An Effective Drought Early Warning System for Sub-
Saharan Africa: Integrating Modern and Indigenous
Approaches
Muthoni Masinde
Central University of Technology, Free
State
+27 51 507 3091
muthonimasinde@yahoo.com
ABSTRACT
Droughts remain the number one disaster in Africa; of all the
people affected by all types disasters, drought is responsible for
over 88% of them. An effective drought early warning system can
support appropriate mitigation and preparedness strategies and
hence minimize these effects. Existing systems tend to ignore the
‘at risk’ community and are faced with a number of
implementation challenges; their utilisation is very low. This
paper describes an effective drought early warning system that
integrates indigenous and scientific drought forecasting
approaches. The system is anchored on a novel integration
framework called ITIKI (acronym for Information Technology
and Indigenous Knowledge with Intelligence). Indigenous
knowledge ensures that the system is relevant, acceptable and
resilient. ITIKI further employs three ICTs (Mobile phones,
wireless sensor networks and artificial intelligence) to enhance the
system’s effectiveness, affordability, sustainability and
intelligence. This paper describes the design, development and
evaluation of the system.
Categories and Subject Descriptors
C.1.4 [Parallel Architectures]: Mobile Processors. C.2.4
[Distributed Systems]: Distributed Applications. H.5.2 [User
Interfaces]: User-Centred Design. J.7 [Computers in Other
Systems]: Early Warning Systems.
General Terms
Measurement, Documentation, Reliability, Human Factors,
Standardization, Language
Keywords
Drought early warning system, Information Technology and
Indigenous Knowledge with Intelligence (ITIKI), Indigenous
knowledge weather forecasting systems, Seasonal Climate
Forecasts, mobile phones, artificial neural networks, wireless
sensor networks
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies
are not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. Copyrights
for components of this work owned by others than ACM must be
honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior
specific permission and/or a fee. Request permissions from
Permissions@acm.org.
SAICSIT2014, September 29 - October 01 2014, Centurion, South Africa
Copyright 2014 ACM 978-1-4503-3246-0/14/09…$15.00
http://dx.doi.org/10.1145/2664591.2664629
1. INTRODUCTION
There is a complex link between droughts and development (or
lack of it); in most African countries, rain-fed agriculture accounts
for over 70% of food production [1]. This form of agriculture is
extremely responsive to weather and when droughts strike,
millions go hungry [2]. In the face of droughts, governments are
forced to re-direct budgets initially allocated for developmental
projects towards supporting the hungry [3].
Effective drought early warning systems (DEWS) have high
potential in making a contribution towards tackling the cycle of
droughts. This is by way of providing timely, relevant and
comprehensible information on impending droughts. This
information could be used to mitigate droughts’ effects and
therefore reduce their negative impacts on the fauna and flora.
Successful DEWS in turn rely on weather forecasting systems;
implementation of such systems in many African countries is
hampered by among other things, inadequate coverage by weather
stations. The second challenge of weather forecasting emanates
from very poor utilisation of the Seasonal Climate Forecasts
(SCFs) by the very farmers that grow the food for the masses.
The content, format and dissemination channels used do not
address the farmers’ needs. SCFs are unreliable (real and
perceived) and their ‘course’ nature makes them irrelevant at the
local (say a village) level. Such farmers have in turn continued to
rely on their local/indigenous knowledge forecasts (IKFs) to
derive critical cropping decisions ([4] and [5]). Though being very
rich and relevant to the farmers’ context, there is evidence that
IKFs are threatened [6].
One of the major weaknesses of the existing drought prediction
tools is the emphasis on macro/international level information.
Further, the design and implementation of drought communication
strategies tend to ignore the ‘at risk’ community who also happen
to be host to very crucial indigenous knowledge on drought, their
environment and indigenous coping mechanisms. In Sub-Saharan
Africa, early warning systems that attempt to address droughts are
multi-hazard with drought being just a small component of the
entire system.
This research was guided by the hypothesis that people-centred
drought early warning systems can empower people at the local
level. Such systems increase people’s sense of ownership and
confidence in using the systems. Consequently, the people
become more resilient to the droughts and are able to protect
themselves. In the African context, one way of achieving this is
by incorporating African indigenous knowledge on droughts,
exploiting the widely available mobile phones and employing
Wireless Sensor Networks (WSNs) to collect micro/local droughts
data. Further, the research was directed by the research question:
Does the incorporation of indigenous knowledge into a drought
60
prediction tool improve the tool’s resilience and relevance to the
countries in Sub-Saharan Africa?
The research started off by designing a novel integration
framework that creates the much-needed bridge (itiki) between
Indigenous Knowledge Forecasts and Seasonal Climate Forecasts.
The Framework was then converted into a drought early warning
system (DEWS) prototype that uses mobile phones as input/output
devices and wireless sensor-based weather meters to complement
the weather stations. This was then tested and evaluated in two
regions in Kenya. The complexity of the resulting system was
enormous and to ensure that these myriad parts worked together,
artificial intelligence technologies were employed: (1) artificial
neural networks (ANNs) to develop forecast models with that had
accuracies of 70% to 98% for lead-times of 1 day to 4 years [7];
(2) fuzzy logic to store and manipulate the holistic indigenous
knowledge; and (3) intelligent agents for linking the prototype
modules. Details of itiki’s conceptualisation are found in [8]
while the working of the ANNs forecasting models that were
developed are found in [7]. The early deployment initiatives are
reported in [9] while the details of the sensor-based weather data
collection sub-system is documented in [10]. The focus of this
paper is the design and implementation details of the DEWS,
details of which are not found in any other publication. The main
role (downscaling and up-scaling forecasts) played by ICTs in this
research is published in [12]. Finally, tests carried out to evaluate
the field-readiness and acceptability (by meteorologists) of
wireless sensor-weather meters for measuring weather parameters
for the DEWS are published in [13].
