Gaplad › Free tools & guides › Sample sensor data
Free datasets · CC0Sample sensor data with gaps, and the answer key.
Realistic logger records with the problems real loggers have: dropouts, outages where everything dies at once, error codes, spikes, frozen sensors and re-installed transducers. Each one comes with a list of exactly what was planted where, so you can check any gap-filling or QC method against it, including ours.
All five are synthetic, generated by a fixed recipe; no real station's data is published here. Free to use for anything (CC0). Every file is small enough for Gaplad's free tier.
Weather station, hourly (6 months)
Air temperature, humidity and wind with random dropouts, a 5-day outage, -9999 codes, spikes and a frozen anemometer.
1 hour · April to September 2025, 4,392 rows
Columns
air_temp_C(°C): diurnal and seasonal cycle plus passing weather systemsrel_humidity_pct(%): anti-correlated with temperaturewind_speed_ms(m/s): daytime peak, weakly linked to the others
What is wrong with it (the answer key)
- about 25-30 short random gaps per sensor (1 to 14 hours)
- a 5-day outage around 62 % of the record where every sensor is missing at once (no predictor can help there)
- the humidity sensor alone out for 7 days around 30 % of the record (temperature can fill it)
- an 18-hour stretch of -9999 error codes in air temperature around 45 %
- a handful of ±25 °C spikes in air temperature
- the wind sensor frozen at 2.4 m/s for 40 hours around 80 %
Download CSV Fill it in Gaplad QC check it List its gaps
Related: Weather station data
River stage and rainfall, 15-minute (60 days)
Stage responding to five storms, a 3-day telemetry dropout, short gaps and a transducer re-installed 25 cm lower.
15 minutes · October to November 2025, 5,760 rows
Columns
stage_m(m): water level, responds to rain with a recessionrain_mm(mm per 15 min): complete; the driver of the stage
What is wrong with it (the answer key)
- 18 short gaps in stage (up to 3 hours)
- a 3-day telemetry dropout at 40 % of the record, in the middle of a storm
- a re-zero at 55 % of the record: the transducer was re-installed and reads 0.25 m lower from then on
Download CSV Fill it in Gaplad QC check it List its gaps
Related: Streamflow and stage
Groundwater level and barometric pressure, hourly (1 year)
A slow seasonal well record with a barometric response, a 23-day gap from a late download visit and a few spikes.
1 hour · July 2024 to June 2025, 8,760 rows
Columns
water_level_m(m): seasonal cycle and a barometric-efficiency responsebaro_kPa(kPa): air pressure at the well head
What is wrong with it (the answer key)
- about 12 short gaps in the level (up to 6 hours)
- a 23-day gap halfway through: the logger filled up before the download visit
- a few +1.2 m spikes
Download CSV Fill it in Gaplad QC check it List its gaps
Related: Groundwater levels
Soil moisture at three depths, 30-minute (120 days)
Volumetric water content at 10, 30 and 60 cm with rain, a 9-day probe outage and a frozen deep probe.
30 minutes · May to August 2025, 5,760 rows
Columns
vwc_10cm(m³/m³): wets quickly after rain, dries steadilyvwc_30cm(m³/m³): a damped, delayed copy of 10 cmvwc_60cm(m³/m³): slower stillrain_mm(mm per 30 min): complete
What is wrong with it (the answer key)
- short random gaps at every depth (up to 5 hours)
- the 30 cm probe out for 9 days at 35 % of the record (the other depths can fill it)
- the 60 cm probe stuck at its last reading for 30 hours at 70 %
- not planted, but worth seeing: a spike detector may flag a few readings where a wetting front or the frozen probe's release moves the value fast. They are real; that is why every QC flag is a suggestion
Download CSV Fill it in Gaplad QC check it List its gaps
Related: Soil moisture
Solar PV, daytime-only logging, hourly (60 days)
PV power and irradiance that are blank every night by design, plus a real 4-day inverter outage.
1 hour · June to July 2025, 1,440 rows
Columns
pv_power_kW(kW): follows irradianceirradiance_Wm2(W/m²): blank at nightpanel_temp_C(°C): recorded around the clock
What is wrong with it (the answer key)
- every night is blank in pv_power_kW and irradiance_Wm2: that is how the logger works, not a defect, and must not be filled
- a 4-day inverter outage halfway through, where pv_power_kW is missing in daylight too
- about 10 short daytime gaps in pv_power_kW
Download CSV Fill it in Gaplad QC check it List its gaps
Related: Any sensor time series
Using them
- Teaching: the answer key makes these good exercises in QC and gap filling: students can check what they found against what was planted.
- Testing your own code: pandas
interpolate(), Rna.approx()or a homemade model; run it and compare with the answer key. - Trying Gaplad: "Fill it in Gaplad" opens the dataset straight in the gap filler; nothing is uploaded, the file comes from this site to your browser.
Have your own record?
Drop it into Gaplad. The free tier runs the whole pipeline (QC, validation, fill, trust report) on up to 20,000 time steps.