Combining Multiple Geospatial Data for Estimating Aboveground Biomass in North Carolina Forests

Mapping and quantifying forest inventories are critical for the management and development of forests for natural resource conservation and for the evaluation of the aboveground forest biomass (AGFB) technically available for bioenergy production. The AGFB estimation procedures that rely on traditional, spatially sparse field inventory samples constitute a problem for geographically diverse regions such as the state of North Carolina in the southeastern U.S. We propose an alternative AGFB estimation procedure that combines multiple geospatial data. The procedure uses land cover maps to allocate forested land areas to alternative forest types; uses the light detection and ranging (LiDAR) data to evaluate tree heights; calculates the area-total AGFB using regionand tree-type-specific functions that relate the tree heights to the AGFB. We demonstrate the procedure for a selected North Carolina region, a 2.3 km2 area randomly chosen in Duplin County. The tree diameter functions are statistically estimated based on the Forest Inventory Analysis (FIA) data, and two publicly available, open source land cover maps, Crop Data Layer (CDL) and National Land Cover Database (NLCD), are compared and contrasted as a source of information on the location and typology of forests in the study area. The assessment of the consistency of forestland mapping derived from the CDL and the NLCD data lets us estimate how the disagreement between the two alternative, widely used maps affects the AGFB estimation. The methodology and the results we present are expected to complement and inform large-scale assessments of woody biomass in the region.


Introduction
The mapping and quantifying of forest inventory dynamics are critical for the management and development of forests for natural resource conservation [1,2] as well as for the evaluation of the aboveground forest biomass (AGFB) technically available for bioenergy production [3,4]. Understanding the inventories is also important for the economic assessment of forests as a source of material for wood products. Two major wood products that could be economically competitive for different geographic regions are saw timber and pulp. Forests in the southeastern U.S. have recently attracted additional attention because of the growth in wood pellet production, and the leading role of the region in the world wood-pellet production capacity [3,[5][6][7][8].
Existing assessments of southeastern U.S. forest growth and removal have mostly been based on U.S. Department of Agriculture Forest Service Forest Inventory and Analysis This study proposes a new approach to the estimation of the AGFB that combines the land use information easily available in land use maps with LiDAR data. Our demonstration of the approach highlights the impact of the potential errors present in alternative, widely-available land use data sources, CDL and the NLCD, on the resulting AGFB estimates. For a selected region of the state, a 2.3 km 2 area randomly chosen in Duplin County, we assess the consistency of forestland mapping derived from CDL and the NLCD; overlay the maps derived from both CDL and the NLCD with LiDAR data to derive tree canopy surface and evaluate tree heights; use the FIA data to statistically estimate functions that relate tree heights to tree diameters and to derive the tree height to the AGFB functions, by forest type; estimate the study-area-total AGFB, while explicitly assessing how the disagreement between the NLCD and CDL affects the estimates. The methodology and the results we present are expected to complement and inform large-scale assessments of woody biomass in the region.

Data and Study Area
According to the latest available North Carolina FIA inventory cycle completed in 2013, forests occupied approximately 60% of the state's land between 2007 and 2013 [14]. FIA surveys divide the state into four regions (units), with the percent of area forested varying between 51% in the central, Piedmont unit, and 76% in the western, Mountains unit. The total forested area in the Southern Coastal Plain unit is the closest to the state average; it remained stable at 61% between 1999 and 2013 [14]. For the purpose of the presentation of the AGFB estimation procedure, we randomly selected a 2.3 km 2 study area in that unit. The area is located in Duplin County ( Figure 1). Remote Sens. 2021, 13, x FOR PEER REVIEW 3 alone are insufficient for the assessment of the AGFB, because the amount of biomass depends on the height and diameter of the trees. This study proposes a new approach to the estimation of the AGFB that combine land use information easily available in land use maps with LiDAR data. Our demon tion of the approach highlights the impact of the potential errors present in alterna widely-available land use data sources, CDL and the NLCD, on the resulting AGFB mates. For a selected region of the state, a 2.3 km 2 area randomly chosen in Duplin Cou we assess the consistency of forestland mapping derived from CDL and the NLCD; o lay the maps derived from both CDL and the NLCD with LiDAR data to derive tree opy surface and evaluate tree heights; use the FIA data to statistically estimate func that relate tree heights to tree diameters and to derive the tree height to the AGFB f tions, by forest type; estimate the study-area-total AGFB, while explicitly assessing the disagreement between the NLCD and CDL affects the estimates. The methodo and the results we present are expected to complement and inform large-scale as ments of woody biomass in the region.