2. RELATED LITERATURE
2.1 Drought Forecasts in Sub-Saharan Africa
A drought is defined as an insidious hazard of nature which
qualifies as a hazard because it is a natural accident of
unpredictable occurrence but of recognizable recurrence [14]. As
a disaster, drought corresponds to the failure of the precipitation
regime, causing the disruption of the water supply to the natural
and agricultural ecosystems.
Droughts are among the most expensive disasters in the world, the
negative impacts of which span economic, social and
environmental aspects of the affected society. It is even more
difficult to quantify the cost of droughts because most of the
effects are indirect. Some direct ones include: (1) effects on
human/animals; (2) decline in food yield, forest and greenbelt; (3)
worsening water/air quality and sanitation; and higher fire
prevention risk. Indirect effects include: (1) price upsurge; (2)
reduction of income; (3) loss of jobs; and (4) degradation of living
standards [15].
Droughts accounted for 50% of the 2.8 billion people affected by
natural disasters between 1967 and 1992 [16]. The study by
Mishra, Singh [17] showed that among all the natural disasters,
droughts have the most impact on a country’s economy. The US,
for example, spent $40 billion to counter the 1988 drought, while
floods a year later cost the government 2-3 times less.
Droughts continue to affect millions of people in Sub-Saharan
Africa (SSA); the uniqueness of the problem in SSA however, is
to be found in the inadequacy and ineffectiveness of the Region’s
preparedness to these disasters [11 and 18]. Further, most
economies in SSA are driven by the notoriously climate-sensitive
agriculture sector and as Virji et al. [19] put it: “...agricultural
production and weather are so highly interrelated that a good
rainy season means a healthy economy, and failure of the rains …
means famine and death” (pg 8). Further: “Of the ten countries
with the highest levels of hunger, and of the ten whose scores have
actually increased since 1990, nine are in Sub-Saharan Africa in
both cases [20].
Despite the challenging contexts they operate in, meteorological
institutions in SSA continue to provide regular climate forecasts
especially in form of Seasonal Climate Forecasts (SCFs).
However, the utilisation of this information by the small-scale
farmers whose crops/livestock depend solely on rainfall is still
below par ([5] and [21].) Studies ([4] and [22]) reveal that over
80% of farmers in some parts of Ethiopia, Kenya, Zambia and
Zimbabwe relied on IKFs. However, IKFs are currently facing
challenges from various quarters especially from climate change,
urbanisation and population growth.
Seasonal forecast methods used to predict SSA rainfall are: (1)
Statistical Analysis of Rainfall: the analysis of the rainfall time
series of a region alone, without linking rainfall variability to the
atmospheric circulation or underlying physical mechanisms; (2)
Numerical Approaches: offers the largest potential for future
improvement. The integration of General Circulation Models
(GCMs), usually with sea surface temperature (SST) forcing,
typically over a period of 50 or more days; (3) multivariate
Empirical Approaches - the ability to predict seasonal rainfall
totals several months in advance is due to the exchanges of energy
between the oceans and the atmosphere, which includes the
weather systems that produce rainfall. In practice, quantification
of the interaction between the ocean and atmosphere in empirical
studies typical of those which lead to forecasting schemes is
assumed to be represented by sea surface temperature anomalies
(SSTAs) alone and (4) numerical and Statistical Approaches [23]
2.2 Knowledge versus Modern Science on
Droughts
In [6], the term indigenous knowledge is used to refer to the
‘place-based knowledge’ that is rooted in local cultures and
generally associated with long-settled communities, which have
strong ties to their natural environments. This is the definition
adopted in this research. The entry point for the indigenous
knowledge forecasts (IKFs) therefore is the amassed knowledge
of exact arrival of the rainy season. The local community has built
this knowledge over the years from their understanding of the
forecasting and the probability of future rain based on variance in
wind, humidity and temperature. Secondly, based on the kind of
social-economic activities the community indulges in,
interpretations of animals, insects, birds, and plants behaviour is
performed. These generally affirm the rainy season indicators
because the plants/creatures observed exhibit subtle fluctuations
in temperature and humidity [24].
Researchers ([4], [25], and [26]) today concur that indigenous
knowledge and modern science complement each other. In [26]),
for example, bridging local farmers’ and scientists knowledge
through participatory research led to more appropriate
technologies. In order to develop sustainable strategies, it is
therefore important to take into account of, and learn from, what
local people already know and do, and to build on this. Scientific
and indigenous knowledge systems have increasingly been
accepted as two areas of expertise that complement each other.
The accumulated knowledge has always worked for the locals
[27]; the debate here can only be on the level of success. On the
other hand, the scientific approaches to weather prediction are
hardly utilised by the mostly illiterate farmers most of whom live
in the remote villages where modern technologies such as
61
televisions and internet are still a foreign concept ([4], [22], [28] conservation ([30], [31], [32], [33] and [34]). Though this
and [29]). Secondly, implementing modern drought prediction integration has potential of bringing numerous benefits, the
technologies is still a costly affair for most African countries diverse nature of the two makes it a daunting task [35]. For
instance, IK is highly informal and tacit; it is often represented
The summary from the reviewed literature above therefore points using vague linguistic variables [36]; on the other hand its
to the fact that Indigenous knowledge (IK) and modern science- scientific counterpart is explicit and highly structured [35]. If IK
based weather forecasts are not mutually exclusive but significant is to be integrated with the scientific knowledge and eventually be
discordance between the two is still apparent. Clear used in decision-making process, there is need to represent IK in
understanding and careful integration of indigenous knowledge some form of explicit-structured way; fuzzy logic is one way of
present opportunities especially in the dissemination process of achieving this. This is because other approaches that employ
weather forecasts to farmers in SSA because this will support subjective and qualitative approaches may not capture the holistic
ways that are culturally appropriate and locally relevant. On the nature of IK whose variables do not have precise
question whether IK needs modern science, there is evidence that non/membership.
