4:00pm - 4:20pm
Spatially-Explicit Sustainable Manufacturing Site Design Using Techno-Ecological Synergy
The Ohio State University, United States of America
A history of separating technology and ecology has led to many unintentional environmental impacts. These impacts affect ecosystems that provide services which we rely on for survival and well-being. Ironically, in search for technology that improves the livelihood of mankind, we risk the well-being that nature provides. Therefore, to make better decisions, there is a need for designing with respect to both technological and ecological systems. Along with decreasing negative environmental impacts, including ecosystems in sustainable design can lead to innovative solutions which utilize the functions and co-benefits of nature as we search for sustainable solutions amid increasing population and demand of ecosystems.
Previous work in techno-ecological process design has lacked inclusion of spatial heterogeneity of ecosystem service supply and demand. Previous research in ecosystem services does provide methods for mapping ecosystem services across different spaces and scales; however, mostly fails to connect these services to specific beneficiaries, such as manufacturing sites. Without this connection, the concept of servicesheds cannot be easily quantified for use in design. Servicesheds are the areas which provide specific ecosystem services to specific beneficiaries. For spatially-explicit sustainable design to be a reality, it is important to understand how, where, and when mass and energy flows across components of the techno-ecological system. Understanding the capacity of local ecosystems to regulate the chemicals released to the atmosphere from industrial processes can provide understanding of absolute sustainability, the condition where our demand on ecosystems is less than their available supply within a scale of interest. This research will both analyze the spatially-explicit interactions between ecosystem services and a manufacturing site and explore design potentials in response.
The ecosystem service that this research initially focuses on is air quality regulation. As a result, it uses physical dispersion models to determine the spatial heterogeneity of air pollution and ecological deposition. As part of the analysis, we provided quantitative definitions for air quality regulating servicesheds to understand the location of ecosystems which provide services to the manufacturing site. These definitions were applied to a case study biodiesel plant along the Ohio River near Cincinnati. Further, we modeled additional scenarios, such as land management and restoration, technological process changes, and manufacturing site location. Using this information, we were able to determine optimal sets of land where management and restoration has the highest potential to increase the capacity of deposition for criteria air pollutants. Further, we were able to use the results of these scenarios to compare techno-ecological design options based on sustainability and financial feasibility. As a step towards including multiple ecosystem services in the sustainability analysis of the scenarios, initial results of including other services will also be discussed. Providing quantitative definitions of servicesheds and exploring spatially-explicit design options for techno-ecological systems enables smarter industrial site design and is one step closer towards bridging ecological knowledge with engineering practice.
4:20pm - 4:40pm
Analysis of Urban Metabolism Models from an Ecological Perspective
1School of Mechanical Engineering, Georgia Institute of Technology, United States of America; 2School of Biological Sciences, Georgia Institute of Technology, United States of America
Nature is often copied or mimicked in science and engineering for novel solutions and unique analysis, e.g., in biologically inspired design of products. Another example of this is the field of Urban Metabolism which is based on the idea that material and energy use in human population centers is similar to metabolism in a living organism. Urban Metabolism looks to characterize what is consumed, how much is consumed, and what is exported (or excreted) within a specified geographic area. Tools such as material flow analysis and input-output analysis are used to track flows into, within, and out of a city. These flows are often a measured material flow such as water or nutrients but can also be energy or the mass of equivalent coal.
While this is a unique way of looking at cities by viewing it as a biological process, it has been criticized for treating cities as an organism when their function is more closely related to a collection of organisms, in other words, an ecosystem. Ecologists have established different ways to analyze ecosystems to gain insight into how they function and are developed. One of the most prominent forms of analysis used by ecologists is Ecological Network Analysis (ENA). ENA, rooted in information theory, is performed by creating a network of connections between organisms within the ecosystem and analyzing the structure and/or flow of material or energy between them. This analysis includes a number of ecological metrics that describe the overall health, maturity, and function of individual organisms as well as the ecosystem as a whole. These metrics look at the relationships between organisms and how those affect overall ecosystem performance. ENA is a tool for analyzing networks and as a result this can be applied to cities, specifically Urban Metabolism models, to understand these systems through the lens of ecology.
