AI Import — loading rainfall series from PDF/Excel

Historical annual series are commonly found in:

• PDFs of regional annuals (dozens of pages, scanned or textual tables);
• Excel files downloaded from hydro-meteorological portals;
• non-standard CSV with Italian headers, mixed separators.

Entering this data manually is slow and error-prone. Runoff Lab NX integrates an AI-assisted importer that automatically recognises the file schema and fills the station.

How it works

1. File → Import → From file with AI… → the modal opens.
2. Drag-drop the file (PDF or Excel) or paste text directly.
3. The AI extracts:
4. station metadata (name, code, municipality, region, lat/lon, observation period);
5. annual maxima by duration (60′, 3h, 6h, 12h, 24h).
6. Preview panel: compare extracted values against the file, edit as needed.
7. Hit Confirm → everything enters the study.

Supported formats

Format	Notes
Textual PDF	Regional annuals, table extracts.
Scanned PDF	Automatic OCR, quality dependent on scan.
Excel .xlsx	One row = one year; columns = durations. Order-independent.
CSV / TSV / TXT	Separators ,, ;, \t or spaces.
Free text (copy-paste)	Useful for excerpts from older annuals.

Ready samples

In the Info modal → Resources you'll find two example files to download:

• ai-rainfall-sample-cosenza.xlsx — realistic Excel format (Calabria annual).
• ai-rainfall-sample-cosenza.pdf — PDF extract from a regional annual.

Open the AI Import modal and drop one of them in to see the procedure live.

What the AI does under the hood

Extraction runs on our backend with a multimodal model (Gemini 1.5 Pro) trained to recognise Italian rainfall tables:

• identifies Italian headers (e.g. "Massimi annuali di pioggia di durata 1h, 3h…");
• distinguishes valid values from footnotes, placeholders (---, n/d);
• recognises metadata even when it's mid-page text;
• normalises decimal separators and numbers.

Only the uploaded file is sent to the backend; it is not stored or indexed. Cost: a few NX credits per operation (visible on the wallet chip before launch).

When verification matters

The AI has good accuracy but isn't infallible. Cases requiring inspection:

• Multi-station tables: the AI tries to identify the requested station, but on unstructured PDFs it may confuse rows.
• Ambiguous decimals: 34,8 (Italian) vs 34.8 (English) — the AI decides from context; check a few values.
• Series with gaps: some series mark gaps with ---; verify they were skipped correctly.

Always confirm the preview before accepting.

Manual alternative

If you prefer, you can:

• copy the annual-maxima column from Excel;
• paste directly into the single-duration box.

The simple parser accepts . and , as decimal separator, one value per line.