A carbon monitoring system for mapping regional, annual aboveground biomass across the northwestern USA

Abstract

This paper presents a prototype Carbon Monitoring System (CMS) developed to produce regionally unbiased annual estimates of aboveground biomass (AGB). Our CMS employed a bottom-up, two-step modeling strategy beginning with a spatially and temporally biased sample: project datasets collected and contributed by US Forest Service (USFS) and other forestry stakeholders in 29 different project areas in the northwestern USA. Plot-level AGB estimates collected in the project areas served as the response variable for predicting AGB primarily from lidar metrics of canopy height and density (R2 = 0.8, RMSE = 115 Mg ha−1, Bias = 2 Mg ha−1). This landscape model was used to map AGB estimates at 30 m resolution where lidar data were available. A stratified random sample of AGB pixels from these landscape-level AGB maps then served as training data for predicting AGB regionally from Landsat image time series variables processed through LandTrendr. In addition, climate metrics calculated from downscaled 30 year climate normals were considered as predictors in both models, as were topographic metrics calculated from elevation data; these environmental predictors allowed AGB estimation over the full range of observations with the regional model (R2 = 0.8, RMSE = 152 Mg ha−1, Bias = 9 Mg ha−1), including higher AGB values (>400 Mg ha−1) where spectral predictors alone saturate. For both the landscape and regional models, the machine-learning algorithm Random Forests (RF) was consistently applied to select predictor variables and estimate AGB. We then calibrated the regional AGB maps using field plot data systematically collected without bias by the national Forest Inventory and Analysis (FIA) Program. We found both our project landscape and regional, annual AGB estimates to be unbiased with respect to FIA estimates (Biases of 1% and 0.7%, respectively) and conclude that they are well suited to inform forest management and planning decisions by our contributing stakeholders. Social media abstract Lidar-based biomass estimates can be upscaled with Landsat data to regionally unbiased annual maps.

Filtering ground noise from LiDAR returns produces inferior models of forest aboveground biomass in heterogenous landscapes

“…AGB models were fit using the ranger R package's implementation of the random forest algorithm (Breiman, 2001;Wright and Ziegler, 2017), a popular machine learning technique for predicting forest biomass across landscapes (see for instance Huang et al, 2019;Hudak et al, 2020). Separate models were fit on predictors calculated using each level of ground noise filtering ("unfiltered," "ground," "0.1m," "1m," and "2m" thresholds)…”

“…However, there exists some disagreement about precisely which returns to aggregate when computing these statistics. While some LiDAR-based AGB models include all returns when calculating summary statistics (Hudak et al, 2020), others first filter out returns below various height thresholds when calculating percentile heights (Ma et al, 2018), density percentiles (Huang et al, 2019), or their entire suite of predictors (García et al, 2010). Filtering is typically described as being done to remove ground noise from return data, in order to avoid having "ground" returns mask any signal contained in the remaining "canopy" returns.…”

Generation of Lidar-Predicted Forest Biomass Maps from Radar Backscatter with Conditional Generative Adversarial Networks

“…While reviewing earlier research applying two sequential regression models in their modelling strategy, we noted a variety of terms describing the same concept in the literature. While we choose to refer to this as a sequential regression approach, we additionally found the following use of terminology: two-step modelling strategy [42], [44], [54], two-stage regression [39], [41], twostage up-scaling method [23], [34], two-phase estimator [36], two-phase (or three-phase) sampling design [31], [38], hybrid and hierarchical model-based inference [37], [40], three-phase design [35]. Additionally, [32], [33], [43], [53], [55], [56] also apply a modelling approach with two sequential regression models without labelling it by any particular term.…”

“…Most of the previous research that we identified focus on relating ground reference data to ALS, and then relate ALS-derived AGB estimates to spaceborne LiDAR data [31], [35], [36], [38], [53], [55] or a combination of different sensors [23], [33], [34], [37], [40], [42], [56]. Some other relate the ALS-derived AGB estimates to a single sensor, such as Sentinel-2 [41], [43], Landsat [39], [44], GEDI Lidar [42], ALOS PALSAR, [54] or SRTM X-band radar [32].…”

Quantifying Drivers of Coastal Forest Carbon Decline Highlights Opportunities for Targeted Human Interventions

“…Carbon has been successfully estimated using derived spectral metrics from multispectral satellite imagery [32][33][34], structural metrics from light detection and ranging (LiDAR) data [35][36][37][38], or a combination of the two [39][40][41] in correlative models linking these metrics with field data and has become a common source for carbon estimation at regional scales. Access to multitemporal remote sensing data provides additional opportunities to measure change in carbon stocks over time; however regional analyses assessing carbon dynamics have focused on upland forests (e.g., [42][43][44][45]), herbaceous marsh (e.g., [46]), and mangrove forests (e.g., [47,48]) and not temperate coastal forests (however, see [49]). These studies also do not quantify the combined effects of the natural and anthropogenic factors driving carbon variability and change at regional scales, nor do they account for the spatial heterogeneity in the strength of the relationships between drivers and aboveground carbon stocks by implementing appropriate spatial modeling approaches.…”