top of page

Mapping in-field spatial maize yield variability in Tanzania using very fine spatial resolution imag

Having been brought up in a rural village in Kenya, our source of livelihoods has been mainly small-scale crop farming. The question of where and why the yield differs in our crop field influenced my career choice of studying GIS and remote sensing. Under the support and guidance of the STARS project team, I was able to carry out a field study in Kilosa district, Tanzania. The study was a proof-of-concept to determine the extent to which very fine spatial resolution remote sensing imagery from Unmanned Aerial Vehicles (UAVs) and satellites could be used to assess field-level maize yield variability. Most of the Sub-Saharan Africa countries have crop fields characterized by small area, fuzzy boundaries and mixed farming systems which complicate most remote sensing studies.

In order to find out how yield varies within the field, I carried out field work with support from the University of Maryland team leader, Dr. Jan Dempewolf and the Sokoine University of Agriculture; Prof. Tumbo and Dr. Mourice. The team supported me throughout my fieldwork where I was able to interview 54 farmers within the two 1x1 km STARS study sites. The fieldwork was carried out in October 2015. The data collected include amount of maize harvested and area planted. I also interviewed farmers on crop management factors. The results of the interview were used to explain the reason for yield variability.

Fig 1. Field digitizing with guidance from one of the farmers, Mr. Semangeni (left)

Using UAV and Worldview (WV) satellite images, I was able to derive various vegetation indices and establish relationships with yield at different stages of maize growth. The results indicated that the best period for assessing maize yield in Kilosa is at the end of May (60-75 days after planting) assuming most farmers sow maize in February/March. The best index was found to be Enhanced Vegetation Index (EVI) which outperformed the commonly used Normalized Difference Vegetation Index (NDVI) as it was able to explain 61% of maize yield variability. The good performance of EVI was attributed to its high sensitivity to high biomass and reduction of atmospheric effects. Regarding management factors, weeding and method of tilling showed to be a major source of yield variability. Interestingly, WorldView imagery proved to be more useful than UAV images. This was not expected as the UAV images have a very high spatial resolution. The plausible reason for the relatively low performance of UAV images was the spectral overlap in bands as well as the high sensitivity of fine spatial resolution (0.05 m) to vegetation structures which are highly heterogeneous in space and time. Fig 2 shows the maize yield variability derived within the two study sites Gongoni (left) and Mbuyuni (right). The black boundaries indicate the sampled fields. Fig 3 Indicates linear relationship (red line) of maize yield relationship derived between EVI and maize yield (ton/ha) from the 54 sampled fields.

Fig 2. In field maize yield variability derived within the two study sites Gongoni (left) and Mbuyuni (right). The black boundaries indicate the sampled fields.

Fig 3. Linear relationship (red line) of maize yield relationship derived between EVI and maize yield (ton/ha) from the 54 sampled fields.

Although with fine spatial resolution imagery we were able to show detailed in-field yield variability, there was also a complicating factor. The presence of weeds and non-maize crops such as pigeon peas and sunflower lead to contamination of maize crop reflectance detected by the sensors. This had a significant effect on the maize yield-vegetation index relationship. Despite this, fine spatial resolution imagery still showed to have great potential for improving field-level maize yield assessment in such heterogeneous landscapes. To improve results further, there is a need for accurate field level production data obtained from destructive sampling, classification of maize crops using such methods as multi-temporal texture analysis, spectral thresholding and inclusion of crop height models in classification. Also, accurate alignment of multi-temporal imagery is highly recommended since a slight misalignment will have a great effect on the final results.

Although the study I undertook was challenging, it gave me a great learning opportunity of conducting field work, data processing, in-depth understanding of field-level yield variability and also technical skills of writing scientific reports. I very much appreciate the support of my supervisors Dr. Anton Vrieling and Dr. Raul Zurita Milla for their guidance during my research period.

Profile: Steve Kibet

Steve is a recently graduated MSc student in the field of Geo-Information and Earth Observation in Natural Resource Management from University of Twente, ITC, Netherlands. His field of interest is in food security studies, focusing on the application of GIS and remote sensing in improving food security situation in developing countries. Prior to his maiden trip to Netherlands for studies, Steve worked as an independent consultant with International Food Policy Research Institute (IFPRI-HarvestChoice) and Spatial Development International (SpatialDev) as a spatial data analyst based in Nairobi. During the course of his studies, together with three other colleagues, he co-founded an NGO, Geo-Information for Natural Resource Management (GeoNAREM)

**This post is originally published on

Featured Posts
Recent Posts
Follow Us
Search By Tags
bottom of page