A farmer data-driven approach for prioritization of agricultural research and development: A case study for intensive crop systems in the humid tropics

Context: Intensive rice-maize sequences in Southeast Asia can include up to three crop cycles per year. Indonesia is the third and fifth largest rice and maize producing country worldwide, and domestic demand for both crops will increase in the future. Novel, cost-effective and less time-consuming approaches are needed to identify causes of yield gap at national level. Objectives: Here, we propose a farmer data-driven approach to prioritize investment in agricultural research and development (AR&D) programs. Methods: We collected data on yield, management practices, and socioeconomic variables from 1,147 smallholders’ fields in intensive rice and maize cropping systems, from 2017 to 2018, across ten provinces in Indonesia, which include a wide range of landscape positions (upland, lowland, tidal), water regimes (irrigated and rainfed), and cropping intensities (from single to three cycles per year on the same piece of land). Separate data were available for each rice and maize cycle included in the annual crop sequence. We used conditional inference trees, random forest regression, and comparisons among high- versus low-yield fields to identify key agronomic and socioeconomic factors explaining yield variation. Results: For a given field and crop species, there was a significant positive correlation between yield in one season and that in subsequent seasons. In contrast, there was poor correlation between rice and maize yields in cropping systems including both crops. Socio-economic factors such as years of farming experience and access to extension services and inputs explain variation in average yield gap across provinces. In turn, agronomic factors such as nutrient input rates, splits and timing, establishment date, and pest control, explained yield gaps in farmer fields. Overall, these findings were not consistent with expectations from local researchers about on-farm yield constraints. Conclusions: Our study shows that a modest investment to gather farmer survey data, together with robust spatial frameworks to guide data collection, proper statistical methods to analyze the data, and crop modeling to estimate yield potential, can help identify yield constraints for areas representing millions of hectares of rice and maize. Significance: Our study provides useful information for guiding investments in AR&D programs at national and sub-national level for improving crop production by closing current yield gaps.