Chapter 3 Methods
3.1 The Coastal Carbon Library
The Coastal Carbon Library is a synthetic dataset built from multiple sources. It can be thought of as a centrally located hub for global Blue Carbon soils data collected from many repositories worldwide. These disparate data come in many formats, and are processed with a transparent workflow that shapes them into a common format before they are deposited in a repository as composited files that can be accessed through a single portal. The data is hierarchical and highly disaggregated, with separate metadata, notes, observations, and measurements for studies, sites, cores, and depth increments. All data within the Coastal Carbon Library is traceable to its original source, which is a citable open source data release, attributable to its original author. In addition to synthesis, the Coastal Carbon Library workflow involves some post-processing, including: automated data quality control, version control and variable name enforcement, the addition of standardized, geography (country and tier-I geography units [states and provinces]) and habitat classifications (marsh, mangrove, swamp, shrub/scrub, seagrass, algal mat, and unvegetated), and the addition of data utility, availability and quality tiers and sub-tiers.
Carbon stocks data are needed for quantifying emissions from wetland loss events or avoided emissions from wetland preservation activities; carbon burial rate data are needed for establishing baselines and projecting removals for restoration projects. For stocks data the highest quality data include cores that contain entire marsh profiles, rather than just shallow samples. For carbon burial rates the highest quality data include full reported age-depth information with associated elevation data needed for fitting models as in Shile et al. (2014).
For data utility, availability, and quality tiers and sub-tiers, data are first classified by what types of analysis they can be used for (carbon stock data, carbon burial rate calculation, and modeling), then what higher level use cases they are appropriate for (full-profile stock assessment, uncertainty propagation, etc.). If dry bulk density and either organic matter or organic carbon are present in depth series, then the core meets the minimal inclusion criteria for carbon stocks (C). If the core is confirmed to include all soil horizons and reach the bottom of the soil profile then it is considered a high quality core (C1 - best). If not, it is classified as lower utility (C2 - good). If any profile age-related or disaggregated data associated with these techniques is present in a core, then the core meets the minimal inclusion criteria for calculating carbon burial rates, aka carbon accretion rates (B). If enough data is provided where derived age-depth models could be replicated by an independent investigator, and error propagated, then data are coded B1 (best). Any missing radioisotope errors or conditional attributes result in the dataset being classified as having incomplete date reporting (B2 - good). Assessments of future carbon sequestration and marsh elevation response to sea level rise require process models that can be fit using dated soil cores that carry precise elevation data (Schile et al. 2014; K. Thorne et al. 2018). Cores with any elevation data meet the minimum inclusion criteria for carbon sequestration modeling (A). Elevation data come in varying quality levels, so we differentiate between elevations that are interpolated from remotely sensed or spatially interpolated digital elevation models (A3 - good) and those that are directly measured using precise real time kinematic GPS data or better (A2 and A1; better and best), depending on the completeness of the age-depth model reporting (Table 3.1).
| Quality Tier | Quality Sub-Tier | Definition |
|---|---|---|
| Carbon Stock | C2 - Good | Carbon stock data complete not confirmed to be a complete profile |
| Carbon Stock | C1 - Best | Carbon stock data complete confirmed to be a complete profile |
| Age-depth | B2 - Good | Dating information present but not complete |
| Age-depth | B1 - Best | Dating information present and complete |
| Elevation | A3 - Good | Elevation data present but of low quality dating info present |
| Elevation | A2 - Better | Elevation data is high quality but dating info present but incomplete |
| Elevation | A1 - Best | Elevation data is high quality and dating info is complete |
| NA | Not available | Data is not available for the core at a given quality tier |
The version of the Coastal Carbon Library used in this analysis was current as of 5 March 2021.
3.2 State-Level Comparison Metrics
In order to help scope efforts at the state level for this project we developed a new system for quantifying four distinct attributes of each state’s Blue Carbon data: data quantity, data quality, spatial representativeness, and habitat representativeness. We also provide the composite ranks across all four of these metrics.
