Using remote sensing to assess the effect of trees on millet yield in complex parklands of Central Senegal. (September 2020)

• Record Type: Journal Article
• Authors: Leroux, L.; Falconnier, G.N.; Diouf, A.A.; Ndao, B.; Gbodjo, J.E.; Tall, L.; Balde, A.A.; Clermont-Dauphin, C.; Bégué, A.; Affholder, F.; Roupsard, O.
• Is Part Of: Agricultural systems. Volume 184(2020)
• Publication Date: 2020-09
• Subjects: Agroforestry -- Landscape -- Crop yield -- Smallholder agriculture -- Remote sensing -- Faidherbia albida -- Africa; Agricultural systems -- Periodicals; Agriculture -- Environmental aspects -- Periodicals; 338.16
• DOI: 10.1016/j.agsy.2020.102918
• Languages: English
• ISSNs: 0308-521X

Abstract

Abstract: Agroforestry is pointed out by the Intergovernmental Panel on Climate Change report as a key option to respond to climate change and land degradation while simultaneously improving global food security (IPCC, 2019 ). Faidherbia albida parklands are widespread in Sub-Saharan Africa and provide several ecosystem services to populations, notably an increase in crop productivity. While remote sensing has been proven useful for crop yield assessment in smallholder farming system, it has so far ignored the woody component. We propose an original approach combining remote sensing, landscape ecology and statistical modelling to i) improve the accuracy of millet yield prediction in parklands and ii) identify the main drivers of millet yield spatial variation. The parkland of Central Senegal was chosen as a case study. Firstly, we calibrated a remote sensing-based linear model that accounted for vegetation productivity and tree density to predict millet yield. Integrating parkland structure improved the accuracy of yield estimation. The best model based on a combination of Green Difference Vegetation Index and number of trees in the field explained 70% of observed yield variability (relative Root Mean Squared Error (RRMSE) of 28%). The best model based solely on vegetation productivity (no information on parkland structure) explained only 46% of the observed variability (RRMSE = 34%). Secondly we investigated the drivers of the spatial variability in estimated yield using Gradient Boosting Machine algorithm (GBM) and biophysical and management factors derived from geospatial data. The GBM model explained 81% of yield spatial variability. Predominant drivers were soil nutrient availability ( i.e. soil total nitrogen and total phosphorous) and woody cover in the surrounding landscape of fields. Our results show that millet yield increases with woody cover in the surrounding landscape of fields up to a woody cover of 35%. These findings have to be strengthened by testing the approach in more diversified and/or denser parklands. Our study illustrates that recent advances in earth observations open up new avenues to improve the monitoring of parkland systems in smallholder context. Highlights: Millet yield in parkland systems were estimated with high resolution remote sensing images. Satellite-based information on parkland structure improves millet yield estimations. Yield spatial variability is mainly driven by soil nutrients. Trees benefit to crop for a landscape woody cover of ~30–40%. Advances in earth observation open new avenues for parkland systems monitoring.