--- language: - en license: cc-by-4.0 tags: - agriculture - crop-simulation - remote-sensing - synthetic-data - timeseries - tsfile - wofost - rice - sri-lanka task_categories: - time-series-forecasting - tabular-regression pretty_name: Sri Lanka Paddy Rice WOFOST Simulation (TsFile) --- # Wofost Synthetic Rice Simulation LK (TsFile) 斯里兰卡水稻 WOFOST 7.2 合成模拟数据集,从 HuggingFace 数据集 [`SanuthK/Wofost-Synthetic-Rice-Simulation-LK`](https://huggingface.co/datasets/SanuthK/Wofost-Synthetic-Rice-Simulation-LK) 转换为 [Apache TsFile](https://tsfile.apache.org/) 格式。 ## 文件 - `wofost_rice_lk.tsfile` —— 578,394 条日records,3,250 条独立模拟序列。 ## TsFile 表模型映射 - **TAG(device)**:`Season` / `Year` / `Lat` / `Lon` / `Irrigation` / `Nitrogen` + 合成的 `replicate` 序号。每条 simulation = 一个 device(共 3,250 个)。这 6 个 scenario 列仅作 tag,不重复存为数值 field。 - **Time**:由 `Date`(YYYY-MM-DD)解析为 INT64 毫秒。 - **FIELD**:`DVS` / `LAI` / `Biomass` / `Height_cm` / `Yield_kgha` / `NDVI` / `NDRE` / `S1_VV_dB` / `S1_VH_dB` / `S1_ratio`(均为 DOUBLE)。`Yield_kgha` 在开花(DVS≥1)前为 NaN,保留为 null。 ## 转换说明 - **未丢弃任何源列**:17 列全部保留(Date→Time,6 个 scenario→tag,10 个数值→field)。 - **`replicate` 为合成列**:同一 scenario 有多个 replicate(天气随机化),但原数据无显式 replicate ID。多个 replicate 顺序拼接,在 Date 回退处分界,据此累计出 replicate 序号。 验证:加上 replicate 后 device 数 = 3,250,`(device, Date)` 完全唯一。 - 转换路径:`script`(`scripts/converters/wofost_rice_lk.py`),时间精度 `ms`。 - 行数转换前后一致(578,394),已通过行数比对 + 读回校验,无去重、无丢行。 > 关于体积:原始 parquet 约 16 MB,转换后 tsfile 约 21 MB。源数据约 51% 的数值单元为 > NaN(出苗/开花前),parquet 用 RLE 把这些 NaN 游程压到近乎为零;TsFile 因按 device 切分 > (3,250 个 device × 10 列 ≈ 3.2 万个小 chunk)元数据开销较高、对 NaN 游程压缩较弱,故略大。 > 两者数据完全一致。 ## 来源 - 原始数据集:[SanuthK/Wofost-Synthetic-Rice-Simulation-LK](https://huggingface.co/datasets/SanuthK/Wofost-Synthetic-Rice-Simulation-LK) - 原作者:Wanniarachchi S. K.(AgroGate Project) - 模型:PCSE / WOFOST 7.2;天气:NASA POWER - 许可证:CC-BY 4.0