Using remote sensing to assess the effect of trees on millet yield in complex parklands of Central Senegal
Introduction
Scientific and political spheres agree on the need to foster the inclusion or upholding of trees in agricultural systems, in order to tackle the social and environmental dimensions of the sustainable development goals (SDGs, United Nations, 2016). Agroforestry, i.e. the combination of trees and crops or pastures on the same piece of land (Nair, 1993) has been acknowledged as an option to respond to climate change and land degradation (IPCC, 2019).
In sub-saharan Africa, around 40% of people in rural areas live in landscapes with more than 10% tree cover, often agroforestry systems (Zomer et al., 2014). In semi-arid West Africa, traditional parklands are characterized by the deliberated retention of trees on agricultural land (Boffa, 1999) due to the socio-ecosystem services they provide (Sinare and Gordon, 2015). Parklands contribute to the conservation of natural resources and biodiversity, and improve soil fertility and agricultural productivity (Baudron et al., 2019; Bayala et al., 2014; Duriaux Chavarría et al., 2018; Peltier, 1996). Trees compete with crops for resources but they can improve nutrient cycling, soil moisture retention and microclimate (e.g. Kho et al., 2001; Sida et al., 2018).
Studies on the impact of tree on crop productivity were generally conducted at tree-scale where crop performance under tree crown was compared with crop performance in a control area without tree influence (Bayala et al., 2015). Tree density in West African parklands is often very high and some tree species can influence crops beyond their crown (sometimes more than 100 m2/tree, Sileshi, 2016). Finding a control area without tree influence can thus be challenging, which can bias the quantification of trees influence on crops. In addition, parklands are composed of combinations of tree species with different densities and spatial arrangement. Synergies or antagonisms occur between trees and trees effect on crop performance is not likely to be additive. The direction and magnitude of the impact of trees on crop productivity depends on the dominant tree and crop species, and management practices. For instance, nitrogen-fixing Faidherbia albida, was found to improve millet and wheat yield (Bayala et al., 2012; Kho et al., 2001; Louppe et al., 1996; Sida et al., 2018) but not groundnut yield (Louppe et al., 1996). In Burkina-Faso, millet performed better under Adansonia digitata than Parkia biglobosa, the latter being a shading-tree (Sanou et al., 2012). The presence of Grevillea robusta in maize and wheat fields decreased fertilizer use efficiency while the presence of F.albida improved it (Sida et al., 2019).
Though parklands have been the focus of researches for several decades, few studies have tackled the question of the landscape-scale effect of parklands on crop productivity. Research in Ethiopia on the effects of F.albida on barley yields according to different land use systems (Hadgu et al., 2009), and agricultural productivity along a forest-agriculture gradient (Baudron et al., 2019; Duriaux Chavarría et al., 2018) are rare example.
The inter-connection of social, environmental and economic challenges as committed by the SDG calls for systemic and integrated approaches in which landscape scale is particularly appropriate to inform decision making (Reed et al., 2016). Remote sensing provides physical measurements of temporal and spatial development of agroforestry systems (e.g. structure, biomass). It could help account for tree-crop interactions and the resulting impacts on crop productivity. Current statistical models establish relationships between remote sensing vegetation productivity indices and in-situ yields measurements or national agricultural statistics. Until recently, crop growth monitoring and crop yield mapping in smallholder agriculture have relied mainly on low spatial resolution images covering large areas (Leroux et al., 2016, Leroux et al., 2019; Maselli et al., 2000; Mkhabela et al., 2005; Rasmussen, 1992). However, in agroforestry parklands across sub-Saharan Africa, accurate estimates of crop yields are hampered by landscape fragmentation, fields being often smaller than one hectare (Fritz et al., 2015). Diversity in soil conditions, crop management and tree conservation practices further amplifies inter and intra-field yield variability. New satellite or low-cost nanosatellite sensors with high spatial resolution (≤ 10 m) and high revisit frequency (< 2 weeks) are more suited to these complex and spatially variable agricultural systems. These new sensors open unprecedented opportunities to predict and map crop yield in smallholder context. A promising crop yield mapping at field level have been obtained for East and West African farming systems using Sentinel-2, Sentinel-1 and PlanetScope data (Burke and Lobell, 2017; Jin et al., 2017, Jin et al., 2019; Lambert et al., 2018). However these studies masked out trees to capture ‘pure cropped pixels’ (Lambert et al., 2018) and thus masked-out the crop below tree crown and neglected the influence of the tree on crops beyond its crown projection. Though promising, these approaches have usually failed to fully reproduce the wide variability in observed crop yield in farmer fields in sub-Saharan Africa (e.g. Jin et al., 2019; Lobell et al., 2019).
