Estimating vegetation height and canopy cover from remotely sensed data with machine learning

Stojanova, Daniela; Panov, Panče; Gjorgjioski, Valentin; Kobler, Andrej; Džeroski, Sašo

doi:10.1016/j.ecoinf.2010.03.004

Cited by 142 publications

(89 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As the target variable, arithmetic mean heights at the plot level were calculated from individual tree-based height data measured from field-surveyed plots using Equation (1). Machine learning approaches, which have been used for various purposes in remote sensing fields [28][29][30][31][32][33][34][35], were used to estimate the forest plot height from the 20 input variables. In forestry-focused remote sensing, machine learning techniques have been commonly applied to estimation of forest parameters such as biomass and leaf area index, tree species classification, and individual tree characterization [36][37][38].…”

Section: Methodsmentioning

confidence: 99%

Machine Learning Approaches for Estimating Forest Stand Height Using Plot-Based Observations and Airborne LiDAR Data

Lee

Kim

et al. 2018

Forests

View full text Add to dashboard Cite

Effective sustainable forest management for broad areas needs consistent country-wide forest inventory data. A stand-level inventory is appropriate as a minimum unit for local and regional forest management. South Korea currently produces a forest type map that contains only four categorical parameters. Stand height is a crucial forest attribute for understanding forest ecosystems that is currently missing and should be included in future forest type maps. Estimation of forest stand height is challenging in South Korea because stands exist in small and irregular patches on highly rugged terrain. In this study, we proposed stand height estimation models suitable for rugged terrain with highly mixed tree species. An arithmetic mean height was used as a target variable. Plot-level height estimation models were first developed using 20 descriptive statistics from airborne Light Detection and Ranging (LiDAR) data and three machine learning approaches-support vector regression (SVR), modified regression trees (RT) and random forest (RF). Two schemes (i.e., central plot-based (Scheme 1) and stand-based (Scheme 2)) for expanding from the plot level to the stand level were then investigated. The results showed varied performance metrics (i.e., coefficient of determination, root mean square error, and mean bias) by model for forest height estimation at the plot level. There was no statistically significant difference among the three mean plot height models (i.e., SVR, RT and RF) in terms of estimated heights and bias (p-values > 0.05). The stand-level validation based on all tree measurements for three selected stands produced varied results by scheme and machine learning used. It implies that additional reference data should be used for a more thorough stand-level validation to identify statistically robust approaches in the future. Nonetheless, the research findings from this study can be used as a guide for estimating stand heights for forests in rugged terrain and with complex composition of tree species.

show abstract

Section: Methodsmentioning

confidence: 99%

Machine Learning Approaches for Estimating Forest Stand Height Using Plot-Based Observations and Airborne LiDAR Data

Lee

Kim

et al. 2018

Forests

View full text Add to dashboard Cite

show abstract

“…Canopy closure (CC) is defined as the percentage of ground within 30 mˆ30 m, covered by the vertical projection of the overlying trees' crown [21]. The CC was visually estimated using a digital aerial photograph acquired in 2006 with 25 cm spatial resolution.…”

Section: Field Datamentioning

confidence: 99%

Estimation of Alpine Forest Structural Variables from Imaging Spectrometer Data

et al. 2015

View full text Add to dashboard Cite

Spatial information of forest structural variables is crucial for sustainable forest management planning, forest monitoring, and the assessment of forest ecosystem productivity. We investigate a complex alpine forest ecosystem located in the Swiss National Park (SNP) and apply empirical models to retrieve the structural variables canopy closure, basal area, and timber volume at plot scale. We used imaging spectrometer (IS) data from the Airborne Prism EXperiment (APEX) in combination with in-situ measurements of forest structural variables to develop empirical models. These models are based on simple and stepwise multiple regressions, while all potential two narrow-band combinations of the Simple Ratio (SR), the Normalized Difference Vegetation Index (NDVI), the perpendicular vegetation index (PVI), the second soil-adjusted vegetation index (SAVI2), and band depth indices were tested. The accuracy of the estimated structural attributes was evaluated using a leave-one-out cross-validation technique. Using stepwise multiple regression models, we obtained a moderate to good accuracy when estimating canopy closure (R 2 = 0.81, rRMSE = 10%), basal area (R 2 = 0.68, rRMSE = 20%), and timber volume (R 2 = 0.73, rRMSE = 22%). We discuss the reliability of empirical approaches for estimates of canopy structural parameters considering the causality of light interaction and surface information.

show abstract

“…Furthermore, the ease of application (e.g., only two model parameters, see Section 2.3.1) and the ability to run efficiently over large datasets makes random forest an ideal choice for large area attribution [32]. A number of studies have utilized random forest for mapping forest attributes with remotely sensed data, including biomass [28,33], species extent [34], forest extent [21,35,36], canopy cover [7,37,38] and canopy height [24,26,[38][39][40][41].…”

Section: Introductionmentioning

confidence: 99%

Mapping Forest Canopy Height Across Large Areas by Upscaling ALS Estimates with Freely Available Satellite Data

et al. 2015

View full text Add to dashboard Cite

Operational assessment of forest structure is an on-going challenge for land managers, particularly over large, remote or inaccessible areas. Here, we present an easily adopted method for generating a continuous map of canopy height at a 30 m resolution, demonstrated over 2.9 million hectares of highly heterogeneous forest (canopy height 0-70 m) in Victoria, Australia. A two-stage approach was utilized where Airborne Laser Scanning (ALS) derived canopy height, captured over ~18% of the study area, was used to train a regression tree ensemble method; random forest. Predictor variables, which have a global coverage and are freely available, included Landsat Thematic Mapper (Tasselled Cap OPEN ACCESSRemote Sens. 2015, 7 2 transformed), Moderate Resolution Imaging Spectroradiometer Normalized Difference Vegetation Index time series, Shuttle Radar Topography Mission elevation data and other ancillary datasets. Reflectance variables were further processed to extract additional spatial and temporal contextual and textural variables. Modeled canopy height was validated following two approaches; (i) random sample cross validation, and (ii) with 108 inventory plots from outside the ALS capture extent. Both the cross validation and comparison with inventory data indicate canopy height can be estimated with a Root Mean Square Error (RMSE) of ≤ 31% (~5.6 m) at the 95th percentile confidence interval. Subtraction of the systematic component of model error, estimated from training data error residuals, rescaled canopy height values to more accurately represent the response variable distribution tails e.g., tall and short forest. Two further experiments were carried out to test the applicability and scalability of the presented method. Results suggest that (a) no improvement in canopy height estimation is achieved when models were constructed and validated for smaller geographic areas, suggesting there is no upper limit to model scalability; and (b) training data can be captured over a small percentage of the study area (~6%) if response and predictor variable variance is captured within the training cohort, however RMSE is higher than when compared to a stratified random sample.

show abstract

Estimating vegetation height and canopy cover from remotely sensed data with machine learning

Cited by 142 publications

References 20 publications

Machine Learning Approaches for Estimating Forest Stand Height Using Plot-Based Observations and Airborne LiDAR Data

Machine Learning Approaches for Estimating Forest Stand Height Using Plot-Based Observations and Airborne LiDAR Data

Estimation of Alpine Forest Structural Variables from Imaging Spectrometer Data

Mapping Forest Canopy Height Across Large Areas by Upscaling ALS Estimates with Freely Available Satellite Data

Contact Info

Product

Resources

About