Short-interval monitoring of land use changes with RADARSAT-1

Abstract

Conventional land use change detections with remote sensing use annual

remote sensing images because of the limitations of optical sensors that cannot

collect data in bad weather and cloudy conditions. This limits its applications in

rapidly developing areas which are cloudy, such as the Pearl River Delta in China.

These areas also need to detect land use changes in short intervals, such as on a

monthly basis, in order to monitor illegal land use changes and prevent

irreversible land use changes that may damage the environment. The objective of

the thesis is to examine short-interval land use change detection, especially the

change from agriculture to built-up areas, using RADARSAT-1 images which can

go through clouds.

This thesis firstly examines the classification of RADARSAT-1 images with

pixel-based and object-based classification methods respectively. Based on the

classification results, post-classification change detection method are conducted in

order to obtain the detailed information of land use changes for the analysis of

short-interval land use change.

Land use change detection accuracies can be improved as the number of the

RADARSAT-1 images used in land use change detection increased. More

images, which represent longer monitoring period, can obtain better results of

land use change detection. For short-interval land use changes detection, four

time periods is the maximum otherwise the period of monitoring will be too long.

Agricultural activities such as planting and harvesting have significant effects

on the monitoring of land use changes. In planting and harvesting months, the

accuracies of the land use change detection are lower than other months because

its land cover is often confused with other land uses, such as water and bare soils.

The process of construction can be considered as a three-stage process and a

combination of two land uses. However, construction sites are often confused

with vegetation and bare soil in RADARSAT-1 images because the values of

backscatter coefficients of construction sites and the two land uses are very similar.

The land cover changes during the planting and harvest seasons are often

confused with the process of construction. It is found that construction sites can

be identified with their two stages of low values of backscatter coefficients, which

is not found in the pattern curves of backscatter coefficients of other land uses.

By the comparison of the accuracies of identifying construction sites using two,

three and four RADARSAT-1 images, it is found that using three time periods can

get better accuracies which is different from the result of general land use change

detection.

This thesis does not try to evaluate land use change detection methods or find

the best method for monitoring land use changes. Instead, it focused on the

analysis of confusions caused by the time periods of land use change detection

and seasonal variation of vegetations. The main contributions of this study are

as follows: 1) it explores the use of multi-temporal RADARSAT-1 images into the

land use change detection to overcome the problems of cloudy conditions, making

short-interval land use change detection possible for areas which are often

covered by clouds; 2) pixel-based maximum likelihood method and the

object-based classification method were compared for their accuracies in land use

classification of RADARSAT-1 images; 3) it examines the optimal time periods

for land use change detection; and 4) it examines the appropriate number of

images that are needed for monitoring land use changes in different seasons in

order to obtain the best accuracies.