SCOUT examples
SCOUT studies three urban questions: how much shadow tall buildings cast, how deep a flood gets with and without nature-based solutions, and which route avoids bad weather. Curio runs each one with packages, models and datasets that ship with it, in dataflows that compare scenarios. SCOUT's data and models are used with the permission of its authors.
Find them
The five dataflows are in the examples every account starts with, under By source on the projects page: "SCOUT shadow model on Chicago Loop buildings" (example 24) under Examples, and four under Tests: "Chicago Loop shadow with towers removed or edited", "Lower Manhattan park shade with towers removed or capped", "Quad Cities flood depth, nature-based solutions and vulnerability" and "Routing around rain and wind in downtown Chicago". The SCOUT tag in the rail lists every dataflow that uses a SCOUT node, model or dataset. On your own computer they come with --with-examples, which also installs the Rasterization and Vega-Lite packages and SCOUT's model packages.
Building rasters and shadows
- Rasterize Buildings, from the Rasterization package (
curio.rasterization@1), turns a buildings layer with heights in metres into SCOUT's zoom-16 height tiles, side by side in one raster: a mosaic an Autark map draws. - Accumulated Shadow, the node of the Deep Umbra model package in the Model Catalog, runs Deep Umbra, SCOUT's shadow model, on that mosaic. It gives the minutes of a day the ground spends in shadow, as a raster on the same grid; its Season widget picks spring, summer or winter. Deep Umbra needs Rasterization, so adding it to a dataflow adds Rasterization too.
- A Raster Calculator, also from Rasterization, then takes the shadow's statistics: its mean and median over the ground.
To use Rasterization in a dataflow of your own, click Add to project on its card in the Node Catalog drawer. Its raster nodes are then under Rasterization in the Tools panel's Node Catalog dropdown. The charts that compare scenarios come with the Vega-Lite package, added the same way.
Example 24, "SCOUT shadow model on Chicago Loop buildings", runs them on the buildings of SCOUT's Chicago Loop example: the buildings in 3D, the height mosaic and the summer shadow on Autark maps, and the mean and median shadow.



"Chicago Loop shadow with towers removed or edited" compares two scenarios of the same buildings over a shared season, Existing and Towers removed, each with an Autark map of its shadow. Both maps set the same colorMapRange, 0 to 720 minutes, so a minute of shadow has one color on both. In Towers removed, a Map Editor node takes SCOUT's 15 buildings out by their building_id, and feeds a buildings map in the scenario's box that it is linked to. Pick more buildings on that map to remove, raise or lower them, and Commit edits runs the shadow again. The Vega-Lite package's Bars charts the two scenarios' mean shadow over the ground, each scenario's statistics on an input of its own, and another Raster Calculator takes Towers removed's shadow minus Existing's, cell by cell, which an Autark map draws on a red to blue scheme centered on 0.



Deep Umbra ships with Curio's repository and its Docker image; a pip install downloads its weights from GitHub the first time Accumulated Shadow runs.
Shade in Lower Manhattan's parks
"Lower Manhattan park shade with towers removed or capped" measures the shade in the parks of the Financial District and City Hall with the towers around them removed or capped:
- an Autark node loads Lower Manhattan's parks, and the buildings of a box about 300 m larger, from an OpenStreetMap extract;
- Python nodes give every building part a height, keep the open-air parks, and mark the parts at least the Tall building height tall within the Distance from a park of any park, two widgets of the Buildings around parks node, set to 60 m and 75 m;
- in three scenarios, Existing, Towers removed and Height cap, Map Editor nodes remove the marked parts or cut them to the Tall building height, each linked to the buildings map it feeds. Rasterize Buildings and Accumulated Shadow give the shadow over the season, which a Season node connected to each Accumulated Shadow sets. A Python node measures it in each park, and Autark maps draw the buildings in 3D and the shadow.
The Vega-Lite package's Dots with a range charts each park's mean minutes of shadow, a dot per scenario on a rule from the park's lowest to its highest, each scenario's parks on an input of its own. For each what-if, a Raster Calculator takes its change in shadow from Existing, cell by cell, and a Python node the change per park, matched by the park's name, which Autark maps draw.

