Dataset Card for colombo-tea-auction-prices

This dataset contains structured historical price data from the Colombo Tea Auction (CTA) in Sri Lanka, synchronized with localized meteorological lag features. It is designed for segment-specific and pooled time-series regression models to investigate how market structures and localized weather patterns drive tea pricing in Sri Lanka.

Dataset Details

Dataset Description

This dataset consists of 12,233 rows compiled from 105 weekly broker market reports spanning from late 2023 through early 2026. The records cover four primary auction market catalogues (High Grown, Low Grown, Off-Grade, and Dust), across multiple tea elevations, grades, and quality tiers. To evaluate climate sensitivity, each auction week is synchronized with historical weather records extracted via the Open-Meteo API, matching major tea-growing zones (such as Nuwara Eliya, Uva, and Udapussellawa) with weekly temperature, total precipitation, and sunshine duration features across multiple lag orders (1, 2 and 3 weeks).

Dataset Sources

• Repository: https://huggingface.co/datasets/hesandism/colombo-tea-auction-prices
• Paper: Forecasting Weather-Driven Price Dynamics Across Sri Lankan Tea Market Catalogues

Uses

Direct Use

• Agricultural Commodity Price Forecasting: Building predictive machine learning and deep learning models ($\texttt{Random Forest}$, $\texttt{XGBoost}$, $\texttt{LightGBM}$, or $\texttt{LSTM}$) to forecast one-week-ahead tea auction mid-prices.
• Climate Impact Studies: Analyzing delayed meteorological impacts on agricultural supply chains using statistical methods like Granger causality tests and Vector Autoregression ($\texttt{VAR}$).
• Segment Sensitivity Analysis: Investigating how premium quality tiers compare against commercial grade or off-grade categories when subjected to weather shocks.

Out-of-Scope Use

• High-frequency daily or real-time intraday trading algorithms (the data is aggregated at a weekly auction level).
• Direct application to other geographic tea centers (e.g., Mombasa or Kolkata auctions) without domain recalibration.

Dataset Structure

The dataset contains the following attributes for each row entry:

Column Name	Type	Description
sale_id	String	Unique identifier for each weekly auction sale (e.g., SALE_2023_42).
catalogue	String	The market category segment: low_grown, high_grown, off_grade, or dust.
category	String	Sub-classification within the catalogue.
elevation	String	The agro-climatic growing zone elevation: low, medium, high.
segment	String	Regional indicator or manufacturing style grouping.
grade	String	Standard tea grade notation (e.g., bop, bop1, dust).
tier	String	Quality classification tier (e.g., select_best, best, below_best, others).
price_mid_lkr	Float	Dependent variable. The recorded mid-point price at auction in Sri Lankan Rupees (LKR).
region	String	Name of the primary sourcing region.
text_crop_change	String	Broader qualitative broker commentary describing changes in harvest volume.
text_keywords	String	Key descriptive phrases extracted from broker notes.
temperature_2m_mean_mean	Float	The mean temperature ($^{\circ}\text{C}$) during the auction week.
precipitation_sum_total_lag1	Float	Total precipitation sum (mm) with a 1-week lag.
sunshine_duration_total_lag1	Float	Total sunshine duration (seconds) with a 1-week lag.
temperature_2m_mean_mean_lag1	Float	Mean temperature ($^{\circ}\text{C}$) with a 1-week lag.
precipitation_sum_total_lag2	Float	Total precipitation sum (mm) with a 2-week lag.
sunshine_duration_total_lag2	Float	Total sunshine duration (seconds) with a 2-week lag.
temperature_2m_mean_mean_lag2	Float	Mean temperature ($^{\circ}\text{C}$) with a 2-week lag.
precipitation_sum_total_lag3	Float	Total precipitation sum (mm) with a 3-week lag.
sunshine_duration_total_lag3	Float	Total sunshine duration (seconds) with a 3-week lag.
temperature_2m_mean_mean_lag3	Float	Mean temperature ($^{\circ}\text{C}$) with a 3-week lag.
sale_number	Integer	Sequenced auction number within the calendar year.
sale_year	Float	Calendar year of the auction event.
sale_date_raw	String	Raw date string indicating the exact auction days.
avg_weather_severity	Float	Computed historical anomaly index reflecting extreme weather.
fx_usd	Float	Weekly exchange rate of USD to LKR.

Dataset Creation

Curation Rationale

Prior tea pricing research often treats the Colombo Tea Auction as a monolithic market, overlooking its structural nuances. This dataset was curated to enable segment-aware modeling, highlighting how weather acts as a subtle, highly lagged driver depending on whether a tea grade is high-grown premium or an off-grade blend.

Source Data

Data Collection and Processing

1. Auction Logs Parsing: Structured catalog tabular prices, tiers, and text summaries were extracted directly from PDF weekly market reports published by Forbes & Walker Tea Brokers (Pvt) Ltd.
2. Weather Synchronization: Geographical coordinates for Sri Lanka’s key tea-growing hubs (Nuwara Eliya, Uva, Udapussellawa, and Low Grown regions) were used to fetch corresponding localized daily climate variables via the Open-Meteo API. Daily figures were aggregated into weekly chunks and aligned with historical auction dates across three sequential lag windows.

Who are the source data producers?

The original auction reports are prepared weekly by professional tea tasters and market auction brokers at Forbes & Walker Tea Brokers (Pvt) Ltd. Meteorological properties originate from the global open-source reanalysis models hosted by Open-Meteo.

Bias, Risks, and Limitations

• Currency Distortions: Prices are listed in nominal Sri Lankan Rupees ($\text{LKR}$). During the 2023-2026 timeframe, macroeconomic factors and currency shifts influenced values; users are encouraged to use the provided fx_usd attribute to evaluate pricing in real dollar terms.
• Data Sparsity: Certain niche grade combinations might hold missing or sparse price values across isolated sales weeks due to seasonal availability fluctuations.