Limits...
Local discrepancies in continental scale biomass maps: a case study over forested and non-forested landscapes in Maryland, USA.

Huang W, Swatantran A, Johnson K, Duncanson L, Tang H, O'Neil Dunne J, Hurtt G, Dubayah R - Carbon Balance Manag (2015)

Bottom Line: A comprehensive understanding of map discrepancies is required to improve their effectiveness in carbon accounting and local decision-making.Discrepancies reduce with aggregation and the agreement among products improves at the county level.There is a net underestimation over high biomass forests and non-forested areas that could impact carbon accounting at all levels.

View Article: PubMed Central - PubMed

Affiliation: Department of Geographical Sciences, University of Maryland, College Park, USA.

ABSTRACT

Background: Continental-scale aboveground biomass maps are increasingly available, but their estimates vary widely, particularly at high resolution. A comprehensive understanding of map discrepancies is required to improve their effectiveness in carbon accounting and local decision-making. To this end, we compare four continental-scale maps with a recent high-resolution lidar-derived biomass map over Maryland, USA. We conduct detailed comparisons at pixel-, county-, and state-level.

Results: Spatial patterns of biomass are broadly consistent in all maps, but there are large differences at fine scales (RMSD 48.5-92.7 Mg ha(-1)). Discrepancies reduce with aggregation and the agreement among products improves at the county level. However, continental scale maps exhibit residual negative biases in mean (33.0-54.6 Mg ha(-1)) and total biomass (3.5-5.8 Tg) when compared to the high-resolution lidar biomass map. Three of the four continental scale maps reach near-perfect agreement at ~4 km and onward but do not converge with the high-resolution biomass map even at county scale. At the State level, these maps underestimate biomass by 30-80 Tg in forested and 40-50 Tg in non-forested areas.

Conclusions: Local discrepancies in continental scale biomass maps are caused by factors including data inputs, modeling approaches, forest/non-forest definitions and time lags. There is a net underestimation over high biomass forests and non-forested areas that could impact carbon accounting at all levels. Local, high-resolution lidar-derived biomass maps provide a valuable bottom-up reference to improve the analysis and interpretation of large-scale maps produced in carbon monitoring systems.

No MeSH data available.


Related in: MedlinePlus

Scatter plots of CMS_RF and NBCD_NCE biomass products against FIA plots and CMS field plots. a CMS_RF vs. FIA, b CMS_RF vs. Field, and c NBCD_NCE vs. Field. The red solid line is the 1:1 line. The blue dashed line is the fitted regression with the filtered dataset, which exclude zero biomass in NBCD_NCE data. R2 and RMSD are calculated based on the filtered dataset.
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4537504&req=5

Fig7: Scatter plots of CMS_RF and NBCD_NCE biomass products against FIA plots and CMS field plots. a CMS_RF vs. FIA, b CMS_RF vs. Field, and c NBCD_NCE vs. Field. The red solid line is the 1:1 line. The blue dashed line is the fitted regression with the filtered dataset, which exclude zero biomass in NBCD_NCE data. R2 and RMSD are calculated based on the filtered dataset.

Mentions: We compared predictions from the CMS_RF and NBCD_NCE maps with biomass estimates from FIA data (average of four sub-plots) (Fig. 7a) and our variable radius field plots (Fig. 7b, c). The Random Forests model used to generate the CMS_RF map explained ~50 % variability in biomass from variable radius field plots (R2 = 0.49, RMSE = 89.3 Mg ha−1, n = 848). A cross-validation of the CMS_RF map with plot level FIA data showed higher agreement, partly due to higher sample number (R2 = 0.69, RMSE = 58.2 Mg ha−1, n = 1,055). On the other hand, a cross validation of the NBCD_NCE map with variable radius estimates resulted in substantially weaker relationships (R2 = 0.14, RMSE = 125.1 Mg ha−1, n = 433).Fig. 7


Local discrepancies in continental scale biomass maps: a case study over forested and non-forested landscapes in Maryland, USA.

Huang W, Swatantran A, Johnson K, Duncanson L, Tang H, O'Neil Dunne J, Hurtt G, Dubayah R - Carbon Balance Manag (2015)

Scatter plots of CMS_RF and NBCD_NCE biomass products against FIA plots and CMS field plots. a CMS_RF vs. FIA, b CMS_RF vs. Field, and c NBCD_NCE vs. Field. The red solid line is the 1:1 line. The blue dashed line is the fitted regression with the filtered dataset, which exclude zero biomass in NBCD_NCE data. R2 and RMSD are calculated based on the filtered dataset.
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4537504&req=5

Fig7: Scatter plots of CMS_RF and NBCD_NCE biomass products against FIA plots and CMS field plots. a CMS_RF vs. FIA, b CMS_RF vs. Field, and c NBCD_NCE vs. Field. The red solid line is the 1:1 line. The blue dashed line is the fitted regression with the filtered dataset, which exclude zero biomass in NBCD_NCE data. R2 and RMSD are calculated based on the filtered dataset.
Mentions: We compared predictions from the CMS_RF and NBCD_NCE maps with biomass estimates from FIA data (average of four sub-plots) (Fig. 7a) and our variable radius field plots (Fig. 7b, c). The Random Forests model used to generate the CMS_RF map explained ~50 % variability in biomass from variable radius field plots (R2 = 0.49, RMSE = 89.3 Mg ha−1, n = 848). A cross-validation of the CMS_RF map with plot level FIA data showed higher agreement, partly due to higher sample number (R2 = 0.69, RMSE = 58.2 Mg ha−1, n = 1,055). On the other hand, a cross validation of the NBCD_NCE map with variable radius estimates resulted in substantially weaker relationships (R2 = 0.14, RMSE = 125.1 Mg ha−1, n = 433).Fig. 7

Bottom Line: A comprehensive understanding of map discrepancies is required to improve their effectiveness in carbon accounting and local decision-making.Discrepancies reduce with aggregation and the agreement among products improves at the county level.There is a net underestimation over high biomass forests and non-forested areas that could impact carbon accounting at all levels.

View Article: PubMed Central - PubMed

Affiliation: Department of Geographical Sciences, University of Maryland, College Park, USA.

ABSTRACT

Background: Continental-scale aboveground biomass maps are increasingly available, but their estimates vary widely, particularly at high resolution. A comprehensive understanding of map discrepancies is required to improve their effectiveness in carbon accounting and local decision-making. To this end, we compare four continental-scale maps with a recent high-resolution lidar-derived biomass map over Maryland, USA. We conduct detailed comparisons at pixel-, county-, and state-level.

Results: Spatial patterns of biomass are broadly consistent in all maps, but there are large differences at fine scales (RMSD 48.5-92.7 Mg ha(-1)). Discrepancies reduce with aggregation and the agreement among products improves at the county level. However, continental scale maps exhibit residual negative biases in mean (33.0-54.6 Mg ha(-1)) and total biomass (3.5-5.8 Tg) when compared to the high-resolution lidar biomass map. Three of the four continental scale maps reach near-perfect agreement at ~4 km and onward but do not converge with the high-resolution biomass map even at county scale. At the State level, these maps underestimate biomass by 30-80 Tg in forested and 40-50 Tg in non-forested areas.

Conclusions: Local discrepancies in continental scale biomass maps are caused by factors including data inputs, modeling approaches, forest/non-forest definitions and time lags. There is a net underestimation over high biomass forests and non-forested areas that could impact carbon accounting at all levels. Local, high-resolution lidar-derived biomass maps provide a valuable bottom-up reference to improve the analysis and interpretation of large-scale maps produced in carbon monitoring systems.

No MeSH data available.


Related in: MedlinePlus