IK has been eroded and is slowly disappearing. Using IK alone
makes it is difficult to forecast beyond a season. In scientific After observing a positive relationship between forecasts based on
approaches, this can be achieved by employing technologies such indigenous knowledge and those based on modern science,
as ANNs. Some terminologies used in IK may sometimes have authors in [4] recommended that the two knowledge systems
more than one meaning, for example ‘abundant rainfall’ may should be integrated. Though not related to this recommendation,
(http://www.africa-
mean rainfall for the day or a season. Finally, unlike scientific such an initiative
adapt.net/aa/ProjectOverview.aspx?PID=PUXVdbXh9bM%3D)
approaches, climate change may be difficult to foretell using IK , spearheaded by the
alone. IGAD (Intergovernmental Authority on Development) Climate
Prediction and Applications Centre (ICPAC), was started in
Table 1. Comparison between Indigenous and Scientific September 2008. The interests of incorporating indigenous
Drought Forecasts knowledge into scientific knowledge has been especially
Indigenous Scientific highlighted within small island developing states (SIDS), due to
Use biophysical indicators of the Use of weather and climate their inherent vulnerabilities and propensity to environmental
environment as well as spiritual models of measurable hazards [24]. In Papua New Guinea, in developing a National
methods meteorological data Disaster Risk Reduction and Disaster Management Framework
Forecast methods are seldom Forecast methods are more
for Action 2005–2015, the need to integrate traditional knowledge
documented developed and documented into disaster management systems was identified but not the how
this may be achieved [37].
Up-scaling and down-scaling are Up-scaling and down-scaling are
usually complex relatively simple Other integration initiatives have been carried out in the
Application of forecast output is less Application of forecast output is assessment of soil types classifications and soil fertility in South-
developed more developed western Burkina Faso [38], East Africa (Uganda and Tanzania)
and Bangladesh [39] and in Nisamabad District of Andhra
Communication is usually oral Communication is usually
written
Pradesh State in India [36]. Using examples from Inuit and other
northern peoples, another integration of IK with scientific
Explanation is based on spiritual and Explanation is theoretical knowledge is in analysing the complexity of ecological
social values knowledge [40]. In [41], integration of the indigenous knowledge
Taught by observation and experience Taught through lectures and on the environment with the scientifically based vegetation
readings analysis for north central Namibia is discussed. Krupnik and Ray
Adapted to local conditions and needs Formulated at a larger scale and [30] tackle the integration in relation to subarctic marine
lacks relevance at local level ecosystems while Lauler, Shankar [31] looked at the integration of
ecological knowledge for marine habitat monitoring in Oceania.
Refers to rainfall duration and Refers to rainfall quantity at a
distribution and it is aligned to crop- regional level Krupnik and Ray [30] compared scientific knowledge (marine
weather indicators biology and ecology) and the knowledge of indigenous hunters on
walrus distribution, abundance, and life cycle to study climate
It is language based and qualitative It is number-based change in the in the Beringian Region. In all these studies, there
It is holistic – it covers large number Covers small number of is a common conclusion that there are fundamental difference
of variables qualitatively variables quantitatively between indigenous and scientific knowledge systems and
It is a way of live – looks at both the Has no social context integrating the two results in more robust systems with the two
process of knowing and the complementing each other.
knowledge itself. It is explicit in its
social-context aspect and it is an Such integration offers one way of dealing with the complexity of
integral part of people’s culture drought monitoring system [42]. However, efforts towards this
integration are rare; this is the main contribution of the system
It has rules of the ‘knowing process’ It is based on rules of science,
that is evidence, repeatability
described in this paper. From a two-year study in Goima Ward,
and quantification Tanzania, [43] a comparison between the scientific agricultural
droughts and farmers’ indigenous knowledge and perception of
droughts is presented. Other brief mentions of this integration are
found in, ([4],[6[] and [44])
2.3 IKFs and SCFs Integration Initiativess
Over the last 10 years, initiatives to integrate these two systems 3. DATA AND METHODS
have increased two-fold especially in the area of environment
3.1 Research Approaches

62
Three research approaches were applied: (1) Applied-descriptive-
structured was carried to analyse the characteristics and patterns
of drought patterns in Kenya. It was also applied in describing the
various aspects of indigenous weather forecasting systems; (2)
Applied-correlational-structured research was used to discover the
similarities and difference between the modern science and
indigenous ways of forecasting weather. This was also used to test
the field-readiness and acceptability (by meteorologists) of
wireless sensor-weather meters for measuring weather parameters;
(3) Applied-exploratory-unstructured was informed by the fact
that the wireless sensors technology is relatively new in the
context (Sub-Saharan Africa and weather monitoring) of its
application in this research. It was also informed by the fact that
that three very diverse domains (computer science, meteorology
and indigenous knowledge) were traversed in the course of the
research. It was applied in investigating the suitability of Effective
Drought Index (EDI) and Artificial Neural Networks (ANNs) in
qualifying/quantifying and forecasting droughts respectively. The
approach was also used in evaluating the effectiveness of ICTs
(mobile phones, wireless sensors and intelligent agents) in
developing an effective Drought Early Warning System (DEWS)
for small-scale farmers in Sub Saharan Africa (SSA). Finally, it
was used to investigate the suitability of fuzzy logic in capturing
and representing the holistic nature of indigenous knowledge.
For research design, the following were applied: (1) Experimental
Design in the testing of the field readiness and the calibration
(against professional weather station) of the wireless sensors-
based weather stations; and (2) Case Study Design in the
validation of the following hypothesises:
i. EDI can be used to qualify and quantify droughts in absolute
terms
ii. ANNs can be used to accurately forecast droughts
iii. The occurrence of an extreme climate event (drought or flood)
during the October-November-December rain season in
Kenya increases the probability that a similar event will occur
during the March-April-May rain season
iv. Integrating IK with SFCs and disseminating the resulting
forecasts through mobile phones in the local languages leads
to improved utilisation of the forecasts.
