The context​

Niger is currently facing its worst food security crisis of the decade. Climate change and the reduction of rainfall are affecting crop cycles, putting agriculture under tremendous pressure. Pearl millet and sorghum, the main food crops in Niger, are rainfed subsistence crops that are increasingly negatively affected by the changing climate. In response to this crisis, Nigerien authorities are seeking data-driven insights to inform policies and interventions. Through the Data Powered Positive Deviance (DPPD) initiative, the GIZ Data Lab, the UNDP Niger Accelerator Lab, GIZ PromAP Niger and the University of Manchester conducted a pilot to find farmers who are particularly successful in cultivating rainfed cereal crops and understand why those farms succeeded.

Using data to identify positive deviants

The project team targeted the southern region of Niger, where sorghum and pearl millet are most cultivated. A two-step approach was followed:

1. Communities were identified in which pearl millet and sorghum yields had been persistently higher than in other communities located within the same agricultural zone. The Humanitarian Data Exchange provided more accurate settlement location data. Copernicus Global Land Cover Layers were used to focus on those communities with recent agricultural activity. Altogether, 12,093 municipalities were analyzed using this method.
2. The communities were manually screened in more detail to identify plots and areas that contributed to the superior performance of the communities. In subsequent field research, farmers in each of these parcels were interviewed to gain greater understanding of any practices or interventions that could be replicated throughout their community.

Accounting for diversity in data

Analyzing the southern region of Niger, the diversity of agricultural zones with specific structural characteristics was accommodated by grouping together different zones with similar agricultural yields of rainfed crops. This approach helped reduce the likelihood of higher performance being driven by structural factors rather than effective farming practices. To facilitate the grouping, Hauswirthet al.'s (2020) classification of agricultural zones has been applied.

Soil-Adjusted Vegetation Index (SAVI) as a Performance Measure

Existing open data sources have been leveraged as proxy measures of performance, being particularly valuable in contexts where data is scarce. Crop biomass has been used as a proxy for the agricultural yield of sorghum and pearlmillet, in this case taking advantage of the Soil-Adjusted Vegetation Index (SAVI). This proved particularly suitable for arid regions with little vegetation, such as Niger. To compute the SAVI Sentinel-2 data was accessed.

Good to know: Sentinel-2 is an Earth observation mission that systematically acquires high spatial resolution visual imagery over land and coastal waters to monitor variability in surface conditions. All data from Sentinel-2 are freely available through the Copernicus Open Access Hub.

The maximum SAVI value was used as a measure of performance, indicating when the crops' biomass and thus yield was highest. To increase accuracy, SAVI values were calculated only for areas classified as "managed or cultivated vegetation" by the Copernicus Global Land Cover Layers. Moreover, SAVI values were calculated only for the rainy season (June to September) when sorghum and pearl millet can be grown as rainfed crops. Limiting the period to the rainy season reduced the likelihood of accidentally sampling other crops grown on the same plot, such as cowpea. Finally, rainy seasons from multiple years (2018-2020) were studied to reduce the impact of individual events that may have had short-term effects on yield but are not representative of long-term performance.‍

‍Comparing against expectations

Having identified the communities to study and calculated SAVI values for them, three more steps remained:

1.     Predicting the average maximum SAVI for each community:

The project team predicted the maximum SAVI value for the 2018-2020 rainy seasons of each agricultural zone using variables such as land cover, soil properties, temperature, rainfall, and soil moisture. Three statistical models were combined for this estimation: (i) a regression model to individually estimate the maximum SAVI for each agricultural zone, (ii) a boosting tree algorithm to capture nonlinear effects on maximum SAVI, and (iii) a neural network.

The results were integrated into a joint prediction to reduce model-specific biases and reduce the susceptibility of the estimates to outliers.

2.    Calculating each community's deviation from this estimate:

The project team then calculated the difference between the maximum SAVI value in each municipality and the expected value from step 1. To standardise these measures the mean maximum SAVI value and the standard deviation of each agricultural zone was determined. This figure, “the residual”, therefore corresponds to higher performance which cannot be explained through environmental factors.

