Open AccessUsing out-of-sample yield forecast experiments to evaluate which earth observation products best indicate end of season maize yields
Author(s) -
Frank Davenport,
L. Harrison,
Shraddhanand Shukla,
G. J. Husak,
Chris Funk,
Amy McNally
Publication year - 2019
Publication title -
environmental research letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.37
H-Index - 124
ISSN - 1748-9326
DOI - 10.1088/1748-9326/ab5ccd
Subject(s) - hectare , yield (engineering) , production (economics) , livelihood , environmental science , product (mathematics) , baseline (sea) , metric (unit) , sample (material) , tonne , work (physics) , agricultural engineering , agricultural economics , environmental resource management , econometrics , geography , mathematics , economics , agriculture , operations management , engineering , materials science , chemistry , chromatography , geology , metallurgy , mechanical engineering , oceanography , geometry , archaeology , macroeconomics
In East Africa, accurate grain yield predictions can help save lives and protect livelihoods. Regional grain yield forecasts can inform decisions regarding the availability and prices of key staples, food aid, and large humanitarian responses. Here, we use earth observation (EO) products to develop and evaluate subnational grain yield forecasts for 56 regions located in two severely food insecure countries: Kenya and Somalia. We identify, for a given region and time of year, which, if any, product is the best indicator for end-of-season maize yields. Our analysis seeks to inform a real-world situation in which analysts have access to multiple regularly updated EO data products, but predictive skill corresponding to each may vary across these regions and throughout the season. We find that the most accurate predictions can be made for high-producing areas, but that the relationship between production and forecast accuracy diminishes in areas with yields averaging greater than one metric ton per hectare. However, while forecast accuracy is highest in high production areas, in many of these regions, the forecast accuracy of models using EO products is not better than a set of baseline models that do not use EO products. Overall, we find that rainfall is the best indicator in low-producing regions and that other EO products work best in areas where yields are relatively consistent, but production is still limited by environmental factors.