(3) Prototyping was applied in developing the various sub-systems
that make up the DEWS; and (4) Framework development was
applied in developing itiki framework.
Data collection methods used are: (1) analysis of relevant
documents; (2) coding (writing computer programs); (3) guided
interviews for identifying IK characteristics and evaluating the
DEWS; (4) laboratory and field sensor experiments; (5)
observation of IK weather forecasting sessions; and (6) focus
groups for collecting IK
For data and error analysis, the two main approaches applied are:
(1) Calibrating for Uncertainty of Meteorological Measurements –
a rigorous calibration exercise carried out to determine and correct
various error factors on the wireless sensors boards used for
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
SAICSIT2014, Sep 28–Oct 01, 2014, City, Pretoria, South Africa.
Copyright 2040 ACM 1-58113-000-0/00/0010 …$15.00.
weather data collection [13]; and (2) Ranking ANNs Models’
Performance – In the case of ANNs that were designed to forecast
droughts, regression (R) and Mean Square Error (MSE) were
mainly used to select the networks models with the best
performance. RMSE and Percentage RMSE were then used to
determine the implications of the errors and the resulting forecasts
3.2 Framework Development
Guided by the 4 main components of an Early warning system
documented in [45], ITIKI [11] is a novel bridge that delivers a
drought early warning system (DEWS) composed of four
elements: (1) Drought Knowledge (2) Drought Monitoring and
Prediction; (3) Drought Communication and Dissemination; (4)
Response Capability. ITIKI; acronym for Information Technology
and Indigenous Knowledge with Intelligence is a bridge that
integrates indigenous drought forecasting approach into the
scientific drought forecasting approach. ITIKI was conceptualised
from itiki which is the name used among the Mbeere people
(found in the Eastern part of Kenya), to refer to an indigenous
bridge made using sticks and was used for decades to go across
rivers. Until mid 90s, this bridge used to be constructed by
‘experts’ who possessed indigenous knowledge on the rivers’
terrain as well as on the strength of the various trees along the
rivers and the trees’ ability to sustain the weight that the bridge
would eventually carry. Such was the accuracy of this knowledge
that during the 1992 floods, a newly constructed modern bridge
was swept away while the itiki nearby was left standing. The
overall goal of ITIKI was to come up with a relevant, affordable,
sustainable, integrated, resilient, useable, effective, generic, and
micro-level early warning system for droughts for the Sub-
Saharan Africa and Africa at large. ITIKI Architecture is shown in
Figure 1 and its components are described below. The framework
was guided by three factors [12]: (1) Using WSNs to Downscale
Weather Forecast; (2) Scaling-Up Indigenous Knowledge; and (3)
Integrating the forecasts into user applications.
3.2.1 Element 1: Drought Risk Knowledge
 1(a): Using wireless sensors that are capable of sensing
temperature, humidity, atmospheric pressure, wind (direction
and speed), precipitation and soil moisture, weather data is
automatically collected and sent to a structured store 1(f) in
form of text messages (SMS).
 1(b): Rainfall data observed from rainfall stations (contains
only rain gauges, the Mbeere case) stations is manually
entered into the system and stored in the same database as
the sensors’ data.
 1(c): Other data elements (IK) are retrieved from various
publications available in print and on-line. These are in form
of limited studies on IK in Mbeere and SCFs by KMD. Out
of these, the structured elements are stored in 1(f) while the
unstructured ones are stored in 1(g).
 1(d): IK on droughts collected during various field studies is
stored in 1(g).
 1(e): This is the real-time IK from the IK Experts.
 1(h) and 1(i): the structured data is stored in a database 1(h)
while the pre-processed indigenous knowledge is represented
as Fuzzy Sets 1(i).
3.2.2 Element 2: Monitoring and Prediction
This was implemented using two sub-components: (1) Drought
Monitoring that pre-processes the data to detect suggestive
patterns as well minimise duplicates and other errors. This is
achieved through EDI Monitor (2(a)) and IK Experts (2(b)); and
(2) Drought Prediction using Artificial Neural Networks ((2(c))
63
and Fuzzy Logic System ((2(d)). In 2(e), the resulting forecasts 3.2.3 Element 3: Forecasts Communication and
are reviewed by both the scientists and IK Experts after which Dissemination
‘reconciled’ forecasts are generated and passed to the
Mobile phones are used to send customized forecasts in form of
Dissemination component. This is partially a manual activity
test message and where possible, free phone calls to the farmers.
where the meteorologists and the IK experts sit to reconcile SCFs
Other forecasts are posted on websites while others are generated
and IKFs. However, the short-term forecasts (a few hours to two
in audio formats that can be broadcasted via community radios
weeks) do not need the manual ‘reconciliation’; the system
stations and visual displays on strategically located village digital
intelligently reconciles the two (from IK and from ANNs) and
billboards. Though not implemented, the Framework is designed
sends them to the Drought Communication and Dissemination
to support natural language processing to allow for translation of
Component. Further, in line with fuzzy system, for purses of
the forecasts into the local languages.
‘recovering’ IK’s original meaning/format, the output 2(e) is
passed through 1(i) for Defuzzification

Figure 1. ITIKI Architecture (source: [13).

