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

Pushbroom Photogrammetric Heights Enhance State-Level Forest Attribute Mapping with Landsat and Environmental Gradients

We demonstrate the potential for pushbroom Digital Aerial Photogrammetry (DAP) to enhance forest modeling (and mapping) over large areas, especially when combined with multitemporal Landsat derivatives. As part of the National Agricultural Imagery Program (NAIP), high resolution (30–60 cm) photogrammetric forest structure measurements can be acquired at low cost (as low as $0.23/km2 when acquired for entire states), repeatedly (2–3 years), over the entire conterminous USA. Our three objectives for this study are to: (1) characterize agreement between DAP measurements with Landsat and biophysical variables, (2) quantify the separate and combined explanatory power of the three auxiliary data sources for 19 separate forest attributes (e.g., age, biomass, trees per hectare, and down dead woody from 2015 USFS Forest Inventory and Analysis plot measurements in Washington state, USA) and (3) assess local biases in mapped predictions. DAP showed the greatest explanatory power for the widest range of forest attributes, but performance was appreciably improved with the addition of Landsat predictors. Biophysical variables contribute little explanatory power to our models with DAP or Landsat variables present. There is need for further investigation, however, as we observed spatial correlation in the coarse single-year grid (≈1 plot/25,000 ha), which suggests local biases at typical scales of mapped inferences (e.g., county, watershed or stand). DAP, in combination with Landsat, provides an unparalleled opportunity for high-to-medium resolution forest structure measurements and mapping, which makes this auxiliary data source immediately viable to enhance large-scale forest mapping projects.

Modeling of Aboveground Biomass with Landsat 8 OLI and Machine Learning in Temperate Forests

An accurate estimation of forests’ aboveground biomass (AGB) is required because of its relevance to the carbon cycle, and because of its economic and ecological importance. The selection of appropriate variables from satellite information and physical variables is important for precise AGB prediction mapping. Because of the complex relationships for AGB prediction, non-parametric machine-learning techniques represent potentially useful techniques for AGB estimation, but their use and comparison in forest remote-sensing applications is still relatively limited. The objective of the present study was to evaluate the performance of automatic learning techniques, support vector regression (SVR) and random forest (RF), to predict the observed AGB (from 318 permanent sampling plots) from the Landsat 8 Landsat 8 Operational Land Imager (OLI) sensor, spectral indexes, texture indexes and physical variables the Sierra Madre Occidental in Mexico. The result showed that the best SVR model explained 80% of the total variance (root mean square error (RMSE) = 8.20 Mg ha−1). The variables that best predicted AGB, in order of importance, were the bands that belong to the region of red and near and middle infrared, and the average temperature. The results show that the SVR technique has a good potential for the estimation of the AGB and that the selection of the model hyperparameters has important implications for optimizing the goodness of fit.

The Importance of Consistent Global Forest Aboveground Biomass Product Validation

Several upcoming satellite missions have core science requirements to produce data for accurate forest aboveground biomass mapping. Largely because of these mission datasets, the number of available biomass products is expected to greatly increase over the coming decade. Despite the recognized importance of biomass mapping for a wide range of science, policy and management applications, there remains no community accepted standard for satellite-based biomass map validation. The Committee on Earth Observing Satellites (CEOS) is developing a protocol to fill this need in advance of the next generation of biomass-relevant satellites, and this paper presents a review of biomass validation practices from a CEOS perspective. We outline the wide range of anticipated user requirements for product accuracy assessment and provide recommendations for the validation of biomass products. These recommendations include the collection of new, high-quality in situ data and the use of airborne lidar biomass maps as tools toward transparent multi-resolution validation. Adoption of community-vetted validation standards and practices will facilitate the uptake of the next generation of biomass products.

Multiscale divergence between Landsat- and lidar-based biomass mapping is related to regional variation in canopy cover and composition

BACKGROUND

Satellite-based aboveground forest biomass maps commonly form the basis of forest biomass and carbon stock mapping and monitoring, but biomass maps likely vary in performance by region and as a function of spatial scale of aggregation. Assessing such variability is not possible with spatially-sparse vegetation plot networks. In the current study, our objective was to determine whether high-resolution lidar-based and moderate-resolution Landsat-base aboveground live forest biomass maps converged on similar predictions at stand- to landscape-levels (10 s to 100 s ha) and whether such differences depended on biophysical setting. Specifically, we examined deviations between lidar- and Landsat-based biomass mapping methods across scales and ecoregions using a measure of error (normalized root mean square deviation), a measure of the unsystematic deviations, or noise (Pearson correlation coefficient), and two measures related to systematic deviations, or biases (intercept and slope of a regression between the two sets of predictions).

