Variability is inherent in the earth's environment at every scale. The traditional views of the scientific method use experimental design to control variability in the presence of responses of interest and explanatory variables, where the three principles of blocking, replication, and randomization are applied. Although this research paradigm applies to some narrowly defined problems in the environmental sciences, the processes of major interest typically exhibit strong spatial, temporal, and exogenous variability. Here statistical challenges abound.

Often environmental processes exhibit important spatial-temporal variability that needs to be characterized in conjunction with purported dependencies on explanatory variables. That is, the "why" question might not be answered sensibly without first answering the "where" and "when" questions. Space and time provide natural ways to stratify (often hierarchically) large problems into more manageable ones.

Hierarchical (Bayesian) statistical modeling offers a very powerful way of representing complex global phenomena through simple local structure. Fitting these models often requires powerful computers. Methods of sample re-use (such as sub-sampling), and Markov chain Monte Carlo (MCMC) for implementing Bayesian methodology, are computationally intensive and demand real innovation in the environmental context. The relevance of environmental statistics to the environmental sciences is in direct proportion to the development of computer-intensive statistical methods on fast workstations and supercomputers. One of the challenges that lies ahead is to develop and adapt statistical methodology for the massive datasets that studies of the environment will engender. The most obvious source of massive environmental data will be from remote-sensing platforms that continue their march toward pinpoint spatial accuracy at high temporal frequency.