SCIMAP Fitted 3: Ensemble Risk Maps¶
The third of the three SCIMAP Fitted tools. Propagates the calibrated land-cover weight sets from SCIMAP Fitted 2 back onto the raster grid and summarises the resulting ensemble as a risk map — so the output carries its own uncertainty rather than presenting a single weight set as the answer.
This is the slow step: it runs one flow-routing pass per land-cover class, per connectivity and land-cover input combination (routing once per class rather than once per weight set is what makes propagating dozens of weight sets affordable). Feed it the optional raster outputs of SCIMAP Fitted 1 so it does not recompute hydrology, and expect tens of minutes on a large grid.
Where to find it: Processing Toolbox → SCIMAP → SCIMAP Fitted 3: Ensemble
Risk Maps; toolbar icon; &SCIMAP menu.
Screenshot:
The Processing dialog with the calibrated weights CSV and connectivity/land-cover raster lists populated.
Parameters¶
| Parameter | Description | Default |
|---|---|---|
| Calibrated weight sets (CSV) | The "Best-fitting weight sets" output of Fitted 2. | — |
| Weight sets to propagate | How many of the calibrated sets to run (0 = all). | 30 |
| Connectivity raster(s) | One or more connectivity rasters. | — |
| Land cover raster(s) | One or more land cover rasters. | — |
| Land cover classification scheme | Same mechanism as elsewhere in the plugin. | — |
| Erosion potential raster (optional) | If omitted, a constant of 1 is used (risk = connectivity × land-cover weight only). | Optional |
| DEM for flow routing (breached) | The DEM used to route the propagated risk. | — |
| Breach depressions first | Tick if supplying a raw (unbreached) DEM. | Off |
| D8 flow accumulation (optional) | Reuse from Fitted 1's optional output to skip recomputing it; enables the river-network outputs. | Optional |
| Rainfall-weighted contributing area (optional) | Reuse from Fitted 1's optional output; without it, routed risk is not diluted and simply grows with catchment area. | Optional |
| Flow routing | D-infinity mass flux (SCIMAP default) or D8 mass flux (SCIMAP-Fitted research scripts). | D-infinity |
| River network threshold (km² upslope) | Defines the "in channel" cells for the network outputs. | 0.8 |
| Ignore cells below this percentile of contributing area (Advanced) | Excludes the smallest-area cells from the ensemble. | 1.0 |
| "No regrets" top tier (%) | The percentage band used for the no-regrets outputs (see below). | 5.0 |
| Concurrent flow-routing runs / row-block workers / memory budget (Advanced) | Performance tuning for large grids. | 2 / 4 / 1500 MB |
| Keep the per-class routed rasters (Advanced) | Retains intermediate per-class rasters instead of deleting them after the run. | Off |
| Colour ramp | Applied to the risk outputs. | SCIMAP default |
| WhiteboxTools executable | Advanced; override the auto-detected binary. | Auto-detected |
Outputs¶
| Output | Description |
|---|---|
| Fitted risk: mean and standard deviation | Two-band raster: mean and standard deviation of routed, diluted risk across the ensemble. |
| Fitted risk: median and IQR | Two-band raster: median and inter-quartile range of the same. |
| Local (unrouted) risk: median and IQR (optional) | The same, but before flow routing — answers "where does the risk come from" rather than "where does it end up". |
| No-regrets priority (in channel, %) | Per cell, the percentage of weight-set combinations that place it in their own top tier — worth acting on whichever calibration turns out to be right. |
| No-regrets priority (landscape, %) | The same, for local (unrouted) risk. |
| River network risk points (optional) | Point layer along the channel network carrying mean/stdev/median/IQR — colour by median and drive opacity from iqr to see risk and confidence together. |
| Per-combination network summary (CSV) (optional) | Mean in-channel risk for every connectivity × land-cover × weight-set combination. |
| Summary histograms folder (optional) | PNG histograms of the combination and network risk distributions (needs matplotlib). |
Screenshot:
The mean/stdev output styled in the QGIS map canvas.
Screenshot:
A no-regrets priority raster, showing the cells worth acting on regardless of which calibrated weight set is correct.
Notes¶
- Every input raster must already share one grid — nothing is resampled here, because resampling would silently change the quantity the weights were calibrated against.
- Unlike the original SCIMAP-Fitted research scripts, this writes a styled point layer for the river network rather than a rendered basemap image, so it can be zoomed and queried over your own basemap.
This completes the SCIMAP Fitted workflow. Return to the guide index.