41. DCAS GeoParquet¶

The Dynamic Crop Advisory System (DCAS) dataset provides daily crop advice, growth-stage monitoring, and climate-related metrics.
It is stored as partitioned GeoParquet files in the GAP bucket.

41.1 Prerequisites¶

Python 3.9+
DuckDB with extensions:

httpfs (S3/HTTP access)
spatial (GEOMETRY support)

Read-only GAP object storage credentials

41.1.1 Environment variables¶

export S3_ENDPOINT_URL="https://fra1.digitaloceanspaces.com"
export S3_ACCESS_KEY_ID="YOUR_READ_ONLY_KEY"
export S3_SECRET_ACCESS_KEY="YOUR_READ_ONLY_SECRET"
export S3_BUCKET_NAME="bucket-name"

41.2 Path layout¶

DCAS is partitioned by country (iso_a3), year, month, and day:

s3://$S3_BUCKET_NAME/staging/dcas_output/iso_a3=KEN/year=/month=/day=*/*.parquet

Example:

s3://gap-products/staging/dcas_output/iso_a3=KEN/year=2025/month=9/day=10/*.parquet

Always pass hive_partitioning=true to make iso_a3, year, month, and day available as columns.

41.3 Column descriptions¶

The following columns are available in DCAS GeoParquet.

Column name	Type	Description
`planting_date_epoch`	UINTEGER	Epoch timestamp of planting date for the farm.
`crop_id`	USMALLINT	Numeric identifier of the crop type.
`crop_stage_type_id`	USMALLINT	Identifier for the type of growth stage.
`group_id`	UINTEGER	Internal group identifier.
`farm_id`	BIGINT	Internal numeric farm identifier.
`farm_unique_id`	VARCHAR	External farm identifier (used for joining/filtering).
`geometry`	GEOMETRY	Spatial geometry of the farm/grid (polygon or point).
`grid_id`	UINTEGER	Numeric identifier for grid cell.
`grid_unique_id`	VARCHAR	Unique identifier for grid cell.
`registry_id`	BIGINT	Identifier from farm registry.
`crop`	VARCHAR	Crop name (e.g. “maize”).
`country_id`	UINTEGER	Country identifier (numeric).
`grid_crop_key`	VARCHAR	Composite key linking grid and crop.
`county`	VARCHAR	County name (administrative boundary).
`subcounty`	VARCHAR	Subcounty name (administrative boundary).
`ward`	VARCHAR	Ward name (administrative boundary).
`preferred_language`	VARCHAR	Preferred language for communication/advisories.
`date`	TIMESTAMP	Record timestamp.
`prev_growth_stage_id`	USMALLINT	Previous growth stage identifier.
`prev_growth_stage_start_date`	UINTEGER	Epoch timestamp when previous growth stage started.
`config_id`	USMALLINT	Identifier for model/config version used.
`growth_stage_start_date`	UINTEGER	Epoch timestamp when current growth stage started.
`growth_stage_id`	USMALLINT	Identifier for current growth stage.
`total_gdd`	DOUBLE	Accumulated Growing Degree Days (GDD).
`seasonal_precipitation`	DOUBLE	Total precipitation recorded in current season (mm).
`temperature`	DOUBLE	Mean temperature for record period (°C).
`humidity`	DOUBLE	Mean relative humidity (%) for record period.
`p_pet`	DOUBLE	Precipitation / Potential Evapotranspiration ratio.
`growth_stage_precipitation`	DOUBLE	Precipitation during current growth stage (mm).
`message`	UINTEGER	Encoded advisory message ID.
`message_2`	UINTEGER	Secondary message code (if available).
`message_3`	UINTEGER	Tertiary message code.
`message_4`	UINTEGER	Quaternary message code.
`message_5`	UINTEGER	Quinary message code.
`is_empty_message`	VARCHAR	Flag if message is empty (text flag).
`has_repetitive_message`	VARCHAR	Flag if advisory is repetitive (text flag).
`final_message`	UINTEGER	Final consolidated advisory message ID.
`prev_week_message`	UINTEGER	Advisory message from the previous week.
`growth_stage`	VARCHAR	Human-readable crop growth stage name.
`__null_dask_index__`	BIGINT	Index column generated by Dask (ignore in most queries).
`day`	BIGINT	Day partition (from file path).
`iso_a3`	VARCHAR	ISO Alpha-3 country code (e.g. KEN).
`month`	BIGINT	Month partition (from file path).
`year`	BIGINT	Year partition (from file path).

41.4 Example queries¶

41.4.1 Stage a subset¶

import os, duckdb

conn = duckdb.connect()
conn.install_extension("httpfs"); conn.load_extension("httpfs")
conn.install_extension("spatial"); conn.load_extension("spatial")

dcas_path = (
  f"s3://{os.environ['S3_BUCKET_NAME']}/"
  "staging/dcas_output/iso_a3=KEN/year=*/month=*/day=*/*.parquet"
)

conn.sql(f"""
CREATE OR REPLACE TABLE dcas AS
SELECT *
FROM read_parquet('{dcas_path}', hive_partitioning=true)
WHERE year=2025 AND month=9 AND day=10
""")

41.4.2 Quick checks¶

-- Count rows
SELECT COUNT(*) FROM dcas;

-- Schema overview
DESCRIBE dcas;

41.4.3 Distinct messages¶

SELECT DISTINCT final_message FROM dcas;

41.4.4 Messages distribution¶

SELECT final_message, COUNT(*) AS n
FROM dcas
GROUP BY final_message
ORDER BY n DESC;

41.4.5 Per-farm history¶

SELECT date, farm_unique_id, crop, growth_stage, final_message
FROM dcas
WHERE farm_unique_id = '4023361'
ORDER BY date DESC
LIMIT 20;