Agroecology, Ecosystem Services and Crop Health and Productivity: A Participatory Geospatial Analysis in Northern Malawi
Abstract
Agroecology (AE) is a cost-effective alternative to increasing productivity and enhancing ecological integrity. Geospatial techniques are cost-effective for near real-time monitoring of earth systems. In smallholder agricultural systems, however, the application of geospatial techniques is limited and understanding of the impact of agronomic practices on ecosystems and crop productivity is underexplored from a geospatial perspective. Therefore, spatial participatory techniques (PPGIS) were applied to explore the relationship between AE, ecosystems, and biodiversity conservation by comparing AE with non-AE farms. Remote sensing techniques were applied to assess the impact of AE on crop health and to prospectively predict crop health using leaf area (LAIs) and vegetation indices (VIs). Machine learning and statistical methods were applied to estimate groundnut productivity from observed yield and satellite-derived VIs and to identify the optimal growth stage for yield estimation. Finally, to overcome the challenge of dense cloud cover in complex heterogeneous agricultural landscapes, optical and radar remote sensing data were integrated to develop a method for mapping crop types and land cover.
The PPGIS activities revealed that the AE farmers understand the linkages between farm-level practices/processes and ecosystem services compared with the non-AE farmers and prioritize ecologically friendly conservation strategies. Farms on which agroecological methods were implemented were healthier, with average seasonal LAIs for maize/pigeon pea (1.28m2/m2) and maize/beans (1.29m2/m2) compared with 0.97m2/m2 and 0.80m2/m2, respectively, for non-AE farms. Random Forest (RF) regressions prospectively predicted crop health for maize/beans (R2=0.90, root mean square error [RMSE] =0.32m2/m2), and maize/pigeon pea (R2=0.88m2/m2, RMSE=0.42m2/m2) on the AE farms, but results for non-AE farms were not statistically significant. The groundnut yield estimation revealed that the RF model (R2=0.96 and RMSE=0.29kg/ha) outperformed other models in estimating groundnut yield and the optimal growth stage for estimation is the R5/development of seeds stage. Finally, integrating Sentinel-1, Sentinel-2 and PlanetScope data produced detailed crop type and land cover maps with an average overall classification accuracy of 85.70%.
The findings demonstrate the importance of spatial participatory approaches for examining the impacts of agronomic practices on ecosystems and crop health and the need for integrative approaches to food security planning and conservation in the Global South.