Agricultural adaptation: Scaling-up the adoption of rainwater harvesting techniques

The basic intervention follows the training program described in Aker and Jack (2023), which involved training farmers on a specific rainwater harvesting technique (demi-lunes) and collected data on adoption outcomes using in person measurement. This trial introduces village-level variation in training and monitoring protocols as described below. The overarching study objective is to refine the training design to facilitate further scale up by the Ministry of Environment.

2023-11-13

2023-12-03

Rainwater harvesting adoption, measured in two ways as described below

Across all villages, we will measure number of demi-lunes and zai constructed, and will measure the area under adoption. We will also ask questions about the labor and tool costs involved during construction, and households’ beliefs about the benefits of these technologies prior to the agricultural season.


Remote measurement: We will also use the in person monitoring data to train a predictive model using remote sensing imagery. (At the moment, we have access to PlanetScope and Sentinel-1 and -2 datasets; if higher resolution data become available, we will use it too.) One predicted outcome will be adoption, which will be consistently available for all villages, regardless of the monitoring treatment. The model will also be used to predict adoption outside of the study sample of 16 households per village.

Agricultural productivity
Land degradation, soil type and soil quality
Climate resilience

ii) Agricultural productivity
- Survey data: At endline, we will ask households to report total agricultural production, by crop, and total revenue from agriculture, also by crop.
- Remote measurement: We will measure NDVI and other agricultural production proxies.

iii) Land degradation, soil type and soil quality
- Soil type data: At midline, we will conduct soil type measures on areas of non-adopted land. As mentioned above, this will involve collecting data on soil texture and soil structure to determine which types are appropriate for RWH techniques. As we cannot measure soil type in all villages with all farmers, we will stratify by region and treatment and randomly select 100 villages and 8 farmers per village from whom to collect soil data.
- Soil quality data. At midline and endline – funding permitting – we will collect data on soil quality on areas where RWH techniques are adopted and not. As we cannot measure soil quality in all villages with all farmers, we will stratify by region and treatment and randomly select a subset of villages from the entire sample.
- Survey data: At endline, we will ask households to report land retirement, new land brought into production and self-reported soil quality. We will also collect soil quality samples.
- Remote measurement: We will predict land degradation from remote sensing imagery using labels collected at baseline.


iv) Climate resilience
- Survey data: At endline, we will follow Chomitz et al. (or any new instruments developed between now and endline) and measure adaptation and resilience using household self-reports. We will also collect survey-based self reports on weather realizations relative to average.
- Remote measurement: We will merge village locations with high resolution temperature and precipitation data to measure deviations from long run weather patterns.

Villages were assigned to a control group or one of the treatment arms. Three main cross cutting treatments are assigned at the village level.

1) Rainwater harvesting technique:

Farmers are trained on where and how to construct rainwater harvesting techniques. Specifically, villages are trained on (a) demi-lunes, (b) zai or (c) both. The training order was randomized among villages trained in both techniques.

2) Monitoring approach

Villages were assigned to receive either prior information about monitoring or not. We originally intended to assign households to receive in-person monitoring or not (with all households being monitored remotely), but, based upon prior remote sensing models, we realized that this was not feasible. As a result, all households will receive in-person monitoring, and the monitoring will either be being announced or not. All households in will be visited by an enumerator and Ministry staff to record adoption in May-June 2024, just before planting. The monitors will visit fields where adoption occurred and count the number and spatial extent (i.e., perimeter) of adoption. Villages assigned to “informed” monitoring are informed of the monitoring protocol at the time of the training. This announcement was made during the end-of-training survey, rather than as a public and village-level announcement.

All treatment villages are monitored remotely, by satellite. Additional detail on satellite-based monitoring is discussed below. Villages assigned to “unannounced” monitoring were not informed about this type of monitoring (as satellite data collection is not at the discretion of the research team), nor were they informed that they would receiving a monitoring visit later in the year.

3) Training size

In small-training villages, only the sampled 16 households are invited to attend the training. In large-training villages, the sample 16 households are invited to attend the training, as well as all households interested. In both cases, invitations are done by the village chief.

4) Goal setting

In addition to the village-level treatment variation, one additional treatment is introduced as part of the training. Randomly-selected households are asked to set a goal for adoption as part of a brief survey administered during the training.

A) Goal – self-select: Households are asked if they would like to set a goal of how much land to cover in demi-lunes or zai or both. The decision to set a goal and the goal are recorded by the enumerator.

B) Goal – requested: Households were asked how many demi-lunes or zai they would like to adopt. This is recorded by the enumerator.

C) Goal – control: No goal set and recorded by enumerator.

In villages where monitoring was announced in advance, the timing of the goal setting treatments are varied to either before or after the in person monitoring is announced.

5) Call-In Hotline

In an effort to determine if and how agricultural extension can assist farmers with adoption, in the second year, we will randomly assign a free, call-in hotline for farmers. Villages will be stratified by geography and treatment status and randomly assigned to the hotline (100 villages) or none (306 villages).

Treatments were assigned within geographic strata (department) using the randtreat command in stata and assigning misfits independently across strata using treatment weights (wstrata). All village level treatments were block randomized, for a fully saturated design.

Village

Yes

The sample consists of 406 villages in the Maradi and Zinder regions of Niger. The village sample was constructed during a listing visit starting in December 2022. To be eligible for the study, villages needed to simply have severely degraded land.

Within each eligible village, we conducted a census of all households with severely degraded land (which was self-reported). Amongst eligible households, we then randomly selected 16 households. The baseline survey was conducted immediately after the random sampling, whereby households were asked a number of questions. In addition, the team visited the household’s most severely degraded plot of land and took the GPS coordinates of the perimeter. Some villages did not have 16 households with severely degraded land; as a result, some villages had fewer respondents. Thus, the intended sample was 6,406 respondents, with a final sample of 6,403 respondents. A spillover sample of 6 additional households was also selected from the randomly ordered list of eligible households. The spillover households were not part of the baseline survey.

Number of clusters by treatment: (Note: "remote" means that households were not informed of monitoring in advance, whereas "in-person" means that households were informed of in-person monitoring in advance).
Control, remote sensing 21
Control, in-person 21
DL, Remote, 16 30
DL, Remote, All 30
DL, In-person, 16 29
DL, In-person, All 32
Zai, Remote, 16 30
Zai, Remote, All 31
Zai, In-person, 16 29
Zai, In-person, All 31
Both, Remote, 16 30
Both, Remote, All 32
Both, In-person, 16 29
Both, In-person, All 31
Total 406