64
33.2.4 Elemeent 3: Forecaasts Communnication and infformation on thhe application. Information on IK Indicattors
DDisseminatioon exppected from thhe community that the interm mediary represeents
waas pre-captured in the database described abovve. After reachhing
MMobile phones are used to send customized forecasts in foorm of
an agreement on tthe observed inddicators for a giiven period (a day,
d
ttest message annd where possibble, free phone calls to the farrmers.
a wweek, a monthh, and so on), the intermediarry logs on to the
OOther forecasts are posted on websites while others are geneerated
moobile applicationn and selects a list of indicattors observed; the
iin audio formaats that can be broadcasted viia community radios r
appplication associiates these withh the period speecified and uplooad
sstations and vissual displays onn strategically loocated village ddigital
theem to the databbase. In order to take care of slow/absencee of
bbillboards. Thoough not implem mented, the Fraamework is dessigned
Intternet, the appliication first saves the records to
t a local databbase
tto support natuural language prrocessing to alloow for translatiion of
beffore transmittinng them to thee remote databbase. The secoond
tthe forecasts intto the local langguages.
funnctionality of thhis application is to allow thhe intermediaryy to
sennd in observedd extreme weatther events. Thhis is an Andrroid
Appplication that captures two pieces of infoormation; extreeme
44. SYSTEM
M PROTOOTYPE weeather events annd indigenous knowledge
k weaather indicatorss. It
DDEVELOP PMENT AN
ND EVALU
UATION is implemented using
u four mainn Java classes and 3 supportting
pacckages: updatees, db and u utils. To ensurre that only the
44.1 System
m Modules autthorised IK expperts’ intermediiary is allowed to use the systeem,
TThis is a comprrehensive systeem made up of several modulees that thee application staarts off with a LLogin Screen for authenticatiion.
aare linked up toogether by intellligent agents that were implem
mented Aft
fter successful login, the Maain Menu withh three options as
uusing the Jaava-based muulti-agent systtems’ developpment shoown below appeears.
fframework called JADE (Java Agent Developpment). The moodules
aare: (1) Sensor--Based Weatheer Monitoring S System prototyppe; (2)
TThe EDI Monnitor which is a FORTRAN pprogram; (3) A ANNs
FForecasting Toool; (4) IK Fuzzyy Sub-System thhat stores Indiggenous
KKnowledge (IIK) drought indicators; (5) Android M Mobile
AApplication to input and outtput IK indicators as well exxtreme
wweather eventss; (6) SMS Gatteway that alloows members oof the
ppublic to interaact with the enttire system and also used to reeceive
wweather readinggs from sensorrs into the systeem; and (7) a user-
ffriendly web portal used for bboth system addministration ass well
ffor displaying detailed
d inform
mation on droughhts and other reelated
ddetails. In linee with the ITIK KI Architecturee shown in figgure 1
((above), these system
s moduless were implemeented in three layers:
oone layer for eaach of the three components.

Drought Monitoring Disseminnation

Knowledge andd and
• Sensor data Forrecasting Communiccation
• Weather • EDDI Monitor • Mobile
Station Data • ANNNs Models Applicatiion
• Indigenous • IK
K Fuzzy • SMS Faccility
Knowledge Syystem • Web Porttal
• Published • Text-to-S
Speech
Weather Dataa • Speech-too-Text
Figurre 3. Mobile Application Main Menu

Figure 2. Architecture of the System Prototype

F 4.22.2 Sensor-B
44.2 Layer 1: Droughtt Knowledgge
SSensor data from sensors is seent to a MySQLL database via a java-
bbased SMS ggateway. Weathher Station Data and Indiggenous
KKnowledge are captured into tthe system via a web interface and a
mmobile phone application resspectively. In iits current form
m, the
DDEWS prototyype does not carry out weaather forecast; such
iinformation is rretrieved from Published Weaather Data espeecially
ffrom meteoroloogical departmeents’ weather reeports. Data intto the
ssystem was capptured using the following systeem modules:
44.2.1 Mobille Applicationn
AAn intermediarry who acts on bbehalf of a focuus group made of IK
eexperts uses thhis application. This individuaal sieves througgh the
ddiscussions by the IK experts,, translates, form
mats and keys in the
Based Weathher Station
Twwo types of deplloyments were set up: (1) Senssor Boards Next to
Weeather Stations: here, sensor boards were pplaced within the
Obbservatory Unitts of selected w weather stationns in Kenya. T The
boaards individuallly sent readingss to a remote dattabase via an SMMS
Gaateway. The ssensors includeed were thosse for measurring
tem
mperature, relatiive humidity annd atmospheric ppressure. In a ffew
of the locations, rainfall, wind speed, wind ddirection, and soil s
mooisture sensors were installed. Apart from moonitoring weathher,
this set up soughtt to further validdate the calibraation decisions. (2)
Staand-Alone Senssor Boards: in oorder to deploy the sensors in the
rurral areas especiially in Mbeeree and Bunyore on which the IK
casse study was based, standd-alone sensorrs mounted w with
tem
mperature, relaative humidity and atmospheeric sensors were w
useed. The sensorss boards were pplaced inside traaditional granarries
whhich provided ennvironment alm most similar to the one supporrted
by the Shenstone Screens. All ddata from the seensors is accessiible

65
from the web portal from where the system administrator is able to Df – for Drought Forecast; Weather Station – represented by
monitor the sensors behaviour. ICAO; ForecastType – either Short Term (1), Medium Term
(2) or Long Term 3); Lead-Time – ranging from 1 to 14
4.2.3 Published Weather Reports days for Short Term, 1 to 12 months for Medium Term and 1
Since weather forecasting was not part of the DEWS described to 4 years for Long Term
here, the various weather forecasts needed for various DEWS’
decisions are currently retrieved from published weather reports. Example:Df;HKEM;2,2 - A request for Medium-Term Drought
For the evaluation of the system prototype, the Seasonal for Embu weather station for two months lead time.
Forecasts were automatically (using agents) retrieved from Kenya
Meteorological Department’s (KMD) website (www.meteo.go.ke)
and used to populate the relevant database tables. Daily, 5-day
and 7-day weather forecasts (also from KMD) are used as input
to the ANNs forecasting models. Some aspects of the Seasonal
Forecasts that deal with advising stakeholders are incorporated in
the IK Fuzzy System to enrich the disseminated forecasts.
Finally, the 10-day Dekad Reports are used to evaluate the extent
to which the EDI monitor is able to quantify droughts.

