There has been much activity in modelling the hydrochemical response of catchments recently (see Hooper, this issue), but our focus is the use of data on chemical composition of waters to model flow of water per se. Chemical data are used to improve models aimed at simulating flows of water along different paths through a catchment by defining ``end members'' (e.g., ``ground water,'' or ``soil water'') according to a unique chemical signature ( Christophersen et al. 1990) and assigning these compositions to the conceptual reservoirs in a watershed model. The composition of stream water can then be simulated through time by mixing the end members in the proportions defined by the purely hydrological portion of the model. (The approach is referred to as ``end-member-mixing analysis,'' or as EMMA for short.) Measured stream chemical data help to constrain the hydrological model ( de Grosbois et al. 1988; Hooper et al. 1988). Even when chemical data are included to constrain model behavior, problems in identifying parameter values through a calibration procedure remain ( Hooper et al. 1988; Christophersen and Neal 1990). Robson et al. (1992) took the approach of calibrating TOPMODEL using flow data alone and compared the flow separations implied by TOPMODEL results with separations based on an EMMA. One consistent interpretation of the TOPMODEL results was based on a hypothesis that saturated overland flow attains the chemical composition of soil water, presumably through a displacement of ``old'' water, water that is in the soils of the catchment prior to the onset of the storm.
Detailed descriptions of flow paths through catchments also can be derived from interpretation of data on the spatial and temporal variation in the concentrations of natural ``tracers'' such as chloride or stable isotopes. Deuterium concentrations in rain water, in water from suction lysimeters, and in water from trenches were used by Stewart and McDonnell (1991) to determine residence times in various parts of the soil matrix in a catchment in New Zealand. They showed that a dispersion model provided the most accurate estimate of mean residence times. Chapman et al. (1993) suggest that the popular mixing models mentioned in the previous paragraphs may not be universally acceptable; these authors found very significant changes in chemical composition of waters along a flow path in an upland catchment. Clearly much remains to be learned from information provided by a variety of tracers ( Peters et al. 1993).