In this study, a number of cities with Urban Metabolism models are analyzed using ENA. The results show that, in terms of the ecological metrics, these cities are lacking in performance when compared to natural ecosystems. The cities have less cycling, lower resiliency, and far fewer connections than the natural ecosystems. If nature is to be used as a benchmark for sustainable design, these results indicate that there is still much room for improvement to reach true eco-cities that mimic the function of ecosystems. Additionally, the cities are compared one to another as well as to other human designed networks to show a ranking of ecological performance, and through this it is seen that most human designed networks are similar. This analysis of flows around urban areas provides a better understanding of the performance of these networks that goes beyond the typical Urban Metabolism model and can be used to further design these systems using nature as the guiding principle.
4:40pm - 5:00pm
Machine learning-based model for estimating carbon losses linked to road expansion in the Peruvian Amazon
Pontificia Universidad Católica del Perú, Peru
Extraction for raw materials in the Amazon has increased in recent years, with cattle ranching, palm oil, urban sprawling or informal mining generating important deforestation rates. For these raw materials to be extracted efficiently, road expansion is needed. Recent studies, in fact, estimate that by 2050 there will be at least 60% more roads than in 2010, most of which will be constructed in tropical areas of the globe. In this context, the main objective of this study is to increase the understanding of the deforestation patterns linked to road planning and expansion in the Peruvian Amazon. For this, four machine learning techniques (e.g., random forest, neural networks) were implemented in order to propose and evaluate adapted deforestation models to the regional characteristics, using variables associated with road and physical parameters. A large-scale analysis was performed to the whole Peruvian Amazon territory using cloud-based tools (e.g., Google Earth Engine). A one hundred meter/pixel resolution map has been generated containing the probability of deforestation and the amounts of released carbon. This map is publically available for its visualization and use. Carbon emissions for future scenarios can be estimated by combining georeferenced carbon density data and predicted deforestation. It is expected that these results will allow stakeholders, namely policy-makers, to quantify the environmental impacts of existing and future road expansion plans in the yet not widely populated Peruvian Amazon.
Keywords: deforestation; machine learning; neural networks; Peru; random forest
5:00pm - 5:20pm
Predicting spatially explicit life cycle environmental impacts of crops under future climate scenarios with machine learning approaches
State University of New York at Albany, United States of America
Agriculture production, as a primary stage for providing essential food and fuel supplies, is associated with a range of environmental challenges spanning from burgeoning greenhouse gas (GHG) emissions to water pollution. Agriculture currently contributes to approximately 20-25% of life cycle GHG emissions, in the United States (US) and globally. Nitrogen and phosphorus from agricultural production are among the leading causes of water pollution. Without immediate and effective mitigation efforts, climate change accompanied with the continuously increasing demand for food and fuel, will further accelerating environmental degradation. Efficiently quantifying environmental releases from agriculture is urgently required to ensure the long term sustainability.
However, our understanding of spatially explicit life cycle environmental impacts from crop production under future climate scenarios is very limited to date. Currently, emission factors and process-based mechanism models are popular approaches for estimating agricultural life cycle impacts. Despite valuable, the emission factors are incapable of describing spatial heterogeneity of agricultural emissions, whereas process-based mechanism models, capturing the heterogeneity, tend to be very complicated, and time-consuming to apply. To address this method challenge, this study develops rapid predictive approaches for estimating future life cycle environmental impacts from agricultural production, by utilizing novel machine learning techniques.
To build the rapid predictive models with best accuracy, we first tested three cutting-edge machine learning techniques, including Boosted Regression Tree (BRT), Random Forest (RF), and Deep Neural Network (DNN), based on the soil, climate, farming practices, topographic information along with historical estimates of life cycle environmental impacts of crop production in the Midwest counties. Then, using the best fitting model, we estimated future life cycle environmental impacts under four under four representative scenarios identified by Intergovernmental Panel on Climate Change (IPCC), including Representative Concentration Pathway scenarios (RCP) 2.6. 4.5, 6.0, and 8.5.
The preliminary results suggested that machine learning models such as BRT, RF and DNN yielded high predictive performance and cross-validated accuracy, with DNN presenting the highest cross validation accuracy of 0.93. The life cycle GHG emissions and nutrient releases of crop production exhibited significant variability across Midwest counties in the US. The life cycle global warming and eutrophication impacts of crops (such as corn and soybean) under RCP 8.5 scenario were significantly higher than RCP 4.5 scenario for years 2020-2100. Overall, this study provides the first machine learning models for rapidly predicting life cycle environmental impacts of agricultural production at county scale. Compared with traditional approaches, our machine learning models have greatly advanced computational efficiency (at least 1000 times faster than process-based approaches), and captured the spatial heterogeneity of life cycle environmental impacts.