4.3 Layer 2: Monitoring and Forecasting

The data collected and stored in the Drought Knowledge layer is
used by the following:
(1) EDI Monitor that takes care of the ‘Monitoring’ aspect; this is
implemented as a stand-alone FORTRAN program. The output Figure 4. Medium term Drought Forecast - SMS
(in form of text files) is then used to run insert queries to populate
4.5.2 Drought Forecasts Via Email
the database. Using the daily precipitation, computed Effective
For each of the requests received, the EmailAgent sends a copy of
Drought Index (EDI) and/or Available Water Resource Index
the drought forecast to the sender. The Agents interacts with the
(AWRI) and Drought Classification, a web-based interface was
WeatherDatabaseAgent to retrieve the email address of the SMS
implemented to demonstrate how EDI could aid in detecting and
sender; an email is then sent to this email address.
preparing for droughts/floods [46]. (2) ANNs models that were
implemented in MATLAB are used to forecast future values of
drought in terms of EDI and AWRI [7]; and (3) IK Fuzzy Logic
System handles the monitoring and forecasting of drought using
indigenous knowledge. This was also implemented in MATLAB;
its output is uploaded to the MySQL database from where it is
linked to the ANNs models’ output.

4.4 Layer 3: Forecasts Communication and

Dissemination
The system is accessible to end users via five input/output
channels: (1) Mobile Application that is used for inputting and
outputting IK indicators and extreme weather events; (2) SMS
Facility for requesting drought and weather forecasts as well for
extreme weather events alerts; (3) Web Portal which is used to
access comprehensive information on droughts, weather and
extreme events. Formats available include text, graphs and audio
files; (4) Text-to-Speech is simple plug-in implemented to
generate audio formats of the drought and weather forecasts; and
(5) Speech-to-Text which is also a plug-in of the android phone;
it is used as an alternative to typing input data via the android
application. Here, a multi-agent system was created to link up all
the DEWS’ components after which the various output interfaces
were created. In line with the overall framework (ITIKI), this is
made up of 3 types of agents; (1) Information Capture Agents, (2)
Monitoring and Forecasting Agents, and (3) Dissemination/
Communication Agents.
4.5 System Testing and Evaluation
4.5.1 Drought Forecasst Via SMS
The system accepts requests in form of text messages (sms) in the
format:
<Df>;<Weather Station>;<ForecastType>;<Lead-
Time>;
Figure 5. Medium term Drought Forecast - Email
4.5.3 Drought Forecasts Via Audio Files
For each of the drought forecast, the SpeechAgent generates an
audio version and stores it on the remote site. The link to the file
is stored in the database table, alertsTable and is available for
download via the web portal. For instance, the requests above are
stored in the paths:
audio/Drought/HKEM/2012-8.wav – representing the audio file
for the medium-term drought forecast for Embu for August, 2012
4.5.4 Drought Forecasts Via Web Portal
All the functionalities of the integrated DEWS are accessible from
the web portal. Examples include:
Real-Time Weather – to check the most current weather for a
given location; this displays readings from both the weather
station (if any) and sensor boards
Drought Monitoring that supports short, medium and long-term
drought forecasts as discussed under Agents

66
 Figure 6. Medium Term Drought Forecast – Webpage Request
Selecting 2 (for 2 months) yields the results below:
Figure 7. Medium term Drought Forecast – Webpage results
5. DISCUSSION AND FURTHER WORK
The overall solution to the drought problem is encapsulated in our
novel integrated drought early warning system called itiki. Itiki is
first an acronym for Information Technology and Indigenous
Knowledge with Intelligence. Secondly, itiki is used among the
Mbeere people (in Eastern part of Kenya) to refer to an
indigenous bridge used until late 90s to go across rivers. The
resilience of this bridge in relation to their modern bridges
counterparts motivated the naming of the drought solution. Using
this analogy, the DEWS presented in this work took the form of a
bridge that provides the much-needed link between the scientific
and indigenous drought forecasting approaches. The raw material
for our bridge is ICTs; in particular, WSNs and mobile phones.
Artificial Intelligence technologies were used to glue these diverse
components into an intelligent, integrated early warning system
for droughts.
The author acknowledges that predicting droughts alone cannot
eradicate droughts in SSA; access to relevant and accurate
information on impending droughts in timely fashion and
comprehensible formats however could go a long way in assisting
all the stakeholders plan for and mitigate effects of the droughts.
This precisely is the contribution of itiki. The presence of sensors
in itiki enables capture of micro weather data and hence, improved
prediction accuracy. Indigenous knowledge on the other hand
helps in improving relevance (culturally and locally), acceptability
and sense of ownership of the forecasts among the small-scale
farmers. The systematic capture and storage of IK on weather that
we implemented is a phenomenal step towards the much-needed
conservation of the endangered IK.
The DEWS presented here is a prototype with lots of extensions
in-waiting as described below:
67
Drought Knowledge Component; first, more sensor boards need
to be acquired and installed. This will improve the current poor
coverage by conventional weather stations.
Drought Monitoring and Predicting component, some work into
how the output from the ANNs could automatically be integrated,
as input into the IK fuzzy logic sub-system is required. Since both
sub-systems are currently implemented using MATLAB
Toolboxes, this integration is feasible. Once this is done, learning
ability needs to be integrated into the resulting sub-system. One
way of achieving this is adoption of Belief-Desire-Intention (BDI)
agents’ model that supports learning.
Drought Dissemination and Communication; innovative
enhancements will be required in order to make the system more
accessible, usable and relevant to the semi-illiterate small-scale
farmers whose only access to ICTs is via mobile phones that are
mostly low-end.
