High-Resolution Mapping of Aboveground Biomass for Forest Carbon Monitoring - A Case Study in Three Mid-Atlantic States, USA Wenli
1* Huang ,
Katelyn Kristofer Jarlath O’Neil 1 1 Ralph Dubayah , George Hurtt 1 Dolan ,
3 Dunne
2 Johnson ,
Introduction
*Contact:
[email protected] 1 University of Maryland, College Park, MD 2 USDA Forest Service, Northern Research Station, PA 3 University of Vermont, Burlington, VA
Study Area & Data
Map Comparison
Location of Field Plots
Background: Accurate mapping of forest aboveground biomass is critical for reducing uncertainties in carbon monitoring and accounting systems. Over the last six years as part of NASA's Carbon Monitoring System program, we have been working towards this direction. Goal: to develop a robust, replicable and scalable framework that quantifies forest structure and aboveground biomass over large areas at fine resolution using LiDAR and Forest Inventory Analysis (FIA) data.
State-wide Empirical Biomass Map
Mean Canopy Height (m)
Methodology
5 10 15 20 25 30 35
University of Maryland (UMD)/ CMS (1193)
TOT=36,600 Mg
Forest Inventory and Analysis (FIA)/ USFS (3268)
Leaf-off LiDAR data were combined with high-resolution leafon agricultural imagery, and the USFS-FIA plot data to estimate forest aboveground biomass over 150,000 square km area in three Mid-Atlantic States (Maryland, Delaware & Pennsylvania).
Non Forest/ Non Measured (2977)
2007 2007
TOT=68,229 Mg
2006
TOT=68,229 Mg
2006
2008
2007 2005 2012
2005 2008
TOT=104,801 Mg TOT=53,257 Mg
2014 2004
2004
LiDAR acquisition year was used to link LiDAR metrics to estimates of biomass from FIA plot measurements that most closely matched.
High resolution (1-m) tree canopy cover maps were generated via objected based approach using NAIP and LiDAR (O'Neil-Dunne et al. 2014)
Pixel-level Comparison Biomass maps compared at 250-m resolution Dubayah, R. (2012). County-Scale Carbon Estimation in NASA's Carbon Monitoring System. Biomass and Carbon Storage.
Total
Forest
Non-Forest
40 20 0
Percentile Metrics
Dencile Metrics
Plot-level Calibration & Validation Random Forest models were calibrated and validated using FIA plot and CMS campaign field data.
Calibration
By-Region
Forest/Non-forest mask was aggregated from 30-m NLCD 2011’ land cover map (Deciduous, Evergreen and Mixed Forest, and Woody Wetland)
Landsat ETM+ 19992002, SRTM 2000, NED
Circa 2000
NLCD 2001
90m Avg. vs. 90m Pred.
Random forest, regression tree (Kellndorfer et al. 2012)
MODIS, Landsat, Circa PALSAR, 2005 GLAS
NLCD 2006
Maximum Entropy, Parametric
183.4
111.9
1200 1000
1
MODIS NLCD 2001, 1992 1990Topo, forest 2003 Climate mask variables 50%
NBCD
Saatchi
Wilson
30m Pred. vs. CMS plot
(Blackard et al. 2008)
*Nominal Year of map determined by year of Remote Sensing dataset CRM=Component Ratio Method
County-level Forest biomass from maps versus FIA_PLOT estimates showing strong correlation (R2 = 0.93–0.96), but coarser resolution maps (Wilson & Blackard) shows consistent negative bias (-3.4 & -3.1 Tg). CMS_RF
Cubist, regression tree
Blackard
268.2
159.3
865.4
895.6
800 600
1243.2
1179.9
1174.2
400
0
(Saatchi et al. 2005)
MODIS NLCD PGNN, 20022001 kNN 2008, 2005percent Topo, 2009 tree (Wilson et Climate canopy al. 2013) variables 25%
Forest
254.5
200
County-level Comparison FIA Plot Biomass (Mg/ha)
Validation
1400
Total AGBM (Tg)
2000 [30-m]
88m Pred. vs. FIA plot
CMS_RF_CRM Biomass (Mg/ha)
Evaluation
Non-forest
Blackard
By-State
90m Avg. vs. FIA plot
1600
1188.5
Map estimates of total biomass are comparable to USFSFIA estimates over Forest regions, however diverse over Non-forest regions.
Conclusion
Wilson
2001 [250-m]
One
2005 [90-m]
60
D05 D15 D25 D35 D45 D55 D65 D75 D85 D95
P00 P10 P20 P30 P40 P50 P60 P70 P80 P90 P100
80
Forest Mapping Mask Approach
Saatchi
2005 [250-m]
100
Precent (%)
Height (m)
30x30 m
40 35 30 25 20 15 10 5 0
National Products Biomass (Mg/ha)
30 x 30 m Lidar Metrics: Percentiles Deciles Mean Heights Canopy Cover
State-level Comparison
NBCD
1 m Canopy Cover 1x1 m
Year* Predictor Field Res. Variables Data Year
Our CMS_RF biomass map provided greater details and much higher total biomass [Mg] over urban/suburban landscapes when compared visually with four national maps and referred to land cover maps and high-resolution imagery (NAIP).
Conclusion Pixel-level comparisons show great differences between estimates of biomass from high [30-m] and coarser [250-m] resolution maps, and LiDAR-derived biomass map reveals more details at finer scales County-level and State-level total biomass estimates from high [30-m] to moderate [90-m] resolution maps are close to USFS-FIA estimates over Forested areas Local, high-resolution LiDAR-derived biomass maps such as ours, provide a valuable bottom-up reference to improve the analysis and interpretation of large-scale mapping efforts, and future development of a national carbon monitoring system
Future Work Investigate the influence from forest cover and tree allometry on biomass product, and cross-validate the empirical result to prognostic biomass output Continue the comparison at representative county/state scales in US, and integrate multi-source datasets to inform carbon monitoring efforts Funded by NASA CMS Grant: NNX12AN07G
References CMS Field Plot Biomass (Mg/ha)
FIA_PLOT Forest Biomass (Tg)
Huang W, Swantaran A, Johnson K, Duncanson L, Tang H, O'Neil Dunne J, Hurtt G, Dubayah R (2015) Local Discrepancies in Continental Scale Biomass Maps: A Case Study over Forested and NonForested Landscapes in Maryland, USA. Carbon Balance and Management 10 (19).