Johannes Rahlf

Research Scientist

(+47) 974 90 945

Ås H8

Visiting address
Høgskoleveien 8, 1433 Ås


The effectiveness of generating virtual transects on unmanned aerial vehicle-derived orthomosaics was evaluated in estimating the extent of soil disturbance by severity class. Combinations of 4 transect lengths (5–50 m) and five sampling intensities (1–20 transects per ha) were used in assessing traffic intensity and the severity of soil disturbance on six post-harvest, cut-to-length (CTL) clearfell sites. In total, 15% of the 33 ha studied showed some trace of vehicle traffic. Of this, 63% of was categorized as light (no visible surface disturbance). Traffic intensity varied from 787 to 1256 m ha−1, with a weighted mean of 956 m ha−1, approximately twice the geometrical minimum achievable with CTL technology under perfect conditions. An overall weighted mean of 4.7% of the total site area was compromised by severe rutting. A high sampling intensity, increasing with decreasing incidence of soil disturbance, is required if mean estimation error is to be kept below 20%. The paper presents a methodology that can be generally applied in forest management or in similar land-use evaluations.


In many applications, estimates are required for small sub-populations with so few (or no) sample plots that direct estimators that do not utilize auxiliary variables (e.g. remotely sensed data) are not applicable or result in low precision. This problem is overcome in small area estimation (SAE) by linking the variable of interest to auxiliary variables using a model. Two types of models can be distinguished based on the scale on which they operate: i) Unit-level models are applied in the well-known area-based approach (ABA) and are commonly used in forest inventories supported by fine-resolution 3D remote sensing data such as airborne laser scanning (ALS) or digital aerial photogrammetry (AP); ii) Area-level models, where the response is a direct estimate based on a sample within the domain and the explanatory variables are aggregated auxiliary variables, are less frequently applied. Estimators associated with these two model types can make use of sample plots within domains if available and reduce to so-called synthetic estimators in domains where no sample plots are available. We used both model types and their associated model-based estimators in the same study area with AP data as auxiliary variables. Heteroscedasticity, i.e. for continuous dependent variables typically an increasing dispersion of re- siduals with increasing predictions, is often observed in models linking field- and remotely sensed data. This violates the model assumption that the distribution of the residual errors is constant. Complying with model assumptions is required for model-based methods to result in reliable estimates. Addressing heteroscedasticity in models had considerable impacts on standard errors. When complying with model assumptions, the precision of estimates based on unit-level models was, on average, considerably greater (29%–31% smaller standard errors) than those based on area-level models. Area-level models may nonetheless be attractive because they allow the use of sampling designs that do not easily link to remotely sensed data, such as variable radius plots.


The use of digital aerial photogrammetry (DAP) for forest inventory purposes has been widely studied and can produce comparable accuracy compared with airborne laser scanning (ALS) in small, homogeneous areas. However, the accuracy of DAP for large scale applications with heterogeneous terrain and forest vegetation has not yet been reported. In this study we examined the accuracy of timber volume, biomass and basal area prediction models based on DAP and national forest inventory (NFI) data on a large area in central Norway. Two separate point clouds were derived from aerial image acquisitions of 2010 and 2013. Vegetation heights were extracted by subtracting terrain elevation derived from ALS. A large number of NFI sample plots (483) measured between 2010 and 2014 were used as reference data to fit linear models for timber volume, biomass and basal area with height metrics derived from the DAP data as explanatory variables. Variables describing the heterogeneous environmental and image acquisition conditions were calculated and their influence on the model accuracy was tested. The results showed that forest parameter prediction using DAP works well when applied to a large area. The model fits of the timber volume, biomass and basal area models were good with R2 of 0.80, 0.81, 0.81 and RMSEs of 41.43 m3 ha−1 (55% of the mean observed value), 32.49 t ha−1 (47%), 5.19 m2 ha−1 (41%), respectively. Only a small proportion of the variation could be attributed to the heterogeneous conditions. The inclusion of the relative sun inclination led to an improvement of the model RMSEs by 2% of the mean observed values. The relatively low cost and stability across large areas make DAP an attractive source of auxiliary information for large scale forest inventories.