The papers in this special issue all agree that dealing with uncertainty is not a quick exercise but rather involves a commitment that is painstaking and long-term. Proper treatment of uncertainty is costly in terms of both time and effort because it forces us to take the step from “simple” to “complex” in order to grasp a wider and more holistic system view. Only after we have taken that step, can we consider simplifications that may be warranted.
To facilitate visualization of the 13 papers within a system context, we group them using a matrix (cf. Table
1). Vertically, the matrix relates to the IPCC’s classification of sectors of emissions and removals (see the first two columns from the left); while horizontally, it lists the systemic features that are most pertinent to the papers (see the first three rows from the top). All papers refer to CO
2 unless indicated otherwise (by “CO
2-eq”). Two groups of papers are distinguished according to whether they follow a bottom-up accounting approach in addressing uncertainty (Group I) or a bottom-up/top-down one (Group II). Five of the 13 papers are content-specific and are clustered under Group III; they provide beyond-inventory or methodological support to improve our understanding and handling of uncertainty (see the footnotes below Table
1).
Table 1
Visualization of the scientific context of the 13 papers in this special issue. All papers refer to CO2 unless indicated otherwise by “CO2-eq”
3.1 Group I: Bottom-up accounting (high resolution or gridded)
The papers by Bun et al. (
2018), Charkovska et al. (
2018,
2019), Danylo et al. (
2019), and Hogue et al. (
2017) address bottom-up accounting, high-resolution or gridded. They contribute to key issue 1 (one-sided perspective), contribute substantially to key issue 3, and advance key issue 4.
Two papers are close in content to this group, although listed elsewhere. The paper by Verstraete (
2018)—listed under Group III (beyond-inventory or methodological support) below—provides particular support for this group of authors by presenting a novel approach to regridding data. The paper by Oda et al. (
2019)—listed under Group II (bottom-up/top-down accounting) below—makes use of the research reported by Bun et al. (
2018).
The research of the authors in Group I is motivated by the desire to achieve a better territorial overview of where emissions appear locally. An increasing number of practical applications require knowledge of where emissions occur at increasingly smaller spatial scales.
However, high-resolution inventories come with much greater uncertainty. The reasons are discussed by Bun et al. (
2018) and Hogue et al. (
2017). There is uncertainty in the geolocation of emission sources/sinks; uncertainty resulting from the aggregation of statistical data; uncertainty underlying proxy and geolocational data; uncertainty regarding how proxy data, in the end, are represented; and last, but not least, uncertainty associated with the choice of emission factor.
The authors analyze the increase in uncertainty and develop methods to reduce it. The two approaches followed are (i) tracing emissions by source and estimating gridded totals (Bun et al.
2018; Charkovska et al.
2018,
2019; Danylo et al.
2019); and (ii) quantifying the uncertainty of gridded emissions as a function of grid size (Hogue et al.
2017).
The group of authors following the first approach (namely under [i]) does not start from a regular grid. Instead, emission (and removal) processes in all categories of human activity, as specified by the IPCC Guidelines for National Greenhouse Gas Inventories (IPCC
2006), are analyzed at the level of sources (and sinks). These are classified as point, line, or areal sources and according to intensity and physical size (with respect to the territory under investigation). The resulting geospatial database contains information about the administrative assignment of each emission source as a vector map object.
The emissions from very diverse emission sources can be combined into a grid, allowing total emissions to be calculated, while the grid size can be chosen arbitrarily depending on analysis and visualization needs. This offers a unique opportunity to calculate total emissions for different levels of administrative unit (settlements, municipalities, districts, provinces) without loss of accuracy, as well as separately by emission category, greenhouse gas, type of fossil fuel, etc. The approach allows uncertainty to be considerably reduced for high-resolution inventories. Starting by assembling statistical data from the lowest available administrative level (ideally municipality) limits disaggregation depth and errors. This is in contrast to handling uncertainty by way of disaggregation, which is applied, for example, in gridded approaches. Here, uncertainty is determined by disaggregation depth and increases with it, when going from large to smaller scales. Note that this important difference is preserved even if emissions estimated at, or aggregated to larger (e.g., national) scales closely agree.
Poland serves as a joint case study. Emissions from various sources are calculated both for grids and administrative units. The results demonstrate the considerable unevenness of spatial distributions of GHG emissions. Distributions and their uncertainty ranges are estimated by applying a Monte Carlo method. Bun et al. (
2018) describe the approach in general; Charkovska et al. (
2019) focus on emissions from industrial processes, Danylo et al. (
2019) on emissions from the residential sector, and Charkovska et al. (
2018) on emissions from agriculture.
