A toolbox to find the best mechanistic model to predict the behavior of environmental systems

Abstract

Reliable prediction of the long-term behavior of environmental systems such as Municipal Solid Waste (MSW) landfills is challenging. While many driving forces influence this behavior, characterization of them is limited by measurement techniques. Therefore, a model structure for reliable prediction needs to optimally combine all measured information with suitable mechanistic information from literature. How to get such an optimal model structure? This study presents a toolbox to find and build the model structure that describes an environmental system as close as possible. The toolbox combines environmental frameworks to include all suitable mechanistic information; it fully couples kinetic and equilibrium reactions and contains multiple resources to obtain biogeochemical parameters. Several possible optimal model structures are quickly built and evaluated with objective statistical performance criteria obtained via Bayesian inference. By applying the novel methodology, we select the best model structure for anaerobic digestion of MSW in full scale landfills. Response to Reviewers: There were no comments from the reviewers.

Theoretical analysis of municipal solid waste treatment by leachate recirculation under anaerobic and aerobic conditions

“…Calibrated bandwidths of maximum rates are a bit lower than the reference values presumably due to mass transport limitations in the experiments which have not been included in the model. The process network closely resembles the network identified for a similar type of lysimeter experiment by vanTurnhout et al (2016). This suggests that the selected processes describe biodegradation of MSW on a general level for lysimeter scale.…”

“…b) X are species of bacteria with the elemental composition CH 1.4 O 0.4 N 0.2 taken from Henze et al (1995). c) Substrate limitation factors (f SL ) and inhibition factors (f NC ) range between 0 and 1 (van Turnhout et al, 2016). Half saturation constants which are not calibrated have low values and are primarily included as switch factors.…”

“…MSW was sieved to a grain size < 20 mm and initially had a water content of 23% and a dry bulk density of 846 kg m 3 . Figure 2: Iterative procedure for finding the biogeochemical process network that describes measured data with minimal ambiguity and uncertainty (van Turnhout et al, 2016). In step 1, the user manually builds the best network with the toolbox which is automatically optimized for the measured data in step 2.…”

“…One time series of each experiment was used for calibration and the other for validation. The calibration dataset was used to find the least ambiguous process network with the toolbox developed by van Turnhout et al (2016). Subsequently, the network was tested on the validation dataset without further parameter fitting.…”

“…If not, the user starts another iteration and adapts the network. Figure 2 shows the approach to select the least ambiguous biogeochemical process network describing the measured data (van Turnhout et al, 2016). First, a model structure is build to be evaluated.…”