Background

Streams and rivers are among the most threatened ecosystems on the planet, with recent assessments showing extensive degradation around the globe (Feio et al. 2022). Human activities such as agriculture are among the leading threats to stream health (Marrochi et al. 2021; Pearce et al. 2021; Arocena et al. 2022; Champagne et al. 2022) because they can have pervasive impacts on hydrology, water chemistry, habitat, and biological structure and function. Thus, clear evidence of how human activity affects freshwater ecosystems is needed for the development of management strategies that can protect and rehabilitate freshwaters. Hynes (1975) famously argued that “the valley rules the stream”, and thus implied that understanding the processes, structure, and function of a stream required knowledge of the natural (e.g. geological) and human (e.g. agricultural) environment of the stream’s catchment area. This stimulated research in catchment land cover as a predictor of freshwater ecosystem conditions (e.g., Lenat and Crawford 1994), but it wasn’t until the mid-1990’s that advances in image processing and geographic information system (GIS) technology (Johnson and Gage 1997) stimulated large scale studies evaluating the effects of human land use on freshwater ecosystems. Such studies typically classified a catchment based on the dominant land cover (e.g. forested, urban, agricultural) or quantified the area or proportion of different types of land cover, including human land use, in the catchment area to at least partially explain variation in stream condition (e.g. Carlisle and Hawkins 2008).

Proportions of a catchment with different types of land cover (e.g. forest versus agriculture versus urban) can help explain stream conditions because as human land use increases, environmental stressors on the land such as sediments, nutrients, and contaminants also increase. As Hynes (1975) noted, these stressors are then transported downslope to the stream, altering physical, chemical and biological conditions. Thus, the amount and type of human activity within the catchment area has a strong, functional linkage to the stream ecosystem and is therefore a useful measure of the effect of this land use on stream conditions. But such simple measures of human activity often leave large amounts of variation in freshwater condition unexplained. For example, the classic study of land use influences on stream nutrient concentrations in the eastern USA by Omernik (1976) showed a wide range of total phosphorus concentrations (approximately 0.018–0.435 mg/L) in developed streams (> 80% human land use), but a very small range (approximately 0.006–0.030 mg/L) in undeveloped (< 10% human land use) streams. Results like these have underscored the need to quantify the position and pattern of land use within the catchment area to better explain human impacts on freshwaters (Fig. 1; Gergel et al. 2002; Gergel 2005).

Fig. 1
figure 1

Four hypothetical catchment areas (a, b, c, and d) with equal amounts of developed (28%, yellow) and natural (72%, green) land cover, but differing positioning of landscape patches in the catchment illustrating how land use proportions may not fully capture effects of human activity. Catchment scenarios represent situations where development is concentrated in: 1) upland (a) versus riparian (b) areas of the catchment, and; 2) upstream (c) versus downstream (d) areas of the catchment. Landscape configuration metrics (LCMs) generate identical values for all catchments (NP = 5; LPI = 7.5%; Ā = 6; PD = 0.05; TE = 60; ED = 0.56, see Table 1). Inverse distance weighted (IDW) and hydrologically active inverse distance weighted (HA-IDW) metrics generate values that vary with position of developed patches

Full size image

Landscape Configuration Metrics (LCMs) describe the spatial arrangement of patches of a given land cover type on a landscape (McGarigal 2001). There are many LCMs, but the most commonly used describe aspects of landscape connectivity and heterogeneity (Table 1). The development of the Fragstats package (McGarigal and Marks 1995) coupled with readily available, high quality land cover data have made LCMs increasingly easy to calculate for large geographic areas (e.g., Hill et al. 2016). Terrestrial ecologists have used LCMs to explain how habitat fragmentation and connectivity in the landscape matrix influence dispersal, habitat selection, predator-prey interactions, and many other aspects of terrestrial ecology (Pereboom et al. 2008; Moqanaki and Cushman 2017; Dominik et al. 2018; Haan et al. 2020). LCMs have also been applied in stream assessment studies to explain the effects of land cover configuration on stream conditions (e.g., Clément et al. 2017; Li et al. 2018; Xu et al. 2019; Shehab et al. 2021; Wu and Lu 2021), where associations between LCMs and several measures of water quality have been detected. However, there are three critical factors that we feel limit the utility of LCMs in stream assessments:

  1. 1.

    Poor Description of Catchment Land Cover: Clément et al. (2017) noted that many LCMs have the same value in catchment areas that differ substantially in land cover. Parabolic relationships between LCMs and the proportion of human land use in a catchment result in maximum values in landscapes with intermediate amounts of human land use (Neel et al. 2004). This will challenge our ability to detect monotonic relationships between stream health measures and LCMs for human land use.

  2. 2.

    Inconsistency: Observed relationships between measures of stream health and LCMs have varied in strength and even sign from study to study (e.g., Uuemaa et al. 2007; Lee et al. 2009; Xu et al. 2019).

  3. 3.

    Missing link between valley and stream: LCMs do not capture the positioning of patches relative to the stream network (Fig. 1), and thus do not reflect the hydrological processes that functionally link a stream to its catchment. Therefore, as a predictor and potential explainer of stream conditions, LCMs lack ecological relevance and are likely to produce spurious explanations of relationships between stream condition and human activity in its catchment area.

