Support for Subnational Entities to Develop and Monitor Land-Based Greenhouse Gas Reduction Activities

Abstract

Land managers across the United States (U.S.) are developing plans to mitigate climate change. Effective implementation and monitoring of these climate action plans require standardized methods and timely, accurate geospatial data at appropriate resolutions. Despite the abundance of geospatial and statistical data in the U.S., a significant gap remains in translating these data into actionable insights. To address this gap, we developed the Land Emissions and Removals Navigator (LEARN), an online tool that automates subnational greenhouse gas (GHG) inventories of forests and trees in nonforest lands using a standardized analytical framework consistent with national and international guidelines. LEARN integrates multiple datasets to calculate land cover and tree canopy changes, delineate areas of forest disturbance, and estimate carbon emissions and removals. To demonstrate the application of LEARN, this paper presents case studies in Jefferson County, Washington; Montgomery County, Maryland; and federally owned forests across the conterminous U.S. Our results highlight LEARN’s capacity to provide localized insights into carbon dynamics, enabling subnational entities to develop tailored climate strategies. By enhancing accessibility to standardized data, LEARN empowers community land managers to more effectively mitigate climate change. Future developments aim to expand LEARN’s scope to cover nonforest landscapes and incorporate additional decision-making functionalities.

1. Introduction

The land sector plays a pivotal role in global greenhouse gas (GHG) mitigation efforts. In 2022, the Land Use, Land-Use Change, and Forestry (LULUCF) sector in the United States (U.S.) had a net increase in carbon stocks of 921.8 million megagrams of carbon dioxide equivalent (million Mg CO₂ eq.), effectively offsetting 14.5% of the nation’s total GHG emissions [1]. While forests play an outsized role in sequestering carbon from the atmosphere, they also cause significant emissions of greenhouse gasses when disturbed, burned, or converted to nonforest land use. Trees on nonforest lands also absorb atmospheric carbon dioxide while providing other ecosystem services, such as reducing urban heat [2] and regulating hydrologic systems [3].

While national GHG inventories offer valuable insights for guiding land management priorities, there is an increasing need for subnational and local GHG data to inform resource management and climate mitigation plans. Subnational entities, including state and local governments as well as private institutions, are instrumental in implementing land-use policies and practices that directly impact carbon emissions and sequestration [4].

However, lack of access to detailed, localized data hinders their ability to make informed decisions tailored to regional conditions. The U.S. Department of Agriculture (USDA) Forest Service (FS) Forest Inventory and Analysis (FIA) Program maintains one of the world’s most robust and comprehensive national forest inventory databases. Despite the wealth of data available, FIA data are often underutilized due to their complexity, scale, and lack of corresponding and accessible geospatial information. While some researchers are producing spatial datasets derived from FIA plot information [5], these datasets are challenging to apply to GHG inventories at local scales without specialized expertise.

Thus, there is a pressing need for standardized geospatial and statistical data that are timely, accurate, and consistent and with sufficient resolution across jurisdictions and scales. Such standardized data can enable monitoring and reporting and facilitate collaboration among different stakeholders. Furthermore, keeping stock of carbon emissions and removals separately rather than combining them to track only the net GHG flux can elucidate key drivers of change and highlight specific areas of improvement [6,7,8].

The Land Emissions and Removals Navigator (LEARN) is a free online tool and open-source codebase that performs automated GHG inventories of forests and trees at any scale within the conterminous U.S. from 2001 to 2023. LEARN is fully operational for use across the conterminous United States, with typical applications demonstrated through the case studies presented here (see Discussion for detailed considerations on appropriate scale and application). Built upon land use, forest inventory, tree cover, and disturbance data from U.S. Department of Agriculture USDA FS, the U.S. Geological Survey (USGS), and other federal agencies, LEARN provides a user-friendly platform for generating localized GHG inventories using nationally consistent data and a standardized approach consistent with USDA and Intergovernmental Panel on Climate Change (IPCC) guidelines [9,10].

This paper builds upon the extensive peer-reviewed methodology with an emphasis on the multiple spatial and statistical data sources underpinning LEARN. The primary goal of this paper is to provide a detailed, expanded, and updated methodological description of the LEARN tool, clearly outlining how LEARN integrates various datasets to estimate greenhouse gas emissions and removals at subnational scales. Additionally, we present two case studies from representative communities and one from federally owned forests to illustrate, contextualize, and analyze LEARN’s application across varied geographic and administrative contexts. By increasing accessibility to standardized geospatial data and statistical analysis, LEARN aims to empower subnational entities and communities to participate more effectively in long-term climate mitigation planning and implementation.

2. Methods

2.1. Conceptual Framework

LEARN’s conceptual framework applies standard USDA and international analytical methods designed to be fit-for-purpose for communities while preserving the detail necessary to provide useful data for informing local climate change mitigation policies. Methods applied in LEARN were reviewed and published originally as an appendix to the ICLEI–Local Governments for Sustainability (ICLEI) U.S. Community Protocol [11] and subsequently expanded for international applicability through the publication of a standalone supplement to the Greenhouse Gas Protocol’s Global Protocol for Community-Scale Greenhouse Gas Inventories [12].

The methods are also aligned with the 2024 update to USDA’s entity-scale GHG inventory guidelines [9]. A technical note [13] documented the data inputs, methods, and processes behind version 1.1 of the LEARN tool. That document, though not peer-reviewed in a journal context, underwent extensive internal and external stakeholder review and is referenced here for transparency, reproducibility, and ease of access to methodological details not fully captured within the constraints of this manuscript.

