This chapter provides a background to the research problem and the significance of
the study. It states the research aim, objectives, questions and scope of the study. The
chapter gives an overview of the perspective of rainfall and its impact on agriculture.
Focus is on seasonal rainfall forecasting technique currently used in Zambia and how
to improve its accuracy. The increasing interest and need by the public for reliable
seasonal rainfall forecast provoked further investigation, resulting in this thesis. A
conceptual framework model is used to illustrate a theoretical overview of the intended
research and show the order of processes, including how the variables to be considered
might relate to each other.
Weather describes the atmospheric conditions at a given place over a short duration of
time 1. It is defined in terms of weather parameters like temperature, pressure, wind
(speed and direction), clouds and rainfall. Weather forecasting is a scientific estimation
of future weather conditions 2. Weather forecasting applies science and technology
to predict the state of the atmosphere for a given time and location 3, 4, 5, 6.
Rainfall is a natural climatic phenomenon resulting from atmospheric, oceanic circulation
systems and complex physical processes that cause an amount of rain to fall
at a place during a particular period 7, 8. It is a stochastic process which depends
on some precursors from other parameters such as temperature, wind, pressure and
other atmospheric parameters 9. Rainfall is one of the weather parameters whose
accurate forecasting has significant implications for agriculture and water resource
management 10, 11. Rainfall is the most important climate variable that affects
agriculture in countries that depend on rain fed agriculture 12.
A forecast describes what will possibly happen in future. Weather forecasting entails
predicting how the present state of the atmosphere will change. Weather forecasting
is a demanding operational responsibilities carried out by meteorological services.
It is a complicated procedure that includes numerous specialized fields of expertise
13, 11. Generating weather forecast is complex because in the field of meteorology
all decisions are to be taken in the visage of uncertainty 11. The chaotic nature of the
atmosphere and systems responsible for events are a culmination of instabilities 14.
Weather forecasts are used in many sectors like agriculture, advisories, and severe
weather alerts. Modern weather forecasting of severe weather alerts and advisories
are done to protect life and property 15.
Weather forecasting is a canonical predictive challenge that has depended primarily
on model-based methods. Making inferences and predictions about weather has been
an omnipresent challenge throughout human history. Challenges with accurate meteorological
modeling brings to the fore difficulties with reasoning about the complex
dynamics of Earth’s atmospheric system 16
Rainfall forecasting is part of weather forecasting and is very essential for various
sectors 17. Of all the weather parameters, rainfall forecasting is the one that is most
complicated and challenging operational task done by meteorological services world
over11, 18. Rainfall prediction is challenging, demanding and complex due to the
various dynamic environmental factors, both spatial and temporal random variations.
Rainfall is a highly non-linear parameter 10, 8, 19, 20, 21.
There are four major different weather forecast types, namely;
Now-casting weather forecast is a very short-period prediction that maps current
weather conditions and changes to predict weather conditions for a period of 0 to 6
hours ahead 22, 23, 20. With now-casting, it is possible to forecast smaller features
with reasonably more accuracy 23.
Short-range weather forecast gives a prediction of the atmospheric condition in each
successive 24 hours for a period of 1 to 3 days. Such forecasting models are set up
to produce daily averages in order to use the same time scale as rain gauges observations.
Most results of short-range forecast show good agreements between the
predicted rainfall and measurements from rain gauge stations for the given period
Medium-range weather forecasts gives prediction for a period of 4 to 10 days. It gives
a forecast of average weather conditions and may prescribe weather on each day with
progressively lesser details and accuracy than for short-range forecasts 25.
Long-range forecast, also known as extended range weather forecast ranges from more
than 10days, a month, a seasonal, and a year to even forecast for longer period. Long
Range Forecasts (LRF), like seasonal rainfall forecasts are even more challenging to
forecast as atmospheric systems may change in time and space 25. Weather forecasts
become less accurate as the difference in current time and the time for which
the forecast is being made (as the range of the forecast) increases 4, 26.
Amongst all weather parameters, rainfall is the one that mostly affects human life
and livelihood in developing countries and least developed countries like Zambia where
majority of the population depends on rain fed agriculture 11, 27, 28, 1, 29. Rainfall
also affects many sectors including but not limited to water resources management,
energy, tourism, health, disaster risk reduction (DRR) and infrastructure development.
These are the core focus areas of the 7 National Development Plan (7NDP)
People wish to know in advance whether there would be normal rainfall in the coming
rainy season. To achieve this requirement, every National Meteorology and Hydrological
Services (NMHS) needs to forecast well ahead the start of the crop season. Such
rainfall forecasts are used by both farmers and government to plan for the ensuing
rainy seasonal 12. Thus, accurate seasonal rainfall forecast is essential for planning
of agriculture, water resources management, and many other sectors 31, 32, 12, 33.
Accurate long-range forecast provides farmers with sufficient time to plan for crop
production by adopting appropriate crops and varieties, that will be suitable for the
expected rainfall 34.
Seasonal rainfall forecast is the prediction of the expected rainfall performance for
the given rainy season. It is usually generated in August and published in September
in the Southern African Development Community (SADC) region, Zambia inclusive.
Empirical statistical forecasting model is developed using Simple Linear Regression
Model (SLRM) to forecast Seasonal rainfall. This model describes a linear relationship
between two variables; X as independent which is Sea Surface Temperature
basins and Y as dependent is rainfall. Statistical models based on regression analysis
and eyeball inspection are used. Current seasonal rainfall forecasting methods used
in Zambia have been proved to be less accurate 7, 8.
This research proposes to use Artificial Neural Networks (ANNs) in order to improve
the accuracy of seasonal rainfall forecast in Zambia, because statistical models have
some inherent limitations over long range rainfall forecasts 35, 36.
1.2 Problem Statement
The current seasonal rainfall forecasting method used in Zambia assumes a direct
correlation between the Sea Surface Temperatures (SST) and station rainfall observations
37. Atmospheric systems are not governed by only these two systems, but
this assumption ignores availability of other systems in influencing rainfall 38. Other
parameters that may have influence on rainfall include Indian Ocean Dipole (IOD),
temperature, wind speed, relative humidity and pressure 11. Moreover, changing
climate has introduced further uncertainties in this assumption of a direct linear correlation
between observed rainfall data and SST 38, 39.
A limitation of high spatial variability of station point rainfall observations increases
the inaccuracy and uncertainty that reduce the skill (accuracy) of the seasonal rainfall
forecasts. Zambia’s area is about 752, 613KM2, but only 33 weather stations which
have enough historical data that is used in generating of the seasonal rainfall forecast.
A common weakness of all statistical rainfall forecasting models is that while the
correlations are assumed to remain constant for the duration of the forecast, they
usually change with time and slowly lose their significance, which makes the rainfall
forecast less accurate 40. Long range weather forecasts like seasonal rainfall forecast
become less accurate as the difference in time between the present moment and the
time for which the forecast is given increases.
Furthermore, some stages in the current seasonal rainfall forecasting process require
expert knowledge through eye ball inspection which is subjective and not easy to pass
on through an educational process 41.
Therefore, the current seasonal rainfall forecast in Zambia is not of high efficacy.
1.3 Aim of the Study
This research is in response to forecasting seasonal rainfall accuracy inefficiency. Thus,
the aim of this research is to improve accuracy of forecasting seasonal rainfall in
Zambia through the use of Artificial Neural Networks.
This chapter provides a background to the research problem and the significance of