Based on these factors, we do not recommend the use of LCMs to measure the effect of human activity on stream ecosystem health.

Table 1 Landscape configuration metrics commonly used to predict impacts of land use on stream water quality and biota (McGarigal and Marks 1995)
Full size table

Inverse-distance-weighted metrics (IDWs) are based on the premise that the proximity of a land use patch to the stream channel increases its effect on stream conditions. The importance of proximity is supported by empirical evidence regarding the disproportionate influence that riparian areas have on stream biogeochemistry, water temperature, and organic matter inputs, among other characteristics (e.g., Cooke et al. 2022; Gregory et al. 1991; Vidon et al. 2010). Early attempts to add proximity of land use to the description of the stream ecosystem quantified land cover categories within zones of varying proximity to the channel (e.g. Strayer et al. 2003; Yates and Bailey 2010). However, the zone widths are arbitrarily defined and lump patches that may differ substantially in their proximity to, and effect on, the stream.

Better GIS tools facilitated the calculation of inverse-distance-weighted metrics (IDWs). IDWs are based on the distance of individual land cover patches to the stream channel or sampling point (Fig. 2a; King et al. 2005; Van Sickle and Burch Johnson 2008), and calculated by applying distance weighting functions (e.g., power, inverse, exponential; Weller et al. 2023), such that greater scores represent increased proximity. IDWs are a substantial improvement over simple proportions of different types of land cover and the arbitrary definition of distance-based zones (Van Sickle and Burch Johnson 2008). Therefore, we recommend the use of IDWs in measuring the effect of human activity on stream health when within-catchment flow direction and accumulation information is unavailable, because IDW metrics do not incorporate the substantial variability in flow direction and flow accumulation along the hydrologic pathways that functionally link streams to catchments.

Fig. 2
figure 2

Representations of distance assignments for catchment (dashed line) locations using the inverse distance weighted (IDW; a) and hydrologically active inverse distance weighted (HA-IDW; b) approaches. For IDW approach, distance is based on cell-based distance (indicated by number in each cell) from the outflow of the catchment. HA-IDW approach determines distance from catchment outflow to locations in the catchment by following topographically defined, hydrologic flow paths along stream channel (blue line) and into upland areas (black to grey lines)

Full size image

Streams are connected to the landscape through a network of surface and subsurface flow paths that extend from the channel to the uppermost parts of the catchment (Fig. 2b). These flow paths contribute varying amounts of water to the channel at rates dependent upon the topography, soils, and vegetation present along the flow path. The chemical and physical components of the water delivered to the stream from these flow paths is a function of character of the lands through which the water flows, so it follows that stream water will reflect the culmination of all its flow paths. However, as flow paths and locations along individual flow paths will differ in the amount of water contributed to the channel, the relative influence of the flow paths on stream conditions will vary with catchment position. Therefore, using land cover descriptors that incorporate the functional connection between a stream and its catchment should be ideal for testing the effect of human activity on stream health.

Peterson et al. (2011) refined IDWs with the development of the hydrologically active inverse-distance-weighted (HA-IDW) metric. The HA-IDW metric scores each land cover patch according to its potential contribution (upslope area contributing to the patch) and the hydrologic distance from the stream (Fig. 2b). Patches with greater contribution and hydrological connection (i.e. smaller hydrological distance) will have a greater effect on stream conditions (Fig. 1). In this way the HA-IDW metric is a significant advancement over the IDW metric because it explicitly links the stream to catchment hydrological processes.

Calculation of the HA-IDW is more computationally complex than IDWs, but there are open-source scripts available that generate HA-IDW values for a vast number of catchments across large landscapes (Peterson and Pearse 2017; Staponites et al. 2019). The data requirements for HA-IDW are also more demanding than for IDWs, since HA-IDWs require a high quality Digital Elevation Model (DEM) to generate an accurate hydrological network within the catchment. Indeed, the sensitivity of the HA-IDW to errors in topographic estimations of flow will likely result in lower explanatory power in studies of basins with low relief and/or with significant subsurface flow. However, through technologies such as LIDAR and remote sensing endeavours, the resolution and availability of topographic information is rapidly improving. When such high quality data are available, we strongly recommend the use of HA-IDWs to measure the effect of human activity on stream ecosystem health.

Conclusions

Accurate, ecologically relevant descriptions of the catchment area are essential for understanding how land use impacts stream ecosystems. Quantifying simple proportions of the catchment area with different types of land cover is helpful but insufficient. Landscape Configuration Metrics (LCMs) that just describe the number and relative position of different types of land cover are not much better, and may actually be misleading. Indeed, LCMs lack stream ecological relevance because they don’t capture the functional linkage between a stream and its valley. Zonal and inverse-distance-weighted (IDW) metrics are better than LCMs because they explicitly incorporate the distance between the land cover patch and the stream, although they don’t consider flow paths. The best alternative that explicitly links the valley with the stream as envisioned by Hynes (1975) is the hydrologically active inverse-distance-weighted (HA-IDW) metric because when the required high quality data are available it incorporates the flow path from a patch of land, whether forest or agriculture or urban, to the stream.