Search

• Old-growth definition criteria greatly impacts amount estimates and maps. • Physically-based old-growth definitions are detectable via lidar remote sensing. • Age-based old-growth definitions are difficult to detect using lidar remote sensing. • Age-based old-growth forests are extremely diverse in physical structure. Old-growth forests have been widely studied for decades. The extreme diversity of old forest characteristics has inspired an equally diverse set of old-growth definitions, and makes mapping old-growth difficult across large areas and different forest types. While the use of remote sensing in old-growth research is not new, there is a growing need for large scale mapping to improve understanding of old forest processes and to support old-growth conservation. Old-growth mapping requires definitions that are ecologically relevant to old forests while also transferable to remote sensing data. In this paper we develop a conceptual framework to evaluate three dimensions of old-growth—a temporal dimension related to tree ages, a physical dimension related to tree sizes, and a functional dimension related to forest processes. In the first part of our analysis, we classify forests throughout the eastern US as old or not with respect to each old-growth dimension using existing old-growth definitions and data from the US Forest Inventory and Analysis (FIA) program. We estimate the proportion of forest classified as old within a hexagon grid, resulting in a unique map of old forest proportion (OFP) for each dimension. Subsequently, we use spaceborne lidar data from NASA's Global Ecosystem Dynamics Investigation (GEDI) to reproduce each OFP map in a modeling framework designed to 1) assess the extent to which each dimension of forest oldness can be mapped at large spatial scales, and 2) identify biophysical GEDI variables related to each dimension of forest oldness. We estimate that only 2% of forest classified as old in any dimension satisfied the old criteria in all three dimensions. We found substantial spatial variation in the mapped OFP estimates across the three dimensions, highlighting how definition criteria impacts old-growth maps. We also found that physically old forests were more effectively mapped using GEDI data than functionally or temporally old forests, and that physically old forests were more structurally similar to one another than temporally or functionally old forests. Our modeling results indicate that while remote sensing may be best suited to mapping physical old-growth characteristics, definitions that rely solely on physical characteristics do not adequately represent old forests throughout the eastern US. We propose that future efforts to map old-growth with spaceborne remote sensing data may maximize utility through collaboration between western and indigenous old-growth experts to determine broad yet nuanced approaches that are appropriately tailored to the target variable of old forests. These efforts should balance explicit and ecologically relevant old-growth definitions specifically for mapping that can be linked to remotely sensed data, 2) appropriate spatial resolutions, and 3) flexible quantitative frameworks that encompass the complexities and heterogeneity of old forests. [ABSTRACT FROM AUTHOR]

NASA's Global Ecosystem Dynamics Investigation (GEDI), to be deployed on the International Space Station (ISS) in 2018, will provide billions of measurements of ground elevation and forest vertical structure. GEDI will acquire data only along transects or tracks, with between-track spacing of about 500 m. To fill the gaps in between these tracks and potentially produce higher spatial resolution products, appropriate fusion strategies between GEDI observations and other spatially contiguous datasets should be explored. One source of global data on canopy structure comes from the TanDEM-X (TDX) mission, which uses the technique of Interferometric Synthetic Aperture Radar (InSAR) to estimate surface structure. The goal of this paper is to explore the fusion of GEDI data with TDX for canopy height retrievals. In particular we examined the improvement in TDX height retrievals from a widely used scattering model – the Random Volume over Ground (RVoG) model using ancillary topographic data from simulated GEDI observations of surface elevation. Our study site is a mountainous, mixed-temperate forest: Hubbard Brook Experimental Forest (HBEF). We started with a wall-to-wall lidar data set acquired by the Land Vegetation and Ice Sensor (LVIS) that provides a close analogue to anticipated GEDI waveforms. We derived a reference canopy height map and a reference bare earth digital terrain model (DTM) using LVIS. We next simulated GEDI ground tracks over HBEF for the nominal one-year period and extracted these observations from the reference DTM. A series of experiments were then conducted to examine the impact of ancillary topographic information. Using two different sets of TDX acquisitions, we compared height from RVoG respectively using no external DTM, the full LVIS DTM, and the DTM derived from simulated GEDI data against reference canopy heights at 90 m spatial resolution. With no external DTM to remove the ground phase, the RVoG model estimated heights with the best RMSE error (of the two TDX acquisitions dates) of 4.3 m and a bias of 2.5 m. Using the full LVIS DTM, results improved to 3.5 m RMSE and a bias of 1.3 m. Using the simulated GEDI DTM, the RMSE was 4.6 m with a bias of 1.8 m. The agreement between predicted and actual heights was good, ranging from an r 2 of 0.39 (GEDI DTM) to 0.71 ( p -value < 0.0001 for all r 2 ) (full resolution DTM). We conclude that for our study area, TDX data used with the RVoG model was an effective means for estimating spatially continuous canopy structure. However, the large biases in height estimation require ancillary topographic data, such as those produced from GEDI, to reduce biases to acceptable levels. [ABSTRACT FROM AUTHOR]

• We compared hybrid and hierarchical model-based (HMB) biomass predictors. • For small study areas, HMB performed best in terms of mean square error. • For large study areas, hybrid inference should be preferred. • The study was intended to provide guidelines for NASA's GEDI mission. NASA's Global Ecosystem Dynamics Investigation (GEDI) mission offers data for temperate and pan-tropical estimates of aboveground forest biomass (AGB). The spaceborne, full-waveform LiDAR from GEDI provides sample footprints of canopy structure, expected to cover about 4% of the land area following two years of operation. Several options are available for estimating AGB at different geographical scales. Using GEDI sample data alone, gridded biomass predictions are based on hybrid inference which correctly propagates errors due to the modeling and accounts for sampling variability, but this method requires at least two GEDI tracks in the area of interest. However, there are significant gaps in GEDI coverage and in some areas of interest GEDI data may need to be combined with other wall-to-wall remotely sensed (RS) data, such as those from multispectral or SAR sensors. In these cases, we may employ hierarchical model-based (HMB) inference that correctly considers the additional model errors that result from relating GEDI data to the wall-to-wall data. Where predictions are possible from both hybrid and HMB inference the question arises which framework to choose, and under what circumstances? In this paper, we make progress towards answering these questions by comparing the performance of the two prediction frameworks under conditions relevant for the GEDI mission. Conventional model-based (MB) inference with wall-to-wall TanDEM-X data was applied as a baseline prediction framework, which does not involve GEDI data at all. An important feature of the study was the comparison of AGB predictors in terms of both standard deviation (SD: the square root of variance) and root mean square error (RMSE: the square root of mean square error – MSE). Since, in model-based inference, the true AGB in an area of interest is a random variable, comparisons of the performance of prediction frameworks should preferably be made in terms of their RMSEs. However, in practice only the SD can be estimated based on empirical survey data, and thus it is important also to study whether or not the difference between the two uncertainty measures is small or large under conditions relevant for the GEDI mission. Our main findings were that: (i) hybrid and HMB prediction typically resulted in smaller RMSEs than conventional MB prediction although the difference between the three frameworks in terms of SD often was small; (ii) in most cases the difference between hybrid and HMB inference was small in terms of both RMSE and SD; (iii) the RMSEs for all frameworks was substantially larger than the SDs in small study areas whereas the two uncertainty measures were similar in large study areas, and; (iv) spatial autocorrelation of model residual errors had a large effect on the RMSEs of AGB predictors, especially in small study areas. We conclude that hybrid inference is suitable in most GEDI applications for AGB assessment, due to its simplicity compared to HMB inference. However, where GEDI data are sparse HMB inference should be preferred. [ABSTRACT FROM AUTHOR]