As of February 2012
Data across all of the 83 HARITA sites from the rain gauges was gathered and organized. Farmers and local experts were also asked about years with a low level of rainfall to create a qualitative list. Moreover, an analysis script was written and applied to query satellite databases. This made it possible to compare 12 years of data of an array of vegetation remote sensing products against yield data, ground rainfall observations and remote sensing of rainfall, enabling IRI to gain valuable insights into alternate data sources: First, satellite vegetation and satellite rainfall measurements can be used to cross check each other to see if, and how often, they capture the same drought events. Although they often lead to similar estimates, they do so less often than expected. This is largely related to the product’s rough approximation of what is happening on the ground, rather than a direct measurement of rainfall or vegetation. When both measurements lead to different estimates, further physical validation may become necessary. An overview of the agreements and disagreements between different products is illustrated in the table below. Second, different satellite products have proven more useful during different times of the year. Satellite vegetation measurements have proven more suitable to capture droughts during certain parts of the year, such as the end of the rainy season, while satellite rainfall measurements are more accurate during other parts of the year, such as the beginning of the rainy season. Third, among the examined vegetation products, the estimates of EVI had in particular reflected those of ARC, so that IRI plans to carry out more in-depth analysis about EVI. Fourth, using satellite products has led to challenges with data processing and manipulation because data errors have sometimes occurred. This reinforces the need for double-checking data when using satellite products.
Average Satellite Ranking for the 12-year period 2000-2012 across all 83 project villages*
Satellite Product
|
Type of Product
|
Early Season
|
Late Season
|
ARC
|
Rainfall
|
1
|
1
|
TAMSAT
|
Rainfall
|
0.625
|
0.68
|
EthRFEadj
|
Rainfall
|
0.611
|
0.45
|
EVI
|
Vegetation
|
0.35
|
0.57
|
NDWI
|
Vegetation
|
0.422
|
0.45
|
* The table shows how the satellite products perform in comparison to ARC. A rank of 1 means that the satellite product captures 100 per cent of the “bad years” measured by the ARC index. A rank of 0 means that the product includes none of the worst years seen by the rainfall index. Disagreements between products do not necessarily signal success or failure. IRI has not tried to determine which metrics are correct indicators of losses. Instead, they seek to understand what insights can be gained from agreements and disagreements between different products.
IRI also seeks to disseminate knowledge about index insurance and to facilitate capacity building. To do so, a first draft of educational material was drafted. The material informs learners about satellite remote sensing, including how it works, its advantages, and its limitations. The material consists of text documents, presentations and hands-on training activities. IRI plans to constantly revise and update the training material, taking into account research findings and feedback from participants gathered during workshops.
IRI organized and held the first remote sensing workshop of the project in Addis Ababa in December 2011, informing a group of stakeholders from the region about index insurance and satellite technology. Major international NGOs, the Ethiopian National Meteorological Agency, local, national and international insurance companies, and international aid organizations participated in the training session. The participants appreciated the capacity building workshop but also highlighted that the training material could be further improved. They called for more illustrative examples, better visual presentations (including actual satellite imagery) and longer series of workshops which explain the entire process of remote sensing (including data acquisition, interpretation and use of data for product design). IRI plans to take this feedback into account to further improve the effectiveness of its training sessions. Yet, the participants of the training session had a wide range of interests and technical ability, which translates into different training requirements. Therefore, IRI will keep on concentrating on educational materials which are suitable for a general audience.
As of February 2013
IRI has continued to conduct research on satellite products and develop a remote sensing validation technology. The initial analysis highlighted that EVI performed particularly well when compared with ARC. Therefore, some diagnostics were performed to check if the performance could be improved. First, the amount of time delayed after the rainfall estimate for when the vegetation images were used was changed. The vegetation was checked a month earlier, the month of, and one, two, and three months after the month of interest. IRI found that the one month delays originally used in the analysis worked best. Second, it was checked whether changing the size of the area over which the vegetative data was averaged led to noticeable differences in the matching. It was found that the results did not significantly change.
The potential of alternative data sources, such as farmer recall, yield data and farmer rain gauges was also explored to validate indexes. It was found that farmer recall and available historical yield data agree on the major drought years in the past and are generally consistent with index payouts and vegetative sensing for major events across most of the region. The problem with farmer rain gauges is that they only have recent data, although they can still be used to diagnose how a season has progressed in a specific location. For the future, IRI plans to focus on to what extend smaller, localized events can be accurately identified by the different data sources and how much smaller events can be targeted by a verified index insurance product.
IRI has also sought to gain a better understanding of what the satellite is looking at, given that a satellite image is a complex mix of many things, including shadows, bare soil, rocks, water, foliage from crops, foliage from grasses and tress, and other vegetation. To do this, the potential of satellite imageries with different resolutions was explored in order to find a cascading series of validations, where the rarest, most expensive, and highest accuracy validations can be used to check less expensive information that is more widely available, covering the widest areas, reserving the most expensive tools to the places where issues have been identified. The analysis of the satellite imageries showed that EVI has a close relationship to vegetative fraction and scalability, which partly explains why it performed well in initial investigations. The findings also suggested that it might be possible to further improve the use of EVI through knowledge of the vegetation fraction within each pixel.
The inclusion of the year 2012 has also strengthened the analysis, since it generated wide-spread payouts of the HARITA index insurance project. In October and November, satellite data triggered payouts worth $322,772 to more than 12,000 farmers in Ethiopia. IRI explored whether the satellite validation techniques are able to identify the places where the index did not perform well in 2012. Overall, the 2012 indices reflected the local experiences and the contracts performed well for the vast majority of farmers. However, complaints from 3 villages complained about the performance of the index. An IRI team visited the villages to follow up on their complaints. The follow-up revealed that the concerns of two villages (Hawelti and Tsegea) were not primarily the result of error in satellite rainfall estimation. Instead, differences in the insurance packages offered between the villages and neighboring villages caused dissatisfaction among farmers. In the third village with concerns, which was Imba Rufael, IRI found that the index needed to be improved.
In parallel to the village visits, IRI flagged the areas of the map where there was the lowest level of agreement between EVI and the satellite rainfall index. The flagged areas included all of the sites with meaningful complaints in 2012, suggesting that EVI offers a useful validation tool. On the contrary, other sources of information, such as historical yield assessments and farmer interviews did not clearly flag concerns for those regions.
The capacity building material has been improved, taking into account the experience of the first year and the feedback received during workshops. The second remote sensing workshop was also organized and held to inform practitioners and other audiences about index insurance.

