The Impact of Landscape Characteristics on Groundwater Dissolved Organic Nitrogen: Insights From Machine Learning Methods and Sensitivity Analysis

Abstract

The effect of groundwater nutrient inputs on river and estuary water quality and the potential impacts of urbanization on groundwater are central concerns in many coastal areas. It has been previously identified that dissolved organic nitrogen (DON) can be the dominant form of total dissolved nitrogen (TDN) in some aquifers. However, there is a paucity of evidence about the sources and flow paths of DON, relative to inorganic nitrogen in groundwater. DON and dissolved organic carbon/DON were first compared against different landscape variables in this study, and no significant relationships were found. However, the relationships became statistically significant when shallow samples (sampling depth < 10 m) were separated from deep samples. A random forest model and sensitivity analysis were then applied to further our understanding of the ecohydrological drivers and seasonal patterns that shape DON variability. The random forest algorithm was built to classify 171 groundwater wellbores into three classes (low: <0.5 mg/L; medium: 0.5–2.5 mg/L; and high: >2.5 mg/L) which achieved 72% classification accuracy using landscape characteristics, hydrological conditions, and temporal information. The results indicated that the effects of landscapes on sandy shallow groundwater DON were controlled both by certain landscape characteristics and depth to groundwater. A conceptual model of groundwater DON is therefore proposed where the balance of exposure and processing time scales from the surface to groundwater is the critical control on the preservation of landscape signatures; we expect that this conceptual model would be applicable for other sandy, shallow groundwater areas.

A Brief Review of Random Forests for Water Scientists and Practitioners and Their Recent History in Water Resources

“…Classification and regression trees (CARTs, [56]) are methods to partition the variable space based on a set of rules embedded in a decision tree (see Figure 1 below), where each node splits according to a decision rule; see e.g., Hastie et al [58] (pp. [305][306][307][308][309][310][311][312][313][314][315][316][317], and the review in Loh [19]. In this way, the variable space is partitioned into a set of rectangles, and a model is fitted to each set, which in the simplest case can be a constant.…”

Woodland Establishment Reduces Nutrient Losses to Waterbodies in Urban Catchments: A Review of the Evidence

“…Nidzgorski and Hobbie (2016) found similar benefits for soil phosphate and studied implications for groundwater quality using hydrological modeling. Nitrogen impacts were inconsistent, seemingly differing between years due to climate variability (Nidzgorski & Hobbie, 2016), although Wang et al (2018) identified some evidence of urban forest benefits for groundwater dissolved organic nitrogen. Whilst Decina et al (2018) studied nutrients in throughfall and soil solution for individual trees, they did not report soil concentrations under contrasting landcover, so data were not included in meta-analysis.…”

ML-SWAN-v1: a hybrid machine learning framework for the concentration prediction and discovery of transport pathways of surface water nutrients

“…The probability with which data points are selected for the next training set is not constant and equal for all data points. The selection probability increases for data points that have been misestimated in the previous iteration; data points that are difficult to classify would receive higher selection probabilities than easily classified data points (Yang et al, 2010;Erdal and Karakurt, 2013). For RF and GBM, the most commonly used base learner is a classification and regression tree (CART).…”

“…Meanwhile, data-driven machine learning (ML) methods are increasingly being applied to quantify relationships between soil, water and environmental landscape attributes (Lintern et al, 2018;Wang et al, 2018;Guo et al, 2019). For instance, random forest (RF), a widely used ML method, was used to model the spatial and seasonal variability of nitrate concentrations in streams (Álvarez-Cabria et al, 2016).…”