Data and Study Area
According to the latest available North Carolina FIA inventory cycle complete 2013, forests occupied approximately 60% of the state's land between 2007 and 2013 FIA surveys divide the state into four regions (units), with the percent of area fore varying between 51% in the central, Piedmont unit, and 76% in the western, Moun unit. The total forested area in the Southern Coastal Plain unit is the closest to the average; it remained stable at 61% between 1999 and 2013 [14]. For the purpose o presentation of the AGFB estimation procedure, we randomly selected a 2.3 km 2 s area in that unit. The area is located in Duplin County ( Figure 1).  (a) The NLCD for the year 2011 was downloaded from the MLRC, 2021 [27]. While the 2016 version of the NLCD is the latest available, it is outside of the range covered by the latest FIA data available for the state; (b) CDL for the year 2011 was downloaded from the USDA, 2021 [26]. We chose the 2011 CDL data to allow for the comparison of CDL and the NLCD in the same year;

Methods
Following convention, we define AGFB as the oven-dry mass of the aboveground portion of a tree, which includes stem wood, stump, bark, top, branches, and foliage [15]. The AGFB estimation procedure we developed and implemented integrates land cover analysis, tree height estimation, and the estimation of region-and forest-type-specific tree diameter functions. The processes involved are summarized in Figure 2.
(a) The NLCD for the year 2011 was downloaded from the MLRC, 2021 [27]. Whi 2016 version of the NLCD is the latest available, it is outside of the range cover the latest FIA data available for the state; (b) CDL for the year 2011 was downloaded from the USDA, 2021 [26]

Methods
Following convention, we define AGFB as the oven-dry mass of the abovegr portion of a tree, which includes stem wood, stump, bark, top, branches, and foliag The AGFB estimation procedure we developed and implemented integrates land analysis, tree height estimation, and the estimation of region-and forest-type-specif diameter functions. The processes involved are summarized in Figure 2.

Land Cover Analysis
Both CDL and the NLCD provide raster-formatted land cover data. Each dataset is classified according to various land cover types, including barren land, water, forest Remote Sens. 2021, 13, 2731 5 of 17 land, and land used for agricultural purposes. Forestland is further broken down into four categories: deciduous forest, evergreen forest, mixed forest, and woody wetlands. The CDL and the NLCD define the four forest categories identically. Deciduous forest is defined as the areas dominated by trees generally greater than 5 m tall, greater than 20% of the total vegetation cover, and more than 75% of the tree species shedding foliage simultaneously in response to seasonal change. Evergreen forests are the areas dominated by trees greater than 5 m tall covering greater than 20% of the total vegetation cover with more than 75% of the tree species maintaining their leaves year-round. Mixed forests are areas with trees greater than 5 m tall and vegetation greater than 20%. Woody wetlands are areas where forest or shrubland vegetation accounts for greater than 20% of the vegetative cover and the soil or substrate is periodically saturated with or covered with water [27]. Because the definitions of the forest categories do not differ between the CDL and the NLCD, in this paper, we use the term CDL/NLCD forest types when referring to the deciduous forest, evergreen forest, mixed forest, and woody wetlands collectively.
In this study, CDL and the NLCD, are processed separately to provide two alternative spatial distributions of forest in the study area. While assessing the accuracy of either land cover map would be desirable, it would require the true land cover data, which we do not possess. Therefore, we only assess the degree to which the NLCD and CDL agree and disagree in identifying specific pixels as forests of alternative types.