Farmers discussing concerns about 2012 experience (left) and a rain gauge at site of Hawelti (right)
As of May 2014
IRI has continued developing the remote sensing technology. Since 2000, two large regional droughts occurred in 2004 and 2009 and the index of the HARITA project would have triggered in nearly all of the project sites, suggesting that the satellite index is doing well. Yet, farmers are not only affected by large regional droughts but also by smaller, more localized droughts. Hence, IRI has sought to gain a better understanding of whether and how an index can effectively target more localized droughts or whether it is only reliable for very large scale events.
IRI has also reviewed and updated its educational material about index insurance. To ensure a high quality, the material was tested at Columbia University and at a workshop in Ethiopia. During these sessions, feedback was gathered to further improve the material. IRI has also disseminated its knowledge about index insurance outside of HARITA. It participated in NASA/SERVIR workshops in Ethiopia, Kenya and Tanzania, and at the 9th Microinsurance Conference in Indonesia.

Locations of HadushAdi July 2013 Field Validation sites

Ground Validation Image, showing fields of barley sown 48 days prior as well as a field of beans and a fallowing field in the distance.

Adi Ha view from ground, looking towards irrigated orchard

Example output from Landsat TM visual imagery (resolution 30m), MODIS EVI (resolution 250m) and ARC rainfall (resolution 10,000m).
Next actions
IRI will seek to gain a better understanding of whether and how an index can effectively target more localized droughts or whether it is only reliable for very large scale events. It will also revise and update its capacity building material and disseminate its findings.