All guidance documents upon which LEARN is based have been extensively reviewed by experts and follow the IPCC Guidelines for National Greenhouse Gas Inventories [10] both in the equations used and in the way land-related GHGs are conceptualized. Of the two equally valid methods offered in the IPCC Guidelines for estimating net CO₂ fluxes from land (stock change and gain–loss), LEARN applies the gain–loss approach. The gain–loss method was chosen for LEARN because few communities have sufficient plot data available to use the stock-change approach. Furthermore, the gain–loss approach allows for the estimation of emissions and removals separately, which is helpful for designing policy interventions. While the term “removals” in forest inventories can sometimes refer to activities like timber harvesting, in this paper, “removals” specifically refers to the process by which trees and plants absorb carbon dioxide from the atmosphere and store it as carbon in biomass and soils—a process also known as carbon sequestration.

The approach summarized below represents an application of the “gain-loss” method described by IPCC and often used in national GHG inventories. Land use and land-use changes have associated emissions (GHGs lost from lands) and removals (carbon accumulated into ecosystem vegetation and soils). Total emissions and removals are calculated by multiplying activity data by the appropriate emission or removal factors:

$$\mathrm{Net}\mathrm{GHG}\mathrm{Flux}=\mathrm{GHG}\mathrm{emissions}+\mathrm{GHG}\mathrm{removals}$$

(1)

where

$$\mathrm{GHG}\mathrm{emissions}=\sum (\mathrm{AD}\times {\mathrm{EF}}_{land})$$

(2)

$$\mathrm{GHG}\mathrm{removals}=\sum (\mathrm{AD}\times {\mathrm{RF}}_{land})$$

(3)

where activity data (AD) describes an area of categorized land use or land-use change and emission factors (EFs) or removal factors (RFs) describe the annual change in CO₂e per area unit. The complete description of the IPCC gain–loss method is presented in the annex to Appendix J [11]. The methods described here directly support IPCC Tier 2 and Tier 3 approaches, as detailed in Appendix B.1.

2.2. Data Sources and Applications

LEARN integrates a range of spatial and tabular datasets to perform inventory analyses, summarized in Table 1 according to their specific applications: land-cover change detection, disturbance identification, and derivation of emission and removal factors. LEARN primarily utilizes datasets at a 30 m resolution, making it most suitable for subnational analyses such as counties, municipalities, and regional landscapes. Although technically possible, applying LEARN at significantly finer (e.g., parcel-level) or broader scales (e.g., multi-state regions) may introduce uncertainties related to dataset resolution limitations and aggregation effects (see Section 4.3 for further details).

Most datasets integrated within LEARN are national in scope and applicable to subnational analyses across the conterminous U.S. A global dataset [14] is specifically used to delineate harvest because no comparable national dataset is available at suitable resolution to capture harvest disturbances comprehensively on both public and private lands. These datasets are integrated using a Python-based (Version 3.12) geospatial modeling framework to quantify land cover and tree canopy changes, identify forest disturbances, and derive accurate carbon emission and removal factors. The methods outlined below summarize the general approach for generating and integrating activity data and carbon factors. Detailed geoprocessing steps and code documentation are publicly available in the accompanying GitHub (Version 3.16) repository (see Data Availability Statement).

Table 1. Data sources for calculation of GHG inventories in LEARN. For more detailed descriptions, see Glen et al., 2024 [13]. Variable categories are in bold, while variable sub-categories are in italics.

Variable	Dataset	Source	Resolution	Years Available in Tool
Land Cover
Land Cover	National Land Cover Database (NLCD)	[15]	30 m	2001, 2004, 2006, 2008, 2011, 2013, 2016, 2019, 2021, 2023
Tree Canopy
Tree Canopy Cover	National Land Cover Database (NLCD)	[15]	30 m	2011, 2013, 2016, 2019, 2021, 2023
Forest Disturbances
Forest Fires	Monitoring Trends in Burn Severity (MTBS)	[16]	30 m	2001–2023 (annual)
Insect and Disease	Insect and Disease Detection Survey	[17]	Varies	2001–2023 (annual)
Timber Harvest and Other	Global Forest Watch Tree Cover Loss	[14]	30 m	2001–2023 (annual)
Estimating Emission and Removal Factors
Removal Factors
Forest Type	FIA Forest Type Groups	[5]	30 m	Single estimate for 2014–2018
Plantations	Spatial Database of Planted Trees (v1)	[18]	Varies	2015
Forest Age	FIA Forest Stand Age	[5]	30 m	Single estimate for 2014–2018
Undisturbed Forests	FIA Database	[19]	Non-spatial	Varies by region and variable combinations
Afforestation or Reforestation	FIA Database	[19]	Non-spatial	Varies by region and variable combinations
Trees outside forests	Urban Trees Emission and Removal Factors	[20]	Non-spatial	2005
Emission Factors
Carbon Stocks	BIGMAP Forest Carbon Pools	[5]	30 meters	Single estimate for 2014–2018
Trees outside forests	Urban Trees Emission and Removal Factors	[20]	Non-spatial	2005
Forest Disturbances	Regionally modeled disturbance database	[21]	Non-spatial	Derived from FIA data (2001–2010)

Table 1. Data sources for calculation of GHG inventories in LEARN. For more detailed descriptions, see Glen et al., 2024 [13]. Variable categories are in bold, while variable sub-categories are in italics.

