TL;DR
This study compares four global above-ground biomass (AGB) change maps over the past decade, finding that while maps generally agree on AGB loss in deforestation hotspots, they often underestimate the magnitude of change. LiDAR data proved to be the most reliable reference for validation, highlighting the need for more comprehensive and diverse ground-truth data, especially in tropical regions, to improve the accuracy of these crucial climate variables.
What Problem Does This Paper Address?
Above-ground biomass (AGB) is a critical component of the global carbon cycle, directly influencing our understanding of climate change mitigation. Accurately tracking changes in AGB (ΔAGB) is essential for climate modeling and national carbon accounting. While satellite-based AGB and ΔAGB products have become more available, their reliability and consistency across different mapping approaches remain a significant challenge. This paper addresses the need to assess and compare these emerging global ΔAGB products, identifying their strengths, weaknesses, and areas of disagreement, particularly in relation to independent reference data. The authors aim to shed light on the biases and uncertainties inherent in current global AGB change maps, guiding both producers and users of this data.
Methods
The study employs a multi-faceted approach to compare four global ΔAGB maps: ESA-CCI, WRI-Flux, JPL, and SMOS-LVOD. The core methodology involves:
- Data Harmonization: Preliminary steps were taken to ensure comparability between the maps, including modifying the original carbon flux model for the Flux product and applying a consistent forest mask (defined as >30% tree cover) to all datasets.
- Reference Data Integration: The researchers utilized three types of reference data:
- National Forest Inventory (NFI) plots: Re-measured plots from Belgium, the Netherlands, Philippines, and Sweden were used, with estimated uncertainties derived from measurement and allometric model errors.
- Airborne LiDAR-based maps: Local ΔAGB maps derived from LiDAR campaigns in Brazil and the USA, as well as other research projects, were employed. These maps often came with associated standard deviation layers.
- Forest Resource Assessment (FRA) country data: Country-level estimates of AGB from FRA reports were used, selected based on countries with robust NFI data.
- Spatial Aggregation and Comparison: Reference data and map products were aggregated to various spatial resolutions (100 m, 500 m, 1 km, 10 km, and 25 km) for comparison. Statistical measures like Mean Difference (MD), Root Mean Square Difference (RMSD), coefficient of determination (R²), and Nash-Sutcliffe Efficiency (NSE) were used to quantify agreement.
- Map-Map Comparison: At a 25 km resolution, the four maps were spatially compared to assess their agreement and identify regions of consistent change (loss, gain, or no change) and disagreement. Cross-correlation analysis was also performed.
Datasets
The study utilizes the following datasets:
- Four Global ΔAGB Maps:
- ESA-CCI Biomass maps: Annual global AGB maps at 100 m resolution, derived from radar data.
- WRI-Flux model: ΔAGB derived from a modified carbon flux model, with 100 m resolution.
- JPL time series: AGB maps at 10 km resolution, derived from a two-step approach combining field plots, LiDAR, and radar data.
- SMOS-LVOD time series: AGB maps at 25 km resolution, derived from Soil Moisture and Ocean Salinity (SMOS) satellite observations.
- Reference Datasets:
- National Forest Inventory (NFI) plot data: From Belgium, Netherlands, Philippines, and Sweden.
- Airborne LiDAR-based AGB maps: From Brazil, USA, Bulgaria, Czech Republic, Costa Rica, Poland, and Spain.
- Forest Resource Assessment (FRA) country data: From countries with well-developed monitoring capacities.
- Ancillary Data:
- Hansen et al. (2013) tree cover data: Used for forest masking.
- IPCC (2019) growth data: Used for adjusting AGB estimates for temporal mismatches.
The study does not introduce a new benchmark dataset but rather focuses on the comparative assessment of existing global maps against diverse reference data.
Key Insights and Ideas
- LiDAR as the Gold Standard (for now): LiDAR-derived AGB maps consistently showed the most reasonable agreement with the satellite-based maps, suggesting their higher reliability as a reference. This underscores the value of high-resolution, direct measurements.
- NFI Limitations: While valuable, NFI plot data showed only moderate agreement with maps up to 1 km, and poor agreement at coarser scales. This is attributed to factors like sparse plot density in grid cells and potential underrepresentation of local heterogeneity.
- Systematic Underestimation of Change: A consistent finding across all comparisons was that the satellite-based maps tended to report smaller magnitudes of AGB loss and gain compared to reference data. This suggests a potential bias in the maps, possibly stemming from baseline map uncertainties or limitations in detecting subtle changes.
- Agreement in Deforestation Hotspots, Disagreement Elsewhere: The maps generally agreed on identifying AGB losses in well-known deforestation areas. However, significant disagreements emerged in complex regions like the Amazon basin, highlighting the challenges in mapping AGB dynamics in these critical carbon sinks.
- Methodological Divergence Matters: The study implicitly highlights how different mapping methodologies (e.g., stock change vs. gain-loss approaches, different remote sensing data) contribute to map disagreements, even when using similar baseline data.
Key Results
- Map-LiDAR Agreement: R² values between map products and LiDAR data ranged from 0.1 to 0.44 across different aggregation levels, indicating a moderate to good agreement.
- Map-NFI Agreement: Agreement with NFI data was moderate up to 1 km (R² ~0.1) and poor at coarser scales (R² ~0.03).
- Map-FRA Agreement: Agreement with FRA data varied, with R² values between 0.03 and 0.28.
- Map-Map Correlation: The overall cross-correlation (r) between the 25 km ΔAGB maps ranged from 0.11 to 0.29, indicating a moderate level of consistency.
- Magnitude Discrepancies: Reported AGB losses and gains from maps were consistently smaller than those from reference data.
- Spatial Agreement: Maps showed consistent identification of AGB losses in deforestation hotspots but significant disagreement in regions like the central Amazon basin.
Limitations and Future Work
- Reference Data Sparsity: A major limitation acknowledged is the underrepresentation of reference data, particularly in tropical regions and across various eco-zones. This restricts the global representativeness of the assessments.
- Future of LiDAR: The authors anticipate that upcoming airborne LiDAR campaigns will significantly alleviate the deficit in tropical reference data.
- NFI Enhancement: The use of NFI data could be improved by developing tools for easier online map assessment and potentially through data crowd-sourcing platforms.
- Uncertainty-Weighted Averages: Future assessments would benefit from reference data with estimated uncertainties, allowing for uncertainty-weighted AGB averaging.
- Map Producer Improvements: Map producers are encouraged to revisit their retrieval processes in areas of high disagreement and to leverage denser datasets (e.g., spaceborne LiDAR) and improved methodologies.
- Addressing Biases: The study emphasizes the need to understand and address the identified biases in AGB maps to increase their reliability for applications like carbon monitoring and ecosystem accounting.
- Ensemble Approaches: The authors suggest that producing new maps from ensembles of existing global AGB maps could be a way to deal with map disagreements, provided a suitable global reference dataset is available.
Summary updated