Tree Heights and LiDAR Analysis
Remote sensing has played an important role in biomass estimation and mapping across different spatial scales [29]. Numerous research studies have demonstrated the Li-DAR's ability to provide information that is useful for forest measurement, with the specific applications ranging from individual tree assessments to forest structure analysis and forest typing [30][31][32]. LiDAR data provides the direct measurement of three-dimensional structures and the underlying terrain and has been found useful for the detection of changes in forest areas due to both tree removal and growth. The canopy height information combined with stand density or canopy cover has been used for biomass estimation [33,34]. The challenges and performance of the methods to detect and delineate individual trees from LiDAR data are a subject of continuous research [35][36][37]. In this study, we apply a widely used method proposed by [38]. Under this approach, the LiDAR information is used to create the digital representations of two surfaces, ground and canopy, and the local high points in the canopy surface are regarded as representations of treetops. Tree heights are then estimated as the vertical difference between the canopy and the ground surfaces at the location of the treetops.
The ground surface or digital terrain model (DTM) is created from the LiDAR's ground points, while vegetation and ground points are used for the canopy surface creation. For creating both surfaces a spline interpolation method is employed. The spline interpolation estimates the elevation of the surfaces at each point using a mathematical function (piecewise polynomial) that creates a smooth surface that passes exactly through the input points (LiDAR points) and minimizes the overall surface curvature. To obtain the treetop points, we employ the extended maxima transformation of the morphological image-analysis method that uses a local spatial search to identify the local high points in the canopy surface [38]. We apply different treetop identification thresholds recommended by this study, assess the results of each threshold visually at different scales (i.e., descriptive for the whole dataset and local comparisons for cell-by-cell), and select the ones that better match the LiDAR data and/or orthophoto available for the area. Then, a raster-based delineation model is applied to estimate the height of an individual tree as the difference between the Z coordinate of the treetop and the Z value of the DTM at the same X and Y position. Finally, each individual tree is assigned a forest type, i.e., deciduous, evergreen, mixed, or woody wetlands, by overlaying the CDL and the NLCD polygons on top of the canopy surface.