All of these metrics required mapping of total area, extent, and areas of various Blue Carbon habitats. We calculated tidal wetland area for each state by layering two products, the National Wetlands Inventory, and the Coastal-Change Analysis Program (C-CAP, 2006-2011) Landcover Change Map (Dobson et al. 1995). The National Wetlands Inventory (NWI) contains detailed information coded for wetland and water units including tidal status and whether or not it is impounded. The Coastal Change Analysis Program has more detailed land cover classifications, including non-wetland systems, changes tracked between time steps, and an accompanying accuracy assessment, so areas of different classes can be statistically estimated from the mapped area, taking into account the fact that some categories are under-mapped and some are over-mapped.
For our state-level tidal wetland habitats we included any wetlands that C-CAP mapped as Estuarine (a.k.a. Brackish to Saline) Emergent, Estuarine Scrub/Shrub, or Estuarine Forested, as well as wetlands classified as Palustrine (a.k.a. Freshwater) Emergent, Palustrine Scrub/Shrub, and Palustrine Forested provided the wetland intersected waters mapped as tidal by NWI. We also included any C-CAP areas classified as Estuarine Aquatic Bed or Palustrine Aquatic Bed if the intersected tidal waters; these included seagrasses, algal mats, and kelp beds.
We estimated areas taking into account the magnitude of inclusion and exclusion errors (Olofsson et al. 2014) for C-CAP using data from an independent accuracy assessment (McCombs et al. 2016) as in Holmquist et al. (2018).
3.2.1 Data Quantity
The Data Quantity metric is the total soil core count normalized by the total estimated tidal wetland area for each state, expressed as cores per 1,000 hectares (ha).
3.2.2 Data Quality
The Data Quality metric is similar to the Data Quantity metric, but it only includes a subset of those cores that are most valuable for coastal greenhouse gas inventorying purposes. These include cores that are confirmed to represent an entire wetland soil profile (C1). It includes cores with dated stratigraphy and enough detailed information so that important derived variables like carbon accumulation rate and vertical accretion rate can be replicated (B1). It also contains cores with precise elevation data, which can be used to fit models that forecast wetland elevation and carbon stock response to sea-level rise (A2 and A1). Similar to the Data Quantity, the Data Quality Metric is expressed as the number of high-value cores per 1,000 ha.
3.2.3 Spatial Representativeness
The Spatial Representativeness metric represents how well the spatial distribution of cores matches the spatial distribution of wetlands in a state. We calculate this spatial representativeness metric (\(R_{s}\)) as the mean distance between cores (\(\mu_{d,c}\)) divided by the mean distance between wetlands (\(\mu_{d,c}\)) seen in Eq. (3.1).
\[\begin{equation} R_{s} = {\mu_{d,c}\over{\mu_{d,w}}} \tag{3.1} \end{equation}\]We calculate the Euclidean distance between cores by converting all latitudes and longitudes of cores to an equal area projection and calculating a distance matrix, using the sf package in R. We filtered the core locations by state and took the mean of the upper triangle of the matrix, to eliminate 0’s of the matrix diagonal, and the doubled entries of the lower triangle. We approximate the mean distance between watersheds in the same way but using the centroids of intermediate-scale watershed units (HUC8’s), containing tidal wetland units according to the NWI. According to this metric a state with a few sites would score lower (Figure 3.1 A), and a state in which cores were evenly spread out (Figure 3.1 B) would score higher and closer to 1.
![Example of how the spatial representativeness metric was calculated in states with different patterns (California [A] and Louisiana [B]). Red points are core locations, light grey shapes are intermediate watershed units (Hydrologic Unit Code 8 [HUC8]). Black dots show the centroids of HUC8’s we used to approximate mean wetlands distance. California has clustered sampling (A), and Louisiana has highly dispersed sampling (B).](figures/Map_Spatial_Rep.jpg)
Figure 3.1: Example of how the spatial representativeness metric was calculated in states with different patterns (California [A] and Louisiana [B]). Red points are core locations, light grey shapes are intermediate watershed units (Hydrologic Unit Code 8 [HUC8]). Black dots show the centroids of HUC8’s we used to approximate mean wetlands distance. California has clustered sampling (A), and Louisiana has highly dispersed sampling (B).
