Frost depth estimate from NOAA daily temperatures

This project fetches NOAA daily summaries and estimates frost depth using a simple net frost index model. It now supports both:

• a command line interface
• a Dash web app with station lookup, charts, warnings, and interactive tables

Default station:

• USW00014734 — Newark Liberty International Airport, NJ

Install

make install

That creates a local .venv and installs all dependencies into it.

If you want to create the environment first, you can also run:

make venv
make install

Run default analysis

make run

.venv/bin/python frost_depth.py

Run just the 2024-2025 winter

make run-2024

.venv/bin/python frost_depth.py --start 2024-07-01 --end 2025-06-30 --winter 2024-2025

Run using a ZIP code or address

.venv/bin/python frost_depth.py --location 07114 --start 2020-07-01 --end 2025-06-30

When --location is provided, the CLI geocodes the location and automatically picks the nearest NOAA daily-summary station it finds for the selected date range.

Run the web app

make run-web

Then open http://127.0.0.1:8050/

Outputs

The CLI writes:

• output/daily_frost_depth.csv
• output/winter_summary.csv
• output/missing_days.csv

The web app shows:

• nearby-station search from address or ZIP code
• sortable/filterable AG Grid tables
• a line chart of frost depth over time
• a bar chart of winter maximum frost depth
• a warnings area for missing days and other issues

Model

Daily values:

freeze_deg = max(0, -Tavg_C)
thaw_deg   = max(0,  Tavg_C)
net_frost_index_today = max(0, net_frost_index_yesterday + freeze_deg - thaw_deg)
depth_cm = k_cm * sqrt(net_frost_index)

Default:

k_cm = 2.0 cm / sqrt(C-day)

That coefficient is a rough value for frost-susceptible wetter soil. It is not a substitute for a proper soil heat-transfer model.

Notes

• NOAA sometimes does not return TAVG. The script fills it using (TMAX + TMIN) / 2 when possible.
• ZIP code lookup uses the ZIP centroid, while street-address lookup uses the U.S. Census geocoder.
• Nearby stations are discovered with NOAA's search API for the chosen date range, then the selected station is analyzed through NOAA daily summaries.
• Missing calendar days in the requested date range are detected, filled from neighboring observed days using linear interpolation, and written to output/missing_days.csv.
• If a missing run touches the requested start or end date, the fill falls back to the nearest observed edge value.
• Winter seasons are grouped July-June and labeled like 2024-2025.
• Frost months used for the summary are October through April.