Variable	Dataset	Source	Resolution	Years Available in Tool
Land Cover
Land Cover	National Land Cover Database (NLCD)	[15]	30 m	2001, 2004, 2006, 2008, 2011, 2013, 2016, 2019, 2021, 2023
Tree Canopy
Tree Canopy Cover	National Land Cover Database (NLCD)	[15]	30 m	2011, 2013, 2016, 2019, 2021, 2023
Forest Disturbances
Forest Fires	Monitoring Trends in Burn Severity (MTBS)	[16]	30 m	2001–2023 (annual)
Insect and Disease	Insect and Disease Detection Survey	[17]	Varies	2001–2023 (annual)
Timber Harvest and Other	Global Forest Watch Tree Cover Loss	[14]	30 m	2001–2023 (annual)
Estimating Emission and Removal Factors
Removal Factors
Forest Type	FIA Forest Type Groups	[5]	30 m	Single estimate for 2014–2018
Plantations	Spatial Database of Planted Trees (v1)	[18]	Varies	2015
Forest Age	FIA Forest Stand Age	[5]	30 m	Single estimate for 2014–2018
Undisturbed Forests	FIA Database	[19]	Non-spatial	Varies by region and variable combinations
Afforestation or Reforestation	FIA Database	[19]	Non-spatial	Varies by region and variable combinations
Trees outside forests	Urban Trees Emission and Removal Factors	[20]	Non-spatial	2005
Emission Factors
Carbon Stocks	BIGMAP Forest Carbon Pools	[5]	30 meters	Single estimate for 2014–2018
Trees outside forests	Urban Trees Emission and Removal Factors	[20]	Non-spatial	2005
Forest Disturbances	Regionally modeled disturbance database	[21]	Non-spatial	Derived from FIA data (2001–2010)

2.3. Activity Data Generation

Detailed methodological steps for generating activity data are provided in Appendix A. This section provides a condensed overview.

2.3.1. Land Cover and Land Use

The terms ’land use’ and ’land cover’ are often used interchangeably but have important distinctions relevant to emissions accounting. Land use refers to the management and intended purpose of land (e.g., cropland and forest land), whereas land cover describes physical vegetation or surface characteristics (e.g., grassland, shrubland, and forest cover). LEARN integrates satellite-derived land-cover data (primarily from NLCD) and includes additional disturbance datasets (such as fire and insect/disease maps), which improve differentiation between temporary disturbances and permanent land-use transitions by providing specific spatial and temporal context for disturbance events. Nonetheless, this integrated approach still has limitations as land-cover maps alone may not fully distinguish temporary changes (e.g., fire or harvest) from permanent conversions (e.g., urban development or agricultural expansion). Therefore, additional contextual analysis and integration of local knowledge or ground-based data are recommended to further refine LEARN’s categorization of emissions and removals (see Section 4).

2.3.2. Land-Use Change Matrices

Activity data are calculated using a land-use change matrix approach [10] in which spatially and temporally explicit data derived from 30 m Landsat satellite imagery are used to track individual pixels consistently through time. NLCD is specifically designed to be harmonized over time so that classification changes in a pixel over a time series are more likely to represent actual changes on the ground. In this approach, area (ha) of total losses and gains in specific land-use categories can be calculated, as well as what these conversions represent. The NLCD datasets for start year and end year of an inventory period are combined to create a land-cover-change matrix. The detailed land-cover change values are recategorized (see

Table A1. Re-classification of NLCD land-cover classes to the six IPCC land-use classes.

IPCC Land-Use Class	Corresponding NLCD Land-Cover Classes
Forest Land	Deciduous Forest; Evergreen Forest; Mixed Forest; Woody Wetlands
Grassland	Shrub/Scrub; Grassland/Herbaceous; Pasture/Hay
Cropland	Cultivated Crops
Wetland	Open Water; Emergent Herbaceous Wetlands
Settlement	Developed, Open Space; Developed, Low Density; Developed, Medium Density; Developed, High Density
Other Land	Perennial Ice/Snow; Barren Land

) to the into six standard IPCC land-use categories (Forest Land, Cropland, Grassland, Settlements, Wetlands, and Other Land) to create a simplified transition matrix (see Figure 1). Transitions identified in the NLCD data are aggregated into four key categories (forests remaining forests, forest to nonforest, nonforest to forest, and nonforest remaining nonforest) based on IPCC standards, which directly inform subsequent emission and removal calculations (see Figure 2). The forests remaining forests and nonforest remaining nonforest categories are further stratified as described in Section 2.3.3 and Section 2.3.4 below.

2.3.3. Forests Remaining Forests

Areas of forest land remaining forest are further stratified as either undisturbed or disturbed (Figure 2), depending on their overlap with disturbance datasets during the inventory period. For each inventory period, spatial information about the location and timing of disturbances are estimated using a hierarchical combination of fire [16], insect and disease mortality [17], and tree cover loss ([14], updated annually on Global Forest Watch). Any areas of tree cover loss during the inventory period that do not overlap with fire or insect and disease mortality are assigned to the “harvest or other” category. This methodology aligns with findings from [22], who determined that from 1986 to 2010, less than 1% of forest disturbances in the United States could be attributed to causes outside of fire, insect/disease (“stress”), or harvest.

Low-severity disturbances, defined as those resulting in less than 50% canopy loss, are excluded from explicit disturbance classifications due to their typically rapid recovery, as corroborated by [21]. These disturbances are thus implicitly considered within the undisturbed forest category as they do not substantially alter long-term forest carbon dynamics within the inventory timeframe. Disturbances in areas outside of forests remaining forests (e.g., fire in grasslands) are excluded.

