Sample independence and correlation

I am examing effects of nutrients on insect communities in lakes. I have 65 lakes but many of these have insect samples from more than one season and/or more than one habitat. If I include all samples for all lakes when looking at responses of insect communities to nutrients, there is an issue of pseudoreplication as samples from seasons/habitats within a lake will be related. I can take the average insect community composition from these samples giving a value for each lake and use standard linear regression or correlation, however this loses much of the collected data.
I am desperately trying to find a way around this. Are there any methods where the effect of season and habitat can be controlled for when looking for a relationship between nutrients and insect communities?