3.1 Gravimetric composition of municipal solid waste arriving at the landfill
Identifying the different components among the MSW that arrive at a landfill allows for predicting the mechanical and biodegradative behavior of the waste. Figure 3 illustrates the gravimetric composition of the MSW received at the LBSR.
Figure 3 illustrates that the categories "organic," "soft plastic," and "sanitary textiles" collectively constitute the most substantial portion, representing approximately 73% of the analyzed material.
Moreover, Fig. 3 highlights the considerable proportion of organic waste, approaching 44%, a value consistent with the national and global averages estimated by Abrelpe (2020) and Waste Atlas (2018), respectively. The estimated production of organic waste worldwide varies depending on assessed socioeconomic parameters. Upon disposal in landfills, the biodegradation of this material induces deformations in the waste mass, resulting in notable settlements, increased gas production, and high leachate generation during the initial landfilling phase.
According to Nanda and Berruti (2021), the proportion of organic waste generated exhibits significant variation based on income level. Therefore, higher income levels correspond to lower percentages of organic waste production. In population segments characterized by low income, such as in poor countries, the organic content in MSW can constitute up to 65% of the gravimetric composition. Conversely, developed and high-income nations typically record values below 30%, whereas in developing countries with moderate to high incomes, this figure surpasses the 50% threshold (Aleluia and Ferrão, 2016).
The quantity of organic matter holds particular significance, as the greater the amount of a specific component, the more closely the overall characteristics of the mass will align with that component. Compared to other constituents, this predominant presence of putrescible organic matter exerts influence over various factors, including leachate generation, biogas production, pore pressures within the mass, water content, resistance, compressibility, and settlement.
The recyclable waste fraction comprises plastics, paper/cardboard, glass, and metal, collectively constituting 33% of the waste fraction, equivalent to 215 tons per month of the waste disposed at the LBSR (Fig. 3). This material category can undergo the recycling process, generating revenue and employment opportunities while significantly diminishing the volume of waste destined for landfill disposal.
As highlighted by Aurpa and Islam (2022), there has been a significant increase of approximately 60% in the disposal of plastic waste in landfills during the COVID-19 pandemic and its aftermath. The proportion of plastics, encompassing both soft and hard varieties, deposited in the LBSR stands at approximately 22%, denoting a substantial presence influenced by shifts in the consumption behaviors of the general population in recent years.
The presence of plastic waste in landfills is desirable, as it serves a reinforcing function, enhancing tensile strength and enabling the optimization of landfill height (König and Jessberger, 1997). However, plastic waste can lead to diverse issues when present in excessive quantities and indiscriminately disposed of. Due to its mechanical characteristics, plastic exhibits a cushion effect during compaction, where it may rebound after compaction, creating the false impression of adequate compaction (Melo et al., 2016; Araújo Neto et al., 2021). Moreover, an abundance of plastic within the landfill mass can form pockets, compromising leachate drainage, the biodegradation process, liquid percolation within the mass, and the stability of landfill slopes (Yu et al., 2022).
The textile and leather (6%) and sanitary textiles (13%) categories emerge as significant constituents contributing to the overall percentage of waste. As indicated by Villalba (2020), the consumption of these materials correlates directly with population growth and per capita income levels. Consequently, it is imperative to conduct dedicated studies to explore the optimal environmental management strategies for these materials, encompassing collection, reuse, recycling, and appropriate final disposal stages.
3.2 Theoretical gravimetric composition of municipal solid waste arriving at the landfill
Statistical metrics (MAE, RMSE, NRMSE, R², and MAPE) were employed to assess the performance of various neural models. These networks were characterized by diverse architectural configurations determined by combinations of activation (or transfer) functions in the input and output layers, the number of neurons in the hidden layer, and training methodologies. Figure 4 depicts the RMSE variation graphs relative to the number of neurons in the hidden layers, associating the lowest RMSE point with other statistical metrics corresponding to the respective training sessions.
Upon examination of each residue category, activation functions, neuron ranges from 1 to 20, and training algorithms, statistical analysis reveals that MAE, RMSE, and NRMSE values fluctuated between 0.17 and 5.3%, while R² consistently exceeded 0.5. Meanwhile, MAPE ranged between 8 and 25%, depending on the presented physical compositions of MSW.
In general, the statistical error parameters MAE, RMSE, and NRMSE exhibit variation based on the dimension of the output data. Therefore, the efficiency of these parameters is relative to the precision of the analysis conducted, with their values representing the average deviation between observed and predicted outcomes. RMSE can identify deviations of large magnitudes, while MAE conducts a more localized analysis, equalizing all deviations. Notably, for this study, all evaluated error parameters proved satisfactory, as their variation was small compared to the values of the studied series.