Tree Diameters and AGFB Estimation
We calculate the region-total AGFB as a sum of the AGFB of all of the individual trees in the region. Following the literature, we estimate the AGFB of individual trees as functions of their diameters [39]. Since LiDAR analysis provides the estimates of the heights of the trees, we begin with the estimation of the tree diameters as functions of their heights.
The standard measure of the diameter of a tree is the diameter at breast height (DBH), which is at 4.5 feet above the ground line [15]. The DBH-tree height (H) relationship varies with tree species, stand age, productivity, competition, and geography, and the functional form best describing this relationship is a subject of continuous research [40][41][42][43]. The topic of modeling DBH as a function of H has received renewed attention with the advancement of airborne laser scanning technologies that measure individual tree heights over large forest areas [44][45][46]. While no single DBH-H model is best for all applications, a log-linear functional form has been found to fit observational data and have desirable statistical properties [41,[44][45][46]. Following this literature, we set: where DBH is in cm, H is in m, α and β are the parameters to be estimated, and ε is the standard linear regression error term. The scaling exponent β represents the ratio of the relative growth rate in DBH with respect to the relative growth rate in H [41]. We estimate four Equation (1), each corresponding to a specific CDL/NLCD forest type on a FIA sample matching the study area to the extent possible, as detailed below. Because of the differences between the FIA and the CDL/NLCD data structures, two aspects of selecting the FIA sample data for estimating Equation (1) warrant a discussion: identification of the smallest FIA spatial unit containing the study area, and matching the four CDL/NLCD forest categories with FIA forest types. With respect to spatial units, the locations of the FIA plots on which individual trees have been measured are approximated. To preserve the privacy of landowners/managers, the actual FIA plot location could be different by up to 1.0 mile from the reported one. Additionally, up to 20% of the plot coordinates are swapped with another similar plot within the same county [15]. Thus, Duplin County is the smallest FIA spatial unit that entirely contains the study area. For that reason, in the estimation of Equation (1) we use only the tree measurements for the 2009-2013 FIA plots located in Duplin County.
With respect to forest categories, the FIA plots are assigned to specific forest types based on the tree species or species groups observed as dominating the location [15]. In contrast, the CDL and the NLCD group forests in more general categories, based on whether the trees in the location shed or do not shed leaves seasonally, and on whether the soils are or are not periodically saturated with water. To break the Duplin County FIA sample into the four sub-samples corresponding to the CLD/NLCD forest types we follow a two-step procedure. First, we use the species-defined FIA plot categorization to assign all of the trees on the plot to the deciduous, evergreen, or mixed forest based on whether the dominant species are deciduous, evergreen, or a combination of the two, respectively. For example, FIA plots of forest type 161 (loblolly pine) are dominated by evergreen species, loblolly pine, while FIA plots of forest type 708 (red maple/lowland) are dominated by deciduous species, red maple. Thus, for the estimation of Equation (1), we initially assign all the FIA trees from the plots of forest type 161 to the evergreen forest sub-sample, while the FIA trees from the plots of forest type 708 to the deciduous forest sub-sample.
In step two, we use the FIA plot information to determine whether to move some of the trees from their initial sub-samples to the woody wetlands sub-sample. Two cases are possible: the FIA forest type definition explicitly states that the forests in question typically occur on wet sites, or the definition allows for both wet and dry sites. In the first case, the corresponding tree observations are moved from the initially assigned sub-sample to the woody wetlands sub-sample. For example, the definition of FIA forest type 708 specifies, "Site-generally restricted to very moist to wet sites with poorly drained soils, and on swamp borders" [15]. Thus, for the estimation of Equation (1), we move all of the FIA trees from the plots of forest type 708 to the woody wetlands sub-sample. In the second case, we attribute the plot to woody wetlands if the plot site is reported as hydric (normally abundant or overabundant moisture all year), and leave the plot in the initial sub-sample if the plot site is not hydric, i.e., xeric (normally low or deficient in available moisture) or mesic (normally moderate but adequate available moisture) [15]. For example, the definition of the FIA forest type 161 allows for both hydric and non-hydric conditions: "Sites-upland soils with abundant moisture but good drainage, and on poorly drained depressions" [15]. Therefore, if the site of a FIA plot of this forest type is reported as hydric, we re-assign all of the FIA trees from this plot to the woody wetlands sub-sample. However, we keep the trees in the evergreen sample if the site of the plot is not hydric.
The AGFB of individual trees is estimated based on Jenkins biomass equations [15,38]. The equations are species-group-specific and are given by: where the tree's AGFB is in kg. Since, in our analysis, we do not observe individual tree species within the forest types, we chose the values of γ and δ corresponding to the tree species that define or dominate the specific CDL/NLCD forest type categories. Once Equation (1) is substituted to Equation (2), we obtain the functional relationship between the tree height and the AGFB as: Function (3) are used to calculate the AGFB of each tree measured by LiDAR analysis. The region-total AGFB is calculated as a sum of the estimated AGFB of all of the individual trees in the region. Table 1 summarizes the land cover data for Duplin County, as reported in the 2011 CDL and 2011 NLCD. We find a noticeable mismatch between the two datasets, suggesting some attributed inaccuracy in either or both datasets. Of particular interest for our application, we find that although the difference between the total forested land areas is 13%, the mismatch for the four specific forest categories is much higher. For example, when the entire Duplin County is considered, the evergreen forest area is estimated to be 35% higher, and the mixed forest area is estimated to be 83% lower according to the CDL when compared to the NLCD.