Flooding and nature-based solutions
"Quad Cities flood depth, nature-based solutions and vulnerability" is SCOUT's flood study of the Quad Cities, built from Curio's own nodes and Rasterization's Raster Calculator:
- two Python nodes hold widgets that other nodes read through connections: Region, the region's two corners, which start at the region of SCOUT's flood example, and Projection period, 2020 - 2050, 2050 - 2080 or 2080 - 2100;
- Data Loading nodes connected to them read only that region's window of SCOUT's whole flood depth rasters, with and without nature-based solutions (NbS), and of its raster of where each solution would go;
- in each of two scenarios, No NbS and NbS, a Raster Calculator takes the depth with NbS where the class is one of the solutions its Nature-based solutions checkboxes pick, and the depth without NbS elsewhere. No NbS picks none, and NbS all four: bioswales and infiltration trenches, permeable pavements, retention ponds and constructed wetlands. A second Raster Calculator takes the depth's median and mean, and an Autark map draws the depth. Both depth maps set the same
colorMapRange, 0 to 11 m.
Outside the scenarios, another Raster Calculator takes the size of the change in depth between No NbS and NbS, with absolute=True, which an Autark map draws, and the Vega-Lite package's Bars charts each scenario's median depth.


The dataflow also weighs the flood against social vulnerability. Two more Data Loading nodes read CDC/ATSDR's Social Vulnerability Index (SVI) for the census tracts and USGS's rivers, lakes and reservoirs from the Data Catalog, and a Python Computation node keeps the tracts the rivers touch, which an Autark map draws by their SVI percentile. In each scenario, a Python Computation node averages the depth over each tract's land, the water left out, which an Autark map draws, and a Raster Calculator averages it over the riverside land in each period. A Python node takes the size of the change in each tract's mean depth, matched by the tract's FIPS code, which an Autark map draws. The Vega-Lite package's Points charts vulnerability against depth, and its Grouped bars the mean depth by period.


A pip install downloads the flood rasters from GitHub the first time a node reads them. The Raster Calculator works on any rasters on one grid, from a Python node or the Data Catalog, cell by cell or as statistics. A Data Loading node reads only part of a large GeoTIFF when its curio_load_data call is given bounds.
Weather-aware routing
Weather Routing, the node of the SCOUT weather GNN model package in the Model Catalog, finds routes between two points over a roads layer. It weighs each road by the weather that SCOUT's weather graph network predicts on it from SCOUT's WRF forecast in the Data Catalog. Its widgets set the origin and destination, the start time, the mode, the number of routes per factor, and the rain and wind weights; the forecast runs from 5 July 2025, 19:00, to 7 July, 19:00, Chicago time. It outputs the routes, which an Autark map draws, and a table of each route's duration, distance and rain, heat, wind and humidity exposure, which the Vega-Lite package's charts compare.
"Routing around rain and wind in downtown Chicago" is SCOUT's weather routing example. An Autark node loads the roads of SCOUT's box in downtown Chicago, and one Weather Routing node finds SCOUT's two routes over them: the fastest, and the one its weather weights favor. Two scenarios, Fastest route and Weather-aware route, each keep one of them. An Autark map draws both routes over the roads, each route's layer colored by its input's scenario chip, such as "color": "[!! scenario input_1 !!]", so each is in its scenario's color. Four of the Vega-Lite package's Bars chart their duration, distance and rain and wind exposure, each with both scenarios' metrics on inputs of its own.


Charts that compare scenarios
The Vega-Lite package (curio.vega-lite@1) holds fifteen charts that compare scenarios: Lines with a band, Curves by group, Bars with error bars, Ensemble strips, Stacked bars with totals, Donuts, Weighted matrix, Dots with a range, Bars, Grouped bars, Lines, Points, Pie, Lollipops and Table. Each is a Vega-Lite node whose spec reads its inputs stacked, through [!! inputs !!], and draws each scenario in its box's color, in the order the dataflow lists the boxes, through its scenario chips (see Compare scenarios).
- Click Add to project on the Vega-Lite card in the Node Catalog drawer. Its charts are then under Vega-Lite in the Tools panel's Node Catalog dropdown.
- Drag a chart onto the canvas and connect a node of each scenario's box to one of its input circles: each connection takes the next circle. Until the first one, the chart's
[!! inputs !!]chip is red. - In its Widgets tab, set the columns it plots: each column widget, such as
yorpart, takes the name of a column of its inputs. A new chart's widgets hold their own names, so the chart says which column its inputs lack until you set them. Combine, in Bars, Grouped bars, Lines, Pie and Lollipops, is how a scenario's rows are combined:mean,sum,median,min,maxorcount.
A new chart opens with its template's spec on its Grammar tab, and the spec is yours to edit. The row above the spec lists the chips it can use, in categories that open with a click (see The chips above the spec).
In "Nine ways to chart scenario comparisons", under Tests on the projects page, nine of these charts each draw one example table (see Scenarios).

