A review of simulation experiment techniques used to analyze wild ﬁ re e ﬀ ects on water quality and supply †

This review addresses the critical knowledge gap of techniques simulating combustion and heating characteristics present in natural wild ﬁ res and their use in assessing post ﬁ re impacts on water quality and quantity. Our assessment includes both laboratory and plot-scale techniques with burn and rainfall simulation components. Studies included focus on advancing understanding of changes in chemical and physical properties of soil, as well as subsequent runo ﬀ changes. Advantages of simulation experiments include: overcoming logistical challenges of collecting in situ wild ﬁ re data, reducing the high spatial variability observed in natural settings ( i.e


Introduction
Forested watersheds are an important source of high quality water for billions of consumers worldwide, providing nearly two-thirds of potable water in the U.S. 1,2 Forests serve to capture and store precipitation, lowering the risk of ooding events. 1 The natural ltration processes that occur in watersheds provide major economic benets, saving an estimated 4.1 trillion USD per year globally in water treatment costs. 2,3 Wildres are a natural disturbance mechanism which support the long-term health of forested ecosystems, 4 but also can degrade stream water quality and alter runoff generation mechanisms. Wildre-driven increases in sediment, dissolved organic matter (DOM), nutrients, and heavy metals 1,2,5-7 can necessitate investments in infrastructure and altered methods of treatment in water treatment plants, 2,[8][9][10][11] diminish reservoir storage due to sediment lling, and disturb freshwater ecosystems. 1,12 Additionally, increased post-re runoff generation can produce high peak ows, increasing ood risks. 14 These effects ( Fig. 1) occur immediately aer a wildre, but impacts can persist for up to 10 years. 13 In recent decades, an increase in wildre size, frequency, and severity has been observed in certain forested regionsa trend predicted to continue. [15][16][17][18] For example, the mean annual burn area in the Western U.S. has doubled since 1984 and is projected to have a 24-169% increase in mean burn area by midcentury. [18][19][20][21] Current understanding of wildre effects on water quality and supply is incomplete, such that research is critically needed to assist water managers in adaptation and mitigation strategies. 1,7,10 Hindrances in the collection of post-wildre in situ dataunstable terrain and road closures immediately aer wildres, lack of comparable pre-burn control data, and high natural spatial and temporal variability-have contributed to a lack of knowledge of wildre effects. 2,10,11 Additionally, post-wildre response varies greatly regionally due to differences in soil and vegetation regimes, as well as climate. 22 Thus, insights from an in situ study are generally not directly transferable to another geographic area. Limited information exists about post-wildre water quality response, in particular, with most studies primarily focusing on DOM, nutrients, and metals-leaving the responses of microbial constituents and anthropogenic contaminants largely unknown. 23 Laboratory-and plot-scale simulation experiments which replicate burning and combustion mechanisms present in natural wildres, as well as raindrop kinetic energies and leaching effects similar to post-wildre precipitation, offer an alternative analytical technique for estimating wildre effects on water quality and supply. 2,7,[24][25][26][27][28] Hereaer, these two types of simulation experiments will be referred to as wildre and rainfall simulators, respectively. These studies primarily advance understanding of burn effects on small-scale soil and water physical and chemical properties. The controlled setting of these simulation experiments can overcome logistical issues associated with in situ studies, as well as provide insight into regionally-specic drivers by allowing for more precise Carli P. Brucker received her B.S. in Mechanical Engineering from the University of Iowa and her Master's degree from the University of Colorado Boulder where she is a current PhD candidate in Civil, Environmental, and Architectural Engineering Department. Her current research addresses the broad challenge of quantifying the impacts of wildres on hydrologic and water quality responses in watersheds in the U.S. West. These efforts include the design and creation of laboratory-scale wildre and rainfall simulation experiments, as well as modeling wildre impacts on water quality and supply using physical models and machine learning techniques. attribution of responses to controls. However, this type of analysis also presents new limitations, including dissimilarities to natural settings and difficulties in upscaling results to the catchment scale. 24,29 2 Scope of review The overarching scope of this review is to provide a critical assessment of existing laboratory-and plot-scale wildre and rainfall simulation techniques used to measure the effects of wildres on water quality and supply. Terminology is dened in Section 3, followed by a review of post-wildre hydrologic and water quality responses in Section 4, including their implications for human and natural systems. Section 5 presents a review of simulation experiment techniques for both wildre and rainfall processes, providing strengths and weaknesses for both laboratory-and plot-scale methods. Based on the merits of each method, recommendations for experimental design are provided in Section 6, followed by overall conclusions in Section 7.
A novel contribution of this review is the focus on methodological techniques-specically, simulation experimentsused to generate and collect wildre response data, rather than the data themselves. 13,30,31 We include brief analyses of common hydrologic and water quality response data from reviewed simulation experiment studies (a summary of the reviewed data is provided in the ESI †). However, this is not the focus of this review due to studies' wide ranges of research goals, temporal and physical scales, and key experimental factors-making useful cross-study comparisons difficult. Overall, this review is distinct from existing reviews, which primarily summarize the state of the art of wildre effects on hydrology, 31-37 sediment transport and erosion, 12,13,[38][39][40] and streamow concentrations of DOM and nutrients (i.e., nitrogen and phosphorous), 2,41-46 as well as metals. 1,30 One notable exception is Ferreira et al. (2008), which discusses limitations of common methods and techniques used to analyze hydrologic and erosional responses in post-wildre settings, from laboratory to catchment scales. However, while Ferreira et al. (2008) does cover rainfall simulation methods, the present manuscript is the rst to provide in-depth review of wildre simulation techniques together with associated rainfall simulators. Additionally, where Ferreira et al. (2008) focuses on hydrologic and sedimentation impacts of re, this review has an additional focus on postre water quality impacts including DOM, nutrient, and heavy metal concentrations. Finally, as over a decade has passed since the Ferreira et al. (2008) review, the time is ripe for a fresh perspective that considers more recent research on simulation experiment techniques, drawing upon a larger pool of studies.
The studies reviewed in this paper were compiled using the search strings of "wildre experiments", "wildre laboratory simulations", and variations of those in the Google Scholar search engine. These were further ltered by laboratory-and/or plot-scale studies, and those which had research goals focused on hydrologic or water quality responses, as well as temperature proles in soils. Studies examining post-re air quality, wildre behavior, and ecosystem restoration were excluded. From these studies, only those analyzing burn effects of natural fuels (i.e., litter and woody biomass), rather than human infrastructure and anthropogenic fuels (i.e., plastics and metals), were included. Additional studies were identied through the internal references across the initial set of publications. In total, 40 studies were included: 23 had a wildre simulation component and 27 had a rainfall simulation component, with 10 studies implementing both elements.