Comparison of CDL and NLCD
In the study area, the difference in the total forest area is relatively smaller than that in the county total, 102.7 ha in the CDL versus 101.0 ha in the NLCD, or 1.7%. The majority of the CDL forest land in the study area, 89%, is marked as forest land in the NLCD ( Table 2). The same is true for the NLCD forestland: 91% of it is marked as forestland in the CDL ( Table 2). The largest share of the mismatch is in shrubland, which is defined as areas dominated by young trees and shrubs that are less than 5 m tall [23]. For the 2.3 km 2 study area, the mismatched forest land occurs mostly at the borders or edges of the forested land areas (Figure 3). While not the focus of this research, the study of edge or boundary effects in the field of GIS, especially at the intersection of raster data, remote sensing, and land cover analysis, is well-documented. Work by [30,[47][48][49] has explored to what extent these boundary effects, essentially spatially-explicit problems, a result of arbitrary or discrete boundaries imposed on spatial data, have on the overall accuracy of land cover-based research. The dark green cells correspond to the forest areas that are identified of the same forest type in both CDL and NLCD. These areas contribute the same amount to the region-total AGFB estimate irrespective of the land cover map used. The light green cells correspond to the areas that are identified as forests in both CDL and NLCD, but of different types. These areas contribute to the region-total AGFB estimate, but the contributions differ between the land cover maps because of the difference in forest type. The yellow cells correspond to the areas that are identified as forest in one and only one of the two datasets. These areas contribute or not to the region-total AGFB estimate depending on which of the cover maps is used. The cells without shading correspond to the areas that are identified as non-forest in both datasets. The non-forest categories are listed in italics. These areas do not contribute to the region-total AGFB estimate irrespective of the land cover map used.
Despite the total forest area estimates being close, there are notable differences between representations of the types of forest in the CDL and the NLCD in the study area ( Figure 3). Remote Sens. 2021, 13, x FOR PEER REVIEW 9 of 17

Tree Heights and LiDAR Analysis
The results of LiDAR analysis are presented in Figure 4. As expected, the finer spatial resolution of the LiDAR data, approximately 0.3-0.5 m as opposed to the 30 m of the CDL and the NLCD, results in the LiDAR data identifying more areas with trees, when compared with either land cover maps. These differences are especially notable when tree cover is sparse, e.g., in the middle right-hand side of the maps.
For the purposes of the present analysis, the comparison of Figures 3 and 4c reveals that differences in the CDL and the NLCD maps lead to some forested areas being included or not in the area-total AGFB estimates (e.g., area marked A in Figure 4c). Another result of the mismatch between the CDL and the NLCD maps is that there are forested areas that have different AGFB estimates because of the differences in the type of forest identified (e.g., area marked B in Figure 4c).

Tree Heights and LiDAR Analysis
The results of LiDAR analysis are presented in Figure 4. As expected, the finer spatial resolution of the LiDAR data, approximately 0.3-0.5 m as opposed to the 30 m of the CDL and the NLCD, results in the LiDAR data identifying more areas with trees, when compared with either land cover maps. These differences are especially notable when tree cover is sparse, e.g., in the middle right-hand side of the maps.
For the purposes of the present analysis, the comparison of Figures 3 and 4c reveals that differences in the CDL and the NLCD maps lead to some forested areas being included or not in the area-total AGFB estimates (e.g., area marked A in Figure 4c). Another result of the mismatch between the CDL and the NLCD maps is that there are forested areas that have different AGFB estimates because of the differences in the type of forest identified (e.g., area marked B in Figure 4c).

