Validating satellite estimates of rainfall

In many regions of the developing world, there is a scarcity of ground-based measuring stations to record environmental conditions such as rainfall and temperature. These data are desperately needed to inform decision making in agriculture, water resource management, energy generation and other sectors.

In the last three decades, institutions have relied increasingly on satellite-derived estimations of environmental conditions. While these data sets are a welcome alternative in areas that have little or no ground-based coverage, their accuracy has not been evaluated properly.

"Take a fairly simple application like measuring rainfall," says Steve Connor, who heads the IRI's Environmental Monitoring group. "Satellite-derived rainfall products have been available for 20-plus years. They're used for food security, famine early-warning systems and now there's an interest to use them for malaria early warning. But what we really don't know is how well these products represent actual rainfall."

So in 2005, Connor brought in Tufa Dinku, an expert on remote-sensing, to evaluate and validate the satellite estimates. Before coming to the U.S. for graduate school, Dinku worked in the Satellite Unit of the National Meteorological Agency of Ethiopia.

Ethiopia relies on a network of stations for rainfall data. But its network, one of the best in Africa, is nevertheless seriously flawed, Dinku says. The country doesn't have enough stations, and the available ones tend to be located in cities, not in rural areas, where their data are most needed.

"This is why satellite data is so important for Ethiopia," he says.

With the cooperation of his former colleagues at the met service, Dinku compared rainfall measurements from 150 stations throughout Ethiopia with that of ten different satellite-based products--including those from NOAA, NASA and the University of Reading in the U.K.

His results show that, while different products vary, satellite estimates for 10-day and monthly rainfall are relatively more accurate than those for daily rainfall. "At the daily scale, the products didn't perform so well," says Dinku, "They're good enough to detect the occurrence of rain, but not very good at indicating the amount of rainfall," he says.

He found a similar relationship for spatial scales. The satellites more closely matched ground measurements at 100- and 200 square- kilometer grids than at 50 square-kilometer grids.

Dinku's work also shows that accuracy is affected by the way in which satellites are calibrated. "The NOAA product, for example, uses the same calibration parameters for all of Africa. But the one from the University of Reading uses different parameters for each region of the continent, and so it performs much better. I've been discussing this with colleagues at NOAA and in the long run, they might try to improve it," he says.

Dinku's work has interested researchers at The Famine Early Warning Systems Network (FEWS NET), which uses satellite and other types of data to identify potential threats to food security around the world. Because of Dinku's results, the organization is considering changing the rainfall estimation products it uses to a TRMM product, says Connor.

Dinku is currently working on a developing a 30-year time series for all of Ethiopia that blends rain-gauge and satellite data. "This will give us a strong historical record for the whole country and provide a method to use going forward."

His ultimate goal is to do this for all of Africa, particularly in the central and sub- Saharan regions, where historical records are patchy at best.

Relevant Links
Validation of satellite rainfall products over East Africa's complex topography

If you would like to learn more about how satellites are used for environmental monitoring, please contact Francesco Fiondella, Communications Officer.