Site-specific fertilizer recommendation using data driven machine approaches enhanced wheat productivity and resource use efficiency

The productivity of agriculture in Africa is low due to limited and/or inefficient use of inputs. The customary use of blanket fertilizer recommendation in the regions further undermines productivity and efficiency. It is thus essential to develop customized recommendations that consider climate and farming systems. Countries such as Ethiopia are seeking site-specific fertilizer recommendations (SSFR), customised to improve yields, profit, and ecological benefits. The objective of this study was to validate the performance of SSFR developed for wheat based on machine learning in comparison to fertilizer recommendations based on local blanket fertilizer recommendations (LBFR) and national blanket fertilizer recommendations (NBFR). On-farm validation trials were established in 2021 on 277 smallholder fields in four major wheat-growing regions of Ethiopia. Replicated trials using randomized complete block design were established on farmers’ fields Farm management history, grain, straw, biomass, fertilizer and grain prices data were collected using Open Data Kit (ODK) tools and analysed using R statistical package. The performance of SSFR for improving wheat productivity, profitability, and nutrient and rainwater use efficiencies were assessed. Results showed that wheat grain, biomass and straw yields were significantly higher with SSFR. Grain yield increased by 16% (0.73 Mg ha-1) and 25% (1.04 Mg ha-1) with SSFR compared to NBFR and LBFR, respectively. SSFR significantly increased straw yields which is valuable as livestock feed and soil cover. Averaged cross all sites, SSFR significantly increased nitrogen use efficiency by 30% compared to NBFR and water use efficiency by 33% compared to LBFR. The partial profit gain per hectare per season due to SSFR was USD 580 compared to the LBFR and USD 412 compared to the NBFR. The results showed that SSFR has very good potential to increase smallholder productivity, profit, and resource use efficiency in wheat production. The steps across the ‘data -analytics-dissemination’ ecosystem is documented and automated for application to other crops and scaling to other countries.