Terminology
A variety of wildre characterization terms are commonplace across studies, despite calls for standardization in recent decades. [47][48][49][50] In this review, wildre is used to describe res which occur in a natural environment, though bushre is an interchangeable term common in Australia. Following are the denitions for wildre intensity and wildre severity as used in this review, as well as other relevant wildre and rainfall simulation terms.
Wildre intensity: is typically a quantitative characterization of energy, such as the amount of fuel burned (e.g., units of g), 22 the rate of fuel burned (e.g., units of g s À1 or W m À2 ), 22,51 or peak temperature (e.g., units of C). 24,48,49,[52][53][54] This allows for explicit quantications of mild to severe burn intensities, however the lack of a standardized scale across studies makes crosscomparisons difficult. 2,22,55,56 Wildre severity: is typically a visual characterization of the response of an ecosystem (i.e., vegetation, soil, water systems, and atmosphere) to re, 5,47,50 such as ash color or amount of Fig. 1 Conceptual illustration of the impacts of wildfire on water quality and supply. The constituent fluxes in streams are driven by burning effects, affecting human and natural systems. Increased sediment and nutrient concentrations (a) drive eutrophication and disturb freshwater resources; increased sediment, DOM, nutrient, and heavy metal concentrations (b) can exceed water treatment plant treatment capacities; and increased sediment loads (c) affect reservoir storage capacity. biomass consumed. 29,57 Similar to wildre intensity, a standard denition of a wildre severity range from low to high does not exist across studies. 38,48,58 Laboratory-scale simulations: of wildre and rainfall are typically applied to smaller soil samples ($0.0045-4 m 2 ) 26,59,60 using experimental apparati. These analyses occur either inside a laboratory or in a designated outdoor setting, 29,61,62 limiting the size of samples to the dimensions of the equipment used.
Plot-scale simulations: of wildre and rainfall typically occur over a larger area of ground or hillslope ($0.5-300 m 2 )-either undisturbed or where a natural wildre has already occurred. [63][64][65] If undisturbed, a prescribed burn is typically applied to the area, then rainfall simulators used to generate runoff. 66,67 These experiments are also referred to as eld-scale or hillslope-scale experiments, but in this review only the term plot-scale is used.
Rainfall intensity: is the ratio of the total rain depth (e.g., units of mm) to the duration of rainfall (e.g., units of minutes, hours, or days).
Slope: is the average topographical inclination or gradient across a terrain surface (e.g., units of degrees or percent).
4 Wildfire-driven effects on water quality and supply

Sedimentation
Wildre-driven increases in suspended sediment in streamow have been reported from 1 to up to 1459 times pre-disturbance levels. 13 Downstream impacts of elevated sediment loading include strain on water treatment plants, reductions in reservoir storage capacity, and disruption to freshwater ecosystems. 1,11,12 Increased sediment loads can require increases in water treatment plant infrastructure and monitoring. 1,11,68 Sedimentdriven reservoir lling reduces the storage capacity, limiting available water for municipalities, and is expensive to mitigate using dredging or tunnels. [69][70][71] Finally, increased turbidity from wildre-driven sediment limits sunlight needed for photosynthesizing organisms that produce oxygen and form the base of aquatic food chains, resulting in sh death. [72][73][74] Sedimentation is enhanced by erosional effects due to loss of vegetation (both ground and canopy cover) and root structure, 27,75,76 exacerbated by high runoff rates in post-re settings. Rill erosion, for example, expands channel networks through high-ow rate runoff owing through streamletsthe eroded soil mobilized into suspended sediment. 76 These high post-re runoff rates can be driven by increased soil water repellency 75,77,78 thought to occur during heating and combustion processes in wildres. 13,43,75,[79][80][81] Organic compounds in litter and topsoil can be volatilized and ll the pores of upper soil layers creating water repellent effects, 29 further increased by the melting and redistribution of waxes and organic molecule polymerization. 81 Sediment is further supplied by an accumulation of combusted vegetation and soil organic material (i.e., ash), mobilized by erosional effects. 77 Initially, ash can absorb and store large amounts of rainfall (in one case, 99% for a $15 mm precipitation storm), thereby limiting runoff generation. 79 In subsequent or larger storms when ash saturation occurs, runoff generation increases, exacerbated by increased water repellency, and ash is mobilized downstream-contributing to high sedimentation rates. 79,82

