Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Korttidsprognoser av Kalixälvens vattenflöde

Carlsson, Mathilda LU (2023) STAH11 20212
Department of Statistics
Abstract
The Kalix River, located in the northern part of Sweden, experiences a slowdown
in water flow during cold winters. However, in spring, when snow and ice melt in
the mountains, the water flow starts to increase. Towards the end of spring, there
is usually a rapid and unpredictable surge in water flow known as the spring flood.
The arrival of the spring flood is stochastic in time and does not occur on the same
day or even the same week each year. Therefore, accurate forecasts for the arrival of
the spring flood are essential and this thesis aims to improve short-term predictions
of water flow.
The seasonal dependence and variability of the water flow brings to mind whether
some recursive and dynamic models would make better... (More)
The Kalix River, located in the northern part of Sweden, experiences a slowdown
in water flow during cold winters. However, in spring, when snow and ice melt in
the mountains, the water flow starts to increase. Towards the end of spring, there
is usually a rapid and unpredictable surge in water flow known as the spring flood.
The arrival of the spring flood is stochastic in time and does not occur on the same
day or even the same week each year. Therefore, accurate forecasts for the arrival of
the spring flood are essential and this thesis aims to improve short-term predictions
of water flow.
The seasonal dependence and variability of the water flow brings to mind whether
some recursive and dynamic models would make better predictions for this river
than a simple autoregressive (AR) model. The predictive ability of the AR model of
order five is compared to three recursive models, relying on this AR(5) model. These
models use the estimation methods; forgetting factor, sliding window and Kalman
filter. The coefficients in the recursive models are dynamically estimated and time
adjusted.
The performance of each model is evaluated, and it is observed that all models
perform well for one-day-ahead predictions, but show slightly lower accuracy for
three-day-ahead predictions. Despite the initial expectation of improved predictions
with dynamic models, there was no significant improvement compared with the more
simple AR(5) model in any case. (Less)
Please use this url to cite or link to this publication:
author
Carlsson, Mathilda LU
supervisor
organization
course
STAH11 20212
year
type
M2 - Bachelor Degree
subject
keywords
ARIMA, RecursiveAR, forgetting factor, sliding window, Kalman filter, short-term predictions, water flows
language
Swedish
id
9113882
date added to LUP
2023-05-04 08:50:09
date last changed
2023-05-04 08:50:09
@misc{9113882,
  abstract     = {{The Kalix River, located in the northern part of Sweden, experiences a slowdown
in water flow during cold winters. However, in spring, when snow and ice melt in
the mountains, the water flow starts to increase. Towards the end of spring, there
is usually a rapid and unpredictable surge in water flow known as the spring flood.
The arrival of the spring flood is stochastic in time and does not occur on the same
day or even the same week each year. Therefore, accurate forecasts for the arrival of
the spring flood are essential and this thesis aims to improve short-term predictions
of water flow.
The seasonal dependence and variability of the water flow brings to mind whether
some recursive and dynamic models would make better predictions for this river
than a simple autoregressive (AR) model. The predictive ability of the AR model of
order five is compared to three recursive models, relying on this AR(5) model. These
models use the estimation methods; forgetting factor, sliding window and Kalman
filter. The coefficients in the recursive models are dynamically estimated and time
adjusted.
The performance of each model is evaluated, and it is observed that all models
perform well for one-day-ahead predictions, but show slightly lower accuracy for
three-day-ahead predictions. Despite the initial expectation of improved predictions
with dynamic models, there was no significant improvement compared with the more
simple AR(5) model in any case.}},
  author       = {{Carlsson, Mathilda}},
  language     = {{swe}},
  note         = {{Student Paper}},
  title        = {{Korttidsprognoser av Kalixälvens vattenflöde}},
  year         = {{2023}},
}