Ana María de Lera Garrido defended her PhD-thesis, “Application of external prediction models in forest management inventories based on airborne laser scanning”, on May 6, 2022.
The title of the trial lecture was: “The use of 3D data from satellite borne sensors in forest applications.”
Summary of the thesis
The use of airborne laser scanning (ALS) data in forest inventories has become operational during the last decades. Since it was introduced, the use of these data has improved forest inventory practices due to the capability of the ALS to describe the three-dimensional structure of the forests. ALS-assisted inventories rely on predictive models, traditionally constructed and applied on data from the same area of interest. In operational forest inventory, old data are usually discarded once new data are acquired. However, data from the past or from other inventories (external data) could potentially be used to obtain up-to-date forest attributes predictions. This thesis reports results from analyses of the use of external data for ALS-assisted forest management inventories as an approach to reduce the field data acquisition requirements and potentially reduce the inventory cost.
The first study presented two approaches to construct forest attribute predictive models based entirely on temporally external field data: (i) external models, constructed from field and ALS data from a previous inventory in the same area, and (ii) forecasted models, for which the field data from the previous inventory are forecasted to the present and used in combination with up-to-date ALS data to construct prediction models. Although both approaches produced adequate predictions judged in terms of the relative root mean squared error (RMSE%) of the predictions, the relative mean differences (MD%) were generally high, and the accuracy of the external models were better than the forecasted models. In the second study, the analyses of using external models continued by locally applying regional Norwegian models constructed with National Forest inventory data. The study went a step further compared to the first study, by applying the models over 33 study areas distributed across southern Norway, and to analyze the effect of climate, topography, forest condition, and other factors on the size of the prediction errors. The range of MD% and RMSE% values between the study areas was broad, and variables characterizing the forest conditions explained the largest proportion of the variation in prediction error. In the third study, three different approaches were used to calibrate the external model predictions using a small number of up-to-date, local field plots. Although the calibration was not able to eliminate all systematic deviations, the results showed that the calibrated predictions using 20 plots were at the same level of accuracy as those resulting from a new inventory.
The studies reported in this thesis has shown the potential of using external data to reduce the cost of forest inventories without reducing the accuracy. Even though further research is still necessary, the work that led to this thesis showed that prediction accuracies in line with those of an ordinary forest management inventory can be achieved, by means of procedures for calibrating external predictions using a reduced number of observations from up‑to‑date field plots from the inventory area. The analyses also revealed several drawbacks related to the calibration approaches that were tested, that need to be overcome, and the results of this thesis constitute an important step further for the operational use of external models in future forest management inventories.
Supervisors were
Ole Martin Bollandsås (NMBU)
Erik Næsset (NMBU)
Terje Gobakken (NMBU)
Hans Ole Ørka (NMBU)
The evaluation committee were
Mª Teresa Lamelas Gracía, Centro Universitario de la Defensa, Academia General Militar, Spain
Mats Nilsson, Department of Forest Resource Management, Swedish University of Agricultural Sciences , Sweden
Maarit Kallio NMBU/MINA