International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012 
XXII ISPRS Congress, 25 August - 01 September 2012, Melbourne, Australia 
418 
In both cases, if more than one layer is involved in the current 
SESQL sentence, we further estimate the number of result rows. 
If it is bigger than the total number of input rows, we consider 
the current SESQL query generating “too many join rows”. In 
Figure 3, the Input hrowser is the data size of the needed input data 
which can only be found on the browser side, while Input server is 
the data size of the rest input data. The rest input data can be 
found on the server side. 
With the above rules, decisions on where to execute a SESQL 
sentence can be made. After execution, the cache structure on 
the browser-side middleware will be updated accordingly. 
4. IMPLEMENTATION, CASE STUDIES AND 
DISCUSSIONS 
4.1 Implementation 
For the server-side middleware, spatial operators, SESQL 
compiler, GML2SVG, and decision-maker are implemented as 
Java Servlets. Specifically, JTS Topology Suite is employed to 
implement the spatial operators. As SESQL is based on the 
original SQL, we adapt the C++ codes provided in Levine et al. 
(1992), and implement our own SESQL compiler by combining 
the spatial operators. For the browser-side middleware, SESQL 
compiler, SVG2GML, decision-maker, and cache structure are 
implemented using JavaScript. We implement spatial operators 
with JTS Topology Suite, and develop them as Java Applet. 
Google Gears API provides an SQLite compiler with 
JavaScript-based API. We therefore implement a JavaScript- 
based SESQL compiler with the Gears API. For spatial 
operators embedded in the SESQL sentences, JavaScript 
invokes the developed Java applet to execute the corresponding 
spatial operations. JavaScript DOM API is also used to 
assess/update SVG documents on the browser side. In order to 
facilitate the interaction between the server-side and the 
browser-side middlewares, we use AJAX (Asynchronous 
JavaScript and XML) technology. 
It is important to note that the load balancing middlewares are 
completely transparent to the end users. Users can access spatial 
analysis functions simply with an SVG-enabled web browser, 
such as Internet Explorer, Firefox, and Google Chrome. 
4.2 Case Studies and Discussions 
We design three case studies to evaluate the proposed solution 
as proof of concept. These case studies use geospatial data of 
Guangdong Province (China), and implement several spatial 
analysis tasks. In the case studies, comparisons among the 
proposed solution, server-side solution, browser-side solution 
and our former solution (Huang et al., 2011b) are also made. 
4.2.1 Case Study 1: Suppose there is some toxic 
contamination throughout river “Rl”. The contamination affects 
the areas that are within 20 km. This task is to list all affected 
administrative districts and calculate their affected area size. 
In order to investigate this issue, we have to create a buffer for 
river “Rl”, and find out which administrative districts are 
overlapped with this buffer, and then calculate the size of 
overlapped area for each district. We carry out this task based 
on the workflow described in Section 2. First, we identify the 
Google has stopped its support for Gears API. However, 
similar functions can be found on HTML 5’s LocalStorage. 
needed data and the evaluation criteria by carefully analysing 
this case study. And then based on the suggested model in 
Section 2, we use GML to represent the needed spatial data 
(river layer) on the server side. We also represent the district 
boundary layer in SVG and deliver it to the browser side as the 
initial User Interface (UI). We then submit SESQL sentences on 
the browser side to carry out the spatial queries by the following 
steps: 1) Calculate a 20 km buffer of river “Rl” (using Buffer 
operator); 2) Find out all the districts which are overlapped by 
this buffer (using Overlap operator); 3) Find out all the affected 
areas in each district (using Intersection operator); 4) Calculate 
the size of affected area in each district (using Area operator). 
The SESQL sentences are listed in the Appendix. 
Figure 4 depicts the results. It lists the names of affected 
districts, and their affected sizes in the listbox at the right- 
bottom comer. These districts are also highlighted in the map. 
Figure 4. Listing all the administrative districts affected by river 
“Rl”, and calculating their affected area size 
To illustrate the advantages of the proposed solution, we 
compare it with different solutions: server-side solution, 
browse-side solution, and layer-based load balancing solution 
(i.e., our former solution in Huang et al. (201 lb)). We mainly 
compare the data amount of the network transmission 
(excluding the request/response sentences) between the server 
and the browser sides. Table 1 depicts the results. The 
difference between layer-based solution and the proposed 
solution is mainly due to their employed granularities (layers 
versus spatial objects). To sum up, the proposed solution has a 
smaller network transmission load between server and browser 
sides, and therefore leads to a better performance. 
Server- 
side 
solution 
Client- 
side 
solution 
Layer- 
based 
solution 
The 
proposed 
solution 
Step 1: 
Buffer 
1,925 
36,559 
36,559 (on 
browser) 
5,082 (on 
browser) 
Step2: 
Overlap 
5,267 
0 
0 (on 
browser) 
0 (on 
browser) 
Step3: 
Intersection 
3,095 
0 
0 (on 
browser) 
0 (on 
browser) 
Step4: Area 
106 
0 
0 (on 
browser) 
0 (on 
browser) 
Total 
10,393 
36,559 
36,559 
5,082 
Table 1. Comparisons of the first case study (data amount is 
measured by Byte)