Reduce Binder cold‑start times and stabilize builds (#8)

* Pin heavy geo deps in env.yaml (numpy, gdal, rasterio, rioxarray, geopandas, shapely, xarray)
* Install package during build via pip in env.yml
* Repo-level .condarc (conda-forge, channel_priority=strict, show_channel_urls)
* Remove binder/postBuild and dev.pixi.toml (unused)
* Edited (removed and added) Binder warmup workflow (push to main, weekly 03:00 UTC, manual)

Author: Jack-Hayes

.condarc

@@ -0,0 +1,4 @@
+channels:
+  - conda-forge
+channel_priority: strict
+show_channel_urls: true

.github/workflows/binder-cache.yml

@@ -0,0 +1,21 @@
+name: Binder cache warmup
+
+on:
+  push:
+    branches: [ main ]
+  schedule:
+    - cron: "0 3 * * 0"   # weekly, Sunday 03:00 UTC
+  workflow_dispatch:
+
+jobs:
+  warmup:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - name: Trigger mybinder.org build
+        run: |
+          set -eux
+          REF="${{ github.ref_name }}"
+          if [ -z "$REF" ] || [ "$REF" = "refs/heads/" ]; then REF="main"; fi
+          curl -LsS -o /dev/null -w "%{http_code}\n" \
+            "https://mybinder.org/build/gh/${{ github.repository }}/$REF" || true

.github/workflows/binder-warmup.yml

@@ -16,34 +16,43 @@ Many thanks to banesullivan's [localtileserver](https://github.com/banesullivan/
   <img src="https://mybinder.org/badge_logo.svg" alt="Launch on Binder" height="28">
 </a>
 
-What happens when you click the button (suggested to right click -> open in new tab):
-- A temporary, free session is launched on mybinder.org (a community, shared service) that builds a Docker image from this repo’s environment files.
-- First launch after a change can take several minutes while the image builds. Subsequent launches are faster (cached), but caches may be evicted at any time.
-- Sessions are ephemeral (typically 1–2 hours) and resources are limited. Your session will shut down if idle or when it times out.
+What happens when you click the button:
+- mybinder.org uses repo2docker to build a Docker image for this repo using binder/environment.yml and runtime.txt (Python 3.11 base)
+- The conda environment is solved/installed with mamba from conda-forge; during the build, pip installs this repo in editable mode (-e .)
+- The built image is cached by mybinder.org for the given Git ref; first builds can take several minutes, later builds are usually faster until the cache is evicted
+- Sessions are ephemeral and resource-limited; they time out when idle (~10 mins) or after a short lifetime
 
 ## Notes and expectations
 
 - First launch: building can take several minutes; please be patient.
-- If you see “Building” or “Pending,” Binder is queuing your session or compiling the environment.
+- If you see "Building' or "Pending," Binder is queuing your session or compiling the environment.
 - If tiles don’t appear immediately after "Fetch & Display Data," wait a few seconds.
+- Don't sample large areas -- this will drop performace. Less than 30km2 is a good bet.
 - The app depends on external open data services; occasional slowdowns or gaps can occur.
 
 ## What you can do
 
 - Pick a date range, draw an Area of Interest (blue) on the map, then click "Fetch & Display Data."
 - Toggle layers in the layer control:
   - RGB composite (Sentinel‑2) [lowest cloud coverage in date range is selected]
-  - NDVI and a basic built-up index [lowest cloud coverage in date range is selected]
   - SAR backscatter (Sentinel‑1) [acquisition closest to Sentinel-2 image is selected]
+  - Coopernicus 30-GLO DEM Multidirectional Hillshade
   - ESA WorldCover (classified land cover) [2021]
 - Draw a red Region of Interest (ROI) to sample statistics from visible layers.
 
 ## How it works (high level)
 
-- You draw an area; the app searches a STAC catalog (Microsoft Planetary Computer) for relevant items.
-- It selects a recent, low‑cloud Sentinel‑2 scene and optionally a time‑aligned Sentinel‑1 scene.
-- On the fly, it computes quicklook products (RGB, NDVI, built‑up index), and prepares tiles from in‑memory COGs.
-- Tiles are served inside the session and displayed via ipyleaflet (no data leaves your browser other than tile requests to the Binder session)
+- You draw an area; the app searches the Microsoft Planetary Computer STAC API for relevant items
+- It selects a low‑cloud Sentinel‑2 scene and a time‑aligned Sentinel‑1 scene
+- It computes quicklook products:
+  - Sentinel‑2 RGB
+  - Sentinel‑1 VV/VH in decibels
+  - Multidirectional hillshade from Copernicus GLO‑30
+  - ESA WorldCover mapped to RGB colors
+- Rasters are written to in‑memory COGs and served by localtileserver running in your Binder session.
+- ipyleaflet requests tiles via Jupyter Server Proxy (e.g., /proxy/{port}); the browser only talks to the Binder pod, while the pod fetches data from the Planetary Computer.
+
+> **_NOTE:_**  All layers / STAC endpoints are oprtional
 
 ## Contributing
 
@@ -56,4 +65,4 @@ What happens when you click the button (suggested to right click -> open in new
   - Conda/mamba: create an environment similar to binder/environment.yml, then `pip install -e .`
   - Or use your own stack and ensure GDAL/Rasterio/PROJ are consistent.
 - Run notebooks locally with JupyterLab or launch the Voila app:
-  - `voila notebooks/app.ipynb
+  - `voila notebooks/app.ipynb`

README.md

@@ -8,16 +8,19 @@ dependencies:
   - voila
   - ipywidgets
   - ipyleaflet
-  - numpy
+  - numpy=2.3.*
   - pandas
-  - xarray
-  - geopandas
-  - rioxarray
-  - rasterio
-  - shapely
+  - xarray=2025.10.*
+  - geopandas=1.1.*
+  - rioxarray=0.19.*
+  - rasterio=1.4.*
+  - shapely=2.1.*
+  - gdal=3.10.*
   - matplotlib
   - odc-stac
   - pystac-client
   - planetary-computer
   - localtileserver
   - jupyter-server-proxy
+  - pip:
+      - -e ..