The authors following the second approach (namely [ii]) look, in particular, at the uncertainty associated with the allocation of point sources. Any misallocation of these emission sources can have important consequences for high-resolution inventories, especially if their emission intensities are high. Such misallocations happen, for example, when databases are combined to merge information using logical rules. Hogue et al. (
2017) analyze how the misallocation of point emission sources impacts inventory uncertainty as a function of grid size. The authors use population density as a proxy to distribute emissions spatially across grids that vary in size. They find that relative uncertainty (total uncertainty divided by total emissions) at grid-cell level decreases with increasingly coarser resolution. In most cases, relative uncertainty also decreases with increasing emissions from point sources. The authors’ results indicate that good data of large point sources are particularly crucial for obtaining reliable, spatially explicit emission inventories.
3.2 Group II: Bottom-up/top-down accounting (high resolution or gridded)
The papers by Lesiv et al. (
2018), Oda et al. (
2019), and Zimnoch et al. (
2018) explore the use of additional, ideally independent, observations as top-down constraints, such as atmospheric measurements and remote sensing, to identify and close potential inaccuracies in bottom-up inventories. Although, all three papers exercise bottom-up/top-down, high-resolution, or gridded, at subglobal scales, they show potential to expand to global scales—which is why, arguably, there are still some degrees of freedom in closing (verifying) bottom-up and top-down. These authors’ research contributes substantially to key issue 1 (two-sided perspective) as well as to key issue 5.
Lesiv et al. (
2018) come up with a “verified” account of carbon in forest ecosystems over larger areas, in this case for Ukraine. The authors present a forest map for 2010 with a spatial resolution as high as 60 m, which is needed to capture Ukraine’s highly fragmented forest landscape. The forest map contains information about dominant tree species, total biomass, and net primary production (NPP). Together with forest inventory statistics and forest-related data collected by applying Geo-Wiki (
https://www.geo-wiki.org/) online, it allows Ukraine’s net carbon flux for 2010 to be determined. To constrain that carbon flux, both a flux-based and a stock-based method are applied. The two methods indicate that Ukraine’s forest serves as a net carbon sink in the range of 11.0 ± 1.4 (stock-based method) to 11.8 ± 3.2 Tg C y
−1 (flux-based method), or 11.4 ± 1.7 Tg C y
−1 (equivalent to 131 ± 20 g C m
−2 y
−1) on average. This sink differs by bioclimatic zone, ranging from 55 in Polissya to 197 g C m
−2 y
−1 in the Carpathians. Above and beyond determining uncertainty, an additional value of the authors’ study is their demonstration of how to compensate for missing knowledge in the accounting of forest ecosystems by proceeding in a spatially explicit manner.
Oda et al. (
2019) compare ODIAC (Open-source Data Inventory for Anthropogenic CO
2) with GESAPU, a high-resolution, spatially explicit emission inventory—here, the one provided by Bun et al. (
2018) for Poland. ODIAC is itself a global inventory with a spatial resolution of 1 km × 1 km, based on the disaggregation of the national annual fossil-fuel CO
2 emission estimates provided by the Carbon Dioxide Information Analysis Center. To achieve that high spatial resolution, ODIAC uses point source information (source points’ geographical location and CO
2 emissions) and satellite nightlight (radiance) data. Because of its greater local “realism”, GESAPU is used as a reference in this comparison. The difference between the two inventories is understood to serve as a proxy for errors and uncertainties associated with ODIAC. This difference is small for total emission estimates of countries (2.2%), point sources (0.1%), and non-point sources (4.5%). However, it increases toward smaller spatial scales, indicating that disaggregation error and uncertainty increase. Oda et al. find a difference (relative at the pixel level) of typically about 30% for urban areas, up to 90–100% for urban-rural transition areas, and 10% for remote areas. The difference decreases with increasing spatial aggregation by approximately 70% for spatial scales, which are typical for global and regional transport models (50 km and greater). Based on their findings for Poland, the authors envisage using ODIAC globally to support monitoring verification and even at subnational levels—it is not unusual for countries to run emission inventories at the state or provincial levels while reporting only national emissions to the UNFCCC. However, as noted by the authors, such a request would need to accompany concerted global actions, ranging from the collection and reporting of data, through monitoring, to international governance.
Zimnoch et al. (
2018) focus on top-down estimation of both CO
2 and CH
4 emissions from the urban area of Krakow, Poland. They present a set of methods based on atmospheric observations of CO
2 and CH
4 mixing ratios and their isotopic composition, the use of additional data relating to the atmospheric concentration of radon and mixing layer height, and atmospheric modeling, to identify and quantify urban emissions. These methods complement each other; they allow a determination not only of the contribution of different emission sources to the total atmospheric load but also of the fluxes of those gases. The methods provide an efficient way of quantifying the surface emissions of major GHGs from distributed sources and thus represent a complementary approach to accounting emission bottom-up. The authors’ approach offers an alternative to validate the effectiveness of potential climate change mitigation strategies at scales of great interest to policy actors, demonstrated here for complex local urban-scale environments.
