REMOTE SENSING

X.X Introduction

We have performed several data collection and analysis exercises via remote sensing technologies in index
insurance. These involve the use of satellite-derived estimates of rainfall for contract design and triggers, diagnostic
level validation of remotely sensed rainfall using ground measurements, and the analysis of various vegetation-
abundance indices for comparison with first-hand accounts or photographs. This analysis is performed for five
villages in Ethiopia - Hade Halga, Geneti, Awat Bikalsi, Adi Ha, and Hadush Adi – for a fifteen year period between
1995 and 2009.

X.X Satellite-Derived Vegetation and Rainfall Indices

One emerging methodology for overcoming sparse field data is to measure levels of vegetation from a remote
sensing platform. We analyze five different satellite-based indices, of which four are derived from the presence of
chlorophyll (or the greenness in vegetation) and one from the presence of water in the vegetation; each of these
products varies spatially and temporally in resolution. These indices include:

1) Normalized Difference Vegetation Index (NDVI-AVHRR) – from the National Oceanic and Atmospheric
Administration (NOAA), which is measured every ten days at an 8km spatial resolution.

2) NDVI-Spot Vegetation Index from the Centre Nationale d’Etudes Spatiales (CNES), which is measured every ten
days at an 1km spatial resolution.

3) NDVI/MODIS Vegetation Index from National Aeronautical and Space Administration’s (NASA), which is a 16-day
composite index measured at a 250m spatial resolution.

4) Enhanced Vegetation Index (EVI) derived from the same sources as NDVI/MODIS, but has a different quantitative
relationship that is more sensitive to canopy structure and type.

5) Normalized Difference Wetness Index (NDWI) derived also from the same sources as NDVI/MODIS, but reflects
changes in the vegetation’s water content.

The MODIS-derived indices (3-5) are only available starting from the year 2000 and NDVI-Spot vegetation from 1998;
we use NDVI-AVHRR starting from 1995.

One remote-sensing product is used for rainfall, known as ARC. These values are compared to the station gauge
data at Maychew, Adishewshu and Alamata.

The comparison of these different indices against the rainfall gauges and satellite-derived rainfall estimates
determines the most appropriate remote sensing tools to incorporate into the design of index insurance products. In
other words, the objective is to find the most representative vegetation index for a given rainfall event or period; as a
result, index insurance products will be triggered by these events which will suggest a commensurately successful or
poor harvest season.

X.X Communications Table

We present communication tables for each village and crop detailing the various satellite and in situ parameters,
contract designs, and first-hand accounts. One example is shown below, for wheat in Hadush Adi. We use a fifteen
year time series (from 1995 to 2009), from which two different insurance products are determined: the first,
representing a medium dry payout with a 1 in 3 year payout (highlighted in yellow); and a second product
representing a very dry payout with a 1 in 5 year payout. Mentioned in previous sections, the payout is awarded
based on a rainfall trigger, which is indicated under two ten-day sections: Jun 11-Jul 10, representing the sowing
season; and Aug 21- Sep 30, representing the late harvest season. For each of these time spans, total rainfall data is
provided from both the ARC and station data sets. Also, when available, the average vegetation index –for each of
the five products- is noted per harvest season.

As expected, a lower rainfall value (such as in 2004) will show a smaller vegetation value, and a possible payout (in
this case, a very dry payout). This is also corroborated or refuted by farmers’ accounts and also by the ARC Water
Resistance Stress Index (WRSI).
 Hadush Rainfall Totals
Adi Wheat with Dekadal Cap Jun 11 - Jul 10 (Jul 20) Aug 21 - Sep 30

NDVI
Frerweni

Frerweni

Farmers

MODIS
Adigrat

Adigrat

SPOT

NDWI
WRSI

NDVI

NDVI
Medium Dry

-Veg
ARC

ARC

ARC

EVI
Payout Very Dry Payout
1995 0 0 50 39 48 27 - 0.59 0.34
1996 0 0 49 58 50 25 - 0.76 0.30
1997 100 100 53 75 16 26 Bad 0.64 0.29
1998 0 0 44 60 62 45 - 0.34 0.35 0.2939355
1999 0 0 30 56 67 35 - 0.38 0.36 0.3154839
2000 0 0 34 95 51 - 0.18 0.31 0.2948387 0.252085 0.07399105 0.1709593
2001 0 0 18 62 35 - 0.29 0.34 0.2948387 0.2638756 0.01227884 0.1754638
2002 100 64 3 75 45 47 9 10 - 0.31 0.33 0.2356129 0.2180059 -0.01019296 0.1566411
2003 0 0 16 74 69 66 47 17 - 0.36 0.31 0.2553549 0.2309738 0.02957663 0.1652018
2004 100 77 24 62 55 20 5 1 Bad 0.20 0.27 0.2407742 0.2256913 -0.00174333 0.1608454
2005 0 0 40 35 71 64 33 30 - 0.47 0.29 0.2294194 0.2231924 0.01756514 0.1536873
2006 56 0 9 57 100 55 50 26 - 0.34 0.31 0.2628387 0.2534537 0.03964578 0.1627275
2007 0 0 50 100 74 56 39 28 Bad 0.66 0.31 0.2490323 0.2445505 0.04092009 0.1700685
2008 0 0 26 89 55 26 Bad 0.47 0.27 0.2668387 0.2503916 0.09351259 0.1667625

Table X.X: Insurance product communication sample table (shown for wheat in Hadush Adi).

x.x.x. Groundtruthing Visits

An important part of analyzing the appropriateness of utilizing remote sensing data is to consider how accurately the
remote sensing data reflects field data. In light of this, we have conducted field visits in March 2009 and October
2009 in order to collect images of the levels of vegetation on the ground.

(Ask for Chris’ input)