A Weekly Multi-Modal Dataset of Soil Physicochemical Parameters, Environmental Conditions, Irrigation Records, Crop Growth Observations, Harvest Parameters, and Field Images from Six Cowpea and Maize Varieties in Sumbrungu, Bolgatanga, Ghana

Published: 22 September 2026| Version 1 | DOI: 10.17632/3wzp9nz3fh.1
Contributors:
,
,
,
,
,
,
,
,
,
,
,
,
,
,
,

Description

This dataset presents weekly multi-modal field trial data collected at the Sumbrungu Experimental Farm, Bolgatanga, Upper East Region, Ghana, under the LSACROFT Project (Development of a Low-cost Soil Analysis Device for Real-time Soil Monitoring, Crop Recommendation, and Fertilizer Optimization). Data were collected over a complete 15-week growing season from 14 March 2026 to 26 June 2026, covering six crop varieties: three cowpea (IT-17K-1367-2-3, SARI-TUGA, TUZIE-VALLENG) and three maize (Sanzal-sima, Wang-daata, Bahilifa), under a Randomised Complete Block Design with three replications and one control treatment. The dataset integrates seven complementary data streams. First, weekly soil physicochemical parameters measured using a low-cost customized soil sensor, including electrical conductivity, pH, nitrogen, phosphorus, potassium, moisture, temperature, salinity, and total dissolved solids, each retained at three processing stages: raw, unit-converted, and laboratory-calibrated (2,713 readings). Second, weekly atmospheric environmental parameters including mean air temperature, relative humidity, and total rainfall. Third, weekly sprinkler irrigation records including number of events, volume per event, total volume applied, and water source. Fourth, all farm management activities per plot per week including sowing, fertilizer application, weeding, thinning, and harvesting. Fifth, weekly cowpea crop growth and harvest observations including plant height, leaf count, stem diameter, growth stage, pod count, pod length, fresh weight, dry weight, harvest index, 100-seed weight, and stover yield (116 records). Sixth, weekly maize crop growth and harvest observations including plant height, leaf count, stem diameter, growth stage, ear length, rows per ear, grains per row, and 100-grain weight (144 records). Seventh, field photographs of all plots from Week 3 to Week 12. The dataset is intended for machine learning model development, TinyML embedded AI deployment, crop yield prediction, soil sensor calibration research, and AI-driven agricultural advisory system design. It is particularly relevant for researchers working on precision agriculture and climate-resilient food systems in the Guinea Savanna and Sudan Savanna zones of sub-Saharan Africa. All files are in Microsoft Excel format. Images are provided as a zipped archive. The dataset is released under CC BY 4.0.

Files

Steps to reproduce

The dataset was collected at the Sumbrungu Experimental Farm, Bolgatanga, Upper East Region, Ghana, over a 15-week growing season from 14 March to 26 June 2026. A Randomized Complete Block Design was used with three replications and one control treatment across six crop varieties: three cowpea (IT-17K-1367-2-3, SARI-TUGA, TUZIE-VALLENG) and three maize (Sanzal-sima, Wang-daata, Bahilifa). Seeds were sourced from CSIR-SARI and sown directly in Week 1 under minimum tillage. Soil physicochemical parameters were measured weekly from Week 3 to Week 15 using a low-cost customized soil sensor (Halisense, Model v2.1-BGA001). The sensor was inserted at a standardized depth at each plot and multiple readings were recorded at 10-second intervals per session. The sensor simultaneously measures moisture, electrical conductivity, pH, nitrogen, phosphorus, potassium, salinity, total dissolved solids, and temperature. All readings were processed through three stages: raw sensor output, unit conversion (mS/cm to dS/m for EC; mg/kg to % for N; mg/kg to cmol/kg for K), and laboratory calibration using a 3-point linear regression. Weekly environmental data including mean air temperature, relative humidity, and total rainfall were recorded using a digital thermohydrometer and a rain gauge installed on site. Weekly irrigation was applied using a sprinkler system drawing from a borehole at a fixed rate of 22 liters per minute for 120 minutes per event, giving 2,640 liters per event. The number of events per week was adjusted based on rainfall received. All management activities including sowing, fertilizer application, weeding, thinning, and harvesting were logged per plot per week with dates, inputs, and doses. Crop growth observations were recorded weekly for all plots. Plant height was measured from soil surface to the tip of the highest leaf using a measuring tape. Leaf count was the total number of fully expanded leaves per sampled plant. Stem diameter was measured at the base using digital calipers. Growth stage was assessed visually. Harvest parameters including fresh weight, dry weight, pod count, seeds per pod, 100-seed weight, harvest index, and stover yield were recorded at physiological maturity for cowpea, and ear length, rows per ear, grains per row, and 100-grain weight for maize. Weekly field photographs were taken at each plot from Week 3 to Week 12 using a smartphone camera under natural lighting. Images were captured from a standardized standing position above the crop canopy. GPS coordinates were embedded in image EXIF metadata. All data were recorded in standardized Excel templates and verified weekly by the field supervisor before entry into the final dataset.

Institutions

Categories

Agronomy, Crop Science, Soil Science, Machine Learning, Agricultural Sensor, Ghana, Maize, Cowpea

Funders

  • Science Granting Councils Initiative
    Grant ID: SGCI-GH-ZM-2025-019
  • Ministry of Environment, Science, and Technology, Ghana
    Grant ID: SGCI-GH-ZM-2025-019
  • National Science and Technology Council
    Lusaka Province, Lusaka
    Grant ID: SGCI-GH-ZM-2025-019

Licence