2.3.4. Nonforest Remaining Nonforest

Within any nonforest areas that remained nonforest during the inventory period, emissions and/or removals from changes from one land-use class to another (e.g., grassland to cropland) are not calculated individually. Rather, in this category LEARN estimates emissions and removals from NLCD Tree Canopy Cover change over the inventory period. Within areas that remained nonforest over the inventory period selected, tree canopy data for start and end year of the inventory period are compared to derive gross tree canopy loss. Tree canopy maintained or gained is estimated by averaging the total tree canopy cover areas in the start and end years of the inventory period. Where tree canopy data do not exactly align with NLCD land-cover data periods, canopy metrics are annualized by averaging available overlapping data points to estimate consistent annual change rates for the inventory period.

2.4. Deriving Emission and Removal Factors

This section provides an overview of underlying assumptions and methods for deriving emission and removal factors. Detailed descriptions for each subsection are provided in Appendix B. Complete lists of emission and removal factors are provided in Supplementary Materials Table S1: Emission and Removal Factors for Forests Remaining Forests and Nonforest to Forest and S2: Emission and Removal Factors for Trees Outside Forests.

2.4.1. Forests Remaining Forests

• Undisturbed Forests: Removal factors for undisturbed forests are generated using the FIA database via EVALIDator [23]. Data stratification includes management regions from [24], FIA forest-type groups, stand origin, and age class distributions. Plots with recent disturbances or atypical stocking levels are excluded.
• Disturbed Forests: Emission factors for forest disturbances within the inventory area are calculated by mapping unique combinations of forest region, forest-type group, disturbance type, and disturbance severity to a database of emission factors developed by the USDA FS and collaborators [25] based on methods reported in [21].

2.4.2. Forests and Land-Use Change

• Forest Converted to Nonforest: For any pixel delineated as forest converted to nonforest land uses, a carbon stock value for the forest is calculated for three aggregated carbon pools: living biomass (aboveground, belowground, and understory), dead organic matter (standing dead, down dead, and litter), and organic soil carbon [5]. Emissions are estimated by applying default national carbon loss percentages provided by the [26] to mapped values for pre-conversion biomass, dead organic matter, and soil carbon pools (

  Table A4. National average percent of carbon (C) stock lost upon conversion of forest to different nonforest land uses by carbon pool [26].

  	Forest Converted to …
  Carbon Pool	Cropland	Grassland	Wetland	Settlement	Other Land
  Biomass C	100	50	100	100	100
  Dead organic matter C	100	100	100	100	100
  Organic soil C	23	0	0	30	100

  Note: C = carbon.

  ).
• Nonforest Converted to Forest: Areas of new forest establishment via reforestation or afforestation are assigned removal factors based on forest region and forest-type group. Removal factors were developed by querying FIA data using the same methods for the 0–20-year age class of undisturbed forests, with additional adjustments for soil organic carbon accumulation during early forest establishment (following [27]).

2.4.3. Nonforest Remaining Nonforest

Emission and removal factors for trees outside forests are derived from [20].

• Tree Canopy Loss: Emission factors associated with canopy loss are determined at the city level based on direct measurements and detailed urban canopy assessments from representative cities within each state or region.
• Tree Canopy Maintained/Gained: Removal factors related to canopy maintained or gained during an inventory period utilize state-level averages due to limited availability of detailed local measurements for canopy growth across all communities.

2.5. Case Studies

LEARN analyses were conducted for three specific areas of interest over four periods: 2013–2016, 2016–2019, 2019–2021, and 2021–2023.

2.5.1. Community Inventories

Two of the analyses are community GHG inventories, selected to demonstrate diversity in community land bases and data applications. Jefferson County, Washington, is a large (5650 km²), sparsely populated county with significant forested area including Olympic National Park and National Forest. Montgomery County, Maryland, is a smaller (1310 km²) but more populated county [28], with a range of settlement densities, patchworks of cropland and pasture, and deciduous forests. Spatial boundaries for both counties were extracted from the US Census TIGER/Line dataset [28].

The community GHG inventories include analyses of forests and trees outside forests (tree canopy extent and tree canopy loss) and associated carbon sequestration and emissions. For trees outside forests, emission and removal factors were selected based on proximity to representative cities and states. Specifically, for Montgomery County, the emission factor from Baltimore, Maryland, and the removal factor for the state of Maryland were applied. Similarly, for Jefferson County, Washington, the emission factor from Seattle, Washington, and the removal factor from the state of Washington were used (see Supplementary Materials Table S2 for a complete list of emission and removal factors applied to trees outside forests). Because the 2023 NLCD Tree Canopy data were not yet available at the time of publication, tree canopy data from 2019 to 2021 were used for the 2019–2023 inventory period.

2.5.2. Federally Owned Forests

The third case study focuses on forests remaining forests (see Section 2.3.2) in federally owned forest lands across the CONUS. The PAD-US database [29] was used to delineate forest ownership. The database was filtered to ownership by USDA FS or the Bureau of Land Management (BLM) and then dissolved by state to correct overlapping or broken geometries. Because this case study is focused on forests remaining forests and excludes trees outside forests, tree canopy data were not included in the analysis.

2.6. Comparison Exercise

To validate our results, we performed independent queries of the U.S. Forest Inventory and Analysis (FIA) database for each of the case study areas using the EVALIDator query system [23] and compared the FIA stock-change estimates with the LEARN “net flux” estimates, which should be of similar magnitude for forests by excluding LEARN’s estimates for trees outside forests since these are not included in the FIA database. The inventory periods, however, are not fully comparable due to differences in remeasurement intervals. FIA remeasurement cycles typically range from 5 to 10 years depending on the region, generally shorter in the eastern United States and longer in the west, while LEARN inventory periods are user-defined and can vary widely but often span intervals aligned with National Land Cover Database (NLCD) releases, typically 2 to 3 years apart. Therefore, some temporal mismatch between FIA remeasurements and LEARN estimates should be considered when interpreting comparison results. For community inventories, comparisons between LEARN and FIA included forests remaining forests and forest change but excluded trees outside forests. For federally owned forests, comparisons included forests remaining forests only.

