Agriculture Reference
In-Depth Information
assesses how climate variability is likely to affect food market prices and functioning, and how
international prices will reduce or exacerbate these impacts.
Food price data used in this model are sourced from the UN FAO food price tool and from
the FEWS NET food price database, include data from rice, wheat, maize, millet and sorghum,
and are from 124 locations in 36 countries. The prices are from the FAO food price tool
( ) and from FEWS NET and its member countries. The series
used here are only from 2008-12 to remove incomplete and missing months, and to have as
comparable series as possible across as many locations as possible. A map of the locations
where the series are from is provided in Plate 1 .
The vegetation index data used in the analysis are the monthly NASA Moderate resolution
Imaging Spectroradiometer (MODIS) Aqua NDVI from 2003 to 2013 (Huete et al ., 2002).
The data used are the MODIS Climate Model Grid monthly dataset at 0.05-degree resolu-
tion. More information on the NDVI product and the data from MODIS sensors more
broadly can be found in Huete et al . (2002). These data are used to derive temporal monthly
anomalies by pixel that quantified the impact of drought, or unusually low rainfall or moisture
conditions due to high temperatures. All vegetation data have uncertainties from cloud con-
tamination, dust and other aerosols in the atmosphere, unusual soil moisture conditions, or
other phenomenon that may not be related to actual vegetation health, thus we did not con-
sider anomalies of less than 10 percent. To remove the influence of non-agriculture areas
from the analysis, we used the NASA MODIS land cover map to identify areas that were
forest, water, desert or urban and masked these pixels (Friedl et al ., 2002).
Empirical methodology
The model presented here uses a state space approach to mathematically model local price
dynamics, which was popularized in the economics literature by Harvey (1989). This approach
has the following advantages:
• itworkswellinasmallsample(Harvey,1989)andusesatime—seriesapproachwhich
can easily incorporate information from the remote sensing data (Shumway and Stoffer,
• itexplicitlymodelspricedynamics,importantforthetheearlywarningcommunity;
• becauseitmodelstheautoregressivenatureofprices,itdoesnotassumestationarity
(Harvey 1989);
• predictiononeorseveralmonthsaheadfollowsnaturallybecausetheestimationpro-
cedure models a dynamic process using a Kalman filter; and
• bydecomposingapriceseriesintotrend,seasonalandtrendcomponents,itallowsthe
analyst to interpret better noisy time series and relate them to environmental dynamics
(Brown et al ., 2008, Cornia et al ., 2012).
By partitioning the data into components that have variations at the same time scale as the
growing season, the seasonal component that is most related to crop production could
be isolated. Figure 7.4 shows a price time series from Nairobi, Kenya and the three
Search WWH ::

Custom Search