As intelligent as scientists are, they can't do everything in their head: software is a necessity of research and data analysis. There's a movement going on with this software towards open source software, no matter what the discipline. Here's an overview.
Unconstrained Data Analysis
One of the biggest problems that scientists have with using proprietary software is with data analysis. It's important to understand how precisely your software is analyzing the data, but with closed source proprietary software, it's impossible to look into the nitty-gritty of the code and make it do exactly as you want it to do. With open source, the code can be tweaked and tailored to exactly what your data analysis methods require.
Supercomputers & Software
Many fields of science require the use of supercomputers to analyze large quantities of data and to complete models. However, proprietary operating systems tend to run too slowly for supercomputers, not nearly as efficiently as their open source counterparts. Also, most parallel processing languages, a necessity for any supercomputer, are all open source. Almost 90% of the world's top 500 supercomputers run some sort of Linux distro, and much of the remaining 10% is filled in by UNIX systems.
Data Mining
As more and more research data becomes publicly available online with the open science movement, some method of pouring through them efficiently to get exactly the data you need will be necessary, and your standard Google search might not be adequate. While large scale data mining software is not yet available, enough people are aware of the eventual need for it that it will probably be developed at some point or another.
Where's The Profit?
Most open source software for science uses the same basic business model for open source. There are usually a few different licenses available depending on the use made of the software, from commercial to educational to individual use. Typically, commercial licenses cost a small sum of money. Many open source software businesses also provide training and support for a subscription fee. However, considering how small of a user base many scientific software applications would have, this isn't always a viable business option.
However, there are also many lone developers who simply create whatever they need for their particular research, and then post it online for anyone to use and develop further. Many professors have the habit of hiring grad students, a source of relatively cheap, yet highly skilled labor.
What's Stopping Open Source Science Software?
One of the biggest bumps in the trend towards more open source is the lack of incentive for many scientists to post their developed code online. Many are simply not aware of the need for it in the greater community, or they don't want the competition getting their hands on anything that could make them lose their edge. Both are rather unhelpful attitudes.
Some funding organizations are beginning to require, as a condition of funding, that scientists post any code or software developed for this project as open source online. This is beginning to encourage those otherwise unwilling scientists to collaborate with their code better. A lot of scientists are more concerned with the idea of intellectual property than of helping the whole field develop. A desire to collaborate is the key to supplying the world's scientists with open source software that they can tailor to their own research needs.
For more information, articles, and links to some open source science software, check out the OpenScience Project.