Smallholders in Mali adjust their crop management decisions thanks to satellite monitoring
Farmers and scientists in Mali are learning first-hand how satellite technology can be used to interpret crop performance and alter decision-making on-farm.
“When I saw on the STARS images my cotton crop’s poor response to the application of the Sabunyuman fertilizer I was completely dismayed,” says Usman Sania Berthe, a cotton planter in Sukumba, Mali. “But during later group discussions, I remembered that I was late in applying the recommended dose at the recommended date. That a satellite way out in the sky could lead me to this realization is flabbergasting. For sure, next season I will do my best to be prepared, if fertilizer is available in due time.”
Sukumba (Mali) women farmers pointing at ICRISAT’s SenseFly eBeeAg unmanned aerial vehicle collecting imagery to calibrate relationships between satellite vegetation indices and plant growth variables sensitive to fertilization levels over a smallholder cotton field (Photo: PCS Traore, ICRISAT).
Smallholder systems are characterized by enormous yield gaps. Variability in environmental conditions and management practices result in high landscape heterogeneity, and monitoring productivity across scales remains a daunting challenge. Modern information technologies including remote sensing can help monitor crop performance at levels of granularity increasingly compatible with smallholder farming. This may open practical support applications for precision agriculture, demonstrating that increased productivity and enhanced livelihoods are also possible in complex cropping systems. ICRISAT research under the Spurring a Transformation for Agriculture through Remote Sensing (STARS) project focuses on analysis of the potential of remote sensing to retrieve crop status information at sub-field scales in heterogeneous smallholder rainfed production systems.
The study measures the sensitivity of satellite sensors to on-farm fertility treatments applied to five locally important crops in Mali’s cotton belt: cotton, maize, peanut, pearl millet and sorghum. Vegetation index sensitivity to fertilization is analyzed at the field scale, along with the amplitude and timing of strongest plant responses. Spatial variance components are quantified with respect to fertilization and also within and between field variability associated with management, soil characteristics and position in a catena or grouping of co-evolving soil types.
Plant growth was assessed in 48 smallholder fields (average size: 1.4 ha) distributed across a catena. In each field, six 225 m2 fertilization plots were installed, within each of which five 4 m2 quadrats were sampled fortnightly for crop condition, development stage, chlorophyll content, fractional cover, leaf area index (LAI) and plant height. Synchronously, time series of very high resolution (VHR) satellite and unmanned aerial vehicle images were collected, ortho-rectified and atmospherically corrected to calculate the Normalized Difference Vegetation Index (NDVI).
Strong relationships between NDVI and plant growth parameters were observed at plot scales, implying that NDVI directly reflects differences in plant conditions associated with soil fertility and management patterns. On average, crops clearly responded to fertilization impacting NDVI, but large within-field and within-plot variability indicated that many other factors influence NDVI response to crop growth. A field-level variance decomposition model applied during the crop-specific time window when NDVI response is strongest (e.g. late August for peanut; early October for sorghum) showed that at best 50% of intra-field variance is attributable to fertilization levels, in only half of the fields monitored.
At the landscape scale, the effect of fertilizer applications on crop NDVI response was proportionally small (< 35% total variation) compared to the effect of other management practices (sowing, weeding) and catena position. Large differences in temporal signatures between fields of a given crop species were observed. NDVI response can therefore be only partially benchmarked against a fertilization reference within the field. Standard precision agriculture methods based on NDVI comparisons within one field are of limited use in these heterogeneous landscapes, and accounting for other sources of variability is a prerequisite before fertilizer response can be monitored at scale.
- a-d: Digital Globe’s GeoEye satellite at 2m resolution (a,b) and the SenseFly eBeeAg UAV at 10cm resolution (c,d) capture crop canopy response to fertility treatments in smallholders’ fields of millet on sandy soil (a,c) and sorghum on loamy soil (b,d), but also reveal the expression of multiple other factors of heterogeneity. e: a seasonal time series from DigitalGlobe satellites displays the NDVI time response to increasing fertilization (B-C-D-E-F) for five dominant crops and illustrates how, on a landscape scale, fertilization practice explains less variation in NDVI than crop type and other management factors.
Towards fertilizer gap estimates from satellite
Landscape-scale variability in smallholder farming practices (other than fertilization proper) and soil properties are overriding, confounding factors to resolve before crop fertilizer response can be monitored at scale, and before inferences can be subsequently drawn from crop spectral signatures about fertilization gaps and nutrient use efficiencies. The advent of concomitant high spatial resolution and high return frequency Earth Observation systems is being exploited to that purpose under the European Space Agency’s Sentinel-2 Agriculture project, to which Mali contributes the first developing country pilot as a spin-off to the STARS project.
Monitoring crop growth in heterogeneous, smallholder agricultural parkland landscapes is now possible with time series of very high-resolution imagery. Fertilization impacts NDVI as a proxy of crop response. However, assessing fertilizer response at scale requires prior stratification of the production environment (e.g. catena classes) and field management (e.g. sowing practices).