Apppublic

snesbitt/pf-grid-tiles-app

sourceHugging Facemitupdated 3mo agoView on Hugging Face

0likes

App README

PF swath-grid climatology — xpublish-tiles map server

![DOI](https://doi.org/10.57967/hf/9189)

Interactive slippy-map of the TRMM PR / GPM Ku precipitation-feature swath-grid climatology. Dataset DOI: [10.57967/hf/9189](https://doi.org/10.57967/hf/9189). Map tiles are rendered on the fly by `xpublish-tiles` from a precomputed Zarr, served by FastAPI, with a small Leaflet front-end at /.

36 fields = 12 quantities × 3 members, annual (summed over all months & UTC hours), on the native 0.05° shared ±68° grid. The quantities cover total precipitation plus a convective / stratiform split (radar 2A typePrecip classification):

quantity	definition	units
`rate`	Σrain / Nviews (unconditional)	mm/hr
`freq`	Nraining / Nviews	fraction
`intensity`	Σrain / Nraining (conditional)	mm/hr
`rain`	annual accumulation = rate × 8766 h/yr	mm/year
`raining_views`	raining pixel-views	count
`views`	radar pixel-views (denominator)	count
`conv_rain`	convective annual accumulation	mm/year
`strat_rain`	stratiform annual accumulation	mm/year
`conv_freq`	convective frequency = Nraining_conv / Nviews	fraction
`strat_freq`	stratiform frequency = Nraining_strat / Nviews	fraction
`conv_intensity`	convective conditional rate	mm/hr
`strat_intensity`	stratiform conditional rate	mm/hr

Members: GPM (Ku, 2014–2026), TRMM (PR, 1997–2015, zero poleward of ±38°), COMBINED (GPM+TRMM pooled). The map viewer exposes 11 of these as a grouped quantity dropdown (All precipitation / Convective / Stratiform); rate is in the store but not the dropdown (rain is the user-facing accumulation field).

How the data is supplied

The Space does not ship the Zarr in the image. At startup app.py pulls pf_tiles.zarr from a HuggingFace Dataset repo named by the HF_DATASET_REPO secret/variable, via huggingface_hub.snapshot_download. The store is ~750 MB.

1. Push the precomputed Zarr to a Dataset repo

bash

pip install huggingface_hub
huggingface-cli login           # needs a write token

huggingface-cli repo create pf-grid-tiles --repo-type dataset
huggingface-cli upload pf-grid-tiles \
    /data/scratch/a/snesbitt/pf_tiles.zarr pf_tiles.zarr \
    --repo-type dataset

(or HfApi().upload_folder(folder_path=".../pf_tiles.zarr", path_in_repo="pf_tiles.zarr", repo_id="<user>/pf-grid-tiles", repo_type="dataset").)

2. Create the Docker Space

bash

huggingface-cli repo create pf-grid-tiles-app --repo-type space --space_sdk docker
git clone https://huggingface.co/spaces/<user>/pf-grid-tiles-app
cp app.py requirements.txt Dockerfile README.md pf-grid-tiles-app/
cd pf-grid-tiles-app && git add -A && git commit -m "tile server" && git push

3. Point the Space at the Dataset

In the Space Settings → Variables, set:

HF_DATASET_REPO = <user>/pf-grid-tiles

(If the dataset is private, also add a HF_TOKEN secret with read access — the huggingface_hub client picks it up automatically.)

The Space builds, pulls the Zarr once, and serves the map at its root URL.

Run locally

bash

pip install -r requirements.txt
TILES_ZARR=/data/scratch/a/snesbitt/pf_tiles.zarr uvicorn app:app --port 7860
# open http://localhost:7860/

Tile endpoint (OGC/XYZ): /tiles/WebMercatorQuad/{z}/{y}/{x}?variables={member}_{quantity}&style=raster/{cmap}&width=256&height=256