RESULTS

Compared to forest inventory data (0.81-ha aggregate-level), lidar and Landsat-based mean biomass predictions exhibited similar performance, though lidar predictions exhibited less normalized root mean square deviation than Landsat when compared with the reference plot data. Across aggregate-levels, the intercepts and slopes of regression equations describing the relationships between lidar- and Landsat-based biomass predictions stabilized (i.e., little additional change with increasing area of aggregates) at aggregate-levels between 10 and 100 ha, suggesting a consistent relationship between the two maps at landscape-scales. Differences between lidar- and Landsat-based biomass maps varied as a function of forest canopy heterogeneity and composition, with systematic deviations (regression intercepts) increasing with mean canopy cover and hardwood proportion within forests and correlations decreasing with hardwood proportion.

CONCLUSIONS

Deviations between lidar- and Landsat-based maps indicated that satellite-based approaches may represent general gradients in forest biomass. Ecoregion impacted deviations between lidar and Landsat biomass maps, highlighting the importance of biophysical setting in determining biomass map performance across aggregate scales. Therefore, regardless of the source of remote sensing (e.g., Landsat vs. lidar), factors affecting the measurement and prediction of forest biomass, such as species composition, need to be taken into account whether one is estimating biomass at the plot, stand, or landscape scale.

Implications of allometric model selection for county-level biomass mapping

BACKGROUND

Carbon accounting in forests remains a large area of uncertainty in the global carbon cycle. Forest aboveground biomass is therefore an attribute of great interest for the forest management community, but the accuracy of aboveground biomass maps depends on the accuracy of the underlying field estimates used to calibrate models. These field estimates depend on the application of allometric models, which often have unknown and unreported uncertainties outside of the size class or environment in which they were developed.

RESULTS

Here, we test three popular allometric approaches to field biomass estimation, and explore the implications of allometric model selection for county-level biomass mapping in Sonoma County, California. We test three allometric models: Jenkins et al. (For Sci 49(1): 12-35, 2003), Chojnacky et al. (Forestry 87(1): 129-151, 2014) and the US Forest Service's Component Ratio Method (CRM). We found that Jenkins and Chojnacky models perform comparably, but that at both a field plot level and a total county level there was a ~ 20% difference between these estimates and the CRM estimates. Further, we show that discrepancies are greater in high biomass areas with high canopy covers and relatively moderate heights (25-45 m). The CRM models, although on average ~ 20% lower than Jenkins and Chojnacky, produce higher estimates in the tallest forests samples (> 60 m), while Jenkins generally produces higher estimates of biomass in forests < 50 m tall. Discrepancies do not continually increase with increasing forest height, suggesting that inclusion of height in allometric models is not primarily driving discrepancies. Models developed using all three allometric models underestimate high biomass and overestimate low biomass, as expected with random forest biomass modeling. However, these deviations were generally larger using the Jenkins and Chojnacky allometries, suggesting that the CRM approach may be more appropriate for biomass mapping with lidar.

CONCLUSIONS

These results confirm that allometric model selection considerably impacts biomass maps and estimates, and that allometric model errors remain poorly understood. Our findings that allometric model discrepancies are not explained by lidar heights suggests that allometric model form does not drive these discrepancies. A better understanding of the sources of allometric model errors, particularly in high biomass systems, is essential for improved forest biomass mapping.

Attribution of net carbon change by disturbance type across forest lands of the conterminous United States

BACKGROUND

Locating terrestrial sources and sinks of carbon (C) will be critical to developing strategies that contribute to the climate change mitigation goals of the Paris Agreement. Here we present spatially resolved estimates of net C change across United States (US) forest lands between 2006 and 2010 and attribute them to natural and anthropogenic processes.

RESULTS

Forests in the conterminous US sequestered -460 ± 48 Tg C year^-1, while C losses from disturbance averaged 191 ± 10 Tg C year^-1. Combining estimates of net C losses and gains results in net carbon change of -269 ± 49 Tg C year^-1. New forests gained -8 ± 1 Tg C year^-1, while deforestation resulted in losses of 6 ± 1 Tg C year^-1. Forest land remaining forest land lost 185 ± 10 Tg C year^-1 to various disturbances; these losses were compensated by net carbon gains of -452 ± 48 Tg C year^-1. C loss in the southern US was highest (105 ± 6 Tg C year^-1) with the highest fractional contributions from harvest (92%) and wind (5%). C loss in the western US (44 ± 3 Tg C year^-1) was due predominantly to harvest (66%), fire (15%), and insect damage (13%). The northern US had the lowest C loss (41 ± 2 Tg C year^-1) with the most significant proportional contributions from harvest (86%), insect damage (9%), and conversion (3%). Taken together, these disturbances reduced the estimated potential C sink of US forests by 42%.

