Automated rainfall data extraction from satellite imagery for crop index insurance
Abstract
Crop insurance uses rainfall as one of the indeces on which farmers can be insured. The
rainfall data can be obtained locally from the weather stations but it is limited to only 30
kilometres radius from a station. Though there are many sources of rainfall data, satellite
data has not been explored in depth within the Kenyan context and by any insurance firm
as a reference point.This project presents an automation prototype that extracts rainfall
data from satellite imagery for crop index insurance. The prototype is implemented using
python, .NET, PHP, HTML and PostGres/PostGIS. The prototype encompasses
download module, data extraction and database storage modules. Mosaic algorithm is
implementedusing the pixel traversal algorithm. The output product is created using the
supplied reprojection information and output parameters such as pixel size and easting
and northing. The input products being satellite image products, the processing
parameter used is a shape file to map out the desired regions and rain gauge stations.
Once an output product has been written it can be updated with an unlimited number of
new input products. The "filling" of the final product follows the principle of fetching
pixels from source products. The Mosaic Processor loops over all cells in the target grid
and determines whether a pixel from a source product may be suitable to read into it. The
pixel RGB value is referenced to an accompanying CLR file to determine the rainfall
value the pixel holds. Rainfall data was extracted from bil images of April 2006 for each
day. The rainfall data extracted from the satellite imagery was compared to the recorded
meteorological data for the same period (mean monthly rainfall) for April 2006. The
research concludes that rainfall can be extracted from satellite imagery and used for
rainfall index insurance. The .bil images provide rainfall information for a radius of 8kms
which can be improved with higher resolution images. With this, the crop insurers can
have a larger pool of farmers to insure as there will be no need of being 30km radius from
a rain gauge sation. This research presents altenative source of rainfall data from the
traditional rain gauges to be provided to farmers and insurance companies. The research
concludes by recommending furtherwork to be done with level 0 imageryto be used as
input and higher resolution images to reduce the 8km by 8km to even a shorter distance