This metric is informed by Tobler’s (1970) first law of geography, which states that “everything is related to everything else, but near things are more related than distant things”. We know from previous meta-analyses that relative sea-level rise is likely a major driver of carbon stocks (Rogers et al. 2019) and carbon accumulation rates (Wang et al. 2019). Relative sea-level rise has strong spatial and especially latitudinal components, affecting marsh function (James R. Holmquist, Brown, and MacDonald 2021). Other drivers that could potentially have mechanistic implications for carbon burial include latitudinal relationships between temperature and belowground productivity (Kirwan, Guntenspergen, and Morris 2009) and temperature and organic matter stabilization (Mueller et al. 2018). If these latitudinal drivers are the dominant drivers of marsh processes, then a sampling strategy maximizing geographic coverage would be optimal. Upon visual inspection, it appears that this metric meets the goal of quantifying the difference between states with geographically clustered sampling and states with evenly distributed sampling. In the case of the California coast the latitudinal gradient, and the correlated relative sea-level rise gradient are not well sampled, where in Louisiana they are (Figure 3.1).
We do not have prior information on how much localized factors affect carbon at large spatial scales, but more localized studies suggest tidal elevation (John C. Callaway et al. 2012b; E. K. Peck, Wheatcroft, and Brophy 2020), salinity (Stagg et al. 2018) and plant community (D. R. Vaughn et al. 2020) can also affect carbon sequestration. As far as we are aware, there has not been a meta-analysis of marsh carbon storage including these drivers as of yet. But, if these localized drivers did account for more variance compared to latitudinal drivers, then a future version of this metric could score states on how well sampling represents a multi-dimensional ‘driver space’ rather than simply how geographically representative sampling is. In this hypothetical example states that more closely followed a stratification by elevation, salinity zone, and/or plant functional type. Until we have a more authoritative meta-analysis of carbon accumulation and both broad geographic and local drivers, we default to Tobler’s law, and assume geographic coverage is the superior sampling strategy.
3.2.4 Habitat Representativeness
The Habitat Representativeness metric ranks states based on how well the habitats sampled by a state’s soil core data inventory matches the proportions of the estimated areas of these habitats. States in which the habitats have been sampled in proportion to their Blue Carbon habitat area rank higher than those where habitats are over or under-sampled.
We calculated habitat representativeness (\(R_h\)) as the Euclidean distance between the proportion of area estimated from maps (\(p_a\)) and the proportion cored (\(p_c\)) for each class (k) of N classes seen in Eq. (3.2). We subtracted this distance from 1. Because 0 represents a perfectly balanced sampling, and values approach 1 as they become less balanced, and we wanted this metric to rank states from better to worse using high to low values in order to be consistent with the other metrics.
\[\begin{equation} R_{h} = 1 - \sqrt{\sum_{k=1}^{N}(p_{a,k}-p_{c,k})^2} \tag{3.2} \end{equation}\]Habitats in our coring database include marshes, scrub/shrub wetlands, swamps, mangroves, seagrass, algal mats, and unvegetated surfaces. The maps that we use to estimate these areas are based on the coastal change analysis program (C-CAP) which does not distinguish between seagrass, algal mats and kelp beds. Since kelp beds are not an in situ carbon sink the way sea grasses are, and are not included in our soils data set, some caution should be used to not over-interpret low scores in this metric for states with extensive kelp beds, such as California.
3.2.5 Composite Ranking
In addition to these four metrics we calculated a Total Score as the average ranking across these four metrics. We visualize the Total Score, as well as the four contributing metrics in a State-Level Blue Carbon Data Report Card, showing which states rank higher and lower. We should note that we are only comparing the Quality and Quantity, and Spatial and Habitat Representativeness of data within states to each other and not to any idealized sampling structure (i.e. grading on a curve).
3.2.6 Area and Soil Carbon Stock Estimates
For each state we present the estimated areas for tidal marshes, mangroves, swamps, and segarasses based on the intersection of C-CAP and NWI tidal areas maps as in Holmquist et al (2018). For marshes, mangroves, swamps, and scrub/shrub wetlands we also present a carbon stock estimate based on an assumption of 1 m soil depth and 27.0 kgC m-3 (James R. Holmquist, Windham-Myers, Bliss, et al. 2018).