Tree Heights and LiDAR Analysis
The results of LiDAR analysis are presented in Figure 4. As expected, the finer spatial resolution of the LiDAR data, approximately 0.3-0.5 m as opposed to the 30 m of the CDL and the NLCD, results in the LiDAR data identifying more areas with trees, when compared with either land cover maps. These differences are especially notable when tree cover is sparse, e.g., in the middle right-hand side of the maps.
For the purposes of the present analysis, the comparison of Figures 3 and 4c reveals that differences in the CDL and the NLCD maps lead to some forested areas being included or not in the area-total AGFB estimates (e.g., area marked A in Figure 4c). Another result of the mismatch between the CDL and the NLCD maps is that there are forested areas that have different AGFB estimates because of the differences in the type of forest identified (e.g., area marked B in Figure 4c).   Tree heights (Figure 4c) are estimated as the vertical difference between the canopy surface ( Figure 4b) and ground surface (Figure 4a) at the location of treetops. Area A contributes to the region-total AGFB estimate if CDL is used, but does not contribute if the NLCD is used. Area B's contribution to the region-total AGFB differs between CDL, which identifies the forest as evergreen and woody wetlands, and the NLCD, which identifies the forest as woody wetlands only. Table 3 summarizes the breakdown of the 2009-2013 FIA sample for Duplin County, the data that we use for estimation of allometric Equation (1). While not exactly matching, the general breakdown of the Duplin County forests is compatible across the three datasets: CDL, NLCD, and FIA as all three datasets show that woody wetlands and evergreen forests dominate the area, followed by mixed and deciduous forests.  Tree heights (Figure 4c) are estimated as the vertical difference between the canopy surface ( Figure 4b) and ground surface (Figure 4a) at the location of treetops. Area A contributes to the region-total AGFB estimate if CDL is used, but does not contribute if the NLCD is used. Area B's contribution to the region-total AGFB differs between CDL, which identifies the forest as evergreen and woody wetlands, and the NLCD, which identifies the forest as woody wetlands only. Table 3 summarizes the breakdown of the 2009-2013 FIA sample for Duplin County, the data that we use for estimation of allometric Equation (1). While not exactly matching, the general breakdown of the Duplin County forests is compatible across the three datasets: CDL, NLCD, and FIA as all three datasets show that woody wetlands and evergreen forests dominate the area, followed by mixed and deciduous forests.  Site-generally restricted to very moist to wet sites with poorly drained soils, and on swamp borders.

No Woody wetlands 23
Total observations 1631 Notes: * Source of information: FIA database manual [15]. For the descriptions and visual references on the tree species, sites, and forest types see [14,50], and www.ncwetlands.org, accessed on 1 June 2021, ** Source of information: FIA data.
The summary of the diameter and height data used in the estimation of Equation (1) is provided in Table 4, and the results of the estimation are reported in Table 5. All four models displayed a good statistical fit, as evidenced by the R 2 ranging from 73% to 84% (Table 5), and all parameter estimates being statistically significant at conventional levels of significance. Table 6 lists the parameter values used for Equation (2), by forest type, along with the original [39] species group names.  Figure 5 shows relationships (3) between the tree heights and the AGFB, which were developed by substituting Equation (1), which expresses tree diameter as a function of tree height, in Equation (2), which expresses the AGFB as a function of tree diameter. Even though Equation (2) does not differ between evergreen and mixed forests, Equation (1) does differ between these two forest types. In consequence, the AGFB function of tree height differs between evergreen and mixed forests. though Equation (2) does not differ between evergreen and mixed forests, Equation (1) does differ between these two forest types. In consequence, the AGFB function of tree height differs between evergreen and mixed forests. The overall spatial distribution of the AGFB ( Figure 6) closely follows the spatial distribution of tree heights (Figure 4), while also reflecting the differences in the location of forests as reported in the CDL and the NLCD (Figure 3). The overall spatial distribution of the AGFB ( Figure 6) closely follows the spatial distribution of tree heights (Figure 4), while also reflecting the differences in the location of forests as reported in the CDL and the NLCD (Figure 3). The overall spatial distribution of the AGFB (Figure 6) closely follows the spatial distribution of tree heights (Figure 4), while also reflecting the differences in the location of forests as reported in the CDL and the NLCD (Figure 3). We calculate the region-total AGFB as a sum of the AGFB of all of the individual trees in the region ( Table 7). The main inputs of the biomass calculation, forest areas, and tree heights, are reported to facilitate the comparison and discussion of the results.

Mixed forest
Woody wetlands We calculate the region-total AGFB as a sum of the AGFB of all of the individual trees in the region ( Table 7). The main inputs of the biomass calculation, forest areas, and tree heights, are reported to facilitate the comparison and discussion of the results.

