Mushabab Alshehri
Department of Civil and Architectural Technology, Technical College of Taif, Taif, Saudi Arabia
Abstract:
Different methods in remote sensing for determining the effect of spatial resolution on marine and coastal areas using different spatial resolution images are presented. The first method is based on principle component analysis. The second method is based on a map from Victoria park website. Also, the reflectance of light using different Ratios and different Scale Factors is used. Unsupervised classification used to classify all images. The output of unsupervised classification is compared with the output images that were produced by using principle component analysis or by using Victoria Park’s map. This study will discuss and analyze the effect of spatial resolution on marine and coastal areas. Furthermore, it will be an attempt to determine the optimal resolution for remote sensing images and for marine and coastal areas.
1. Introduction
During the last three decades the spatial resolution of the remotely sensed images has been considerably developed as a result of the improvement of technology. In the 1970s the spatial resolution of the Landsat satellites was 80 m [1]. Now less than 1 m high spatial resolution has come to be available providing detailed information of the ground. For example, 60 cm spatial resolution could be available by the QuickBird satellite. It can acquire high-quality satellite imagery for map creation and changes detection and image analysis. Many fields have benefited of the availability of the high spatial resolution images such as geology, environment, urban planning, agriculture and others [1].
Mumby et al. have compared the efficiency of different remote sensing platforms in different spatial resolution. They found that the high resolution digital airborne instrument (Compact Airborne Spectrographic Imager) CASI was significantly more accurate than other sensors that were less in spatial resolution [2]. Moreover, according to Legleiter et al. the benefit of data that acquired by remote sensing sensors for distinguishing between under water (spatially rivers) environments depended strongly on the sensors spatial resolution and the channel shape [3]. In addition, in some application according to Frank et al. the remotely sensed images with a moderate spatial resolution (20 – 30 m) is insufficient, for example in the vegetation cover especially in the situation of shrubs collection [4]. Frank et al. in their study compared between the measurement of the vegetation by using different high spatial resolution 0.6, 1 and 2 m. They found that the spatial resolution has a considerable impact on the measurement of the area of some kind of vegetation such as shrubs[4]. The influence of the spatial resolution was more noticeable for the smaller shrubs[4]. Also their results demonstrated that less than 1 m spatial resolution was necessary to calculate accurately the area of shrubs from the remote sensed images [4].
On the other hand, despite the fact that the high spatial resolution images are more detailed information, the classification results were discouraging as expected [1]. The classification accuracy was not necessarily improved by the increasing of the spatial resolution. The increase in spatial resolution caused the decrease of the class spectral separability. The consequence of this reduction in the class spectral separability was the rise in the classification errors [1].
In the field of studying the impacts of land use and climate change in arid and semi arid regions, Sprintsin and his colleagues found that the difference between high and moderate spatial resolution was almost non-existent [5]. They also discussed that they had analyzed the histograms of stand leaf area index (LAI) of a forest located the desert edge by using the two different spatial resolution [5]. Their results indicated that the overall distribution of (LAI) did not change considerably in the both high and moderate spatial resolution [5].
The big question which needs to be answered, is there any affect of variation of spatial resolution on the marine and coastal areas? If yes, to what extent the final results will be affected. Although many studies have discussed widely the effect of spatial resolution of remote sensing based studies on different features, up until now the marine and coastal areas require more attention studying and researching. This study will discuss and analyze the effect of spatial resolution of remote sensing on marine and coastal areas. Furthermore, it will be an attempt to determine the optimal resolution for remote sensing images also for marine and coastal areas.
2- Study area and data
The study area, Swan Island and a small part of Swan bay, represents many kind of feature such as land, water and sea grass. Swan Island is a small island, 1.4 km² in size, separates Swan Bay from Port Philip Bay. Port Phillip, also called Port Phillip Bay, is a large bay in southern Victoria, Australia, Figure 1. Geographically, Port Phillip is a large marine bay 1,930 km² in area which has a coastline length of 264 km. The bay is extremely shallow for its size, but mostly navigable. The deepest part is only 24 m, and half the region is shallower than 8 m. Habitat types found within the park include sea grass beds, sheltered intertidal mudflats, intertidal sandy beaches and rocky shores, sub tidal soft substrata and rocky reefs, as well as the open water environment. Port Phillip contains many bays and beaches. One of them is Swan Bay. It is shallow, 30 km² marine area. Swan Bay considered environmentally one of the most diverse ecosystems on the coast of Victoria. It includes marsh plants and sea grass. It is also home to a variety of plants, no invertebrates, fish and birds. Sea grass is source of life for almost everything in the Gulf. Fish and birds all use this sea grass as a shelter and food supplies.
