We developed a new algorithm for COntinuous monitoring of Land Disturbance (COLD) using Landsat time series. COLD can detect many kinds of land disturbance continuously as new images are collected and provide historical land disturbance maps retrospectively. To better detect land disturbance, we tested different kinds of input data and explored many time series analysis techniques. We have several major observations as follows. First, time series of surface reflectance provides much better detection results than time series of Top-Of- Atmosphere (TOA) reflectance, and with some adjustments to the temporal density, time series from Landsat Analysis Ready Data (ARD) is better than it is from the same Landsat scene. Second, the combined use of spectral bands is always better than using a single spectral band or index, and if all the essential spectral bands have been employed, the inclusion of other indices does not further improve the algorithm performance. Third, the remaining outliers in the time series can be removed based on their deviation from model predicted values based on probability-based thresholds derived from normal or chi-squared distributions. Fourth, model initialization is pivotal for monitoring land disturbance, and a good initialization stability test can influence algorithm performance substantially. Fifth, time series model estimation with eight coefficients model, updated for every single observation, based on all available clear observations achieves the best result. Sixth, a change probability of 0.99 (chi-squared distribution) with six consecutive anomaly observations and a mean included angle < 45° to confirm a change provide the best results, and the combined use of temporally-adjusted Root Mean Square Error (RMSE) and minimum RMSE is recommended. Finally, spectral changes (or “breaks”) contributed from vegetation regrowth should be excluded from land disturbance maps. The COLD algorithm was developed and calibrated based on all these lessons learned above. The accuracy assessment shows that COLD results were accurate for detecting land disturbance, with an omission error of 27% and a commission error of 28%.
In interior Alaska’s 115 million acres of boreal forest, white and black spruce are the dominant tree species. Climate models suggest that the region is becoming warmer and drier, resulting in declining growth of black and white spruce, according to some researchers. These drier conditions also may lead to greater risk of stand-replacing wildfires, resulting in forests dominated by birch and aspen, which are early-successional tree species.
To compare long-term growth trends of the dominant coniferous and deciduous tree species, a team of researchers with the USDA Forest Service Pacific Northwest Research Station and the University of Alaska Anchorage analyzed tree cores collected from the Tanana Valley and measured tree-ring widths of these four tree species over the past 150 years. They also compared growth against monthly temperature and precipitation data to determine if there is a correlation between climate and growth.
The team found that white and black spruce have not experienced as rapid a growth decline as earlier studies suggested; instead, their annual growth remains near the long-term mean. Of the four species examined, aspen showed the greatest recent growth decline, likely reflecting a widespread insect outbreak. Among the climate variables that will affect the future growth of these species, summer rainfall was identified as a significant factor.
The United States national inventory program measures a subset of tree heights in each plot in the Pacific Northwest. Unmeasured tree heights are predicted by adding the difference between modeled tree heights at two measurements to the height observed at the first measurement. This study compared different approaches for directly modeling 10-year height increment of red alder (RA) and ponderosa pine (PP) in Washington and Oregon using national inventory data from 2001–2015. In addition to the current approach, five models were implemented: nonlinear exponential, log-transformed linear, gamma, quasi-Poisson, and zero-inflated Poisson models using both tree-level (e.g., height, diameter at breast height, and compacted crown ratio) and plot-level (e.g., basal area, elevation, and slope) measurements as predictor variables. To account for negative height increment observations in the modeling process, a constant was added to shift all response values to greater than zero (log-transformed linear and gamma models), the negative increment was set to zero (quasi-Poisson and zero-inflated Poisson models), or a nonlinear model, which allows negative observations, was used. Random plot effects were included to account for the hierarchical data structure of the inventory data. Predictive model performance was examined through cross-validation. Among the implemented models, the gamma model performed best for both species, showing the smallest root mean square error (RSME) of 2.61 and 1.33 m for RA and PP, respectively (current method: RA—3.33 m, PP—1.40 m). Among the models that did not add the constant to the response, the quasi-Poisson model exhibited the smallest RMSE of 2.74 and 1.38 m for RA and PP, respectively. Our study showed that the prediction of tree height increment in Oregon and Washington can be improved by accounting for the negative and zero height increment values that are present in inventory data, and by including random plot effects in the models.