Combining information on vegetation productivity and parkland structure derived from high spatial resolution, satellite images offers the opportunity to capture the variability in crop yield in parkland systems and to identify where and how crop productivity could be improved. Remote sensing have been extensively used to identify and analyzed yield gap (i.e. the difference between observed actual yields and water-limited yields) (e.g. Jain et al., 2017; Jin et al., 2019; Löw et al., 2017; Zhao et al., 2015). In Kenya, Jin et al. (2019) explained more than 70% of maize yield variability by edaphic drivers using remote sensing, crop process-based modelling and machine learning. In parkland systems, analyzing drivers of yield spatial variability could help assess relevant opportunities to optimize parkland management.
The main aim of this study was to assess the role of trees in explaining spatial variations in millet yield in a case-study agroforestry parkland dominated by Faidherbia albida, in the Groundnut Basin of Senegal. To do so, we used high spatio-temporal resolution images (Sentinel-2, PlanetScope and RapidEye) and ground-observations. More specifically, we addressed three questions: (i) Does information on parkland structure (i.e. number of trees per field, tree density, and percentage of tree cover) help improve the accuracy of millet yield prediction in parklands of central Senegal? (ii) What are the main drivers of the predicted spatial variability in millet yield?, and (iii) What is the relative influence of trees compared with the other identified drivers?
We thus propose an original approach combining remote sensing, field data and statistical modelling. This approach was tested for an agroforestry parkland dominated by Faidherbia albida, in the Groundnut Basin of Senegal.
Section snippets
Study area
The study was conducted in 2017 and 2018 in Senegal. The study area (~17 km2) is located in a village named Diohine. The village is at the centre of the main rainfed agriculture area of Senegal (Fig. 1a), the “Old Groundnut Basin”. This name refers to the economic importance of groundnut in the region, since colonial times.
The climate is sudano-sahelian, with annual rainfall ranging from 400 mm to 650 mm. An increasing trend in annual rainfall has been observed since the 1990's (Lalou et al.,
Effects of parkland structure and vegetation productivity proxies, and integration period on millet yield
Proxies of Vegetation productivity explained at least 50% (i.e. R2 > 0.50) of millet yield variability (except NDWI) (Fig. 3a). NDVI and GDVI were the VI with the highest explanatory power corresponding respectively to 32% and 27% of models with R2 > 0.50. Greater accuracy was achieved when proxies for parklands structure (i.e. number of trees, tree density and woody cover) were combined as explanatory variables in the linear regression models (excepted for GDVI where some models based only on
Integrating information on parkland structure improves yield prediction
Our study combined for the first time parkland structure variables with vegetation productivity proxies. We found that a model combining GDVI index integrated over 50–65 days after emergence and within-field number of trees explained 70% of millet yield variability (RMSE = 348 kg/ha). Regardless of the vegetation productivity proxies considered, including proxies of parkland structure improved the accuracy of remote sensing based models (Fig. 3a and Fig. 4c). A major challenge in agroforestry
Conclusion
Agroforestry attracted the attention of policies as an entry point to address climate change and food security challenges (IPCC, 2019). Reliable assessment of crop yields under parkland systems are urgently needed to inform global debates and foster local policy interventions. Few studies have tackled the challenge to assess the effects of agroforestry parklands on crops production beyond tree scale. By adopting landscape scale as an entry point and using cutting-edge remote sensing images,
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgement
This study was supported by the LYSA project (DAR-TOSCA 4800001089) funded by the French Space Agency (CNES), the SERENA project funded by the Cirad-INRA metaprogramme GloFoodS and the SIMCo project (agreement number 201403286–10) funded by the Feed The Future Sustainable Innovation Lab (SIIL) through the USAID AID-OOA-L-14-00006. Sentinel-2 data were obtained from the Theia processing center at CNES (https://theia.cnes.fr/atdistrib/rocket). We acknowledge Planet's Ambassador Program for the
References (80)
Effect of organic matter input on the water balance and yield of millet under tropical dryland condition
F Crop. Res.
(1995)- et al.
The yield gap of major food crops in family agriculture in the tropics: assessment and analysis through field surveys and modelling
F Crop. Res.
(2013) - et al.
Parklands for buffering climate risk and sustaining agricultural production in the Sahel of West Africa
Curr. Opin. Environ. Sustain.
(2014) - et al.
Advances in knowledge of processes in soil–tree–crop interactions in parkland systems in the West African Sahel: a review
Agric. Ecosyst. Environ.
(2015) - et al.
