A hydrochemically guided landscape classification system for modelling spatial variation in multiple water quality indices: Process-attribute mapping

Abstract

Spatial variation in landscape attributes can account for much of the variability in water quality relative to land use on its own. Such variation results from the coupling between the dominant processes governing water quality, namely hydrological, redox, and weathering and gradients in key landscape attributes, such as topography, geology, and soil drainage. Despite the importance of ‘process-attribute’ gradients (PAG), few water quality models explicitly account for their influence. Here a processes-based water quality modelling framework is presented that more completely accounts for the role of landscape variability over water quality – Process-Attribute Mapping (PoAM). Critically, hydrochemical measures form the basis for the identification and mapping of effective landscape attributes, producing PAG maps that attempt to replicate the natural landscape gradients governing each dominant process. Application to the province of Southland (31,824 km²), New Zealand, utilised 12 existing geospatial datasets and a total of 28,626 surface water, groundwater, spring, soil water, and precipitation analyses to guide the identification and mapping of 11 individual PAG. The ability of PAGs to replicate regional hydrological, redox, and weathering gradients was assessed on the accuracy with which the hydrochemical indicators of each dominant process (e.g. hydrological tracers, redox indicators) were estimated across 93 long-term surface water monitoring sites (cross-validated R² values of 0.75–0.95). Given hydrochemical evidence that PAGs replicate actual landscape gradients governing the dominant processes, they were combined with a land use intensity layer and used to estimate steady-state surface water quality. Cross-validated R² values ranged between 0.81 and 0.92 for median total nitrogen, total oxidised nitrogen, total phosphorus and dissolved reactive phosphorus. Models of particulate species E. coli and total suspended sediment, although reasonable (R² 0.72–0.73), were less accurate, suggesting finer-grained land use, landscape attribute, and/or flow normalised measures are required to improve estimation.