Knowledge Discovery From Satellite Images For Drought Monitoring
No Thumbnail Available
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa
Abstract
Drought is one of the most impo rtant challenges facing the planet. When it
happens, it usually re sults in serious econom ic. environmental, and social cr ises.
Despite the growi ng number of freely available biophysical, climate. and satellite
data for characterizing and modeli ng drought, research efforts have been
constrained to using only meteo rological point data, such as the amount of rainfall,
for drought monitoring information. This po int data is insufficient for representing
diversified ecosystems, and the data has coarse reso lut ion levels (lim ited spatial
coverage). Researchers also have limited tools for data retrieval and integration for
improved drought identification and model ing. which usua lly results in a time de lay
fo r informat ion to reach dec is ion makers. Taking this into account, this dissertation
researc h has three objectives: I) identify the most re levant attr ibutes for effic ient ly
implementing drought monitoring, 2) develop a new approach for extracting
knowledge from sate ll ite imageries for improved ident ifica tion and pred iction of
drought, and 3) evaluate the new approach for national and regio nal dro ught
prediction appl ications. Using an exploratory research approach and modeli ng
research method, different data co llect ion and analys is techniques were executed
using knowledge d iscovery in a database approach. The data mi ning models
developed using art ificial neural network and regress ion tree models were able to
predict DroughtObject with accuracy of 0.70 - 0.95 co rrelat ion coefficients. in a neta
four month s' time lag. The develo ped DroughtObject model was evaluated for its
application in showing drought severity and food defic it status. There were positive
relat ionships between DroughtObjecl products and crop yield data up to 0.91 R2
values. The results confirmed that the model can direct ly be used by those who are
currently responsible for drought monitoring and ri sk management. The new
concept developed in this research was prolotypcd and demonstrated in an easy-tousc
approach. with a focus on demonstfaling the concept of DroughtObjecl
characterization and identification fro m a group of pixels. This demonSiration also
revealed poss ible future system deve lopments.
This di ssert ation research could he lp deci sion makers use advanced satellite
technology fo r crrcctive drought monitoring and early warning systems in va rious
regio ns. Combined with proper pol ic ies. Ihese systems can he lp to prevent famine
and starvat ion in food-insecure reg ions. Up to now, satellite technologies have been
used primarily in areas of meteoro logical applications. In this research. the main
emphas is is on mining knowledge from satell ite images for dro ught ri sk assessment
and sa ving the lives of individua ls who are affected by recurring drought s. The
findings of this research can help decision makers take time ly and appropriate
actions to save lives in drought-affected areas using advanced satellite techno logy.
Description
Keywords
Drought Monitoring; Drought Prediction; Geo-spat ial Informat ion; Knowledge Discovery; Satell ite Image