Team: Göran Ståhl, Terje Gobakken, Svetlana Saarela, Henrik Persson, Magnus Ekström, Sean P. Healey, Zhigiang Yang, Johan Holmgren, Eva Lindberg, Kenneth Nyström, Emanuele Papucci, Patrik Ulvdal, Hans Ole Ørka, Erik Næsset, Zhengyang Hou, Håkan Olson, Ronald E. McRoberts.
Remotely sensed data are frequently used for predicting and mapping ecosystem characteristics, and spatially explicit wall-to-wall information is sometimes proposed as the best possible source of information for decision-making. However, wall-to-wall information typically relies on model-based prediction, and several features of model-based prediction should be understood before extensively relying on this type of information. One such feature is that model-based predictors can be considered both unbiased and biased at the same time, which has important implications in several areas of application.
The authors showed that standard regression or machine learning predictors, which are approximately model-unbiased, are at the same time design-biased. For populations that remain more or less stable across long periods, such as forests, this implies that the same type of systematic errors repeatedly will be observed when forest ecosystem characteristics are assessed using data from a certain RS sensor type. This has important negative implications in several areas of application.
The discussion about the relevance of model-based inference in comparison to design-based inference is far from new. However, conclusions tend to vary depending on area of application. In this article, the authors discussed and highlighted important issues to consider in studies applying RS data for assessing forest ecosystem characteristics. In some cases, calibration to remove design-bias trends of model-based predictors should be considered.