No abstract

Forest inventory and planning decisions are frequently informed by LiDAR data. Repeated LiDAR acquisitions offer an opportunity to update forest inventories and potentially improve forest inventory estimates through time. We leveraged repeated LiDAR and ground measures for a study area in northern Idaho, U.S.A., to predict (via imputation) — across both space and time — four forest inventory attributes: aboveground carbon (AGC), basal area (BA), stand density index (SDI), and total stem volume (Vol). Models were independently developed from 2003 and 2009 LiDAR datasets to spatially predict response variables at both times. Annual rates of change were calculated by comparing response variables between the two collections. Additionally, a pooled model was built by combining reference observations from both years to test if imputation can be performed across measurement dates. The R2 values for the pooled model were 0.87, 0.90, 0.89, and 0.87 for AGC, BA, SDI, and Vol, respectively. Mapping response variables at the landscape level demonstrates that the relationship between field data and LiDAR metrics holds true even though the data were collected in different years. Pooling data across time increases the number of reference observations available to resource managers and may ultimately improve inventory predictions.

Since the mid-1800s pinyon-juniper (PJ) woodlands have been encroaching into sagebrush-steppe shrublands and grasslands such that they now comprise 40% of the total forest and woodland area of the Intermountain West of the United States. More recently, PJ ecosystems in select areas have experienced dramatic reductions in area and biomass due to extreme drought, wildfire, and management. Due to the vast area of PJ ecosystems, tracking these changes in woodland tree cover is essential for understanding their consequences for carbon accounting efforts, as well as ecosystem structure and functioning. Here we present a carbon monitoring, reporting, and verification (MRV) system for characterizing total aboveground biomass stocks and flux of PJ ecosystems across the Great Basin. This is achieved through a two-stage remote sensing approach by first using spatial wavelet analysis to rapidly sample tree cover from very high-resolution imagery (1 m), and then training a Random Forest model which maps tree cover across the region from 2000 to 2016 using temporallysegmented Landsat spectral indices obtained from the LandTrendr algorithm in Google Earth Engine. Estimates of cover were validated against field data from the SageSTEP project (R 2 =0.67, RMSE=10% cover). Biomass estimated from cover-based allometry was higher than estimates from the Forest Inventory and Analysis Program (FIA) at the plot-level (bias=5 Mg ha −1 and RMSE=15.5 Mg ha −1 ) due in part to differences in tree-level biomass allometrics. County-level aggregation of biomass closely matched estimates from the FIA (R 2 =0.97) after correcting for bias at the plot level. Even after many previous decades of encroachment, we find forest area (i.e. areas with 10% cover) increasing at a steady rate of 0.46% per year, but 80% of the 9.86 Tg increase in biomass is attributable to infilling of existing forest. This suggests that the known consequences of encroachment such as reduced water availability, impacts to biodiversity, and risk of severe wildfire may have been increasing across the region in recent years despite the actions of sagebrush steppe restoration initiatives.

No abstract