Review of yield gap explaining factors and opportunities for alternative data collection approaches
Eur. J. Agron.
(2017) - et al.
Estimating corn leaf chlorophyll concentration from leaf and canopy reflectance
Remote Sens. Environ.
(2000) - et al.
Retaining forests within agricultural landscapes as a pathway to sustainable intensification: evidence from Southern Ethiopia
Agric. Ecosyst. Environ.
(2018) - et al.
Dry land tree management for improved household livelihoods: farmer managed natural regeneration in Niger
J. Environ. Manag.
(2011) - et al.
Smallholder maize area and yield mapping at national scales with Google earth engine
Remote Sens. Environ.
(2019) - et al.
Assessing the potential of multi-seasonal WorldView-2 imagery for mapping west African agroforestry tree species
Int. J. Appl. Earth Obs. Geoinf.
(2016)
Maize yield estimation in West Africa from crop processes-induced combination of multi-domain remote sensing indices
Eur. J. Agron.
Assessing gaps in irrigated agricultural productivity through satellite earth observations—a case study of the Fergana Valley, Central Asia
Int. J. Appl. Earth Obs. Geoinf.
Field-scale modeling of tree–crop interactions: challenges and development needs
Agric. Syst.
Regional carbon stocks and dynamics in native woody shrub communities of Senegal’s Peanut Basin
Agric. Ecosyst. Environ.
Multi-phenology WorldView-2 imagery improves remote sensing of savannah tree species
Int. J. Appl. Earth Obs. Geoinf.
Spatial carbon, nitrogen and phosphorus budget of a village in the West African savanna - I. Element pools and structure of a mixed-farming system
Agric. Syst.
Early maize yield forecasting in the four agro-ecological regions of Swaziland using NDVI data derived from NOAA’s-AVHRR
Agric. For. Meteorol.
Ring management of soils and crops in the west African semi-arid tropics: the case of the mossi farming system in Burkina Faso
Agric. Ecosyst. Environ.
Are subsidies to weather-index insurance the best use of public funds? A bio-economic farm model applied to the Senegalese groundnut basin
Agric. Syst.
Climate-smart agroforestry: Faidherbia albida trees buffer wheat against climatic extremes in the central Rift Valley of Ethiopia
Agric. For. Meteorol.
The magnitude and spatial extent of influence of Faidherbia albida trees on soil properties and primary productivity in drylands
J. Arid Environ.
Ecosystem services from woody vegetation on agricultural lands in Sudano-Sahelian West Africa
Agric. Ecosyst. Environ.
Sources of water used by trees and millet in Sahelian windbreak systems
J. Hydrol.
Effect of landscape diversity and crop management on the control of the millet head miner, Heliocheilus albipunctella (Lepidoptera: Noctuidae) by natural enemies
Biol. Control
Carbon stock and sequestration potential of traditional and improved agroforestry systems in the West African Sahel
Agric. Ecosyst. Environ.
Maize crop nutrient input requirements for food security in sub-Saharan Africa
Glob. Food Sec.
Inorganic fertilizer use efficiency of millet crop varies with organic fertilizer application in rainfed agriculture on smallholdings in Central Senegal
Agric. Ecosyst. Environ.
Red and photographic infrared linear combinations for monitoring vegetation
Remote Sens. Environ.
Using satellite remote sensing to understand maize yield gaps in the North China Plain
F Crop. Res.
Agriculturally productive yet biodiverse: human benefits and conservation values along a forest-agriculture gradient in southern Ethiopia
Landsc. Ecol.
Cereal yield response to conservation agriculture practices in drylands of West Africa: a quantitative synthesis
J. Arid Environ.
Agroforestry Parklands in Sub-Saharan Africa (No. 34)
Random forest
Mach. Learn.
Satellite-based assessment of yield variation and its determinants in smallholder African systems
Proc. Natl. Acad. Sci. U. S. A.
The potential of satellite-observed crop phenology to enhance yield gap assessments in smallholder landscapes
Front. Environ. Sci.
A Perfect Smoother
Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography
Commun. S ACM
Trend change detection in NDVI time series: effects of inter-annual variability and methodology
Remote Sens.
Greedy function approximation: a gradient boosting machine
Ann. Stat.
Multiple additive regression trees with application in epidemiology
Stat. Med.
Cited by (25)
Calibrating the STICS soil-crop model to explore the impact of agroforestry parklands on millet growth
2024, Field Crops ResearchExploring the use of Sentinel-2 datasets and environmental variables to model wheat crop yield in smallholder arid and semi-arid farming systems
2023, Science of the Total Environment