Dissolved organic matter
Post-re runoff DOM response varies widely across studies, from slight decreases in concentration to levels in the 95th percentile of pre-re conditions. 1,10,13,28,43,60,[83][84][85] Downstream, DOM is the main substrate in the formation of carcinogenic disinfection byproducts (DBPs) during the chlorination stage of water treatment. 60 Elevated levels can require water treatment plants to implement expensive alternate disinfectants, precursor (i.e., DOM) and DBP removal strategies, or even force them to shut down. 84,86,87 Alterations in the load and chemistry of DOM in post-re settings is driven by thermal reactions during burning. 87,88 However, isolating wildre-driven DOM response is challenging because of natural background sources of DOM, contributed by other hydrological, topographical, physicochemical, and microbiological processes. 2,56 DOM levels monotonically decrease 89 or remain the same 61 with increasing burn severity in some studies, while others report peak DOM concentrations under moderate burn severity conditions. 5,26,30,83,90 Lower DOM concentrations are frequently reported at high burn intensities, likely due to DOM vaporization (transformation of organic material into carbon dioxide and water vapor) at extreme temperatures. 6,43,89

Nutrients
Wildre-driven increases in the nutrients phosphorous and nitrogen have been reported up to 250 and 400 times pre-burn conditions, respectively. 1,6,13 These dramatic increases in nutrients can lead to eutrophication in aquatic ecosystems. 91 Eutrophication is a process where excess nutrients lead to accelerated growth of aquatic plants and benthic communities, but also algal blooms which produce toxins and deplete oxygen from the ecosystem. 1,91,92 These effects, in addition to the impacts of sediment discussed earlier, result in the mortality of sh and other aquatic species, as well as loss of biodiversity. 85,91,92 Many complex mechanisms contribute to changes in nutrients in post-re settings. 1,43,[93][94][95] However, studies have generally shown higher rates of nitrogen concentrations aer burning due to deposition and changes in chemical structure. 1,43,93 Increases in phosphorous can result from constituent transport through increased post-wildre sedimentation rates. 68

Heavy metals
Though heavily dependent on the nature of the forest and climatic conditions, increases in dissolved heavy metal concentrations in runoff driven by wildres can range from 2 to at least 2500 times pre-disturbance levels. 30,96 These increases can exceed federal regulations, posing health hazards as some metals are carcinogenic or can cause anemia or heart failure. 97 Increases in magnesium are of particular concern due to toxicity, volatility, and persistence in the environment. 98,99 Iron, manganese, arsenic, chromium, aluminum, barium, and lead have all been observed at statistically signicantly higher concentrations in post-re runoff. 13 Moderate to high intensity res can alter soil properties, releasing sequestered metals in organic matter which are transported downstream by high post-re owrates. 30,100 5 Experimental techniques to observe and quantify post-fire impacts The strengths of wildre and rainfall simulation experiments are in the reduction of the logistical challenges associated with collecting in situ data, the ability to generate baseline pre-re data as well as replicate samples, and to provide greater control over factors in post-re systems. 10,13,24,29 Greater control over the timescale and number of samples collected is conducive to isolating wildre responses from background sources. 10,38 Limitations of simulation experiments include deviation from natural conditions, e.g. vegetation characteristics and lateral ow paths, as well as signicant differences in hydrologic and chemical processes at different scales. 7,29,67 Interpretation of results from laboratory-and plot-scale simulation experiments for catchment scale impacts should incorporate uncertainties, such as landscape heterogeneity, associated with the upscaling process. The following section will rst discuss key differences in plot-versus laboratory-scale simulation techniques, then present the strengths and weaknesses of specic methodologies for wildre and rainfall simulation experiments. Fig. 2 shows how laboratory-, plot-, and catchment-scale studies are related in terms of size, as well as common simulation apparati used at the laboratory and plot scales.

