Estimation of crown biomass of Pinus pinaster stands and shrubland above-ground biomass using forest inventory data, remotely sensed imagery and spatial prediction models

Spatially crown biomass of Pinus pinaster stands and shrubland above-ground biomass (AGB) estimation was carried-out in a region located in Centre-North Portugal, by means of different approaches including forest inventory data, remotely sensed imagery and spatial prediction models. Two cover types (pine stands and shrubland) were inventoried and biomass assessed in a total of 276 sample field plots. We compared AGB spatial predictions derived from Direct Radiometric Relationships (DRR) of remotely sensed data; and the geostatistical method Regression-kriging (RK), using remotely sensed data as auxiliary variables. Also, Ordinary Kriging (OK), Universal Kriging (UK), Inverse Distance Weighted (IDW) and Thiessen Polygons estimations were performed and tested. The comparison of AGB maps shows distinct predictions among DRR and RK; and Kriging and deterministic methods, indicating the inadequacy from these later ones to map AGB over large areas. DRR and RK methods produced lower statistical error values, in pine stands and shrubland, when compared to kriging and deterministic interpolators. Since forest landscape is not continuous variable, the tested forest variables showed low spatial autocorrelation, which makes kriging methods unsuitable to these purposes. Despite the geostatistical method RK did not increase the accuracy of estimates developed by DRR, denser sampling schemes and different auxiliary variables should be explored, in order to test if the accuracy of predictions is improved.

► Spatially above-ground biomass estimation throughout forest and shrubland cover was carried-out.
► Inventory data, remotely sensed imagery and spatial prediction models were used.
► Radiometric Relationships, Regression-kriging, IDW, Thiessen Polygons and Kriging methods were evaluated.
► Remote sensing was the most feasible approach to map forest variables.
► Regression-kriging has good potential to improve AGB predictions.