it4lia/soil_moisture_dataset

Soil Moisture Dataset

Dataset Description

A dataset of soil moisture measurements collected from on-field tensiometers and volumetric sensors, correlated with irrigation records, weather observations, satellite-derived vegetation indices, and static soil and crop characterisation. Covers an anonymised agricultural area in Trentino, Italy, across three regional consortiums.

• Producer: Fondazione Bruno Kessler (FBK) — OpenIoT research unit
• Project ID: fbk.aif.soil_moisture_dataset
• Repository: it4lia/soil_moisture_dataset
• Access: Openly accessible via HuggingFace; pandas is sufficient to read all Parquet files
• Useful for: agronomists, irrigation managers, researchers, and developers working on irrigation decision-support and crop water management

---

Dataset Structure

Repository file tree

soil_moisture_dataset/
├── field_sensor_data_consortium{0,1,2}.parquet       # IoT sensor measurements
├── irrigation_data_consortium{0,1,2}.parquet         # Irrigation data
├── locations_ids_consortium{0,1,2}.parquet           # Location reference
├── historical_weather_data_consortium{0,1,2}.parquet # Historical weather data
├── forecasted_weather_data_consortium{0,1,2}.parquet # 7-day weather forecast
├── weather_data_consortium{0,1}.parquet              # On-site weather station
├── soil_type_data_consortium{0,1}.parquet            # Soil properties
├── remote_sensing_data_final_consortium{0,1}.parquet # Satellite spectral
└── crop_type_data_consortium{0,1,2}.pickle           # Crop information

Consortium summary:

Sub-dataset	C0 (66 loc)	C1 (50 loc)	C2 (26 loc)
field_sensor_data	✓	✓	✓
irrigation_data	✓	✓	✓
locations_ids	✓	✓	✓
historical_weather_data	✓	✓	✓
forecasted_weather_data	✓	✓	✓
weather_data (on-site)	✓	✓	—
soil_type_data	✓	✓	—
remote_sensing_data	✓	✓	—
crop_type_data (pickle)	✓	✓	✓

Join keys

Key	Use
datastream_name	Primary join key across all sub-datasets
datastream_id	Numeric alternative where available

---

Dataset Creation

Data sources (from data provider)

Data type	Origin
Soil moisture & irrigation	On-field IoT sensors owned by the FBK OpenIoT research unit
Weather (on-site)	Data from on-field weather station sensors
Weather (historical/forecasted gridded)	Public weather data
Satellite imagery	Public satellite data
Soil properties	Soil type information
Crop parameters	Crop type information

Collection

• Provider: Fondazione Bruno Kessler (FBK), OpenIoT research unit, Trentino, Italy
• Geographic coverage: Anonymised agricultural area in Trentino (exact locations not disclosed)
• Temporal coverage: 2023 and 2024 growing seasons
• Frequency: Daily
• Format: Apache Parquet (tabular data) + Python Pickle (crop model objects)

Observed date ranges (from files)

Sub-dataset	C0	C1	C2
field_sensor_data	2023-01-03 → 2024-12-03	2024-01-01 → 2024-11-14	2024-04-10 → 2025-08-25
irrigation_data	2023-07-01 → 2024-10-15	2024-05-07 → 2024-12-31	2024-04-15 → 2025-08-25
historical_weather_data	2023-01-01 → 2025-09-30	2023-01-01 → 2025-09-30	2023-01-01 → 2025-09-30
forecasted_weather_data	2023-01-01 → 2025-09-29	2023-01-01 → 2025-09-29	2023-01-01 → 2025-09-29
weather_data (on-site)	2023-01-01 → 2024-12-31	2023-01-01 → 2024-12-31	—
remote_sensing_data	N/A	N/A	N/A—

---

Dataset Statistics

Row counts

Sub-dataset	C0	C1	C2	Total
field_sensor_data	13,346	4,705	4,981	23,032
irrigation_data	1,192	2,587	549	4,328
locations_ids	66	50	26	142
weather_data (on-site)	731	731	—	1,462
historical_weather_data	66,264	50,200	26,104	142,568
forecasted_weather_data	66,198	50,150	26,078	142,426
soil_type_data	66	50	—	116
remote_sensing_data	24,072	20,060	—	44,132
Grand total				358,206

File sizes (compressed Parquet on disk)

Sub-dataset	C0	C1	C2	Sub-total
field_sensor_data	134 KB	38 KB	55 KB	227 KB
irrigation_data	7 KB	12 KB	6 KB	25 KB
locations_ids	5 KB	5 KB	4 KB	14 KB
weather_data	89 KB	45 KB	—	134 KB
historical_weather_data	1,515 KB	1,444 KB	142 KB	3,101 KB
forecasted_weather_data	9,064 KB	10,771 KB	1,372 KB	21,207 KB
soil_type_data	11 KB	11 KB	—	22 KB
remote_sensing_data	1,152 KB	4,497 KB	—	5,649 KB
Total Parquet				~30.4 MB

---

Limitations

• Geographic coverage is a single Italian region (Trentino). Generalisation to other climates is not validated.
• Exact field locations are anonymised; coordinates cannot be used for spatial analysis.
• Sensor placement reflects operational decisions of the FBK OpenIoT unit, which may introduce selection bias toward actively managed fields.
• Crop model parameters (crop_type_data) are defined at consortium level, not per location.

---

Usage

import pandas as pd
import glob

# Load and concatenate field sensor data across all consortiums
# Note: cast ground_offset to float to handle int64 vs float64 difference in C2
dfs = []
for f in sorted(glob.glob("field_sensor_data_consortium*.parquet")):
    df = pd.read_parquet(f)
    df["ground_offset"] = df["ground_offset"].astype(float)
    dfs.append(df)
sensors = pd.concat(dfs)

# Filter to Water Content sensors only (not available in C1)
wc = sensors[sensors["sensor_type"] == "Water Content"]

# Load historical weather (note: datastream_name is the last column)
weather = pd.concat([
    pd.read_parquet(f) for f in sorted(glob.glob("historical_weather_data_consortium*.parquet"))
])

# Load remote sensing (columns are lowercase: ndvi, grvi, etc.)
rs = pd.concat([
    pd.read_parquet(f) for f in sorted(glob.glob("remote_sensing_data_final_consortium*.parquet"))
])
# Replace Inf before use
import numpy as np
rs = rs.replace([np.inf, -np.inf], np.nan)

For crop model objects (requires aquacrop):

import pickle
with open("crop_type_data_consortium0.pickle", "rb") as f:
    crop = pickle.load(f)