As shown in Figure 2, four images representing different spatial resolutions were used to examine and analyze the effect of Spatial Resolution of remote sensing on marine and coastal areas. Data of the study include different spatial resolution aerial photographs cover all the area of the Swan Island and a part of Swan Bay.
Figure 1. The study area – Melbourne, Australia.
Figure 2. Four images of the study area with different spatial resolutions: a) Bottom Image with medium spatial resolution; b) high resolution image; c) Screen image with low spatial resolution and d) Top-Screen image with medium spatial resolution.
3- Images enhancement and classification technique
There are many technique could be used to enhance the image. In this study, operations such as rationing, stretching and principle component analysis were used to achieve the best results. Firstly, the rationing or the image division is an enhancement process and one of the most important methods that were used to transform image data [6]. The general concept of rationing is that the procedure of dividing the data from two different bands. One benefit of rationing is that variations in the slopes of the spectral reflectance will be masked. Another benefit of rationing is that the variations in brightness will reduce and the image contrast will increase.
In this study a series of operations using image ratio at optimum scale factor was implemented. Ratio used to highlight a particular feature class in the images. Also Ratios used to enhance images. The following are common ratios that were used:
- Dividing Green Band / Red Band and the Scale Factor SF was 50 to identify Sea grass vegetation.
- Dividing Green Band / Blue Band and the Scale Factor SF was 30 to identify intertidal vegetation.
- Dividing Red Band /Blue Band and the Scale Factor SF was 10 to identify Water depth.
The histograms were used to identify which scale factor provides a better contrast. Secondly, principle component analysis is one of most useful analysis in the remote sensing. The result of using the principle component analysis is a band or a few bands. These contain a big percentage of the information in the original bands [6]. The aim of the principle component analysis is that all information of the image could be obtained in the less number of bands [6]. Finally, the classification also is a method that was used in the research. The objective of image classification is to group together pixels that have similar patterns of brightness values across a series of image bands or information channels [6]. There are two general methods: unsupervised and supervised classification. In unsupervised classification, a statistical technique called k-means cluster analysis is used [6]. In this procedure, the analyst specifies the number of classes required. Pixels are initially distributed to classes at random. Once all pixels have been distributed to a class, group means are calculated for each class. Each pixel is compared to each class and is renamed to the class with which it has the highest similarity [6]. Once all pixels have been renamed, class means are recalculated and the process iterates until no further changes in class membership occur [6]. The output is a new image in which each pixel is represented by its class identifier [6]. Unsupervised classification is useful for exploring what cover types can be discovered using the available imagery [6]. However, the analyst has no control over the nature of the classes [6]. The final classes will be similar to some extent but may not be consistent to any useful land cover classes [6].
4- Methodology
The methodology for determining the effect of spatial resolution of remote sensing on marine and coastal areas requires images in different spatial resolutions. The four different spatial resolution images contain the study areas, marine and coastal areas, were processed through a series of digital operations using TNT program.
The following steps were involved.
- Determining the spatial resolution for every image.
- The undesirable areas were masked.
- Using the principle component analysis for initial classification.
- The images were enhanced by using different methods such as Rationing and Stretching.
- Identifying the cover types using unsupervised classification.
A detailed description of each operation is given in the following sections.