3.3 Group III. Beyond-inventory or methodological support
The papers by La Notte et al. (
2018), Verstraete (
2018), Gusti et al. (
2018), Jarnicka and Żebrowski (
2019), and Jonas and Żebrowski (
2018) provide beyond-inventory or methodological support on aspects of uncertainty that have not been addressed to date, have been overlooked, or have emerged over time. The research described in these papers contributes equally to key issues 1 (one-sided perspective), 3, and 4.
La Notte et al. (
2018) focus on combining, at the regional level, economic accounts with environmental data on atmospheric emissions (GHGs and air pollutants), with the atmospheric emissions combined into groups and expressed in terms of various potentials: their global warming potential (GWP), their potential acid equivalent (PAE), and their tropospheric ozone formation potential (TOPF). By considering uncertainty in an on–off mode (instantaneous learning), policy recommendations at regional and provincial levels can be made.
Verstraete (
2018) proposes a new method that does not lead to an increase in uncertainty during the process of overlaying data sets mapped to different grids. This so-called regridding process is an important preprocessing tool in handling spatially resolved datasets, offering considerable potential, particularly for authors compiling high-resolution spatial inventories (cf. Group I). Such researchers frequently face the problem of having to rely on data represented using different grids (e.g., proxy data). These can differ in terms of cell size, or they can be displaced latitudinally and/or longitudinally or even be rotated relative to each other; and grids can also be irregular. The approach offered by Verstraete can be used for remapping, for example, a grid onto administrative borders (or vice versa). Fuzzy rule-based methods are elaborated and tested for regridding using additionally available knowledge in order to obtain better results, particularly during spatial disaggregation processes.
Gusti et al. (
2018) address the uncertainty underlying marginal abatement cost curves (MACCs) derived for the LULUCF sector, by studying their sensitivity to uncertainty in the price of agricultural land and forestry commodities and to uncertainty in the quality of governance—stability, effectiveness, and assertiveness (simplified and summarized by means of a so-called “corruption” coefficient)—in the regions studied, ranging from individual countries to the global level. The uncertainty in MACCs, if not considered, may crucially influence the trade of emission permits or undermine decisions based on MACCs to mitigate GHG emissions. The authors’ results indicate that MACCs are especially more sensitive to the quality of governance than to the price of agricultural land. MACCs appear more robust for high CO
2 prices, while they are more sensitive to the variation in these parameters for low CO
2 prices. The authors conclude that considering the quality of governance is key if medium-term mitigation policies, usually designed for low CO
2 prices, are developed.
Jarnicka and Żebrowski (
2019) quantify (historical or diagnostic) learning—defined as the decrease in uncertainty (inaccuracy and imprecision) in the estimates of GHG emissions with the focus here on country CO
2 emission totals (excluding LULUCF emissions)—reported in national inventory reports. The authors demonstrate how knowledge of a change in uncertainty can be gained from analyzing annually revised emission estimates in retrospect. In cases of pronounced learning, the reduction in uncertainty can be well described by the coefficient in an exponential model. The authors’ results allow the conclusion to be drawn that continuous efforts (i.e., time series of 20 years and longer of continuously revised emissions) are necessary to determine a half-time of this reduction (of the order of 5 years for the EU15) that is sufficiently robust. Their approach goes beyond that favored by the IPCC of estimating and monitoring uncertainty to help prioritize efforts to improve the accuracy of inventories and guide decisions on methodological choice (IPCC
2000: Chapter 6). Understanding what it takes to decrease uncertainty over time is crucial, on the one hand for evaluating the quality of compliance under which countries meet their emission reduction targets and, on the other, for setting future emission reduction targets more skillfully, that is, from an emission change-versus-uncertainty perspective rather than from an emissions change-only perspective.
Jonas and Żebrowski (
2018) focus on the memory and persistence of forced, causally linked systems—such as population, GHG emissions, atmospheric concentrations, and average surface air temperature—with GHG emissions into the atmosphere serving as their case in point (although the authors still prefer working with synthetic data). Memory allows reference to be made to how strongly the past can influence the “near-term future” of the system, its so-called explainable outreach. In contrast to Jarnicka and Żebrowski, the mode of learning underlying these authors’ approach to analyzing data in retrospect could be termed “diagnostic learning under controlled prognostic conditions”. (Data are subdivided for testing and learning.) In light of the continued increase in emissions globally vis-à-vis, the reductions urgently needed until 2050 and beyond, the authors conjecture that, being ignorant of memory and persistence, the “inertia” with which global GHG emissions will continue on their increasing path beyond today is underestimated and thus, that the amount of reduction achievable in the future is overestimated. This inertia is initially caused by human behavior (leading to emissions of GHGs into the atmosphere), but it will become increasingly “geophysical” the more Earth processes (e.g., the manifestation of emissions as concentrations of GHGs in the atmosphere) are disturbed. The authors anticipate that persistence is a powerful system characteristic, by which they mean that the system’s explainable outreach appears determinable even under incomplete knowledge of memory and imperfect understanding of how the system is forced.