Plot-versus laboratory-scale simulation techniques
Plot-scale experiments are conducted on in situ hillslopes or plots and are generally subject to higher spatial variability in soil properties and vegetation, as well as spatial and temporal variability in burning and rainfall, than laboratory-scale experiments. 101 Though high variability can hinder the attribution of responses to drivers, this type of analysis also minimizes areato-edge ratios (i.e., limiting edge effects) and allows for largerscale vegetation (e.g., trees) and hydrologic processes to be captured. 67,76 Larger-scale hydrologic processes, i.e. rill formation and other erosional processes, can be key drivers of post-wildre sedimentation and runoff responses 76 and thus important to consider in post-wildre analyses. A wide range of replicates are typically tested in plot-scale analyses, from 0 to approximately 15. 76,102 Laboratory-scale experiments typically use excavated, intact soil cores or homogenized samples which have low spatial variability, due in part to their smaller size. 54,67,76,89,103 Samples are typically excavated by hammering lysimeter boxes or PVC cylinders into soil 89,90 or collected from loose soil and litter on the ground-oen homogenized to minimize variability. 26,54,102,104 Laboratory-scale samples are oen less representative of natural conditions in terms of wildre intensity, erosional processes, vegetation type, and soil structure, but allow for precise control over experimental factors. This is conducive to more direct attribution of the impacts of burning and other factors in the system (e.g., vegetation, rainfall intensity, etc.) to observed responses. 29 Another key feature of laboratory-scale analyses is they allow for the measurement of inltration through the bottom of samples and subsurface later ows, which is difficult or impossible in plot-scale in situ analyses. 29,90,104 A comparison of the pros and cons of plot-and laboratory-scale analyses are presented in Table 1. The number of replicate samples in laboratory-scale analyses typically range from 2-5. 56,106

Wildre simulation
Prescribed and slash burns, propane torches and heat lamps, litter burns, and muffle furnaces have all been used to study the effects of burning on soil and runoff physical and chemical properties. 22,26,54,67,83,90,103 The key features which differ across these techniques are the range of simulated wildre intensities, methods of burn characterization, and spatial variability of the combustion and burn properties. These studies are tabulated in Table 2 and are described in the following sub-sections.
5.2.1 Prescribed and slash burns. Prescribed and slash burns are used as wildre simulation techniques for plot-scale analysis of erosional, hydrologic, and sedimentation response to burning. 63,76,107 They involve the application of an incendiary device to in situ accumulated fuels under conditions conducive to the control of a re, i.e. low winds and high soil moisture. 63,66,[107][108][109][110] Either aerial or ground-based incendiary devices are used to start these res, such as a Plastic Sphere Dispenser 111 or a drip torch, 112,113 respectively. Prescribed burning, also referred to as control or experimental burns, 113,114 typically span a 40-200 ha (100-500 acre) area and most oen burn at a low intensity, only consuming fuels with small diameters (e.g., pine needles and small branches). 115 Slash burns, also referred to as pile burns, 115 involve rst collecting surplus woody debris (i.e., trees or brush) into a concentrated areatypically burning at a high intensity. [115][116][117] Prescribed and slash burns are most oen qualitatively characterized by visual characteristics, such as biomass consumed (i.e., wildre severity). 112,118 A wide range of responses are typical for this type of technique, with runoff increasing $20-300%, inltration decreasing $2-40%, sediment yield increasing $3-200 000%, and soil chemical composition decreasing $7-15% with burning (typically analyzed at just one burn severity increment). 63,108,113,119 However, some studies reported no statistical signicance in certain results due to natural variability exceeding post-burn response. 66, 119 5.2.2 Propane torches and heat lamps. Propane torches and heat lamps are common wildre simulation techniques used in laboratory-scale analyses of burn impacts on soil and runoff chemical composition, as well as sedimentation response and changes to soil structure. 61,90,106 These techniques involve a steady heat ux concentrated at a point ($50-100 mm) on the soil surface. 54,59,61,106,120,121 Propane torches typically produce peak soil surface temperatures of 500-600 C 54,61 whereas heat lamps produce peak temperatures of 200-400 C 90 from 5 to up to 220 minutes of exposure, 54,61 representing a range in burn intensities. The temperature gradient extending down through the soil surface in these experiments is typically measured using thermocouples placed 0-15 cm below the soil surface. 22,24,29,54,61,103 These studies report a wide range of signicant responses to burning, with soil and water chemical constituents varying by $10-700% and hydraulic and water repellent properties by $60-200%, typically at 1-3 increments of burn intensities. 62,89,121 Kral et al. (2015) showed that propane Table 1 Comparison of pros and cons of plot-and laboratory-scale simulation experiments based on the following factors: spatial and temporal variability, control over environmental factors, representation of natural environments, edge effects, and subsurface flows. An N/A entry indicates that no significant pro or con exists for that category  Table 2 Summary of major wildfire simulation studies included in this review, listed in alphabetical order. The characteristics included are scale, size and shape of samples, wildfire simulation technique, peak burning temperature, duration of the burn, and a summary of key results from each study. The provided summaries are reported with respect to unburned conditions unless otherwise stated, e.g. a percent decrease indicates the percent decrease of the response after burning from the control or unburned case in the study. An N/A entry denotes an unreported study characteristic, WR is water repellency, TOC is total organic carbon, PyC is pyrogenic carbon, O horizon is organic soil horizon, Ah horizon is soil mineral horizon, KCl is potassium chloride, DOC is dissolved organic carbon, and DON is dissolved organic nitrogen Scale Size and shape of samples   prong res had disparate time-temperature curves compared to prescribed res, with longer durations near peak temperatures, but the least difference in total heat dosage.