4.1. Determining the Spatial resolutions
The spatial resolution of the sensors plays a large role in the clarity of the imagery’s details. The spatial resolution depends on the distance between the sensors and the target being imaged. Sensors with a little distance from their targets view great details and higher spatial resolution. However, there are many studies do not require high spatial resolution such as the weather applications. Most of the remote sensing sensors use digital system to store the images data. The images are formed of a matrix of pixels. Each pixel represents a cretin area of the image and the reflectance of the target on the ground. It is usual that the spatial resolution of the sensor and the pixel size are the same. In this study the first step was determining the spatial resolution for the four images. It is important that the spatial resolution of the image is must be known before the beginning in the analyzing especially if there are many images with different spatial resolution. The method that used to calculate the spatial resolution was dividing the real distance between two points on the earth (from Google earth) by the number of pixels between the same points in the digital image. The following tables show the method that used to calculate the spatial resolution for the images.
Table 1. Measurement of three different distances for every image. D1, D2 and D3 were different distances on the land using Google Earth. d1, d2 and d3 were the number of pixels equivalent to the real distances on the land.
Table 2. The spatial resolution for every image after dividing the distances on the land by the numbers of pixels.
4.2 Masking
One of the main objectives in this study was determining the effect of spatial resolution on the marine and coastal areas. In this case it is more effectively to work with feature separately. So, using the masking tool was used to abstract the areas of interest. The masking tool requires manual procedure by using the computer mouse tracing around the part of the image. To find the boundary of water the Principle Components Analysis was used to ensure more accurate results. Figure 3 (1a-1d) shows masking the land out using Principle Components Analysis. Marine and coastal areas remain in all images the regions of interest and will be used in the next analysis steps.
4.3 Initial classification using Principle Components Analysis (PCA)
As shown in Figure 3 many colours were obtained using Principle Component Analysis such as green, red, yellow and blue. Initially, this illustrates that there are different features. The principle components images can assist clearly to distinguish between the different features. However it is difficult in this stage to make final interpretation of features kind in the images. So, it can be useful to benefit from the other experiences. For example there are maps in Victoria Park website illustrating all features in the study area. As shown in Figure 4 there are two kind of water law water and another which is deeper. These two kinds of water will be classified as Deep water and Shallow water. Also the map indicates that there are two types of vegetation. The first is Zostera / Heterozostera Dominant Seagrass and Macroalgae. The second is Zostera / Heterozostera Dominant Seagrass. They will be classified as Seagrass 1 and Seagrass 2. There are still small areas of Sediment and Macroalgae and they will be classified by their names.
Figure 3. The Principle Components images and the land masked out.
4.4 Image enhancement and false color Classification
According to the previous step it can be seen that there are three kinds of features; water, vegetation and sediment. So, for every image a series of operations was implemented to obtain initial information of all features in the images. The first operation was dividing Red Band / Blue Band and the Scale Factor was 10. Also the color palette was changed to rainbow .The reason to use this method was to identify the water depth. The second operation was dividing Green Band / Red Band and the Scale Factor was 50, Figure 4. Also the color palette was changed to rainbow .The reason to use this method was to identify the underwater vegetations. An attempt was made to identify sediment by using the Green Band / Blue Band and the Scale Factor was 30. Also, the color was changed to be Rainbow. The benefit of using this method is that this kind of procedures which depend on the reflectance of light can help of distinguishing between features in the images.
Figure 4. Map of the study area shows the variations in features and water level.
Figure 5. Dividing Green Band / Red Band and the Scale Factor = 50.
4.5 Unsupervised Classification
As shown in the Figure 6, all images were possessed by using unsupervised classification procedure. As a result of the number of classes in the study area which was 6, the number of classes was adjusted to 12 in the program. The method that used was Simple One – Pass Clustering.
Figure 6. Output of the images after unsupervised classification.
4.6 Comparisons between the four methods
In this section the four methods were compared. The principle component images, the false color images, Victoria Park map and unsupervised classification output. Every image was compared with the similar images that have the same spatial resolution. Figure 7 shows groups of images. Every group presents a different spatial resolution. The groups of images are Bottom – screen image with 6 m spatial resolution, high resolution image with 2 m spatial resolution, screen image with 8 m spatial resolution and Top – screen image with 6 m spatial resolution.
