storymap.js Modelling at different scales
Authors: Miguel Arias, Gabriela Torchio, Hugo Thierry
Introduction
Decision-making and research questions are meaningful at specific scales and often need scale-specific models.
Selecting the appropriate extent and resolution to study natural phenomena poses a significant challenge for scientists (Oreskes et al., 1994 [18]). The chosen extent dictates the area the model can encompass, while the selected spatial resolution determines the level of detail and accuracy the model achieves. Hence, the spatial extent and resolution of a study significantly affects model complexity, accuracy (and/or precision), and generalizability (figure on the right) (Brimicombe, 2010 [19]).
Spatial models must address trade-offs between spatial resolution, complexity, accuracy (and/or precision), and generalizability, especially when applied across varying spatial extents. For models covering large extents (e.g., planetary/global models), simplicity is often favored to maintain generalizability and reduce computational demands. However, this can come at the cost of local accuracy; as such, large extent models may overlook fine-scale processes and heterogeneities. Conversely, models developed for smaller extents (e.g., provincial or municipal models) can afford to be more complex, incorporating detailed biophysical and socio-ecological variables to achieve higher accuracy and capture site-specific dynamics. Yet, these finely tuned models risk reduced generalizability when applied beyond their original context due to overfitting and the presence of unique local factors that do not scale up effectively.
Case study
The human footprint is a cumulative index that quantifies the extent and the intensity of human pressures on ecosystems and biodiversity. We compared the Human footprint at three scales: global (Williams et al., 2020 [20]), national (Hirsh-Pearson et al., 2022 [21]), and provincial [22].
The interactive maps on the right represent the state of human pressures in the northeast of British Columbia. From coarse (Global Human Footprint) to detailed (Provincial Human Footprint) spatial resolution (scales), we can easily identify how roads are represented at different extents. The global model captures a broad perspective of human disturbances in which roads and settlements are the most visible features. The resolution of this global map is improved with the national model (National Human Footprint), which integrates 12 layers of information, providing more levels of detail such as oil and gas infrastructure and mining. The provincial map (Provincial Human Footprint), which incorporates 16 layers of information, shows a more comprehensive state of the human footprint, including seismic lines, oil and gas wells, human infrastructure, and even recreational features. By turning on and off the different spatial layers on the right, differences in spatial resolution can be seen.
The static maps below show the same human footprint models but zoomed in and centered on Fort Nielson, British Columbia. These maps show the same spatial extent for a global (a) [20], a national (b) [21] and a provincial footprint model (c) [22]. A gradient of increased spatial resolution is shown from a coarser global model (a) with less detail to a finer provincial model (c) with more details. The map in (a) shows less detail but the original source map covers the entire globe. The map in (b) shows more details than (a) but the original source map only covers Canada. The map in (c) reveals more details than (b) but the original source map only covers the province of British Columbia. A similar spatial resolution gradient can be replicated in the spatial layers on the right by zooming in and turning on and off the different spatial layers.
Best practices and opportunities
Models are very sensitive to input data, so the more accurate the input data, the higher the reliability of the model. On the other hand, models are developed to be applied at a certain scale (spatial and temporal), and most times they should not be directly applied at other scales. A model that is acceptable at one scale might not be at another. Another important consideration is that model evaluation can be a complex process in itself. Hence, the larger the spatial and temporal scale of the phenomena being studied the more complex and challenging the evaluation of the model.
Regarding these aspects, we suggest the following to improve model reliability:
- Be clear about the purpose of the model and verify that it applies to the scale considered.
- Document all decisions made regarding the input data (sources, decisions made to incorporate it, etc.).
- Create clear and explicit criteria for every step taken in model evaluation. Always document what you did and why.
- Conduct a sensitivity analysis to test the consistency of the model under a range of input parameters.