CONCLUSION

The framework presented here allows for the integration of ground and space observations to more fully inform US forest C policy and monitoring efforts.

Attribution of net carbon change by disturbance type across forest lands of the conterminous United States

BACKGROUND

Locating terrestrial sources and sinks of carbon (C) will be critical to developing strategies that contribute to the climate change mitigation goals of the Paris Agreement. Here we present spatially resolved estimates of net C change across United States (US) forest lands between 2006 and 2010 and attribute them to natural and anthropogenic processes.

RESULTS

Forests in the conterminous US sequestered -460 ± 48 Tg C year^-1, while C losses from disturbance averaged 191 ± 10 Tg C year^-1. Combining estimates of net C losses and gains results in net carbon change of -269 ± 49 Tg C year^-1. New forests gained -8 ± 1 Tg C year^-1, while deforestation resulted in losses of 6 ± 1 Tg C year^-1. Forest land remaining forest land lost 185 ± 10 Tg C year^-1 to various disturbances; these losses were compensated by net carbon gains of -452 ± 48 Tg C year^-1. C loss in the southern US was highest (105 ± 6 Tg C year^-1) with the highest fractional contributions from harvest (92%) and wind (5%). C loss in the western US (44 ± 3 Tg C year^-1) was due predominantly to harvest (66%), fire (15%), and insect damage (13%). The northern US had the lowest C loss (41 ± 2 Tg C year^-1) with the most significant proportional contributions from harvest (86%), insect damage (9%), and conversion (3%). Taken together, these disturbances reduced the estimated potential C sink of US forests by 42%.

CONCLUSION

The framework presented here allows for the integration of ground and space observations to more fully inform US forest C policy and monitoring efforts.

Rapid, High-Resolution Forest Structure and Terrain Mapping over Large Areas using Single Photon Lidar

Single photon lidar (SPL) is an innovative technology for rapid forest structure and terrain characterization over large areas. Here, we evaluate data from an SPL instrument - the High Resolution Quantum Lidar System (HRQLS) that was used to map the entirety of Garrett County in Maryland, USA (1700 km²). We develop novel approaches to filter solar noise to enable the derivation of forest canopy structure and ground elevation from SPL point clouds. SPL attributes are compared with field measurements and an existing leaf-off, low-point density discrete return lidar dataset as a means of validation. We find that canopy and ground characteristics from SPL are similar to discrete return lidar despite differences in wavelength and acquisition periods but the higher point density of the SPL data provides more structural detail. Our experience suggests that automated noise removal may be challenging, particularly over high albedo surfaces and rigorous instrument calibration is required to reduce ground measurement biases to accepted mapping standards. Nonetheless, its efficiency of data collection, and its ability to produce fine-scale, three-dimensional structure over large areas quickly strongly suggests that SPL should be considered as an efficient and potentially cost-effective alternative to existing lidar systems for large area mapping.

Integrating LIDAR and forest inventories to fill the trees outside forests data gap

Forest inventories are commonly used to estimate total tree biomass of forest land even though they are not traditionally designed to measure biomass of trees outside forests (TOF). The consequence may be an inaccurate representation of all of the aboveground biomass, which propagates error to the outputs of spatial and process models that rely on the inventory data. An ideal approach to fill this data gap would be to integrate TOF measurements within a traditional forest inventory for a parsimonious estimate of total tree biomass. In this study, Light Detection and Ranging (LIDAR) data were used to predict biomass of TOF in all "nonforest" Forest Inventory and Analysis (FIA) plots in the state of Maryland. To validate the LIDAR-based biomass predictions, a field crew was sent to measure TOF on nonforest plots in three Maryland counties, revealing close agreement at both the plot and county scales between the two estimates. Total tree biomass in Maryland increased by 25.5 Tg, or 15.6%, when biomass of TOF were included. In two counties (Carroll and Howard), there was a 47% increase. In contrast, counties located further away from the interstate highway corridor showed only a modest increase in biomass when TOF were added because nonforest conditions were less common in those areas. The advantage of this approach for estimating biomass of TOF is that it is compatible with, and explicitly separates TOF biomass from, forest biomass already measured by FIA crews. By predicting biomass of TOF at actual FIA plots, this approach is directly compatible with traditionally reported FIA forest biomass, providing a framework for other states to follow, and should improve carbon reporting and modeling activities in Maryland.