Table 3 shows the number of cells and their percentage in all images. It can be seen that the image with the lowest spatial resolution was the lowest in the number of cells. Also the table illustrates that the percentage of cluster no.2, which in green color, is more than 50 % . According to the available information of this area and by a visual comparison of the Principle Component image, this area is mixed of tow kind of sea grass. However in the Bottom – Screen image the two areas were classified as one area. This was a brief remark of cells percentages of the images which were produced by unsupervised classification. These percentages are important when applying supervised classification.
Table 3. The number of cells and their percentages in all output of unsupervised classification.
Figure 7. A comparisons between the four methods ; the principle component images, the false color images, Victoria Park map and unsupervised classification output.
5- Results
5-1 The effect of spatial resolution on the classification accuracy
In this study, different spatial resolution images have been compared. By comparing the images in Figures 7, yes is the simple answer of the question that is there any effect of variation of the spatial resolution on the classification accuracy. However, to what extent the spatial resolution affects the results. In the high spatial resolution image, it can be seen that there is clarity to identify different features and distinguishing between them easily. It can be clearly identified the intertidal in the med of the image. Also the area of seagrass in the upper right corner of the image can be identified. Finally the area of deep water can be seen in green color in the image. In the Top – screen image which has a moderate spatial resolution, it can be seen clearly the deep water area and some locations of seagrass. In the lowest spatial resolution image the seagrass is difficult to be identified.
The final result in this study indicates that the high spatial resolution image displays more features than low/coarse resolution images. Also features can be differentiated clearly using high spatial resolution image.
5-2 The optimal resolution for marine and coastal study
It is difficult to decide whether the high, moderate or low spatial resolution is the optimal for marine and coastal area for many reasons. First, high spatial resolution images are more detailed than the low spatial resolution images. Second, some features can be identified clearly in high special resolution whereas some features can be identified in low or moderate spatial resolution much clearer. Finally, the coverage using high spatial resolution is smaller than the coverage that introduced by using low spatial resolution images. So, in my opinion the optimal resolution for marine and coastal studies depends on the objectives of these studies.
6- Conclusion
In this study, different methods in remote sensing for determining the effect of spatial resolution on marine and coast areas using different spatial resolution images were presented. The first method is based on principle component analysis. The principle component images are useful in the initial classification. Variation in colours guided to know that there is variations in features. The second method is based on a map from Victoria park website. The map is very useful to show information of the study area such as level of water and an overview of existing features and creatures. The last method is based on reflectance of light using different Ratios and different Scale Factors.
Unsupervised classification used to classify all images. The output of unsupervised classification was compared with the output images that were produced by using principle component analysis or by using Victoria Park’s map.
The result indicates that high spatial resolution image is more detailed than the lower spatial resolution. Low spatial resolution covers more areas. The need is still for more research to identify accurately the effect of spatial resolution on the classification accuracy and the optimal resolution for marine and coastal areas.
Acknowledgment
The author would like to thank Dr. J. Leach the professor of remote sensing in the Department of Geometics for his support and preparation of this study. Also, the author acknowledges that this study was prepared in most of its stages using data, softwares and labs of the University of Melbourne.
References
- Hsieh, P.; Lee, L. and Chen, N. Effect of Spatial Resolution on Classification Errors of Pure and Mixed Pixels in Remote Sensing. IEEE Transactions on Geoscience and Remote Sensing 2001, 39, 12.
- Mumby, P.; Green, E.; Edwards, A. and Clark, C. Measurement of seagrass standing crop using satellite and digital airborne remote sensing. Marine Ecology Progress Series 1997, 159, 51-60.
- Legleiter, C. and Roberts, D. Effect of channel morphology and sensor spatial resolution on image-derived depth estimates. Remote Sensing of Environment 2005, 95, 231-247.
- Frank, T. and Tweddale, S. The effect of spatial resolution on measurement of vegetation cover in three Mojave Desert shrub communities. Journal of Arid Environments 2006, 67, 88-99.
- Sprintsin, M.; Karnieli, A.; Berliner, P.; Rotenberg, E.; Yakir, D. and Cohen, S. The effect of spatial resolution on the accuracy of leaf area index estimation for a forest planted in the desert transition zone. Remote Sensing of Environment 2007, 109, 416-428.
- Nicholas, M. Remote Sensing Tutorial 2007 ,Available online:
