A recent development in corpus linguistics has been the integration of critical discourse methodologies, which allow in-depth contextual and qualitative analyses, with corpus linguistic methodologies, which allow broader quantitative analyses. Our study is a contribution to this approach. We present the methods used in a study of vocabulary pertaining to the environment, undertaken as part of the UK National Ecosystem Assessment. A clear and replicable methodology was developed and applied to three custom-built specialised web corpora and a reference web corpus; automatic analysis of collocations found using the Sketch Engine was complemented by manual analysis; and a small-scale replicability check was carried out to ensure that investigator divergence was minimal. We outline the approach and some of the key findings, and we also suggest areas for further refinement/investigation.