6. REFERENCES
[1] United Nations Office for Disaster Risk
Reduction(UNISDR), Economic Commission for Africa
(UNECA) and United Nations Development Programme
(UNDP), 2008. 3rd African Drought Adaptation Forum
Report 17-19 September 2008. Addis Ababa, Ethiopia:
UNISDR; UNECA; UNDP. Available online:
http://www.unisdr.org/files/8339_3rdAfricanDroughtAdaptat
ionForumReport1.pdf; accessed on 14 May 2014
[2] Haile, Menghestab 2005. Weather Patterns, Food Security, and
Humanitarian Response In Sub-Saharan Africa. Philosophical
Transactions: Biological Sciences 360(1463): 2169-2182.
JSTOR, http://www.jstor.org , accessed August. 8, 2014.
[3] Mutua, S., 2011. Strengthening Drought Early Warning At
The Community And District Levels: Analysis Of Traditional
Community Warning Systems In Wajir & Turkana Counties.
Oxfam GB, PO Box 40680, 00100 Nairobi, Kenya: Oxfam
GB, Kenya.
[4] Mugabe, F.,T., Mubaya, C.,P., Nanja, D., H., Gondwe, 2010.
Use of indigenous knowledge systems and scientific methods
for climate forecasting in southern Zambia and north western
Zimbabwe. Zimbabwe Journal of Technological Sciences, 1
(1)
[5] Ziervogel, G. and Calder, R., 2003. Climate Variability and
Rural Livelihoods: Assessing the Impact of Seasonal Climate
Forecasts in Lesotho. Area, 35(4), pp. 403-417
[6] Orlove, B., Roncoli, C., Merit, K. and Abushen, M., 2009.
Indigenous climate knowledge in southern Uganda: the
multiple components of a dynamic regional system. Climate
Change, 100 (2), pp. 243-265.
[7] Masinde, M, 2013. Artificial Neural Networks Models for
Predicting Effective Drought Index: Factoring Effects of
Rainfall Variability. Mitigation and Adaptation Strategies for
Global Change Journal, DOI: 10.1007/s11027-013-9464-0
ISSN:1381-2386, Springer Netherlands.
[8] Masinde, M. and Bagula, A., 2012. ITIKI: bridge between
African indigenous knowledge and modern science of
drought prediction. Knowledge Management for
Development Journal, 7(03), pp. 274-290
[9] Masinde, M. and Bagula, A., 2013. Downscaling Weather
Forecasts using ITIKI in Mbeere: Implementation Roadmap,
Proceedings of ITU Kaleidoscope 2013: Building
Sustainable Communities, K-2013, held in Kyoto Japan
between 22-24 April 2013, IEEE Xplore, pp. 1-8
[10] Masinde, M. and Bagula, A., 2013. SenseWeather: Sensor-
Based Weather Monitoring System for Kenya, In: The IST-
Africa 2013 Conference Proceedings, Nairobi, Kenya, 29-31
May 2013, IEEE Xplore, pp. 1-13
[11] Masinde, M., 2012, ITIKI: Bridge between African
Indigenous Knowledge and Modern Science on Drought
Prediction', PhD thesis, University of Cape Town
[12] Masinde, M., Bagula, A. and Muthama, N., 2012. The Role
of ICTs in Downscaling and Up-scaling Integrated Weather
Forecasts for Farmers in Sub-Saharan Africa, M. Best L., Z.
Ellen, D. Jonathan, G. Beki and M. Gary, eds. In: The Fifth
International Conference on Information and
Communication Technologies and Development, March 12-
15 2012, ACM Digital Library, pp. 122.
[13] Masinde, M, 2014. An Assessment of Field Readiness for
Wireless Sensor Based Weatherboards– A Calibration
Report, In: 12th IEEE International Conference on Industrial
Informatics (INDIN 2014) Proceedings, held in Porto Alegre,
Brazil 27-30 July 2014
[14] Elsa, E., Moreira, Carlos, A., Coelho and Ana, A., Paulo.
2008. SPI-Based Drought Category Predication Using
Loglinear Models. Journal of Hydrology 354, 116-130.
[15] Kung, H., Hua, J. and Chen, C. 2006. Drought forecast
model and framework using wireless sensor networks.
Journal of Information Science and Engineering 22, 751-769.
[16] Mishra, A.,K. and Desai, V.,R. 2006. Drought Forecasting
Using Fee-Forward Recursive Neural Network. Ecological
Modelling 198, 128-138.
[17] Mishra, A.K. and Singn, V.P. 2010. A Review of Drought
Concepts. Journal of Hydrology 391, 202-216.
[18] Patt, A., G. and Jordan, W., 2007. Applying Climate
Information in Africa; An Assessment of Current Knowledge.
http://www.iiasa.ac.at/Publications/Documents/XO-07-
006.pdf: Written in support of NOAA, Boston University and
the International Institute for Applied Systems Analysis (
IIASA).
[19] Virji, H., Cory, F., Amy, F. and Mayuri, S., 1997. Climate
Variability, Water Resources and Agricultural Productivity:
Food Security Issues in Tropical Sub-Saharan Africa.
Workshop on Climate Variability Prediction:
START/WCRP/OSTROM/SCOWAR.
[20] IFPRI, 2011-last update, 2011 GHI - Facts and Findings:
Sub-Saharan Africa [Homepage of International Food Policy
Research Institute], [Online]. Available: [May 20, 2012].
http://www.ifpri.org/publication/2011-ghi-facts-and-
findings-sub-saharan-africa
[21] Ziervogel, G. and Calder, R., 2003. Climate Variability and
Rural Livelihoods: Assessing the Impact of Seasonal Climate
Forecasts in Lesotho. Area, 35(4), pp. 403-417.
[22] Luseno, W.,K., Mcpeak, J.,G, Barrett, C.,B., Little, P.,D and
Gebru, G., 2003. Assessing the value of climate forecast
information for pastoralists: evidence from southern Ethiopia
and Northern Kenya 31, pp. 1477-1494.