3. Results

3.1. Community Inventories

3.1.1. Jefferson County

Jefferson County had significant net GHG removals by forests and trees in all four inventory periods (see Figure 3 and

Table 2. Greenhouse gas inventories for four successive time periods in Jefferson County, Washington. Area units are in ha, and flux units are in Mg CO₂/year. Positive values are gross emissions, and negative values are gross removals, except in the total net flux line, which is the sum or all emissions and removals per inventory period. All values are annual averages across the inventory period. Bold text denotes categories.

	2013–2016		2016–2019		2019–2021		2021–2023
Category	Area	Flux	Area	Flux	Area	Flux	Area	Flux
Forest Change
To Forest	4862	−67,644	7181	−102,224	5430	−78,037	4717	−65,397
To Cropland	0	0	0	0	0	0	0	0
To Grassland	5851	397,840	5677	533,030	2645	619,481	3182	792,061
To Other	42	5109	20	4477	40	13,973	10	3537
To Settlement	87	12,413	109	13,399	33	6397	42	9155
To Wetland	322	29,910	155	26,467	122	27,970	210	43,567
Forest Remaining Forest
Fire	477	41,890	17	1478	0	0	484	66,616
Harvest/Other	1325	271,996	2064	425,303	1848	590,065	1843	588,617
Insect/Disease	22,557	−377,125	18,572	−193,646	0	0	0	0
Undisturbed	354,328	−4,580,890	356,932	−4,635,858	380,079	−4,951,419	381,584	−4,973,477
Trees Outside Forest
Maintained/gained	16,333	−168,885	15,971	−165,141	15,788	−163,251	14,887	−153,933
Tree-canopy loss	558	65,502	643	75,471	484	85,268	1975	347,263
Total Emissions		447,535		885,979		1,343,154		1,850,816
Total Removals		−4,817,419		−4,903,223		−5,192,707		−5,192,807
Net Flux		−4,369,884		−4,017,244		−3,849,553		−3,341,991

). The magnitude of net removals averaged about −3.89 million Mg CO₂/year from 2013 to 2023, roughly equivalent to the emissions from driving 900,000 gasoline-powered passenger vehicles for one year [1]. The net sink decreased about 24% from 2013–2016 to 2019–2023, driven by an increase in emissions from forest conversion and disturbances.

Gross removals are approximately −5.03 million Mg CO₂/year, on average, and average gross emissions roughly 1.13 million Mg CO₂/year. Most of the carbon removals in the Jefferson County Inventory (95%) came from undisturbed forests. Forests disturbed by insects and disease, while sequestering less carbon than an undisturbed forest, remove an average of roughly −142.7 thousand Mg CO₂/year. Nonforest to forest change accounts for roughly 2% of gross removals. Trees outside forests contribute about 3% of gross removals.

The largest cumulative sources of emissions are forest-to-grassland changes and disturbances in the harvest/other category, comprising an average of 52% and 41% of gross emissions, respectively. Emissions from forest-to-grassland changes and harvest disturbances were spatially clustered, predominantly in areas identified as private or state forest lands. Areas identified as forest-to-grassland change occur in larger blocks compared to a more dispersed pattern in areas identified as disturbed by harvest/other (Figure 4).

A comparison exercise between LEARN and FIA estimates showed that LEARN estimated similar net CO₂ flux for Jefferson County as FIA estimates (−3.88 million Mg CO₂/year during 2019–2021 compared to FIA’s −3.97 million Mg CO₂/year during a similar period, respectively, excluding trees outside forests). The forest area estimated by LEARN was nearly identical to FIA, about 2% greater (Table 5).

3.1.2. Montgomery County

Montgomery County’s gross emissions decreased steadily over the four inventory periods from 2013 to 2023, showing a reduction of 39% from the first inventory to the last (see Figure 5 and

Table 3. Greenhouse gas inventories for four successive time periods in Montgomery County, Maryland. Area units are in ha, and flux units are in Mg CO₂/year. Positive values are gross emissions, and negative values are gross removals, except in the total Net Flux line, which is the sum or all emissions and removals per inventory period. All values are annual averages across the inventory period. Bold text denotes categories.

	2013–2016		2016–2019		2019–2021		2021–2023
Category	Area	Flux	Area	Flux	Area	Flux	Area	Flux
Forest Change
To Forest	96	−693	85	−609	66	−474	83	−599
To Cropland	7	181	6	202	4	225	6	117
To Grassland	32	203	38	717	22	1444	28	1551
To Other	0	0	0	13	0	0	0	0
To Settlement	92	1,603	87	4093	43	6493	40	4655
To Wetland	10	115	17	157	18	255	6	144
Forest Remaining Forest
Fire	0	0	0	0	0	0	0	0
Harvest/Other	17	1337	43	3498	19	2665	15	2097
Insect/Disease	0	0	0	0	0	0	0	0
Undisturbed	38,398	−249,973	38,319	−249,456	38,340	−249,592	38,328	−249,520
Trees Outside Forest
Maintained/gained	23,430	−303,268	23,477	−303,873	24,062	−311,452	24,079	−311,671
Tree-canopy loss	1356	170,810	947	119,291	489	92,374	519	98,094
Total Emissions		174,249		127,971		103,456		106,658
Total Removals		−553,934		−553,938		−561,518		−561,790
Net Flux		−379,685		−425,967		−458,062		−455,132

). This reduction was due largely due to decreasing emissions from nonforest tree canopy loss, which contributed 94% of cumulative gross emissions on average. Most of this nonforest tree canopy loss in Montgomery County from 2013 to 2023 occurred in settlement land use areas (81%). Other emission sources such as forest conversion and forest disturbances accounted for only about 6% of gross emissions, with the harvest/other disturbance category as the largest forest emissions source.

