Precision Agriculture using ICT
How to cite
On, Robert. 2016. "Precision Agriculture using ICT." AEA RCT Registry. AEARCTR-0001574. https://doi.org/10.1257/rct.1574-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
This is a randomized-controlled trial that I am currently running with One Acre Fund. This study seeks to understand if heterogeneity in underlying agronomic conditions is a barrier to adoption of agricultural inputs that are widely understood to have high returns. Suri and Tjernström find evidence for this heterogeneity in both fertilizer and seeds, respectively. To address this heterogeneity, a mobile phone information system will disseminate location-specific agricultural advice to farmers based on statistically interpolated soil chemistry content. This intervention leverages technological scale in two ways: (1) using geospatial statistics and modern machine learning techniques to turn a relatively small number of expensive soil tests into a predictive map of continuous soil characteristics, and (2) scaling the dissemination of this precise information over the mobile phone in a way that would be cost prohibitive for human extension agents. Previous research by Casaburi and Cole have found mobile phones to be an effective medium to disseminate agricultural advice to positively affect farmer behavior.
One Acre Fund has been doing soil tests in all areas of operation to better understand what inputs would be most appropriate for their farmers. Several thousand soil tests are performed using soil spectroscopy (<$1 per sample) which are calibrated using Random Forests with a 10% percent sample of wet chemistry tests ($10-$50 per sample). These soil chemistry results can then be interpolated using Kriging to create a continuous field of soil chemistry predictions. Some of these soil properties interpolate better than others. We can reliably spatially interpolate nitrogen and potassium but have a difficult time explaining variation in phosphorus spatially. This is evident in the semi-variograms produced when estimating the parameters for the Kriging procedure.
Soil acidity (pH) is one characteristic that is spatially well-interpolated and happens to be a major issue for farmers in western Kenya. Under a critical threshold (pH < 5.5), the maize plant will suffer from aluminum photo-toxicity. High acidity also inhibits the absorption of other beneficial nutrients that affect plant growth. This can be remedied by the application of agricultural lime (ground limestone) to the soil and was the key to ending rural poverty and unlocking the green revolution in rural Brazil with similar soils. Agricultural trials run by One Acre Fund sees yield increases of up to 40% by microdosing agricultural lime. Despite this high return and the low cost of agricultural lime, there is very low uptake of agricultural lime (<2%). Our interpolated soil maps show a fairly high degree of variation in soil pH which makes agricultural lime hugely profitable in some areas but less so in others.
This trial seeks to provide this information at scale, targeting poor farmers with specific lime application recommendations over the mobile phone. There is a group of ~4000 One Acre Fund farmers who will be randomly assigned into treatment or control. The control group will receive no messaging and the treatment group will receive messages with location-specific recommendations for the application of agricultural lime or a general recommendation to use lime. These messages will be sent this month (August 2016). In September, farmers will choose which agricultural products they will purchase from One Acre Fund, including the uptake of lime. We will compare the appropriate uptake of agricultural lime in the treatment group with the control group controlling for information spillovers and group leader effects.
Intervention
SMS providing targeted lime recommendations to farmers in Nambale District in Busia County. One SMS is geo-targeted at the farmer, the other is a general nudge to use lime.
Experimental design
RCT randomizing SMS messages to farmers. 1/3 will be control, 1/3 will receive a general SMS recommendation to adopt lime, 1/3 will receive a geo-targeted SMS recommendation to adopt lime.
Design details
Cited by
No citation data available for this registration yet.