Canadian Forest Service Publications
Updating stand-level forest inventories using airborne laser scanning and Landsat time series data. 2018. Bolton, D.K., White, J. C., Wulder, M.A., Coops, N.C., Hermosilla, T., Yuan, X. Int J Appl Earth Obs Geoinformation 66, 174–183.
Issued by: Pacific Forestry Centre
Catalog ID: 38996
Availability: PDF (request by e-mail)
Available from the Journal's Web site. †
† This site may require a fee
Vertical forest structure can be mapped over large areas by combining samples of airborne laser scanning (ALS) data with wall-to-wall spatial data, such as Landsat imagery. Here, we use samples of ALS data and Landsat timeseries metrics to produce estimates of top height, basal area, and net stem volume for two timber supply areas near Kamloops, British Columbia, Canada, using an imputation approach. Both single-year and time series metrics were calculated from annual, gap-free Landsat reflectance composites representing 1984–2014. Metrics included long-term means of vegetation indices, as well as measures of the variance and slope of the indices through time. Terrain metrics, generated from a 30 m digital elevation model, were also included as predictors. We found that imputation models improved with the inclusion of Landsat time series metrics when compared to single-year Landsat metrics (relative RMSE decreased from 22.8% to 16.5% for top height, from 32.1% to 23.3% for basal area, and from 45.6% to 34.1% for net stem volume). Landsat metrics that characterized 30-years of stand history resulted in more accurate models (for all three structural attributes) than Landsat metrics that characterized only the most recent 10 or 20 years of stand history. To test model transferability, we compared imputed attributes against ALS-based estimates in nearby forest blocks (> 150,000 ha) that were not included in model training or testing. Landsat-imputed attributes correlated strongly to ALS-based estimates in these blocks (R2 =0.62 and relative RMSE = 13.1% for top height, R2 = 0.75 and relative RMSE =17.8% for basal area, and R2 =0.67 and relative RMSE =26.5% for net stem volume), indicating model transferability. These findings suggest that in areas containing spatially-limited ALS data acquisitions, imputation models, and Landsat time series and terrain metrics can be effectively used to produce wall-to-wall estimates of key inventory attributes, providing an opportunity to update estimates of forest attributes in areas where inventory information is either out of date or non-existent.
Plain Language Summary
Recent advances in mass data processing of Landsat imagery combined with operational applications of airborne laser scanning (ALS) have provided opportunities to utilize both long time series of landscape-level change and detailed forest structure data to augment forest inventories. In particular, sample-based applications of ALS data offer opportunities to fulfill certain strategic forest inventory information needs with lower costs than full wall-to-wall ALS acquisitions. Landsat data can provide the required wall-to-wall coverage of calibrated surface reflectance data that can be combined with samples of ALS data via an imputation approach to provide spatially extensive characterizations of forest vertical structure. These findings suggest that in areas containing discontiguous ALS data, imputation models and Landsat time series and terrain metrics can be effectively used to produce wall-to-wall estimates of key inventory attributes, which provides an opportunity to update estimates of forest attributes in areas where inventory information is either out of date or non-existent.