Montgomery County maintained an average net carbon sink of roughly −429.7 thousand Mg CO₂/year from 2013 to 2023, with notable variation across inventory periods. The net sink increased by approximately 20% from 2013–2016 to 2019–2021, driven primarily by decreased emissions from tree canopy loss. However, there was a 6% uptick in tree canopy loss area and associated emissions between the 2019–2021 and 2021–2023 periods. Gross removals were roughly 55% from nonforest tree canopy and 45% from undisturbed forests.

LEARN estimated the net flux of CO₂ for Montgomery County forests at less than half compared with an independent FIA query (−242.3 thousand Mg CO₂/year vs. FIA’s −629.4 thousand Mg CO₂/year, excluding trees outside forests). The area of forest land was similar for both estimates, with minimal influence from disturbances or harvesting (Table 5). Possible reasons for these differences are explored in Section 4.1.

Patches of tree canopy loss (see Figure 6) are widespread across Montgomery County and occur primarily in urban land use areas. The pattern of tree canopy loss is decentralized in many small patches as opposed to large blocks of loss.

3.2. Federally Owned Forests

The net carbon sink from federally owned forests was roughly −133.91 million Mg CO₂/year from 2013 to 2023, with a decrease in net removals in the 2019–2021 period (Figure 7,

Table 4. Greenhouse gas inventories for four successive time periods in federally owned forests in the forest remaining forest category. Area units are in ha, and flux units are in Mg CO₂/year. Positive values are gross emissions, and negative values are gross removals, except in the total Net Flux line, which is the sum or all emissions and removals per inventory period. All values are annual averages across the inventory period.

	2013–2016		2016–2019		2019–2021		2021–2023
Category	Area	Flux	Area	Flux	Area	Flux	Area	Flux
Fire	160,102	7,186,885	91,328	4,178,937	429,712	38,001,265	116,570	10,810,266
Harvest/Other	198,490	18,023,850	267,781	26,788,129	185,121	30,936,174	174,497	29,760,304
Insect/Disease	3,699,556	3,615,148	2,896,261	−4,256,666	201	3,369	483	5,806
Undisturbed	46,840,576	−173,071,224	47,003,598	−172,334,863	48,752,295	−178,174,491	48,394,716	−177,105,604
Total Emissions		28,825,883		26,710,400		68,940,808		40,576,376
Total Removals		−173,071,224		−172,334,863		−178,174,491		−177,105,604
Net Flux		−144,245,341		−145,624,463		−109,233,683		−136,529,228

). Gross removals remained relatively stable at an average of −175.17 million Mg CO₂/year. The observed decrease in net removals was driven primarily by increased emissions from fire disturbances, which increased by more than 400% from the 2013–2016 period to the 2019–2021 period. Emissions from the harvest/other category also increased from 2013 to 2023, with annual average emissions rising about 65% from the 2013–2016 to 2021–2023 periods. In some cases, disturbances from insect/disease can result in overall net emissions (2013–2016 period), while in other cases, they can result in overall net removals (2016–2019 period), though these are always reported within the emissions category in the inventory.

Land ownership and forest disturbances have high spatial heterogeneity across CONUS. Figure 8a provides an overview of federal land ownership, which is highly concentrated in the Western States. Figure 8b shows forests remaining forests within federally owned lands in Western Colorado, where disturbance from fire and insect/disease is prevalent. LEARN estimated the net flux of CO₂ for federally owned forests at approximately −133.91 million Mg CO₂/year, similar to the FIA’s independent estimate of −129.01 million Mg CO₂/year. Forest area differed significantly between estimates: LEARN estimated around 49.80 million hectares versus FIA’s 76.22 million hectares. Both estimates indicated significant emissions from fire and harvesting, though direct comparisons are complicated by differences in inventory periods and the resolution of small-scale disturbances (

Table 5. Comparison of LEARN and FIA area and net flux estimates. Area units are in ha, and flux units are in Mg CO₂/year. Net flux is the sum or all emissions and removals per inventory period. For Jefferson County and Montgomery County, values are annual averages for the 2019–2021 period. Values exclude trees outside forests but include forests remaining forests and forest change. For federal forests, values are annual averages for the 2013–2023 period. Inventory periods were chosen to best align with FIA measurement dates. Values include forests remaining forests only.

Case Study	LEARN Area	FIA Area	LEARN Net Flux	FIA Net Flux
Jefferson County	390,712	384,469	−3,875,242	−3,966,447
Montgomery County	38,566	38,440	−242,288	−629,448
Federal forests	49,802,822	76,224,457	−133,908,179	−129,009,128

).

4. Discussion

The LEARN tool provides a robust approach to developing subnational GHG inventories by integrating publicly available spatial datasets with ground-sourced forest and tree inventory data. This standardized yet flexible framework bridges a critical gap between national-level aggregation and local-level action, enabling jurisdictions to effectively monitor changes in carbon sources and sinks from forests and trees and to inform climate strategies at a local level. Case studies from Jefferson County, Montgomery County, and federally owned forests demonstrate the practical utility and adaptability of LEARN across varying scales and land-management contexts.

