28 thousand in 2014 [32] to 330 thousand in 2019 [25]. The loading of fecal sludge was underestimated in sMFA compared with the iMFA, as shown in Fig. 5. This difference stems from IS and LEJ estimation differences on the proportion of fecal sludge discharged to the environment in Fig. 2c: 0.80 in IS, and 0.18 in LEJ, possibly caused by the active informal fecal sludge collecting business [15] not mentioned by the local experts. However, human excreta and fecal sludge were minor in both the iMFA and sMFA and did not significantly impact the sMFA overall.

The seven most sensitive parameters for the total pollution loading to the environment in the sMFA are summarized in Table 1. The unit loading in human excreta (C_2(N,P)) was the highest for p10:p50 for nitrogen (ratio: 0.80) and p90:p50 for phosphorus (1.54), and the unit loading in greywater (C_3(N,P)) was p90:p50 for nitrogen (1.35) and p10:p50 for phosphorus (0.81). The proportion of on-site sanitation effluent/leakage going to the environment (β_2,9) was the second highest for p10:p50 and p90:p50 for nitrogen (0.84 and 1.39) and p10:p:50 for phosphorus (0.82). Among the 47 parameters based on PDFs, sensitivity analysis highlighted the importance of the unit pollution loadings in human excreta and greywater and the proportion of on-site sanitation effluent/leakage to the environment. Such findings are in line with those reported in the previous studies conducting sensitivity analysis for MFA [10, 11, 18], wherein the unit pollution loadings in human excreta and/or greywater were the most sensitive parameters for nutrient flow in each study area.

Although the parameter variability of the unit loading in human excreta and the proportion of on-site sanitation effluent/leakage were relatively small (Table S2), there was high sensitivity to the total environmental load. This sensitivity is due to the loading of effluent/leakage from on-site sanitation, which received human excreta, was large, accounting for 57% and 46% of the total load to the environment for nitrogen and phosphorus, respectively (Figure S3). In contrast, the parameter variability of the unit loading in greywater was large (Table S2), and the loading of greywater to the environment was also high, accounting for 22% and 23% of the total loading for nitrogen and phosphorus, respectively (Figure S3). Therefore, the high sensitivity of the unit loading in greywater to the total environmental load was caused by the variability of the parameter itself and the large size of the flow relevant to greywater. The present study could not find the local data of the unit loading in greywater in Myanmar, and therefore used the data from the previous studies in other Southeast Asian countries (n = 12 for nitrogen and n = 8 for phosphorus) (Table S2), potentially increasing the uncertainty of the parameter. Although LEJ could complement local statistics on waste and wastewater management and provide good estimation in the MFA, supplementary surveys on sensitive and largely variable parameters, such as unit loading in greywater, would increase the accuracy of the sMFA. Such a focused and limited-resource survey would effectively decrease the uncertainty and increase the accuracy of the material flow.