Optimal Model Complexity for Sediment Assessment in Mangrove Restoration: A One-Dimensional Framework Under Data Constraints
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Excessive suspended sediment concentration (SSC) is a primary driver of mangrove seedling dieback in canal networks dug to provide hydrology for mangrove restoration. However, SSC prediction under the strict data gathering capacity constraints of resource-limited and community-based restoration projects remains challenging. This study develops and evaluates a one-dimensional (1-D) advection model across a branching mangrove canal network in Sri Lanka, calibrated against 264 manual samples from 23 instances across four ponds. The model solves steady-state advection and requires only boundary SSC and routine hydraulic inputs. Against withheld observations (n = 96), the calibrated model achieved RMSE = 0.053 g L − 1 , Pearson r = 0.605, and, sensitivity = 84.6%, correctly identifying approximately four in five instances that exceeded an empirically validated 0.08 g L − 1 SSC dieback threshold. In contrast, higher-order and data-driven models spanning the complexity spectrum failed at classification despite comparable RMSEs. Evaluated alternatives were: explicit FOCDMS, implicit TDMA, rating curve, radial basis function, and ensemble machine learning. Both FOCDMS and TDMA models failed due to grid Péclet numbers of O(10 4 –10 5 ) at manual sampling spacings. Data-driven models failed through geometric overfitting under leave-one-plot-out cross-validation, with Random Forest sensitivity falling to 34.6% and Pearson r=-0.025. The results demonstrate an analytically locatable optimal model complexity that is sufficient to encode governing physics without requiring data volumes unavailable through manual data gathering. Further, the developed sampler is deployable at USD 43.50, a 96.5% cost reduction relative to conventional instrumentation. The framework provides a scoping-level tool that enables sediment-aware restoration under data-constraints, transitioning away from “design by intuition”.
Fields of Study & Subjects
Coastal wetland ecosystem dynamics
Aquatic and Environmental Studies
Coastal and Marine Dynamics
Publication details
- Journal / source
- Research Square
- Publication date
- 30 March 2026
- Language
- EN
- Document type
- Preprint