# Hi-res salt intrusion modeling

# Project details:

**Applicant:** Marlein Geraeds
**Faculty:** Civil Engineering
**DCC members** Raúl Ortiz
**Support period** January - September 2024
**Repository/Archive:** https://github.com/mgeraeds/hires-processing

**Research Background**:

Extreme salt intrusion events are becoming more prevalent as a result of climate change. Recent droughts have made this issue very visible. In 2022, salt intrusion as a result of the extreme drought almost caused a shortage of fresh water in The Netherlands. Currently, there is a lack of 3D models that are able to predict salt intrusion accurately. Contributing factors to this are that existing models often don't have enough resolution, don't include the non-hydrostatic effects, or aren't able to represent the physics in both the offshore region and the estuarine region.

**Project Description:**

This project addresses current knowledge gaps by developing a detailed coupled coast-delta model that incorporates parameterisations of non-hydrostatic effects. Issues that may be addressed with the model are the influence of extreme weather conditions, human interventions, and climate change on the frequency and severity of salt intrusion events. Better predictions of salt intrusion are a crucial requirement for short-term salt related decision-making as well as long-term strategies to mitigate the risks of salt intrusion through countermeasures.

The researcher faced different memory usage constraints as datasets have to be stored and transferred, but also processed. Briefly, these datasets are structured according to the D-FLOW-FM model. While this model aims for high-resolution representations of such measurements, their computational requirements have, so far, been higher than the resources available.

**Project goals**

- Test and set different data storage, transfer and compression options.
- Optimize memory usage
- Optimize software performance

## Project Results

- Isolated workflow into a [new repo](https://github.com/mgeraeds/hires-processing):
- Documented with Sphinx
- Implemented branching convention
- Used a project board to track issues
- Generated and used smaller datasets
- Reduced memory usage by selecting variables and reducing float precision
- Parallelized and partitioned I/O operations using Dask
- Optimised resource allocation with Dask configuration files

- Generated a [test environment](https://github.com/rortizmerino/dask-playground) using a modified version of [Remote Sensing Deployable Analysis environmenT (RS-DAT)](https://github.com/RS-DAT/RS-DAT) roginally developed by the eScience Center

- [Documented learning](https://hackmd.io/h7kiVSzLR9-50XU7YgyRFw?view)

- Consulted [Ou Ku](https://www.esciencecenter.nl/team/ou-ku/), eScience Center Research Software Engineer experienced with similar datasets and tools

**Links to output**
- [Software repository](https://github.com/mgeraeds/hires-processing)
- [Dask playground](https://github.com/rortizmerino/dask-playground)
- [Dask knowledgebase](https://hackmd.io/h7kiVSzLR9-50XU7YgyRFw?view)

### Feedback from researcher

>"Allocating time to work together helped focusing"

>"Having a DCC member as collaborator helped understading problems better by explaining them to each other"

>"My expectations were only partially met, was expecting to have more domain and technical knowledge"

>"Seems the DCC support would've worked better for more concrete, simpler tasks"

### Lessons learned
>"I underestimated the complexity of Dask and overestimated my ability to get into it"

>"The domain-specific knowledge (unstructured datasets) did become a bottleneck as the data model is very complex even for domain experts"