Uganda is one of Africa’s leading coffee exporters and producers. With 70% of Ugandans dependent on agriculture for their livelihoods, maximizing coffee yields is one pathway to improving livelihoods of smallholder farmers in the country.
The Uganda Coffee Agronomy Training (UCAT) program, implemented by the non-profit foundation Hanns R. Neumann Stiftung and TechnoServe, aims to maximize yields by improving the use of good agricultural practices among coffee farmers.
To understand whether training improves yields, we need to be able to measure them. Common yield measurement methods range from asking farmers how much they harvested in a given period, to harvesting and weighing the full yield on a plot, to data-driven approaches using machine learning to count coffee cherries. Each method brings trade-offs in terms of precision, rigor, and cost. To date there is no agreement on the best way to measure coffee yields.
In this context, our team in Uganda tested a new protocol to measure coffee yields across 3,264 farms as part of the International Food Policy Research Institute’s impact evaluation of UCAT. This protocol offers a cost-effective alternative to full yield measurement by harvesting and measuring yields from a random sample of coffee trees on each farm, and extrapolating these results to the full size of the farm.
Laterite’s Data Operations Associate Job Obuya and Senior Field Supervisor Josh Wegoye outline the protocol, and share lessons learned. By sharing this protocol with other researchers, we hope to contribute to this growing field of research, and to collaboratively test and improve this approach.
Coffee measurement protocol
Our protocol was led by 124 enumerators divided into three teams to survey farmers and measure on-farm yields in the Kyenjojo, Kagadi and Kibaale districts of Uganda. The multi-team approach fostered accountability and facilitated quality checks.
Team A – sampling coffee trees
Team A were trained on the agronomic practices taught to farmers through the UCAT program. Once trained, Team A visited farmers to seek their consent to participate in the survey, and to identify their coffee farm plots.
Team A then walked the perimeter of the plot, measuring its perimeter and calculating the acreage using a GPS device.
Next, the enumerator walked the longest transect of the plot, which they marked with a ribbon (see below). The enumerator recorded how many steps this took in the tablet, which then automatically divided the number of steps into four and instructed enumerators how many steps to take to reach each of the three sampling points. At each sampling point, the enumerator selected a mature coffee tree closest to the sampling point to randomly sample.
Enumerators marked the sampled trees with paint and ribbon and entered the tree’s approximate coordinates into the tablet using the GPS device.
Team B – harvesting the coffee
The next day, Team B visited the farmers who had consented to coffee cherry picking, reconfirmed their consent to participate and repeated the aims of the study.
Team B then located and harvested the three identified trees of both red (mature) and green (non-mature) coffee cherries. They weighed the coffee on the spot, and reimbursed the farmer for the harvested coffee.
Team C – yield measurement
Finally, Team C collected the three bags of coffee cherries per farm from Team B for measurement. This happened in several steps.
First, Team C weighed the three bags of coffee cherries separately using weighing scales. They then put the cherries from each tree into sieves to separate coffee into grades by size. Weighing cherries by grade allowed us to avoid under-counting the weight of small (green) cherries, which would normally have ripened and increased in weight before the harvest.
Once the full harvest from each farmer was weighed and sorted, enumerators randomly selected 100 cherries from each grade and weighed them independently by grade. This allowed us to estimate the number of cherries of each grade and inflate the weight of the small ones to full size.
Team C recorded this information electronically for analysis and gave the harvested coffee to local farmers and traders, who used the cherries for compost.
Challenges and lessons learned
Data collection timing. It’s important to visit farmers before they start harvesting, but not so early that the cherries are left over from the previous season. Before planning our visits, we first asked farmers when they planned to harvest via a phone survey.
Variation in plot sizes. Our assumptions about plot sizes did not always match reality. In practice, plots varied significantly: some were up to 5 acres in size with more than 100 trees, and others had less than 20. This meant that our sampling protocol needed to be flexible. For example, on plots with no coffee trees, we surveyed the farmer but did not collect yield data.
Farmer compensation for coffee. We needed to pay farmers double the market price in compensation for harvesting their green cherries. The increased price was to account for the fact that we expected the cherries to weigh more if they were left to ripen.
Long data collection period. For this project, our enumerator teams worked in study villages for three consecutive months, compared to 1-2 weeks for an average data collection project. This led to some fatigue, especially towards the end of data collection. We made sure to have daily debriefs, regular breaks, and hold team building activities to maintain morale.
Do you want to test this protocol in your research, or suggest ways to improve it? Get in touch!
Job Obuya is a Data Operations Associate, and Josh Wegoye is a Senior Field Supervisor, both of Laterite Uganda.