Exploring the spatial relationship between NDVI and rainfall in the semi-arid Sahel with geographically weighted regression
(2016) In Student thesis series INES NGEM01 20161Dept of Physical Geography and Ecosystem Science
- Abstract
- The Normalized Vegetation Difference Index (NDVI) is frequently used as a surrogate for vegetation properties and is often correlated with climatic variables such as rainfall. However, studies have shown that conventional regression models used to study the spatial relationship between NDVI and rainfall are often plagued by non-stationarity and are scale dependent. This thesis employed a spatial disaggregation modelling technique to tackle this issue – Geographically Weighted Regression (GWR) allows measured relationships to vary in space. GWR was applied in the Sahel of Africa for the growing seasons of 2002 and 2012 (June-September).
The results highlighted areas which were particularly sensitive to variations in rainfall and which... (More) - The Normalized Vegetation Difference Index (NDVI) is frequently used as a surrogate for vegetation properties and is often correlated with climatic variables such as rainfall. However, studies have shown that conventional regression models used to study the spatial relationship between NDVI and rainfall are often plagued by non-stationarity and are scale dependent. This thesis employed a spatial disaggregation modelling technique to tackle this issue – Geographically Weighted Regression (GWR) allows measured relationships to vary in space. GWR was applied in the Sahel of Africa for the growing seasons of 2002 and 2012 (June-September).
The results highlighted areas which were particularly sensitive to variations in rainfall and which seemingly form large clusters that connect humid and arid climatic zones. In these areas, rainfall appears to be the dominant determinant in understanding the distribution of vegetation. Moreover, regions mainly located around wetlands were shown to have a very weak relationship with rainfall indicating the need for incorporating additional variables to explain the NDVI variation. Finally, temporal variations were showcased as the spatial relationships would often change from a drier year to a more humid one.
In comparison with traditional linear regression modelling such as Ordinary Least Squares (OLS), GWR model performed significantly better in both years, by producing more accurate predictions, reducing autocorrelation in the regression residuals and allowing for local inferences to be made due to a large output from GWR results being a set of maps showcasing the local situation between NDVI and rainfall. The results were validated by conventional regression diagnostics and local tests to assess the significant and degree of non-stationarity in the data. Therefore, GWR is suggested as an accurate, informative technique both for exploratory and explanatory reasons to treat non-stationarity in heterogeneous areas in an ecological context. (Less) - Popular Abstract
- The spatial relationship of rainfall and vegetation in the Sahel
The Normalized Vegetation Difference Index (NDVI) is frequently used as a surrogate for vegetation properties and is often correlated with climatic variables such as rainfall. In a large scale analysis, the relationship between vegetation and rainfall often emerges with characteristics such as non-stationarity and scale dependency. In this thesis, the spatial variation and scale dependency of that relationship was revealed in the Sahel of Africa, a prominent transition zone between arid to humid environments which have been in the spotlight of scientific research since the past decades due to its sensitive spatiotemporal dynamics. By taking into account the existence of... (More) - The spatial relationship of rainfall and vegetation in the Sahel
The Normalized Vegetation Difference Index (NDVI) is frequently used as a surrogate for vegetation properties and is often correlated with climatic variables such as rainfall. In a large scale analysis, the relationship between vegetation and rainfall often emerges with characteristics such as non-stationarity and scale dependency. In this thesis, the spatial variation and scale dependency of that relationship was revealed in the Sahel of Africa, a prominent transition zone between arid to humid environments which have been in the spotlight of scientific research since the past decades due to its sensitive spatiotemporal dynamics. By taking into account the existence of different types of land cover, and the extreme heterogeneity in climatic conditions from the southern Sahara boundaries to the humid areas of the South, the effect of rainfall upon the spatial patterns of vegetation was the main interest of this study.
By undertaking a spatial disaggregation modelling technique named Geographically Weighted Regression (GWR) and after finding the most appropriate scale to examine the spatial relationship, the results highlighted areas which were particularly sensitive to variations in rainfall and which seemingly form large clusters that connect humid and arid climatic zones. In these areas, rainfall appears to be the dominant determinant in understanding the distribution of vegetation. Moreover, regions mainly located around wetlands were shown to have a very weak relationship with rainfall indicating the need for incorporating additional variables to explain the NDVI variation. Finally, temporal variations were showcased as the spatial relationships would often change from a drier year to a more humid one.
In comparison with traditional linear regression modelling such as Ordinary Least Squares (OLS), GWR model performed significantly better in both years, by producing more accurate predictions, reducing autocorrelation in the regression residuals and allowing for local inferences to be made due to a large output from GWR results being a set of maps showcasing the local situation between NDVI and rainfall. The results were validated by conventional regression diagnostics and local tests to assess the significant and degree of non-stationarity in the data. Therefore, GWR is suggested as an accurate, informative technique both for exploratory and explanatory reasons to treat non-stationarity in heterogeneous areas in an ecological context. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8885965
- author
- Georganos, Stefanos LU
- supervisor
-
- Hakim Abdi LU
- David Tenenbaum LU
- organization
- course
- NGEM01 20161
- year
- 2016
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- geomatics, rainfall, Sahel, scale dependency, non-stationarity, physical geography and ecosystem analysis, geographically weighted regression
- publication/series
- Student thesis series INES
- report number
- 394
- language
- English
- id
- 8885965
- date added to LUP
- 2016-06-30 12:59:48
- date last changed
- 2016-06-30 12:59:48
@misc{8885965,
abstract = {{The Normalized Vegetation Difference Index (NDVI) is frequently used as a surrogate for vegetation properties and is often correlated with climatic variables such as rainfall. However, studies have shown that conventional regression models used to study the spatial relationship between NDVI and rainfall are often plagued by non-stationarity and are scale dependent. This thesis employed a spatial disaggregation modelling technique to tackle this issue – Geographically Weighted Regression (GWR) allows measured relationships to vary in space. GWR was applied in the Sahel of Africa for the growing seasons of 2002 and 2012 (June-September).
The results highlighted areas which were particularly sensitive to variations in rainfall and which seemingly form large clusters that connect humid and arid climatic zones. In these areas, rainfall appears to be the dominant determinant in understanding the distribution of vegetation. Moreover, regions mainly located around wetlands were shown to have a very weak relationship with rainfall indicating the need for incorporating additional variables to explain the NDVI variation. Finally, temporal variations were showcased as the spatial relationships would often change from a drier year to a more humid one.
In comparison with traditional linear regression modelling such as Ordinary Least Squares (OLS), GWR model performed significantly better in both years, by producing more accurate predictions, reducing autocorrelation in the regression residuals and allowing for local inferences to be made due to a large output from GWR results being a set of maps showcasing the local situation between NDVI and rainfall. The results were validated by conventional regression diagnostics and local tests to assess the significant and degree of non-stationarity in the data. Therefore, GWR is suggested as an accurate, informative technique both for exploratory and explanatory reasons to treat non-stationarity in heterogeneous areas in an ecological context.}},
author = {{Georganos, Stefanos}},
language = {{eng}},
note = {{Student Paper}},
series = {{Student thesis series INES}},
title = {{Exploring the spatial relationship between NDVI and rainfall in the semi-arid Sahel with geographically weighted regression}},
year = {{2016}},
}