[23] Washington, R. and Downing, T., E., 1999. Seasonal
forecasting of African rainfall: Prediction, responses and
household food security. Geographical Journal, 165, pp.
255-274.
68
[24] Mercer, J., Kelman, I., Taranis, L. and Suchet-Pearson, S.,
2010. Framework for integrating indigenous and scientific
knowledge for disaster risk reduction. Disasters, 34 (1), pp.
214-239.
[25] Ziervogel, G. and Opere, A., 2010. Integrating
meteorological and indigenous knowledge-based seasonal
climate forecasts for the agricultural sector: Lessons from
participatory action research in sub-Saharan Africa. 2010.
http://web.idrc.ca/uploads/user-
S/12882908321CCAA_seasonal_forecasting.pdf: IDRC.
[26] Dondeyne, S., Emanuel, L., B. and Deckers, J., A., 2003. Mr
Napite’s Botanical Knowledge: Bridging Farmers’ and
Scientists’ Insights During Participatory Research. African
Journal Of Indigenous Knowledge Systems, 2 (2), pp. 45-57.
[27] Aparna, P. and Trivedi, P.,C., 2011. Cultural Values and
Indigenous Knowledge of climate change and disaster
prediction in Rajasthan, India. Indian Journal of Indigenous
Knowledge, 10 (1), pp. 183-189.
[28] Grasso, V., F., 2007. Early Warning Systems: State-of-Art
Analysis and Future Directions. UNEP.
[29] Masinde, M., Bagula, A. and Muthama, N., 2012. Wireless
Sensor Networks (WSNs) use in Drought Prediction and
Alerts. In: A. Finlay and E. Adera, eds, Application of ICTs
for Climate Change Adaptation in the Water Sector:
Developing Country Experiences and Emerging Research
Priorities. APC, IDRC, CRDI; Association for Progressive
Communication (APC) and the International Development
Research Centre (IDRC), South Africa, pp. 107-108
[30] Krupnik, I., I. and Ray, G., Carleton, 2007. Pacific Walruses,
Indigenous Hunters, and Climate Change: Bridging
Scientific and Indigenous Knowledge. Deep Sea Research II,
54 (2007), pp. 2946-2957.
[31] Lauler, M. and Shankar, A., 2008. Integrating indigenous
ecological knowledge and multi-spectral image classification
for marine habitat mapping in Oceania. Ocean & Coastal
Management, 51 (2008), pp. 495-504.
[32] Voinov, A.,A. and Brown, Gaddis, E.,J., 2008. Lessons for
successful participatory watershed modelling : a perspective
from modelling practioners. Ecological modelling, 216
(2008), pp. 197-207.
[33] Rajaram, T. and Ashutosh, D., 2010. Modelling of
interactions among sustainability components of an agro-
ecosystem using local knowledge through cognitive mapping
and fuzzy inference system. Expert Systems with
Applications, 37 (2010), pp. 1734-1744.
[34] Voinov, A.,A. and Bousquet, F., 2010. Modelling with
stakeholders. Environmental modelling and software, 25
(2010), pp. 1268-1281.
[35] Giordanoa, R. and S., L., 2011. A fuzzy GIS-based system to
integrate local and technical knowledge in soil salinity
monitoring. Environmental Modelling & Software, XXX
(2011), pp. 1-15.
[36] Sicat, R.S., Emmanuel, John, M., Carranza and Uday, B.,
Nidumolu, 2005. Fuzzy modelling of farmers’ knowledge for
land suitability classification. Agricultural Systems, 89
(2005), pp. 49-75.
[37] Apua NEW Guinea National Disaster Centre, 2005. Papua
New Guinea National Disaster Risk Reduction and Disaster
Management Framework for Action 2005–2015. National
Disaster Centre, Port Moresby.
[38] Gray, L.C. and Morant, P., 2003. Reconciling indigenous
knowledge with scientific assessment of soil fertility changes
in southwestern Burkina Faso. Geoderma, 111 :, 111 (2003),
pp. 425-437.
[39] Payton, R.W., Barr, J., Martin, A., Sillitoe, P., Deckers, J.F.
and Gowing, J.W., 2010. Contrasting approaches to
integrating indigenous knowledge about soils and scientific
soil survey in East Africa and Bangladesh. Geoderma, 111
(3-4), pp. 4-386.
[40] Berkes, F. And Mina, K., Berkes, 2009. Ecological
complexity, fuzzy logic, and holism in indigenous
knowledge. Futures, 41 (1), pp. 6-12.
[41] Verlinden, A. and Dayot, B., 2005. A comparison between
indigenous environmental knowledge and a conventional
vegetation analysis in north central Namibia Journal of Arid
Envirnoments, 62 (1), pp. 143-175.
[42] Pahl-Wostl, C., 2007. Requirements for Adaptive Water
Management. In: C. Pahl-Wostl, P. Kabat and J. Möltgen,
eds, Adaptive and Integrated Water Management. Coping
with Complexity and Uncertainty. Heidelberg, Germany:
Springer Verlag, pp. 1-22.
[43] Slegers, M.F.W., 2008. “If only it would rain”: Farmers’
perceptions of rainfall and drought in semi-arid central
Tanzania. Journal of Arid Envirnoments, 72 (2008), pp.
2106-2123.
[44] Roncoli, C., 2006. Ethnographic and participatory
approaches to research on farmers' responses to climate
predictions. Climate Research, 33 (1), pp. 81-99.
[45] ISDR, 2006. Developing Early Warning Systems: A
Checklist. Third International Conference on Early
Warning: From concept to action, EWC III, pp. 1-13.
[46] Masinde, M. and Bagula, A., 2011. The Role of ICTs in
Quantifying the Severity and Duration of Climatic Variations
– Kenya’s Case, Proceedings of ITU Kaleidoscope 2011:
The Fully Networked Human? - Innovations for Future
Networks and Services K-2011, 12-14 December 2011, IEEE
Xplore, pp. 1-8.
69