There is growing interest in using Digital Aerial Photogrammetry (DAP) for forestry applications. However, the performance of pushbroom DAP relative to frame-based DAP and airborne lidar is not well documented. Interest in DAP stems largely from its low cost relative to lidar. Studies have demonstrated that frame-based DAP generally performs slightly poorer than lidar, but still provides good value due to its reduced cost. In the USA pushbroom imagery can be dramatically less expensive than frame-camera imagery in part because of a nationwide collection program. There is an immediate need then to understand how well pushbroom DAP works as an auxiliary data source in the prediction of key forest attributes including basal area, volume, height, and the number of trees per ha. This study compares point clouds generated from 40 cm pushbroom DAP with point clouds from lidar and 7.5 cm, 15 cm, and 30 cm frame-based DAP. Differences in point clouds from these data sources are readily apparent in visual inspections; e.g. DAP tends to measure canopy gaps poorly, omit individual trees in openings, is typically unable to represent the ground beneath canopy, and is susceptible to commission errors manifested as points above the canopy surface. Frame-based DAP provides greater canopy detail than pushbroom DAP, which becomes more apparent with higher image resolution. Our results indicated that DAP height metrics generally have a strong linear relationship with lidar metrics, with R2 values ranging from 83 – 90% for cover, and 47–80% for height quantiles. Similarly, lidar auxiliary variables explain the greatest variation in forest attributes, e.g., volume (84%), followed closely by 30 cm frame-based DAP (81%), with the poorest results from pushbroom DAP (75%). While DAP resolution had a visible effect on canopy definition, it did not appreciably affect point cloud metrics or model performances. Although pushbroom DAP explained the least variation in forest attributes, it still had sufficient explanatory power to provide good value when frame-based DAP and lidar are not available.
Warming in arctic and boreal regions is increasing shrub cover and biomass. In southcentral Alaska, willow (Salix spp.) and alder (Alnus spp.) shrubs grow taller than many tree species and account for a substantial proportion of aboveground biomass, yet they are not individually measured as part of the operational Forest Inventory and Analysis (FIA) Program. The goal of this research was to test methods for landscape-scale mapping of tall shrub biomass in upper montane and subalpine environments using FIA-type plot measurements (n = 51) and predictor variables from imagery-based structure-from-motion (SfM) and airborne lidar. Specifically, we compared biomass models constructed from imagery acquired by unmanned aerial vehicle (UAV; ~1.7 cm pixels), imagery from the NASA Goddard's Lidar, Hyperspectral, and Thermal Airborne Imager (G-LiHT; ~3.1 cm pixels), and concomitant G-LiHT small-footprint lidar. Tall shrub biomass was most accurately predicted at 5 m resolution (R2 = 0.81, RMSE = 1.09 kg m−2) using G-LiHT SfM color and structure variables. Lidar-only models had lower precision (R2 = 0.74, RMSE = 1.26 kg m−2), possibly due to reduced model information content from variable multicollinearity or lower data density. Separate models for upper montane zones with trees and shrubs and subalpine zones with only shrubs were always chosen over single models based on minimization of Akaike's Information Criterion, indicating the need for variable sets robust to overhanging tree canopy. Decreasing point density from UAV (5000–8000 pts. m−2) to the G-LiHT SfM point cloud (500–2000 pts. m−2) had little impact on model fit, suggesting that high-resolution airborne imagery can extend SfM approaches well beyond line-of-sight restrictions for UAV platforms. Overall, our results confirmed that SfM from high-resolution imagery is a viable approach to estimate shrub biomass in the boreal region, especially when an existing lidar terrain model and local field calibration data are available to quantify uncertainty in the SfM point cloud and landscape-scale estimates of shrub biomass.