Discussion
Overall, we find that the two land cover maps agree in the qualitative description of the study region showing that woody wetlands and evergreen forests dominate the area, with mixed and deciduous forests occupying smaller portions. However, quantitatively, the CDL and the NLCD disagree on both the total forested area and on the forest composition. We find that the CDL-based forest area estimate is 13% higher than that based on the NLCD, but the area-total AGFB estimate derived with the NLCD is 22% higher than the CDL-based one. This difference points to the sensitivity of the area-total AGFB estimation to the breakdown of total forest land by type. One contributing factor for the higher NLCDderived biomass estimate is that this land cover product identifies a larger portion of the area as deciduous forest. This, combined with the estimated tree height-AGFB relationship being higher for the deciduous forest when compared with the other three types considered (over a relevant range of tree heights) ( Figure 5), results in higher NLCD-derived AGFB estimates for a relatively large, northeastern portion of the study area ( Figure 3).
In addition to the potential errors in land cover classification, there are three other major sources of potential errors that can impact the AGFB estimates obtainable using our method: errors in the temporal mismatch of the three datasets used in the method (land use datasets, LiDAR, and data that goes in the estimation of DBH-height functions), errors in LiDAR estimates of tree heights, and errors in the DBH-height functions. For the proposed approach to work, LiDAR data would need to be collected at the same time as the land use maps. To minimize the impact of the temporal misalliance uncertainty, we match the years of the data to the extent that is possible, as detailed in Section 2. The specific method of estimation of tree heights from LiDAR data we use was chosen for its proven wide applicability and use. While other methods of tree height estimation might be better suited under specific situations [35][36][37], there is no indication that alternatively derived tree height estimates would have altered our main conclusion about the disagreement in the CDL and the NLCD impacting the area-total AGFB estimates.
We find that the fit of the models (1) quantifying the DBH-height relationship, as measured by R 2 , is on the lower side of the models presented in the literature. The reason for this tendency is that the majority of the models presented in the literature were estimated with a goal that was different from ours, namely, to show how the DBH-height relationship varies with tree species, stand age, productivity, competition, and/or differences in a geographic region [43,[52][53][54][55]. In contrast with this literature, the purpose of the statistical estimation in our case is to only demonstrate that the DBH-height relationship could be reliably estimated from the publicly available FIA data. The data we use for the estimation of each of the four Equation (1) covers multiple species and/or narrowly defined forest types, as detailed in Table 3. This variety of species/forest types leads to greater variation in the tree DBH-height relationship within each of the four sub-samples when compared to the samples containing only a single tree species or a single narrowly defined forest type. Additionally, we model tree DBH as a function of a single independent variable; height. However, the relationship has also been modeled using plot or stand-level covariates to account for the influence of stand density or competition, and with accounting for the spatial correlation between observations [43,54]. Future research could extend our analysis in such a direction, contributing to the improved understanding of how the AGFB varies with tree height for different types of forests Future applications of the biomass estimates obtained using our method could explore the economic potential and the environmental implications of harvesting the AGFB of specific forest types. We found that woody wetlands occupy a significant portion of Duplin County in general and the chosen study area in particular. The soil or substrate in such forests is periodically saturated with or covered with water. This provides an impetus to conduct additional analysis to evaluate what portion, if any, of the woody wetlands could be harvested in a cost-effective and environmentally sustainable manner.

Conclusions
This study introduced a new approach of combining the information available in land use maps and LiDAR data for the estimation of the AGFB and has highlighted the impact of the disagreement in widely available land use maps on AGFB estimates. We use the information from a FIA sample to develop study-region-specific allometric equations, but we are not aiming at improving FIA metrics. We view the proposed approach as a complementary way to obtain the AGFB estimates by combining the types of data (digital land use maps and LiDAR) that have not previously been used for such a purpose. The comparison of the accuracy of the traditional, FIA-based approach and our proposed approach on larger study areas is important but requires the concurrent LiDAR data that covers such larger areas that we do not have. Such a comparison constitutes an intriguing proposal for future research.
The approach we propose is a step towards supplementing sample-based regional AGFB estimates with the estimates based on the population representation of forests via remotely sensed land cover data. Continuing progress with the methodology design of the NLCD and CDL is contributing to the improved accuracy of the land cover data [55]. As the accuracy of the remotely sensed data improves with time, the contribution of the potential land cover data error to the AGFB estimation approach proposed is expected to decline.
With the advances in the new remote sensing technology and the availability of high spatial and temporal resolution of land use geospatial data, such combined use of multiple data sources opens up new possibilities for the proposed methodology. Both alternative