Litter burns.
Litter burns are a laboratory-scale wild-re simulation technique used to study burn effects on soil and runoff composition, as well as the effects of soil composition on heating proles. 29,59,103,120 The method involves igniting small amounts of litter spread evenly on top of soil samples, reaching temperatures of 230-867 C. 29,120,122 This measured amount of fuel allows for a direct and quantitative measurement of burn intensity. 22,29 Additionally, similar to propane torches and heat lamps, thermocouples placed 0-15 cm below the soil surface are used to measure heating proles during the burn process. 29,59,122 Study results show substantial changes in water repellency, runoff and inltration rates (up to a $6000% increase and $90% decrease, respectively), and sedimentation rates, 29 as well as clear soil heating proles, analyzed at typically 1-3 increments of burn intensities. 59,103,120 Hogue and Inglett (2012) found signicant changes in some soil chemical properties (i.e., total carbon and nitrogen) aer litter burning, however no clear trend with increasing burn intensity was observed due to heterogeneity of burn residues. Kral et al. (2015) showed that litter burns had analogous time-temperature curves to prescribed res, however oen had signicantly higher heat durations and dosages.

Muffle furnaces.
Muffle furnaces are a laboratoryscale wildre simulation technique typically used to analyze burn impacts on soil chemical composition. 26,52,113,123 Samples are placed in an oven which is typically raised to temperatures between 100-570 C for as little as 65 seconds up to 2 hours to simulate a range of burn intensities. 26,52,83,113 Results from muffle furnace studies show generally signicant monotonic and parabolic-shaped trends in leached chemical constituents and water repellency, typically measured at 1-9 increments of burn intensities. 22,26,52,83,113,123 Hogue and Inglett (2012) found that, unlike litter burns, muffle furnace burning did not produce heterogeneous burn residuescommonly observed in natural wildres.

Comparison of key wildre simulation technique characteristics
Each of the wildre simulation experiment techniques described above has benets and limitations which future researchers may take into consideration in their own designs. Table 3 shows a broad description the pros and cons of each technique.
The following paragraphs discuss common pros of wildre simulation experiments as listed in Table 3: Heterogeneous combustion (WP1) is a key strength of prescribed res and slash burns, as well as litter burns. The variability in spatial distribution of heating and volatilization created by these types of burns produce variable combustion residues analogous to natural wildres. This variability is due to spatial heterogeneity of fuel types and amounts, as well as variable wind speed, direction, and air temperature across the burn area in plot-scale experiments. 22 Similar intensity and duration to a natural wildre (WP2) is a strength of prescribed res and slash burns. As a comparable amount of fuel is available in a prescribed burn as a natural wildre, prescribed res typically match the peak temperatures and duration of a low intensity wildre. 112 Slash burns can reach the extreme temperatures reached by severe wildres, sometimes as high as $2200 C-oen unachievable by laboratoryscale simulation techniques. 117 Precise control over burn intensity and spatial distribution (WP3) is a strength of propane torches and heat lamps, as well as litter burns. 121 This technique allows for more direct attribution of burning effects to specic intensities, given the controlled range and spatial distribution of burn intensities. 29 Low variability in heating (WP4) is a benet of propane torches and heat lamps, as well as muffle furnaces. Low spatial and temporal variability created by the consistent heating distribution of these methods is conducive to quantitative analysis, as it limits variability-driven uncertainty in responses. 113 Allowance for measurement of heating proles (WP5) is a key aspect of propane torches and heat lamps, litter burns, and muffle furnace methods. Propane torches and heat lamp methods, as well as litter burns, achieve this using thermocouples, and muffle furnaces achieve this through digital temperature readings. This allows for consistent characterization of burn intensity, conducive to quantitative post-wildre analyses.
Control over duration of heating (WP6) is an attribute of propane torches and heat lamps, as well as muffle furnaces. This allows for further control and precision of burn intensitysometimes characterized by burn duration. 24 Incremental control over burn intensity (WP7) is a benet unique to muffle furnaces. This feature allows for analysis of changing burn effects over a range of burn intensitiesconducive to analyzing response trends with increasing burn intensity. 113 The following paragraphs discuss common cons of wildre simulation experiments as listed in Table 3: Qualitative burn severity characterization (WC1) is a limitation for prescribed res and slash burns. 67,101 Qualitative wild-re characterizations (i.e., burn severity) are less precise than temperature or fuel measurements (i.e., burn intensity), hindering quantitative analyses and precise replication of simulated burn severities.
Tradeoff between high intensities and burn coverage (WC2) is another limitation for prescribed res and slash burns. As these burns must be managed in a safe, controlled way, larger prescribed burns (40-200 ha) are typically limited to a low burn intensity. Slash burns, which can achieve extreme temperatures, are therefore limited to smaller areas.
Uniformity in spatial heating (WC3) is considered a limitation for propane torch and heat lamp methods, as well as muffle furnaces. While this characteristic may assist in the attribution of burn effects to specic drivers, it typically does not produce the heterogeneous combustion residues present aer a natural wild-re. This is a key source of uncertainty in the representativeness of burned soil structure and composition as compared to natural combustion (i.e., burning fuel). 22 Some studies have addressed this shortcoming by using controlled heating methods coupled with igniting vegetation or litter on the soil surface. 59,90,120 Limitations to low-intensity burns (WC4), due to safety and other logistical considerations, is a drawback of litter burns. 103 The even distribution of fuels is oen inconsequential in substantively heating soils, which may not exceed a low intensity burn. 122 Heating from all sides (WC5) is a disadvantage of muffle furnaces. These heating mechanics are categorically different from a natural wildre, which only heat the side of exposed soil surface. 26,52

