In this article, we give an overview of applications of machine learning in the field of environmental sciences. Environmental sciences are possibly the largest grouping of sciences, drawing heavily on life sciences and earth sciences, both of which are relatively large groupings themselves. A typical representative of environmental sciences is ecology, which studies the relationships among members of living communities and between those communities and their abiotic environment. We focus on applications of machine learning in ecology, and in particular ecological modelling.

Ecological modelling is concerned with the development of models of the relationships among members of living communities and between those communities and their abiotic environment. These models can then be used to better understand the domain at hand or to predict the behavior of the studied communities and thus support decision making for environmental management. Typical modelling topics are population dynamics of several interacting species and habitat suitability for a given species (or higher taxonomic unit).

Population dynamics studies the behavior of a given community of living organisms (population) over time, usually taking into account abiotic factors and other living communities in the environment. For example, one might study the population of phytoplankton in a given lake and its relation to water temperature, concentrations of nutrients/pollutants (such as nitrogen and phosphorus) and the biomass of zooplankton (which feeds on phytoplankton). The modelling formalism most often used by ecological experts is the formalism of differential equations, which describe the change of state of a dynamic system over time.

To learn population dynamics models from measured behaviors of studied populations, e.g. concentrations of algae in a lagoon, we apply methods of computational scientific discovery, more specifically equation discovery. Our machine learning algorithms use a process-based modelling approach, which focusses on the ecological processes taking place in the studied ecosystem. They take into account both measured data and existing expert knowledge about the studied system and population dynamics modelling in general. The applications of our methods for modelling algal growth in the Lagoon of Venice and other ecosystems will be described.

Habitat-suitability modelling is closely related to population dynamics modelling. Typically, the effect of the abiotic characteristics of the habitat on the presence, abundance or diversity of a given taxonomic group of organisms is studied. For example, one might study the influence of soil characteristics, such as soil temperature, water content, and proportion of mineral soil on the abundance and species richness of springtails, the most abundant insects in soil.

To build habitat-suitability models, machine learning techniques can be applied to measured data. Techniques such as decision tree induction are most commonly used. We apply such techniques to build models of habitat suitability for several kinds of organisms. These include habitat models for springtails and other soil organisms in an agricultural setting, which explain the influence of agricultural activities on the abundance and diversity of soil fauna, as well as habitat suitability models for sea cucumbers in a sustainable fishing setting.