3.    Identifying the communities which exceeded the estimate by the largest amount within their respective agricultural zones:

The residuals of all the communities were compared. Following this approach, 180 communities were identified that performed particularly well. These were then validated and further investigated, identifying those properties that are likely to have contributed to the above-average performance.

Once identified, satellite imagery was used to manually examine each community with local experts to exclude false positives. Incorrect identification may occur as a result of missing or inaccurate data. As an example, some communities have been located where pastures, shrubs, and trees (which score high on SAVI) were misclassified as agricultural land.

Finally, a two-stage semi-qualitative field study was conducted:

1. First, the project team visited various sites to gather initial information about and from the local farmers that had been identified as potential positive deviants.
2. Second, a local agricultural expert proceeded with a more in-depth investigation in eighteen villages to further decipher some of the environmental factors that could explain above-average yields.

What were the findings of the field study?​

The agricultural expert interviewed 179 local stakeholders growing pearl millet and sorghum in four regions (Tahoua, Dosso, Maradi, and Zinder). 35 producers were identified as positive deviants due to their combination of above-average yields and their use of various behaviours and techniques that likely explain them. In particular, the following approaches correlated to particularly strong crop returns:

1. Rainfall and irrigation: Sowing at times likely to result in higher yields and lower seed losses, as well as utilising zaï holes or stone strings to harness rainfall and avoid rainwater runoff.
2. ‍Fertilisers and pest control: Use of advanced technical equipment through field schools for farmers, use of organic fertiliser to supplement mineral fertiliser, and use of pesticides.
3. ‍Field clearing practices: Rather than clearing fields after harvest, millet stalks were left on the field to protect the soil from wind erosion and restore organic matter.
4. ‍Natural regeneration: A mix of active planting (e.g., Gao trees that help regenerate and fertilise the soil as they lose their leaves) and leaving the land to rest, leading to improved water infiltration and recovery of degraded land.

How can better data contribute to better policy? ​

By identifying groups that have outperformed others, policymakers can replicate successful interventions and adjust ineffective ones.

More specifically, several opportunities exist for policymakers to leverage new data relating to the rainfed economy in Niger and thus address the population's food insecurity. Among these are the following:

1. ‍Developing targeted interventions: Analyzing the structural drivers of performance, such as precipitation patterns and soil characteristics, can help policymakers identify specific areas where targeted interventions are needed. For instance, policymakers could provide additional resources to targeted areas to balance unfavourable structural characteristics such as poor soil quality, lack of precipitation and low adoption of efficient irrigation techniques.‍
2. Investing in weather forecasting and early warning systems: Given that high precipitation in the early months of crop cultivation is one of the structural drivers of high yield, policymakers can invest in weather forecasting and early warning systems to enable farmers to make better-informed decisions about when to plant their crops. This could help reduce crop failure due to poor timing of planting and ultimately lead to better yields.‍
3. Supporting farmer field schools: The positive deviants identified in the study utilised farmer field schools to learn and adopt improved techniques and products. Policymakers could support such schools and other farmer-led initiatives to help spread knowledge and best practices across communities.‍
4. Encouraging agroforestry practices: The study also identified the use of assisted natural regeneration as drivers of high yield. Policymakers could encourage agroforestry practices by providing incentives for planting trees and preserving existing vegetation cover. This could help increase soil fertility, reduce soil erosion, and improve the overall health of the ecosystem.

Where do we go from here?​

Evidence from this study on positive deviations in Niger's rainfed agriculture provides policymakers with a valuable evidence-based foundation for combating a worsening hunger crisis. Through dissemination of best practices to other farmers in the same or similar regions, scalable and sustainable solutions can be created that have the potential to address malnutrition and, by boosting the agricultural economy, severe poverty. As most of the data used in this study is open and readily available, policymakers in neighbouring countries facing similar challenges such as Mali, Chad and Burkina Faso can similarly benefit from the approach. Leveraging the power of data and evidence-based solutions, a brighter future may be shaped for agriculture across the wider region, grounded in resilience, innovation and success.