Understanding change in forest carbon (C) is important for devising strategies to reduce emissions of greenhouse gases. National forest inventories (NFIs) are important to meet international accounting goals, but data are often incomplete going back in time, and the amount of time between remeasurements can make attribution of C flux to specific events difficult. The long time series of Landsat imagery provides spatially comprehensive, consistent information that can be used to fill the gaps in ground measurements with predictive models. To evaluate such models, we relate Landsat spectral changes and disturbance interpretations directly to C flux measured on NFI plots and compare the performance of models with and without ground-measured predictor variables. The study was conducted in the forests of southwest Oregon State, USA, a region of diverse forest types, disturbances, and landowner management objectives. Plot data consisted of 676 NFI plots with remeasured individual tree data over a mean interval (time 1 to time 2) of 10.0 years. We calculated change in live aboveground woody carbon (AWC), including separate components of growth, mortality, and harvest. We interpreted radiometrically corrected annual Landsat images with the TimeSync (TS) tool for a 90 m X 90 m area over each plot. Spectral time series were divided into segments of similar trajectories and classified as disturbance, recovery, or stability segments, with type of disturbance identified. We calculated a variety of values and segment changes from tasseled cap angle and distance (TCA and TCD) as potential predictor variables of C flux. Multiple linear regression was used to model AWC and net change in AWC from the TS change metrics. The TS attribution of disturbance matched the plot measurements 89% of the time regarding whether fire or harvest had occurred or not. The primary disagreement was due to plots that had been partially cut, mostly in vigorous stands where the net change in AWC over the measurement was positive in spite of cutting. The plot-measured AWC at time 2 was 86.0 ± 78.7 Mg C ha−1 (mean and standard deviation), and the change in AWC across all plots was 3.5 ± 33 Mg C ha−1 year−1. The best model for AWC based solely on TS and other mapped variables had an R2 = 0.52 (RMSE = 54.6 Mg C ha−1); applying this model at two time periods to estimate net change in AWC resulted in an R2 = 0.25 (RMSE = 28.3 Mg ha−1) and a mean error of −5.4 Mg ha−1. The best model for AWC at time 2 using plot measurements at time 1 and TS variables had an R2 = 0.95 (RSME = 17.0 Mg ha−1). The model for net change in AWC using the same data was identical except that, because the variable being estimated was smaller in magnitude, the R2 = 0.73. All models performed better at estimating net change in AWC on TS-disturbed plots than on TS-undisturbed plots. The TS discrimination of disturbance between fire and harvest was an important variable in the models because the magnitude of spectral change from fire was greater for a given change in AWC. Regional models without plot-level predictors produced erroneous predictions of net change in AWC for some of the forest types. Our study suggests that, in spite of the simplicity of applying a single carbon model to multiple image dates, the approach can produce inaccurate estimates of C flux. Although models built with plot-level predictors are necessarily constrained to making predictions at plot locations, they show promise for providing accurate updates or back-calculations of C flux assessments.
Forest land managers rely on predictions of tree mortality generated from fire behavior models to identify stands for post-fire salvage and to design fuel reduction treatments that reduce mortality. A key challenge in improving the accuracy of these predictions is selecting appropriate wind and fuel moisture inputs. Our objective was to evaluate post-fire mortality predictions using the Forest Vegetation Simulator Fire and Fuels Extension (FVS-FFE) to determine if using representative fire-weather data would improve prediction accuracy over two default weather scenarios. We used pre- and post-fire measurements from 342 stands on forest inventory plots, representing a wide range of vegetation types affected by wildfire in California, Oregon, and Washington. Our representative weather scenarios were created by using data from local weather stations for the time each stand was believed to have burned. The accuracy of predicted mortality (percent basal area) with different weather scenarios was evaluated for all stands, by forest type group, and by major tree species using mean error, mean absolute error (MAE), and root mean square error (RMSE). One of the representative weather scenarios, MeanWind, had the lowest mean error (4%) in predicted mortality, but performed poorly in some forest types, which contributed to a relatively high RMSE of 48% across all stands. Driven in large part by over-prediction of modelled flame length on steeper slopes, the greatest over-prediction mortality errors arose in the scenarios with higher winds and lower fuel moisture. Our results also indicated that fuel moisture was a stronger influence on post-fire mortality than wind speed. Our results suggest that using representative weather can improve accuracy of mortality predictions when attempting to model over a wide range of forest types. Focusing simulations exclusively on extreme conditions, especially with regard to wind speed, may lead to over-prediction of tree mortality from fire.