Rainfall simulation
Nozzle-based and drip-based rainfall simulators, water drop penetration time (WDPT) tests, and leaching are typically used in conjunction with one of the wildre simulation techniques mentioned above or they are implemented in situ, over an area already burned by a wildre. 29,78,104,105,[124][125][126] The effects of consecutive rainfall events, with drying periods anywhere from 30 minutes to 1 year, are examined by some rainfall simulation studies, as antecedent moisture content and weathering over time can greatly inuence post-wildre hydrologic response. 10,64,79 Runoff collection chambers in these simulations are typically located at the lower end of a sloped plot or sample with a guard to deect the simulated rainfall, with sampling frequencies ranging from 20 seconds to 20 minutes. 29,64,76,104 To facilitate discussion, various rainfall simulation techniques are divided into four categories: xed nozzle-based simulators, dynamic nozzle-based simulators, drip-style rainfall simulators, and WDPT tests and leaching. The key features which vary between these simulation techniques are range of rainfall intensities, precision of the droplet size and kinetic energy, and spatial distribution. Published wildre studies which employ these techniques are listed in Table 4 and described in the following subsections.
5.4.1 Fixed nozzle-based rainfall simulators. Fixed nozzlebased rainfall simulators have been used in both plot-and laboratory-scale analyses of wildre impacts on soil and runoff physical and chemical changes. 65,90,104,[127][128][129] Most of these simulators use a single, stationary nozzle $2 m above the ground which points downward, covering areas of 0.08-1 Table 3 Pros and cons of wildfire simulation techniques covered in the review, as well as the studies referenced and their scales. WP1, WP2, etc. represents wildfire simulation pro 1, wildfire simulation pro 2, etc. Similarly, WC1, WC2, etc. represents wildfire simulation con 1, wildfire simulation con 2, etc One exception is the plot-scale Field Efficient Colorado State Rainfall Simulator which has ten 3 m risers covering 300 m 2 , each with 1-2 nozzles pointing upwards. 65,130 Nozzles are rated to produce a droplet size and kinetic energy similar to natural rainfall at a specied distance beneath the nozzle, as well as an even distribution of rainfall intensity (e.g., the Full-Jet® and VeeJet© nozzles produced by the Spraying Systems Company). 104,127,129 Droplet sizes range from 0.8-4.0 mm and kinetic energies from 0.1-28 J m À2 mm À1 , with rainfall intensities ranging from 5-203 mm h À1 and durations from 5 min to 2 h. 90,108,[127][128][129]131,132 Results from studies which implemented this type of simulator show a wide range of signicant responses in runoff solute concentrations ($20-700% increases), runoff ($10-20 000% increases), and sediment ($300-20 000% increases) aer burning, typically measured at 1-3 increments of rainfall intensities. 62,65,90,102,126 However, some studies also reported little statistical signicance in runoff and inltration rate response. 65

Dynamic nozzle-based rainfall simulators.
Although dynamic nozzle-based rainfall simulators have been used to analyze similar wildre effects as xed nozzle simulators, their use is more common in plot-scale analyses. 64,66,76,78,101,105,133 These simulators incorporate horizontal rotation 29,105 or sweeping motions $3 m above the plot or samples, 76,78,101,134 covering large areas of up to $30 m 2 . 66 The nozzles used are rated to produce natural rainfall kinetic energy and droplet size, same as the xed nozzle simulators, and produce rainfall intensities ranging from 33-127 mm h À1 for 15 min up to 4 hours. 105 Studies with this type of simulator report wide ranges of generally signicant responses in runoff ($5-600% increases), inltration ($30-90% decreases), and sediment ($40-7000% increases) aer burning, analyzed at 1-2 increments of rainfall intensities. 29,64,66,76,78,82,133 Studies analyzing responses several months or years aer burn events tended to not see signicance in results. 66,76,78,82,133 5.4.3 Drip-style rainfall simulators. Drip-style rainfall simulators have been used to study the same processes as nozzle-style simulators, typically at a plot-scale. 63,124,135 In these simulators, water is channelled to a large number ($168 to 2209) of ne tubes or needles which periodically release droplets due to gravitational forces, 125,135 covering areas from 0.4-9 m 2 . 119,135,136 This technique produces droplet sizes ranging from 2.6-3.3 mm in diameter and rainfall intensities ranging from 20-203 mm h À1 (ref. 63 and 124) for durations of 0.5-1 hour. 133,136 Results from studies using this type of simulation technique show generally signicant changes in runoff chemical constituents ($20-49% increases), 124 as well as inltration rates ($2-11% decreases) and sedimentation rates ($3-300 000% increases) aer burning, typically analyzed at 1-3 rainfall intensity increments. 63,119,136 5.4.4 Water drop penetration time tests and leaching. WDPT tests and leaching do not simulate the mechanics of rainfall, but are important laboratory-scale techniques used to assess wildre impacts on soil water repellency and changes in chemical composition, respectively. 52,61,83,89,123 WDPTs involve placing droplets of water or a water-ethanol mixture on burned soil and recording the duration of time for each drop to inltrate as a measure of soil water repellency. 61,121,137 Leaching involves dissolving water-soluble chemical constituents in burned soil or litter into water, then analyzing the water for chemical composition. 26 Studies which used this type of analysis show generally signicant changes aer burning in water repellency and hydraulic conductivity ($60-200% increases), as well as leached chemical constituents ($10-60% increases and decreases). 29,89

