Perhaps we should re-consider the status quo? A new article showing how we can improve the efficiency and transparency of land use model calibration.
Over the last decade the methods applied to modelling Land Use and Land Cover (LULC) change have become increasingly complex, especially with the mainstreaming of machine learning techniques to capture relationships from data. On the one hand this heightened complexity can allow for more realistic outputs but at the same time it can make the process of confirming the accuracy of these models more time-consuming and less transparent.
For cellular automata type models in particular, accuracy is typically tested only by comparing the simulated maps of LULC change against real maps instead of directly assessing the probabilistic predictions of change generated by the model. In our new paper we highlight the limitations of this status quo approach to assessing models and demonstrate that focusing on probabilistic predictions allows modellers to more efficiently test different specifications as well as gain insights into model behaviour that are not possible from map comparison alone.
The full paper is published Open Access in Environmental Modelling and Software:
Black B., van Strien M., Adde A., Grêt-Regamey, A. (2023) Re-considering the status quo: Improving calibration of land use change models through validation of transition potential predictions. Environmental Modelling & Software, external page 159. https://doi.org/10.1016/j.envsoft.2022.105574.
And the research data and scripts will also be made available:
external page https://doi.org/10.5281/zenodo.6912914