English, Kiswahili, Chichewa, Chitumbuka, & Gikuyu

CountryNamed PlacesLast UpdateLink
Burundi11918August-15-2026 14:02 EATBurundi - August-15-2026.xls
Ghana20531August-15-2026 13:06 EATGhana - August-15-2026.xls
Kenya30894August-15-2026 13:10 EATKenya - August-15-2026.xls
Malawi10214August-15-2026 13:11 EATMalawi - August-15-2026.xls
Mozambique85707August-15-2026 13:23 EATMozambique - August-15-2026.xls
Rwanda16481August-15-2026 13:28 EATRwanda - August-15-2026.xls
Tanzania18492August-15-2026 13:31 EATTanzania - August-15-2026.xls
Uganda12301August-15-2026 13:32 EATUganda - August-15-2026.xls
Zambia38297August-15-2026 13:38 EATZambia - August-15-2026.xls
Zimbabwe31658August-15-2026 13:42 EATZimbabwe - August-15-2026.xls

Welcome!


Background: this page displays weather forecasts that have been formatted for SMS messages destined for use by smallholder farmers in Africa, community radio stations, and community knowledge workers. The first test of these forecasts in the field was conducted in Kenya, with the following comment from the person in charge of content for that test (Natalia Pshenichnaya, mAgri Business Development Manager, GSMA):
"We were providing weather information by SMS for almost two months by now. Overall feedback is positive - most of the farmers say the information was accurate, while few mentioned discrepancies. For me however the main indicator of how information is useful - is the willingness to pay for the info. That is the point where you can actually see how valuable the information is to the farmers. 29 out of 42 (70%) farmers are ready to pay for the service, and the rest of them say they don't have money for that, with two of them mentioning they couldn't assess the usefulness of the service (didn't see it being useful). Honestly speaking, those numbers seem to be very exciting for me, and its not a hypothetical survey but an assessment of a real service by users."
Methodology: the method used for the creation of phrases rely on satellite-based forecasts (combined with ground-level information) and the use of econometric/statistical modeling which is evolving. The most important aspect of the modeling is the exploitation/detection of strong positive spatial autocorrelations that persist in weather data. Such autocorrelations are positive to proximate longitudes and latitudes. As proximity reduces (especially following a longitude, but not a latitude), spatial autocorrelations are reduced. On a global basis, climates themselves have negative spatial autocorrelations by longitude, and largely positive spatial autocorrelations following a latitude. As we proceed and learn from the field, we are fine tuning the methodology to improve performance.

Place names: Place names are "geo-tagged" locations. We emphasize populated places, markets (places where goods and services are bought and sold), administrative locations (capitals of regions, etc.), and similar locations that might be typed into an automated radio broadcasting system, chat bot or similar system. Given the proximity of many of these place names to each other (e.g. within a kilometer of each other, etc.), we exploit positive spatial autocorrelations to generate local forecasts.

Thanks: My thanks to a number of partner organizations for their inputs, including the Grameen Foundation, Farmer Voice Radio, Farm Radio International, and the GSM Association. Special thanks are for financial support offered to the Bill and Melinda Gates Foundation, and INSEAD (funding from the INSEAD Chair Professorship for Management Science). Finally, thanks are offered to the team of programmers involved in this effort from INSEAD and ICON Group International.

Philip Parker, PhD
INSEAD Chair Professor of Management Science