Florian Hartig

← Publications

Estimating over- and understorey canopy density of temperate mixed stands by airborne LiDAR data

Hooman Latifi, Marco Heurich, Florian Hartig, Jörg Müller, Peter Krzystek, Hans Jehl & Stefan Dech

Forestry, 89(1), 69–81 (2016)
Cite this
@article{latifi2015assessing,
  author = {Hooman Latifi and Marco Heurich and Florian Hartig and Jörg Müller and Peter Krzystek and Hans Jehl and Stefan Dech},
  title = {Estimating over- and understorey canopy density of temperate mixed stands by airborne LiDAR data},
  journal = {Forestry},
  volume = {89},
  number = {1},
  pages = {69--81},
  year = {2016},
  doi = {10.1093/forestry/cpv032},
}

DOI: 10.1093/forestry/cpv032
Citation statistics: 96 (Google Scholar) · 75 (OpenAlex), as of 29 October 2026

Google Scholar citations per year, as of 20 October 2026 (yearly counts are fetched separately from the total)

View article (DOI)

Abstract

Estimating forest structural attributes is one of the essential forestry-related remote sensing applications. The methods applied so far typically concentrate on the structure of the overstorey. For various conservation and management applications, however, information on lower layers is also of great interest. Detecting understorey cover by remote sensing is challenging, as passive sensors do not penetrate to the forest ground layer. An alternative to these is 3D metrics from active light detection and ranging (LiDAR). Here, we evaluate this technique for describing the vegetation density of multiple stand layers within the temperate stands of a large protected area in south-eastern Germany. We combined LiDAR metrics and information on forest habitat types with regression models to investigate LiDAR metrics that are significantly correlated with vegetation density. The top canopy and the herbal layer showed strong correlations with the applied LiDAR metrics, whereas the predictive power was lower for the intermediate stand layers. Moreover, our results suggest that the relationship between LiDAR predictors and vegetation density depends on the forest type. A comparison of the regression models with random forest predictions showed no major improvement in predictive error. In conclusion, this study highlights the value of the LiDAR metrics for characterizing the structural properties of lower forest layers, which has implications for wildlife and forest management applications, especially in protected areas.

What the paper shows and why it matters

(AI-generated)

Remote sensing of forest structure has typically concentrated on the overstorey, the upper tree canopy. For wildlife, conservation and management the layers beneath it, down to the herb layer, matter too, but passive sensors do not see through the canopy to them. Airborne LiDAR, which records where laser pulses are reflected within the vegetation, can in principle reach them. Yet its signal reflects local leaf density, whereas field ecologists record the proportion of ground covered by each layer, and earlier studies with discrete-return LiDAR reported limited power for shrubs. The paper therefore asks how well LiDAR metrics, including those from waveform systems that record the full return of each pulse, describe cover in several vertical layers, and whether that relationship holds in all forest types.

The paper shows that LiDAR metrics describe vegetation cover in the top tree layer and the herb layer better than in the intermediate tree and shrub layers and, the results suggest, that the link between LiDAR metrics and cover differs between forest habitat types. It tests this in the Bavarian Forest National Park on up to 166 field plots where cover had been estimated by eye in five height layers, with waveform LiDAR flown four years later. Regression models gave r² of 0.63 for the top layer and 0.53 for the herb layer, against 0.19 to 0.40 for the three intermediate layers; under cross-validation the two best values were 0.56 and 0.44. Habitat type was offered to every model as a main effect and through interactions with LiDAR metrics, and the selected models for the top, second tree and herb layers retain it. The same LiDAR signal therefore went with different cover in different habitat types. Random forests gave no major improvement in r² over the regressions.

The paper reads the results as supporting LiDAR-based mapping of top-layer and herb-layer cover, while the intermediate layers carry less LiDAR information. It advises including habitat type as auxiliary information in LiDAR-based models of forest structure. It restricts its results to a protected area and notes that the four-year gap between field and LiDAR data may have affected the fits. Venier et al. (2019) cited it for forest type sometimes affecting model accuracy and compared their shrub-layer result with it.

References