has several outputs, including price forecasts during the 2009-11 period and maps of which
NDVI pixels are related to observed price changes.
Maps of NDVI time series correlated with price changes
When using NDVI time series to estimate changes in food prices, the question arises, where
should the NDVI be from? Each market can draw from a variety of export locations, which
may change dynamically after each harvest. As some regions have excess grain and have rela-
tionships with traders, some connections may be static and others may be active depending
on the region and the year's production. Here we present maps that show the relationship
between the price seasonality and NDVI anomalies, to demonstrate the areas that are more
related to changes in food prices for a particular commodity and market. We can then make
maps of the NDVI pixels that are most correlated to the local food prices. Although this does
not give us exact information about where food may be coming from, it is a big improvement
over using only the NDVI around each market. In some regions, such as East and Southern
Africa, there are significant movements of grain in cross-border trade.
Plate 18 shows the pixels in Western Africa most related to maize prices in Njamena, Chad
during the 2003-08 periods. The plate shows a clustered band of pixels along the Sudanian
region that are approximately regions that have NDVI most related to the seasonality of maize
prices in Njamena. Since NDVI in the region is highly correlated to rainfall, and rainfall is
likely to be similar across latitudinal bands (Nicholson et al ., 1998), the horizontal band of
pixels is to be expected. Thus the pixels shown in Plate 18 are all highly correlated with each
other and with the growing conditions seen in the export zone.
Plate 19 shows FEWS NET's Markets and Trade Flow Map for Chad. The map provides
a summary of experience-based knowledge of market networks significant to food security.
Maps are produced by USGS in collaboration with other FEWS NET staff, local government
ministries, market information systems, NGOs, and network and private sector partners. The
maps are the result of interviews with experts and local traders, and are meant to represent
normal trading patterns during a normal year. They are also static in time and do not show
seasonal impacts of trade or specific annual differences. Because food price spikes often occur
in years that are not average or normal, these flows are not very useful to assess the impact of
extreme events. These maps are also not available digitally, so they cannot be easily quantified
and used in a model, as they are meant only as a guide to how trade usually moves. These
characteristics reduce the maps' utility for model development, but their usefulness for com-
parison to the quantitatively assessed maps remains.
The pixels highlighted in the Njamena maize map roughly match up with the location of the
Bol export zone, but are much more widespread across the area. This may occur because of the
similarity in the NDVI signal between the pixels in the Bol zone and the other areas surround-
ing the country. On the other hand, since the FEWS NET map does not include areas outside
Chad, thus it may be that maize is transported from these other regions into Njamena.
Plate 20 shows the pixels related to maize prices in Nampula, Mozambique are far more wide-
spread than those identified for Njamena. Mozambique has a much more developed regional
trade network as well as a larger maize export zone which takes up nearly the whole country, as
seen in the trade flow map from FEWS NET ( Plate 21 ). The source locations for Nampula thus
are quite extensive, with pixels highlighted both north and south of the market, as well as from
surrounding regions outside Mozambique. For this particular analysis, Zimbabwe was excluded,