4.1. Interpretation of Case Studies

The case studies of Jefferson and Montgomery counties illustrate LEARN’s practical applications and implications for local climate strategies. In Jefferson County, with its expansive forests and relatively low population density, the significant average net sink of approximately −3.89 million Mg CO₂/year highlights the critical role that forests and trees play in regional climate mitigation strategies. However, substantial temporal variation in emissions from forest disturbances and land-use changes emphasizes the importance of understanding and managing local drivers of emissions. Primary sources of emissions for the county are forest disturbances and forest-to-grassland transitions. Given this context, local management strategies could focus on limiting forest conversions through zoning restrictions or incentives promoting forest retention. Additionally, the county could enhance disturbance management by addressing risks related to pests, disease outbreaks, and human-driven disturbances, alongside implementing measures to restore or enhance forest resilience.

A significant portion of emissions in the Jefferson County inventory (52%) were found to be from forest-to-grassland changes. Challenges exist in differentiating between emissions resulting from land use and land-cover changes, especially for forests. Major forest disturbances, such as fires, can be classified as changes in land cover from forests to grasslands. In contrast, areas designated as grasslands may appear inaccurately as forest transitions during the early recovery phases after disturbances. Thus, it is vital to clearly distinguish temporary disturbances from permanent changes. Jurisdictions like Jefferson County that find large changes between forest and grassland should conduct supplemental analyses to correctly identify and categorize emissions and removals.

Notably, FIA retrievals for Jefferson County yielded similar results for both forest area (390.7 thousand ha for LEARN and 384.5 thousand ha for FIA) and net flux (−3.88 million Mg CO₂ for LEARN and −3.97 million Mg CO₂ for FIA). However, careful interpretation is advised regarding this alignment, particularly given the fundamental methodological differences between LEARN and FIA. LEARN primarily uses remotely sensed spatial data combined with FIA emission and removal factors to estimate carbon fluxes, which provides broad geographic coverage but may overlook fine-scale variability. In contrast, FIA utilizes detailed ground-based plot measurements, offering higher precision but potentially missing broader landscape-level dynamics. Also, temporal mismatches affect results, with remote sensing observations being more aligned with specific dates of data collection compared with FIA that spans longer and older inventory periods. These methodological distinctions inherently limit direct comparability and may contribute to discrepancies observed in other case studies, such as Montgomery County and federally owned forests. Therefore, while the Jefferson County results suggest promising consistency, users should recognize these methodological constraints and pursue supplemental validation to ensure accurate and context-specific interpretation for local climate strategies.

Montgomery County features pockets of dense population and suburban sprawl. The steady decrease in emissions from tree canopy loss within settlement areas could suggest effective local management practices. However, the recent slight uptick in emissions indicate the need for continued vigilance and monitoring to sustain progress. Land managers may want to assess the effectiveness of local tree canopy initiatives and strengthen funding for successful initiatives. Such programs might aim to specifically target areas vulnerable to urban heat island effects, bringing the shade and cooling benefits of tree cover to communities that need it most.

This landscape illustrates distinct challenges given its smaller forested area and higher development density. LEARN estimated the net flux of CO₂ for Montgomery County at less than half compared with an independent FIA query, excluding trees outside forests. The area of forest land was similar for both estimates, with neither showing significant influence from disturbances or harvest. The difference is likely due to other factors, such as different dates of inventory data, regional removal factors potentially set too low given Montgomery County’s position at ecological boundaries between the Northeast and Southeast regions, and uncertainties inherent in estimating fluxes for relatively small, forested areas. Users encountering similar discrepancies should carefully review and consider revising local removal factors to enhance accuracy.

Federally owned forests provide insights into broader-scale dynamics affecting carbon sinks, notably highlighting significant increases in emissions from fire disturbances and timber harvesting activities. LEARN estimated the net flux of CO₂ for federally owned forests similarly to an independent FIA query (−133.90 million Mg CO₂/year vs. FIA’s −129 million Mg CO₂/year). However, significant differences in estimated forest area exist: LEARN estimated approximately 49.80 million hectares compared to FIA’s 67.6 million hectares. These discrepancies are primarily because NLCD data classifies some areas without trees as nonforest land, while FIA classifies these areas based on land use as forests. Since these areas lack substantial tree cover, this discrepancy in forest area has limited influence on the net CO₂ flux. Both estimates indicate significant emissions from fire and harvesting activities, though direct comparisons are challenging due to differences in inventory periods and LEARN’s limited representation of small-scale disturbances. These area discrepancies reinforce the need for explicit local verification and interpretation of results.

4.2. Methodological Considerations and Recommended Improvements

Despite LEARN’s strengths, several methodological considerations and uncertainties remain. First, LEARN currently assesses land-use changes involving only forests and trees outside forests and does not address other significant land-use types such as cropland and pasture. Expanding LEARN to include soil carbon stock changes in nonforest landscapes would significantly enhance its utility for jurisdictions with diverse land-use profiles. Within forest analyses, LEARN currently focuses exclusively on CO₂ due to the greater availability and standardization of spatially explicit datasets and emission/removal factors. However, we acknowledge the significant emissions of other greenhouse gases such as methane (CH₄) and nitrous oxide (N₂O), particularly from disturbances like fires or specific land-management practices. Future developments of LEARN aim to integrate these additional greenhouse gases once reliable, spatially explicit methodologies and datasets become broadly available.

