deploy-streamlit-flood-dashboard

name	deploy-streamlit-flood-dashboard
description	Deploy the Streamlit flood vulnerability dashboard with pydeck maps, altair charts, and Snowflake brand styling. Use when deploying, redeploying, or updating the Streamlit app from the workspace.

Deploy Streamlit Flood Dashboard

This skill deploys the interactive Streamlit flood vulnerability dashboard to Snowflake with pydeck building polygons, H3 hexagonal heatmap, altair charts, and Snowflake brand styling.

Prerequisites

• Database FLOOD_ANALYTICS and schema FLOOD must exist
• Warehouse FLOOD_WH must exist
• Tables required: PARISH_FLOOD_SUMMARY, BUILDING_FLOOD_RISK, BUILDINGS_LA, H3_FLOOD_RISK_MAP
• Role must have CREATE STREAMLIT and CREATE STAGE privileges

Files to Deploy

The app consists of 3 files in the streamlit/ directory:

streamlit/flood_dashboard.py

Must include:

• import pydeck as pdk and import altair as alt
• 4 tabs: Parish Overview, Building Explorer, H3 Heatmap, AI Insights
• Snowflake brand colours: #29B5E8 (cyan accent), #11567F (dark blue), #6E7681 (grey), #E4584F (alert red)

streamlit/environment.yml

name: sf_env
channels:
  - snowflake
dependencies:
  - pandas
  - pydeck
  - plotly
  - altair

streamlit/.streamlit/config.toml

[theme]
primaryColor = "#29B5E8"
backgroundColor = "#FFFFFF"
secondaryBackgroundColor = "#F4F7FA"
textColor = "#11567F"
font = "sans serif"

Key Implementation Details

Building Explorer — Pydeck PolygonLayer (NOT GeoJsonLayer)

CRITICAL: Use PolygonLayer with a flat dataframe — NOT GeoJsonLayer. This is because Streamlit's pydeck tooltip {field} syntax only resolves fields from the dataframe columns, not from nested GeoJSON properties.

geo_df["coordinates"] = geo_df["GEOJSON"].apply(lambda g: json.loads(g)["coordinates"])
# Add tooltip columns as flat dataframe columns
geo_df["name"] = geo_df["BUILDING_NAME"].fillna("Unknown")
geo_df["vuln"] = geo_df["COMPOSITE_VULNERABILITY_SCORE"].round(1)

polygon_layer = pdk.Layer(
    "PolygonLayer",
    geo_df[["coordinates", "color_r", "color_g", "name", "vuln", "zone", "bclass", "par"]],
    get_polygon="coordinates",
    get_fill_color="[color_r, color_g, 60, 200]",
    pickable=True,
    extruded=False,
)

tooltip = {
    "text": "Building: {name}\nClass: {bclass}\nParish: {par}\nFlood Zone: {zone}\nVulnerability: {vuln}",
    ...
}

The geometry comes from BUILDINGS_LA.GEOMETRY (GEOGRAPHY type) via:

ST_ASGEOJSON(b.GEOMETRY) AS GEOJSON
FROM BUILDING_FLOOD_RISK r
JOIN BUILDINGS_LA b ON r.BUILDING_ID = b.ID

H3 Heatmap — H3HexagonLayer (2D, flat)

• Use uppercase column names in the dataframe (COLOR_R, COLOR_G, COLOR_B)
• extruded=False, pitch=0 for flat 2D rendering
• Tooltip uses "text" format (NOT "html") with {COLUMN_NAME} matching dataframe columns exactly

Charts — Altair (NOT st.bar_chart)

• Parish bar chart with gradient fill and interactive tooltips
• Bubble scatter plot (NRI vs SVI)
• Vulnerability histogram in Building Explorer
• Donut chart for H3 risk distribution

Streamlit API Limitations (Snowflake Runtime)

• st.pydeck_chart() does NOT support height kwarg — omit it
• st.map() does NOT support latitude/longitude kwargs — use pydeck instead
• st.set_page_config() must be first Streamlit command

Deploy via SQL

CREATE OR REPLACE STAGE FLOOD_ANALYTICS.FLOOD.STREAMLIT_STAGE
  DIRECTORY = (ENABLE = TRUE)
  ENCRYPTION = (TYPE = 'SNOWFLAKE_SSE');

COPY FILES INTO @FLOOD_ANALYTICS.FLOOD.STREAMLIT_STAGE/
FROM 'snow://workspace/USER$.PUBLIC."flood-resilience"/versions/live/'
FILES=('streamlit/flood_dashboard.py', 'streamlit/environment.yml', 'streamlit/.streamlit/config.toml');

CREATE OR REPLACE STREAMLIT FLOOD_ANALYTICS.FLOOD.FLOOD_VULNERABILITY_DASHBOARD
  ROOT_LOCATION  = '@FLOOD_ANALYTICS.FLOOD.STREAMLIT_STAGE/streamlit'
  MAIN_FILE      = 'flood_dashboard.py'
  QUERY_WAREHOUSE = FLOOD_WH
  TITLE          = 'Flood Vulnerability Dashboard';

Verify Deployment

SHOW STREAMLITS LIKE 'FLOOD_VULNERABILITY_DASHBOARD' IN SCHEMA FLOOD_ANALYTICS.FLOOD;

Data Quality Notes

• PARISH_FLOOD_SUMMARY.PCT_IN_SFHA: Must be calculated from tract-level FLOOD_ZONES table (not the parish-level IN_SPECIAL_FLOOD_HAZARD_AREA boolean which is the same for all buildings in a parish)
• PARISH_FLOOD_SUMMARY.TOTAL_EXPECTED_ANNUAL_LOSS: Must use MAX() not SUM() — the EAL value is parish-level and duplicated across all buildings in a parish

Troubleshooting

• Tooltips show {field} as literal text: You are using GeoJsonLayer — switch to PolygonLayer with flat dataframe
• st.pydeck_chart() got unexpected keyword argument 'height': Remove height param
• TypeError: MapMixin.map() got unexpected keyword argument 'latitude': Use pydeck instead of st.map
• All parishes show 100% flood zone: Recalculate from tract-level FLOOD_ZONES, not parish-level boolean
• Annual loss in trillions: You used SUM instead of MAX on the parish-level EAL
• Stage copy fails: Verify workspace name matches USER$.PUBLIC."flood-resilience"