Comparison of key rainfall simulation technique characteristics
The rainfall simulation techniques described above all have benets and limitations which future researchers may take into consideration. Table 5 shows a broad description the pros and cons of each technique.
The following paragraphs discuss common pros of rainfall simulation experiments as listed in Table 5: Simplicity in design (RP1) is a key benet of xed nozzlebased rainfall simulators, WDPT tests, and leaching. 90,127,129,130 A stationary nozzle is a relatively simple and inexpensive mechanism to construct or purchase, and is oen sufficient in terms of coverage area and semblance to natural rainfall. WDPT tests and leaching typically only require simple laboratory equipment.
Transportability and adaptability to steep terrains (RP2) is a benet of xed nozzle-based rainfall simulators. This allows these simulators to be tested on otherwise inaccessible sampling locations and on steep terrains up to 45 . 65,127,129 Rainfall intensities and droplet sizes similar to natural rainfall (RP3) is a key benet of xed nozzle-based, dynamic nozzle-based, and drip-style rainfall simulators. The ranges of intensities and droplet sizes for these types of simulators make them representative of typical natural precipitation. 90,108,[127][128][129] A large area of coverage (RP4) is a benet unique to dynamic nozzle-based rainfall simulators. Larger plot-scale analyses allow for larger-scale hydrologic processes to occur which cannot be observed on smaller scales, [138][139][140] and minimizes edge effects. 67 Spatial and temporal variability in droplet distribution similar to natural rainfall (RP5) is an important attribute of dynamic nozzle-based rainfall simulators. The horizontal rotation and sweeping motion of the nozzles used in these simulators may create spatial and temporal variability which is more representative of natural rainfall than stationary nozzles. 125 Increased control and precision of droplet size (RP6) is a benet of drip-style rainfall simulators. Droplet size can be altered by changing the gage of the tubes and needles used, allowing for control over droplet diameters 125 and subsequently the kinetic energy of raindrops produced.
Direct measurement of water repellency and chemical changes (RP7) is a benet unique to WDPT tests and leaching. This can allow for more precise attribution of burn effects, as opposed to the indirect measurements through runoff generation and chemical composition in other rainfall simulation techniques. 26,89,123 The following paragraphs discuss common cons of rainfall simulation experiments as listed in Table 5: Table 4 Summary of major rainfall simulation studies included in this review, listed in alphabetical order. The characteristics included are study subject, scale, size and shape of samples, rainfall simulation technique, rainfall intensity, duration and scheduling of the simulated rainfall, and a summary of key results from each study. The provided summaries are reported with respect to unburned conditions unless otherwise stated, e.g. a percent decrease indicates the percent decrease of the response after burning from the control or unburned case in the study. An N/A entry denotes an unreported study characteristic. WR is water repellency, TOC is total organic carbon, and DOC is dissolved organic carbon Scale Size and shape of samples