Second, the geospatial data integration approach employed by LEARN provides a scalable and consistent methodology but introduces uncertainty. Although LEARN utilizes the latest publicly available datasets, these datasets vary in accuracy and often lack comprehensive uncertainty documentation. Additionally, uncertainties associated with scale and resolution affect the interpretation of LEARN’s outputs. Applying LEARN at scales finer than intended, such as individual parcels, may amplify data uncertainty due to limitations in dataset resolution. Conversely, aggregating data to broader scales can obscure local dynamics and introduce classification challenges. Establishing standardized methods for uncertainty documentation and reporting across LEARN’s components could increase user confidence and transparency.

Third, accurately delineating forest disturbances—especially timber harvests—remains challenging due to the lack of dedicated national datasets explicitly tracking harvest activities. LEARN applies a hierarchical rule to classify disturbances (fire > insect mortality > harvest or other disturbances), but events such as windthrow or weather-related disturbances can be mistakenly classified as harvests. Given that harvest, insect, and fire disturbances constitute most forest disturbances in the conterminous United States [22], LEARN relies on a process of elimination based on assumptions. Developing robust delineations of forest disturbances and improving the classification of temporary versus permanent land-use changes could reduce misclassification errors. Land managers with significant areas of harvest or other disturbances should evaluate contextual data to accurately identify disturbance causes and refine their policy decisions accordingly.

4.3. Implications and Guidance for Users

Considering the methodological challenges and uncertainties described above, users should verify automated outputs against local knowledge and supplemental data to distinguish temporary disturbances from permanent land-use changes. Contextual analyses are essential to accurately differentiate types of disturbances, particularly distinguishing harvests from other forms of forest disturbance. A particular challenge users should actively address is differentiating between temporary land-cover changes due to disturbances (e.g., forest harvests or fires) and genuine, permanent land-use transitions (e.g., permanent deforestation or urban development). Because LEARN relies on satellite-derived land-cover data (primarily NLCD) as proxies for land-use changes, misclassification can occur, especially in areas recently disturbed by forest harvests or stand-replacing events, where temporary grassland or shrubland cover might appear.

To mitigate potential misclassification, users are encouraged to incorporate supplemental analyses using aerial imagery, local forest-management records, or ground-based inventories. Such verification can confirm whether significant changes detected by LEARN represent temporary disturbances or actual land-use conversions. Additionally, users should consider cross-validating LEARN results with complementary datasets that explicitly classify land use, such as the FIA land-use classifications or emerging national land-use mapping products. Integrating these additional data sources can significantly refine the accuracy and reliability of GHG inventories generated by LEARN.

It is crucial that users align the spatial scale of their analyses with the resolution of input datasets to avoid magnifying inaccuracies or overlooking critical local details. Although LEARN is fully operational across the entire conterminous United States, its optimal application is at subnational scales—such as counties, municipalities, and regional landscapes. LEARN primarily uses datasets provided at a 30 m resolution, effectively supporting analyses at these scales. While the tool can technically produce inventories at finer scales (e.g., parcel-level) or broader scales (e.g., multi-state regions), users should exercise caution with such applications. Finer-scale analyses can amplify uncertainties inherent in input datasets, while overly broad analyses might obscure critical local variations. Carefully matching analytical scale with data resolution ensures more meaningful and reliable results.

The LEARN tool’s reliance on 30 m resolution data from NLCD aligns well with recommendations from the recent literature [30], which highlights that extremely high-resolution satellite-based forest carbon maps (0.3–5 m) may significantly increase uncertainty and be less reliable for carbon monitoring. Spatial resolutions exceeding typical tree crown diameters (approximately 10–20 m for temperate forests) are most appropriate, confirming LEARN’s resolution as suitable for accurately capturing forest carbon dynamics at typical plot scales used in carbon accounting.

In addition to spatial considerations, temporal resolution and the timing of dataset updates are also important factors influencing LEARN’s utility and accuracy. LEARN provides annualized estimates of emissions and removals, designed explicitly for long-term monitoring and strategic climate planning rather than immediate disturbance response. While rapid events like wildfires will not be reflected instantly, their longer-term impacts on forest cover and tree canopy health are fully integrated over subsequent inventory periods. This design aligns with cumulative, longer-term climate mitigation objectives.

When significant discrepancies occur between LEARN outputs and independent datasets, users should evaluate and recalibrate local emission and removal factors accordingly. Subnational entities—cities, counties, states, or tribal nations—should clearly define jurisdictional boundaries when establishing net-zero or net-negative targets, recognizing contributions from lands beyond direct management control, such as federally managed areas. Clear delineation of jurisdictions, distinguishing locally managed from federally managed lands, is crucial for effectively setting and achieving climate targets. Proactive collaboration with data producers and funders can help identify emerging data needs, refine methodologies, and enhance support for local climate policies and land-use management strategies. By proactively addressing these methodological and interpretative challenges, users can significantly improve the reliability, accuracy, and policy relevance of GHG inventories derived using LEARN.

5. Conclusions

The LEARN tool has been applied in multiple subnational contexts across the conterminous United States, including but not limited to the examples presented here, reflecting its broad utility in various land-use and climate mitigation initiatives. These diverse applications illustrate how land-based greenhouse gas inventories can inform strategies aimed at reducing emissions and enhancing carbon sequestration. LEARN’s alignment with USDA and IPCC guidelines, combined with its accessibility and standardized framework, supports jurisdictions seeking to incorporate land sector dynamics into their climate action planning, particularly those with limited technical resources. By identifying key factors influencing local carbon balances—such as forest management, tree planting, and canopy maintenance—LEARN provides information that can help guide effective land-management decisions. Continued user engagement and iterative refinement based on practical feedback will be essential to address remaining methodological limitations and ensure the tool continues to effectively support climate mitigation efforts.