Datasets:

smilegeng
/

Wofost-Synthetic-Rice-Simulation-LK

language:
  - en
license: cc-by-4.0
tags:
  - agriculture
  - crop-simulation
  - remote-sensing
  - synthetic-data
  - wofost
  - rice
  - sri-lanka
  - tabular
  - climate-variability
  - yield-prediction
  - tsfile
  - format:tsfile
  - timeseries
task_categories:
  - tabular-regression
  - time-series-forecasting
pretty_name: Sri Lanka Paddy Rice WOFOST Simulation (TsFile)
size_categories:
  - 100K<n<1M

Wofost Synthetic Rice Simulation LK (TsFile)

A WOFOST 7.2 synthetic simulation dataset for paddy rice in Sri Lanka, converted to Apache TsFile format from the Hugging Face dataset SanuthK/Wofost-Synthetic-Rice-Simulation-LK.

Files

wofost_rice_lk.tsfile — 578,394 daily records across 3,250 independent simulation series.

TsFile table-model mapping

TAG (device): Season / Year / Lat / Lon / Irrigation / Nitrogen plus the synthesized replicate index. Each simulation = one device (3,250 devices in total). These 6 scenario columns are kept only as tags and are not duplicated as numeric fields.
Time: parsed from Date (YYYY-MM-DD) into INT64 milliseconds.
FIELD: DVS / LAI / Biomass / Height_cm / Yield_kgha / NDVI / NDRE / S1_VV_dB / S1_VH_dB / S1_ratio (all DOUBLE). Yield_kgha is NaN before flowering (DVS≥1) and is kept as null.

Conversion Notes

No source columns dropped: all 17 columns are retained (Date→Time, 6 scenario→tag, 10 numeric→field).
replicate is a synthesized column: the same scenario has multiple replicates (weather randomization), but the source data has no explicit replicate ID. Multiple replicates are concatenated sequentially and split at the points where Date resets, from which the replicate index is accumulated. Validation: after adding replicate, device count = 3,250 and (device, Date) is fully unique.
Conversion path: script (scripts/converters/wofost_rice_lk.py), millisecond precision.
Row count is identical before and after conversion (578,394), verified by row-count comparison + read-back validation; no dedup, no dropped rows.

On size: the original parquet is ~16 MB and the converted tsfile is ~21 MB. About 51% of the source numeric cells are NaN (before emergence/flowering); parquet uses RLE to compress these NaN runs to nearly zero, while TsFile has higher metadata overhead from per-device chunking (3,250 devices × 10 columns ≈ 32k small chunks) and weaker NaN-run compression, hence slightly larger. The data is identical in both.

Source

Original dataset: SanuthK/Wofost-Synthetic-Rice-Simulation-LK
Original author: Wanniarachchi S. K. (AgroGate Project)
Model: PCSE / WOFOST 7.2; weather: NASA POWER
License: CC-BY 4.0