Predictive model for coastal subsidence and elevation change

Columbia University collaborators Chadwick and Steckler built a predictive model for coastal subsidence and elevation change (Fig. 1). The model was designed to resolve observed variations and disentangle the effects of climate change and human activities. The model is built upon fundamental principles of effective stress, conservation of mass, and Darcy flow as well as constitutive relations for sediment porosity and shallow-subsurface ecology (e.g., tree roots, animal burrows) (Fig. 2). As input, the model uses stratigraphic data for the recent (Holocene) sedimentary deposits, near-surface soil properties, and sedimentation rate for a designated site. The model generates a time-series of subsidence and surface-elevation change, which can be directly compared to field measurements and projected into the future.

Figure 1. Overview of the model framework, including A) key model inputs of vertical accretion, water-table depth, surface ecology, and Holocene stratigraphy; B) schematic of the one-dimensional vertical framework; C) model-output depth-profile of the subsidence rate; and D) the model’s depth intervals for comparison to field observations. 

Figure 2. An overview of the fundamental principles that govern the subsidence model.

The team validated the model region of Polder 32 in coastal Bangladesh, which boasts a dense clustering of field measurements (Fig. 3). Results show the model can accurately resolve twenty-first-century subsidence dynamics within +/- 5 mm/yr., including local amplification of subsidence hazards in response to land use and climate change.


Figure 3. A) Predicted versus measured rates of subsidence from GNSS (red circles), strainmeters (blue squares), and SETs (dark yellow triangles), as well as SET-derived surface elevation change (light yellow triangles), underlain with line of perfect agreement (thick black) and zone of agreement within +/- mm/yr (light gray). Horizontal error bars show reported measurement precision (Table 2). Vertical error bars show 90% confidence interval for model predictions propagated from uncertainty in model-input parameters (Tables 1–2; Section 3.1). B) Box plots showing mean, median, 25–75% range, and 0–100% range of discrepancies between mean predicted and observed rates from Panel (A) for each instrument type. Note subsidence is defined positive-downwards, and surface elevation change is positive-upwards. 

A manuscript presenting the model and its publicly available, open-source components is in preparation for publication at a peer-reviewed scientific journal. The outcome of this work will allow for the generation of new, dynamic Digital Elevation Models (DEMs), which can be utilized across the Jameel Observatory CREWSnet efforts to improve predictions of where water will flow during floods and where waterlogging will occur during intense rains (Fig. 4).

Figure 4. Estimated height of embankments needed to defend against a 3-m coastal storm surge by 2100 before (left) and after (right) accounting for the effect of subsidence using the model. Modeling subsidence allows us to more accurately predict future flood hazards, and potential hazard hotspots (red areas) associated with land-use and climate changes. 

Watch JO-CREWSnet PI Austin Chadwick share more about the predictive model for subsidence in the video below.