Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa
Adoption of CA in smallholder farmers in Africa is (s)low partly due to poor spatial targeting. Mapping the crop yield from different CA systems across space and time can reveal their spatial recommendation domains. Integration of machine learning (ML) and free remotely sensed big data have opened h...
| Main Authors: | , , , , , |
|---|---|
| Format: | Conference Paper |
| Language: | Inglés |
| Published: |
Institute of Electrical and Electronics Engineers
2021
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10568/119867 |
| _version_ | 1855518119994327040 |
|---|---|
| author | Muthoni, Francis K. Thierfelder, Christian L. Mudereri, B.T. Manda, J. Bekunda, Mateete A. Hoeschle-Zeledon, Irmgard |
| author_browse | Bekunda, Mateete A. Hoeschle-Zeledon, Irmgard Manda, J. Mudereri, B.T. Muthoni, Francis K. Thierfelder, Christian L. |
| author_facet | Muthoni, Francis K. Thierfelder, Christian L. Mudereri, B.T. Manda, J. Bekunda, Mateete A. Hoeschle-Zeledon, Irmgard |
| author_sort | Muthoni, Francis K. |
| collection | Repository of Agricultural Research Outputs (CGSpace) |
| description | Adoption of CA in smallholder farmers in Africa is (s)low partly due to poor spatial targeting. Mapping the crop yield from different CA systems across space and time can reveal their spatial recommendation domains. Integration of machine learning (ML) and free remotely sensed big data have opened huge opportunities for data-driven insights into complex problems in agriculture. The objective of this study was to estimate the spatial-temporal variations of maize grain yields from 13-year multi-location on-farm trials implemented across four countries in southern Africa. The agronomic data from the long-term CA trials is used together with gridded biophysical and socio-economic variables. A spatially explicit random forest (RF) algorithm was developed. Spatial variation of yield advantage or loss from CA practices was compared with conventional tillage practices (CP) during seasons with above and below-normal precipitation. The out-of-bag accuracy of the RF model was R 2 = 0.63 and RMSE = 1.2 t ha -1 . The variable importance analysis showed that the altitude, precipitation, temperature, and soil physical and nutrients conditions variables explained most of the variation in maize grain yield. Maps were generated to identify the locations where CA had a yield advantage over CP during seasons with below and above-average precipitation. The CA showed yield gains of up-to 1 t ha -1 during the season with drought compared to CP. In contrast, the CA returned yield losses of similar magnitude during the season with above-normal precipitation, except in Mozambique. The maps on yield advantage will support the spatial targeting of CA to suitable biophysical and socioeconomic contexts. Results demonstrates that multi-source remotely sensed data, coupled with advanced and efficient machine learning algorithms can provides accurate, cost-effective, and timely platform for predicting the optimal locations for the upscaling sustainable agricultural technologies. |
| format | Conference Paper |
| id | CGSpace119867 |
| institution | CGIAR Consortium |
| language | Inglés |
| publishDate | 2021 |
| publishDateRange | 2021 |
| publishDateSort | 2021 |
| publisher | Institute of Electrical and Electronics Engineers |
| publisherStr | Institute of Electrical and Electronics Engineers |
| record_format | dspace |
| spelling | CGSpace1198672025-08-15T13:21:21Z Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa Muthoni, Francis K. Thierfelder, Christian L. Mudereri, B.T. Manda, J. Bekunda, Mateete A. Hoeschle-Zeledon, Irmgard data climate variability conservation agriculture machine learning forest remote sensing Adoption of CA in smallholder farmers in Africa is (s)low partly due to poor spatial targeting. Mapping the crop yield from different CA systems across space and time can reveal their spatial recommendation domains. Integration of machine learning (ML) and free remotely sensed big data have opened huge opportunities for data-driven insights into complex problems in agriculture. The objective of this study was to estimate the spatial-temporal variations of maize grain yields from 13-year multi-location on-farm trials implemented across four countries in southern Africa. The agronomic data from the long-term CA trials is used together with gridded biophysical and socio-economic variables. A spatially explicit random forest (RF) algorithm was developed. Spatial variation of yield advantage or loss from CA practices was compared with conventional tillage practices (CP) during seasons with above and below-normal precipitation. The out-of-bag accuracy of the RF model was R 2 = 0.63 and RMSE = 1.2 t ha -1 . The variable importance analysis showed that the altitude, precipitation, temperature, and soil physical and nutrients conditions variables explained most of the variation in maize grain yield. Maps were generated to identify the locations where CA had a yield advantage over CP during seasons with below and above-average precipitation. The CA showed yield gains of up-to 1 t ha -1 during the season with drought compared to CP. In contrast, the CA returned yield losses of similar magnitude during the season with above-normal precipitation, except in Mozambique. The maps on yield advantage will support the spatial targeting of CA to suitable biophysical and socioeconomic contexts. Results demonstrates that multi-source remotely sensed data, coupled with advanced and efficient machine learning algorithms can provides accurate, cost-effective, and timely platform for predicting the optimal locations for the upscaling sustainable agricultural technologies. 2021-07 2022-06-17T08:40:48Z 2022-06-17T08:40:48Z Conference Paper https://hdl.handle.net/10568/119867 en Limited Access application/pdf Institute of Electrical and Electronics Engineers Muthoni, F.K., Thierfelder, C., Mudereri, B.T., Manda, J., Bekunda, M. & Hoeschle-Zeledon, I. (2021). Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa. 9th International Conference on Agro-Geoinformatics (Agro-Geoinformatics), 26-29 July 2021, Shenzhen, China: IEEE, (p. 1-5). |
| spellingShingle | data climate variability conservation agriculture machine learning forest remote sensing Muthoni, Francis K. Thierfelder, Christian L. Mudereri, B.T. Manda, J. Bekunda, Mateete A. Hoeschle-Zeledon, Irmgard Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title | Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title_full | Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title_fullStr | Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title_full_unstemmed | Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title_short | Machine learning model accurately predict maize grain yields in conservation agriculture systems in southern Africa |
| title_sort | machine learning model accurately predict maize grain yields in conservation agriculture systems in southern africa |
| topic | data climate variability conservation agriculture machine learning forest remote sensing |
| url | https://hdl.handle.net/10568/119867 |
| work_keys_str_mv | AT muthonifrancisk machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica AT thierfelderchristianl machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica AT mudereribt machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica AT mandaj machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica AT bekundamateetea machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica AT hoeschlezeledonirmgard machinelearningmodelaccuratelypredictmaizegrainyieldsinconservationagriculturesystemsinsouthernafrica |