MAPE indicates the average of absolute percentage errors, where the lower values, the greater the precision. According to Vivas et al. (2020), MAPE < 10% signifies highly accurate predictions, while values between 10% and 20% indicate good predictions. Predictions ranging between 20% and 50% suggest reasonable accuracy, whereas MAPE > 50% indicates low accuracy. In this context, it is noteworthy that the organic waste category yielded a MAPE of 8%, indicating a highly accurate forecast. The paper fraction exhibited a MAPE of 19%, indicative of a good prediction, while the remaining categories provided reasonable predictions. Despite having intermediary MAPE values, it is essential to highlight that the models maintained good precision, as MAPE should be evaluated in conjunction with other statistical metrics, all of which yielded low error values.
The dispersion of MAPE values suggests that the proportion of organic waste remains relatively consistent across municipalities within the output database. Conversely, categories such as paper, metal, glass, and plastic exhibit elevated MAPE values, indicating potential disparities in category percentages among municipalities. These fluctuations are intricately tied to the socioeconomic, cultural, regional, and climatic characteristics of each municipality.
The performance of R² remains largely independent of the linearity of the ANN adjustment model, with values closer to 1 indicating a higher degree of agreement between the predicted and observed data. Therefore, when selecting the optimal models for each waste category, assessing R² in conjunction with other performance metrics such as MAE, RMSE, NRMSE, and MAPE is imperative. For instance, while the R² value for the paper fraction was relatively low in this study, the statistical analysis of errors revealed minimal dispersion, concluding that the models were satisfactory.
Figure 4 illustrates that the models evaluated in this research exhibit adequate outcomes compared to previous studies on gravimetric composition predictions using ANN. In a similar study involving gravimetric data from Johannesburg, South Africa, Adeleke et al. (2021) reported comparable RMSE values ranging between 2% and 5%, R² values between 0.8 and 0.9, and MAPE values of 12–20%. Their analysis encompassed four waste categories (organic, paper, plastic, and textiles), employing solely climate data from the region, such as wind speed, humidity, and maximum and minimum temperature as input variables.
In another investigation by Ma et al. (2020), gravimetric composition served as the output variable, with the only validation parameter applied to their models being R². They identified the most effective training scenarios with R² values exceeding 0.6.
3.3 Actual versus theoretical gravimetric composition of waste arriving at the Landfill in the Brazilian Semiarid Region
The neural training models generated predictions regarding the compositions of municipal solid waste from all municipalities depositing waste in the LBSR. Subsequently, the weighted arithmetic mean was calculated by leveraging percentages of MSW categories and the mass of waste from each city, yielding the average theoretical gravimetric composition reaching the facility. Thus, it was possible to compare the theoretical MSW composition with the gravimetric composition of the waste received at the LBSR (Fig. 5).
As illustrated in Fig. 5, the statistical parameters MAE, RMSE, and MAPE were approximately 2.1, 2.8, and 15, respectively. These findings indicate minimal dispersion when contrasting the theoretical gravimetric composition data with the actual values. Statistical parameters were employed to corroborate the close alignment between the two compositions by comparing theoretical and actual values.
According to Fig. 5, the Plastics category exhibited notable disparities between the predicted value (16%) and the actual value (22%). This variance could be attributed to the inherent limitations of the selected model for predicting this category, as models can only estimate the values they are trained on, potentially not fully capturing the actual value. The quantity of plastic content is intricately correlated to the socioeconomic attributes of a given area. Thus, when examined in the original database, considerable variations are evident, contingent upon the specific location under analysis.
Moreover, it is noteworthy that certain municipalities in Brazil, as depicted in the ANN database, may have implemented reverse logistics and selective collection initiatives. These initiatives are likely instrumental in diminishing the presence of recyclable waste within their gravimetric compositions, potentially leading to more significant variability and dispersion of data across these categories.
Another factor contributing to the variance between predicted and observed values can be attributed to the nature of the database compilation. The dataset employed in this study was sourced from national-level scientific literature, encompassing socioeconomic indicators that might partially depict the municipalities under consideration for gravimetric composition predictions. Moreover, there are indications that Brazilian socioeconomic data are outdated, indicative of challenges in data collection and tabulation by the official Brazilian authorities. Consequently, these factors introduce potential inaccuracies, leading to discrepancies between the predicted values and the actual conditions.