Can drone images predict within-field variability in soil fertility? -A Case Study in the Northern Region of Ghana

Yussif Baba Kassim^1,2,3*Francisco Pinto Espinosa¹Dilys Sefakor MacCarthy⁴Prem Bindraban⁵Ngonidzashe Chirinda²Tjeerdjan Stomph¹Paul Christiaan Struik¹

• ¹Centre for Crop Systems Analysis, Wageningen University and Research, Wageningen, Netherlands
• ²Agricultural Innovation and Technology Transfer Center, Mohammed VI Polytechnic University, Benguerir, Morocco
• ³CSIR-Savanna Agricultural Research Institute, Nyankpala, Ghana
• ⁴Soil and Irrigation Research Centre, College of Basic and Applied Sciences, University of Ghana, Accra, Ghana
• ⁵International Fertilizer Development Center, Muscle Shoals, Alabama, United States

Background Soil fertility varies within fields of smallholder farmers in Africa. Drone-based field mapping may quantify this within-field variability with high resolution. This study analyzed if variation in spectral vegetation indices from early season weed cover could offer criteria to quickly assess heterogeneity in soil fertility. We tested (i) whether within field spatial patterns in early season weed cover and soil organic matter could be correlated and (ii) whether predicted soil organic matter could indicate within-field heterogeneity in crop yields.We collected images of early season weed cover using a DJI Phantom 4 proV2 drone and data on maize and soybean final above-ground biomass from on-farm experiments, conducted in Bognaayili and Gauwogo (northern Ghana), during 2022 and 2023 cropping seasons. There were eight experiments in total, i.e., two of each crop at each site and in each year. In these experiments, we varied planting density, variety, mulching, ridging, and fertilizer application, as management options to increase productivity. Spectral vegetation indices extracted from early season weed cover were used to predict soil organic matter.Variation in spectral vegetation indices from early season weed cover was higher in Bognaayili than in Gauwogo. Predicted soil organic matter from a model built with spectral vegetation indices had a significant relationship (Radj 2 = 0.54) with measured soil organic matter in Bognaayili, but not in Gauwogo. In Bognaayili, predicted soil organic matter was significantly and positively correlated with soybean above-ground biomass in 2022 (r = 0.53) and 2023 (r = 0.65). There was no relationship observed between predicted soil organic matter and maize above-ground biomass.The use of spectral vegetation indices from early season weed cover images as proxy for within-field variation in soil organic matter is a promising option although it still requires some soil sampling for organic matter analysis. Incorporating drone-based early season weed cover assessment into field crop experimentation by researchers, to explore inherent soil organic matter variations would lead to detailed understanding of the characteristics and delineation of areas with lower or higher productivity in regions where soil organic matter is limiting crop productivity.