---
site:
outline_maxdepth: 2
---
# Split-panel comparisons
Comparing two datasets side by side with a draggable slider
---
[](https://colab.research.google.com/github/HendrikWulf/sds320-jb/blob/main/book/04_lessons/L03/05_split-panel-comparisons.ipynb)
---
## 1. Motivation
A large share of GeoAI work is fundamentally comparative: {term}`change detection ` compares two dates, model evaluation compares predictions against {term}`ground truth `, and data selection compares imagery from different sensors or seasons. Viewing two layers stacked on top of each other, one at a time, makes it hard to judge fine spatial differences. A split view keeps both layers visible at the same extent and zoom level.
---
## 2. Core idea
A **split-panel map** places two layers side by side with a draggable divider, so both sides always share the same geographic extent and zoom, which keeps the comparison spatially honest.
---
## 3. Workflow
### A. Prepare sample data
This example compares SWISSIMAGE imagery, a height raster and building footprints.
```{code-cell} python
from pathlib import Path
import geoai
import leafmap.foliumap as leafmap
DATA_DIR = Path("data/raw")
DATA_DIR.mkdir(parents=True, exist_ok=True)
swissimage_url = "https://data.source.coop/giuz/sds320/L03/data/willisau_2024_swissimage_rgb_subset.tif"
hag_url = "https://data.source.coop/giuz/sds320/L03/data/willisau_height_above_ground_ndsm.tif"
buildings_url = ("https://data.source.coop/giuz/sds320/L03/data/willisau_overture_buildings_subset.geojson")
swissimage_path = geoai.download_file(swissimage_url, output_path=str(DATA_DIR / Path(swissimage_url).name))
hag_path = geoai.download_file(hag_url, output_path=str(DATA_DIR / Path(hag_url).name))
buildings_path = geoai.download_file(buildings_url, output_path=str(DATA_DIR / Path(buildings_url).name))
```
### B. Build a basic split comparison
`split_map()` accepts common layer sources, including raster and vector files, with `left_args` and `right_args` for per-side rendering options.
```{code-cell} python
m = leafmap.Map()
m.split_map(
left_layer=swissimage_path,
right_layer=hag_path,
left_args={"indexes": [1, 2, 3]},
right_args={"vmin": 0, "vmax": 10, "colormap": "plasma"},
left_label="SWISSIMAGE (true color)",
right_label="Height Above Ground",
)
m
```
Dragging the slider left and right reveals each layer. Areas with tall structures in the imagery should line up with high values in the height raster; if they consistently do not, that is a sign the two datasets are misaligned.
### C. Overlay comparison for quality control
With the standard, ipyleaflet-based `leafmap.Map`, `split_map()` handles a vector file such as this building-footprint GeoJSON exactly like a raster file, no different code path needed. This page uses `leafmap.foliumap` instead, so the map renders correctly on the built book website, and that backend's `split_map()` only synchronizes raster tile layers. If you are working in your own notebook with plain `import leafmap`, you can pass `buildings_path` straight into `split_map()` just like the raster example above; here, we overlay both layers on one map instead, which is just as effective a check before trusting a set of labels.
```{code-cell} python
m2 = leafmap.Map()
m2.add_raster(swissimage_path, layer_name="SWISSIMAGE")
m2.add_geojson(
buildings_path,
layer_name="Building labels",
style={"color": "red", "fillOpacity": 0.2},
)
m2
```
```{admonition} What a split view is good for
:class: hint
Reach for a split-panel comparison whenever you are asking "do these two things actually match up in space," rather than "what does this one thing look like." The former is a comparison question; the latter is answered by the single-layer techniques from the earlier pages in this lesson.
```
---
## 4. Key takeaways
- Split-panel maps keep two layers at the same extent and zoom, which makes spatial comparisons more reliable than toggling layers on and off.
- `split_map()` supports both raster and vector layers with the standard, ipyleaflet-based `leafmap.Map`; the `leafmap.foliumap` backend used on this page only synchronizes raster layers, so vector-vs-raster checks here use an overlay instead.
- Split views are well suited to alignment checks; they do not replace quantitative evaluation metrics.
- Consistent styling on both sides is necessary for a comparison to reflect real differences in the data rather than differences in how it was drawn.