Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions
Crop yield predictions and monitoring are important in understanding key challenges in crop production and management to ensure the effective utilization of resources to enhance food security. Over the years remote sensing data and machine learning models have been employed with the help of ground t...
| Autor principal: | |
|---|---|
| Formato: | Tesis |
| Lenguaje: | Inglés |
| Publicado: |
Universidade NOVA de Lisboa
2024
|
| Materias: | |
| Acceso en línea: | https://hdl.handle.net/10568/141816 |
| _version_ | 1855515913667739648 |
|---|---|
| author | Msangi, F.M. |
| author_browse | Msangi, F.M. |
| author_facet | Msangi, F.M. |
| author_sort | Msangi, F.M. |
| collection | Repository of Agricultural Research Outputs (CGSpace) |
| description | Crop yield predictions and monitoring are important in understanding key challenges
in crop production and management to ensure the effective utilization of resources to
enhance food security. Over the years remote sensing data and machine learning models
have been employed with the help of ground truth data as reference in the estimation of
crop yields across space and time. However, the common machine learning methods often
overlook the spatial heterogeneity inherent in regions leading to sub-optimal estimations.
Moreover, the transferability of the machine-learning model to new environments is rarely
addressed during spatial-temporal predictions.
This study integrates spatial heterogeneity by utilizing the Geographically weighted random
forest model(GWRF). It investigates whether accounting for heterogeneity can improve
spatial-temporal predictions of crop yields and estimate the area of applicability of
the models. The models are tested with maize yield data from farms practising conservation
agriculture(CA) and another group applying the farmers’ conventional practices(CP)
in Zambia and Malawi. The GWRF is compared to the ordinary Random Forest(RF)
model using environmental blocking cross-validation.
The overall performance of the GWRF was better compared to the standard RF model
with RMSE of 1587.731 kg/ha and 1389.206 kg/ha for the CA and CP respectively. The
coefficient of determination (R2) was 0.171 and 0.234 for CA and CP respectively. |
| format | Tesis |
| id | CGSpace141816 |
| institution | CGIAR Consortium |
| language | Inglés |
| publishDate | 2024 |
| publishDateRange | 2024 |
| publishDateSort | 2024 |
| publisher | Universidade NOVA de Lisboa |
| publisherStr | Universidade NOVA de Lisboa |
| record_format | dspace |
| spelling | CGSpace1418162025-01-27T15:00:52Z Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions Msangi, F.M. spatial analysis crop yield conservation agriculture environmental factors food security maize Crop yield predictions and monitoring are important in understanding key challenges in crop production and management to ensure the effective utilization of resources to enhance food security. Over the years remote sensing data and machine learning models have been employed with the help of ground truth data as reference in the estimation of crop yields across space and time. However, the common machine learning methods often overlook the spatial heterogeneity inherent in regions leading to sub-optimal estimations. Moreover, the transferability of the machine-learning model to new environments is rarely addressed during spatial-temporal predictions. This study integrates spatial heterogeneity by utilizing the Geographically weighted random forest model(GWRF). It investigates whether accounting for heterogeneity can improve spatial-temporal predictions of crop yields and estimate the area of applicability of the models. The models are tested with maize yield data from farms practising conservation agriculture(CA) and another group applying the farmers’ conventional practices(CP) in Zambia and Malawi. The GWRF is compared to the ordinary Random Forest(RF) model using environmental blocking cross-validation. The overall performance of the GWRF was better compared to the standard RF model with RMSE of 1587.731 kg/ha and 1389.206 kg/ha for the CA and CP respectively. The coefficient of determination (R2) was 0.171 and 0.234 for CA and CP respectively. 2024-02 2024-05-13T07:50:16Z 2024-05-13T07:50:16Z Thesis https://hdl.handle.net/10568/141816 en Limited Access Universidade NOVA de Lisboa Msangi, F.M. (2024). Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions. Lisboa: Portugal, Universidade NOVA de Lisboa, (p. 50.). |
| spellingShingle | spatial analysis crop yield conservation agriculture environmental factors food security maize Msangi, F.M. Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title | Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title_full | Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title_fullStr | Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title_full_unstemmed | Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title_short | Integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| title_sort | integrating spatial heterogeneity to enhance spatial temporal crop yield predictions |
| topic | spatial analysis crop yield conservation agriculture environmental factors food security maize |
| url | https://hdl.handle.net/10568/141816 |
| work_keys_str_mv | AT msangifm integratingspatialheterogeneitytoenhancespatialtemporalcropyieldpredictions |