View Article Online
A small area of coverage (RC1) is a limitation of many xed nozzle-based rainfall simulators, WDPT tests, and leaching. 127 The smaller area covered limits analyses to the laboratory-or smaller plot-scale for xed nozzle-based rainfall simulators, and typically only laboratory-scale samples for WDPT tests and leaching. This oen means that these simulation experiments do not capture larger-scale hydrologic processes, such as rill erosion (as discussed in Section 5.1). Rainfall kinetic energies lower than typical natural rainfall (RC2) is a limitation for xed nozzle-based rainfall simulators. Rainfall kinetic energies for these simulators tend to be lower than natural rainfall due to simulated droplets oen not reaching terminal velocity before impact, as nozzle heights are constrained by equipment. 127 Complexity and expense of design (RC3) are limitations for dynamic nozzle-based rainfall simulators and drip-style rainfall simulators, which may result in logistical and nancial challenges for studies.
Difficulty in transportation (RC4) is another drawback for dynamic nozzle-based rainfall simulators and drip-style rainfall simulators. This can limit plot-scale study sites to ones with accessible roads, as well as relatively at terrain.
Lack of rainfall impact on soils (RC5), and therefore lack of droplet kinetic energy, is a limitation of WDPT tests and leaching. These methods do not capture the physical processes of rainfall impact on soil surface and therefore cannot simulate natural constituent transport through runoff. 26 Table 5 Pros and cons of rainfall simulation techniques covered in the review, as well as the studies referenced and their scales. RP1, RP2, etc. represents rainfall simulation pro 1, rainfall simulation pro 2, etc. Similarly, RC1, RC2, etc. represents rainfall simulation con 1, rainfall simulation con 2, etc Decisions about which experimental methodologies to choose should largely depend on study scope: the hydrologic or chemical responses being analyzed, as well as the temporal and physical scales of the analysis. Researchers should also carefully consider the geographical setting of their study and incorporate specic regional characteristics into their experimental design, such as soil type, vegetation cover, climate regimes, and terrain slopes. In general, we recommend that experimental design elements should be optimized based on their strengths in analyzing important study elements, while weighing time, logistic, and nancial constraints. For example, a study analyzing the effects of different vegetation types on post-wildre hydrologic processes may want to focus on plot-scale techniques in order to capture larger-scale erosional processes such as rill erosion. In such a scenario, a prescribed burn method could provide the necessary burning scale, burn the intended fuel type (i.e., the different types of vegetation in question), and also partially represent heterogeneous combustion patterns, intact soil structure, and largerscale vegetation present in a natural wildre. In this example, a dynamic nozzle-based rainfall simulator may be the best choice of rainfall simulator to allow for a large area of coverage. Alternatively, if the interaction between burn intensity and vegetation characteristics was of primary interest in the above example, then a more appropriate experimental set-up may use a burn simulation technique which allows for greater incremental control over wildre intensity. Heat lamps, for example, allow for analysis of targeted vegetation burning at specic intensity levels. However, this type of analysis sacrices some representation of natural burning, due to low spatial variability in combustion. Additionally, as with any laboratory method, the soil sampling process involved in heat lamp simulation techniques introduces edge effects, potentially disturbed soil structure, and can only represent small-scale hydrologic processes and vegetation. A xed nozzle-based rainfall simulator may be the best choice in this scenario, as only a small area of coverage would be required.
Researchers should also take the results from previous simulation studies into consideration for their study designs. For example, plot-scale studies in this review which implemented prescribed burning tended to most frequently produce results that were not statistically signicant relative to control samples. This is likely due to a combination of the spatial heterogeneity of burn intensities in prescribed burning methods and the high variability of plot-scale natural settings typically subjected to prescribed burning. Conversely, laboratory-scale studies which used muffle furnace heating tended to produce results that were highly statistically signicant and were able to assess responses at a high number (up to nine) of burn intensity increments. This level of granularity allowed these studies to infer a more fundamental character of the heating effect, for example a monotonic versus negative parabolic response to heating. However, these types of analyses were typically limited to water quality constituents and water repellency, as larger-scale erosional and hydrologic responses were not captured.
Using precedents set by previous simulation studies as a guide, researchers may choose appropriate methods to t their research goals. To aid discussion, we categorize strengths and weaknesses of simulation experiments into four important factors: (1) representation of natural processes and settings, (2) analysis of multiple post-wildre water quality and supply drivers simultaneously and independently, (3) observation of responses on different temporal and spatial scales, and (4) mitigation of uncertainty of results. The following paragraphs discuss the tradeoffs that exist between these four key design factors.

Representation of natural processes
Replicating natural processes improves the representativeness of experimental results-furthering understanding of wildre effects on soil and runoff characteristics. However, this goal must be weighed against logistical challenges of in situ collection, as well as the increased spatial and temporal variability inherent with natural features-which can create difficulties in attribution. Studies typically address these tradeoffs by only choosing natural or unperturbed features most important to the subject and scale of the study. For example, Hogue and Inglett (2012) examined carbon and nitrogen concentrations in naturally combusted residue, using litter burning with spatially variable combustion to replicate natural wildre mechanisms. 22 Similarly, Benavides-Solorio and MacDonald (2005) and Johansen et al. (2001) analyzed wildre's role in increased rill erosion, a plot-scale erosional process, by employing plot-scale wildre and rainfall simulation techniques. Therefore, future studies are recommended to rst identify the subject and scale(s) of greatest interest, then focus efforts on replicating natural processes for those elements.

Incorporation of multiple key drivers
Incorporation of multiple drivers-i.e. burn severity, rainfall intensity, terrain slope, vegetation type, and soil characteristics-at multiple increments and categories, is oen sought to gain a more comprehensive understanding of their relative importance and system interactions. However, studies must evaluate the benets of including these characteristics, since they can limit the number of replicate samples useful for characterizing uncertainty, as well as require large numbers of samples to be collected. Most studies in this review include burn severity and rainfall intensity in their analyses. 7,13,38,79,90 Soil structure and composition, terrain slope, climate, vegetation type, and antecedent moisture content are less commonly incorporated, but can have comparable impacts on hydrology and water quality. 93 For example, Johansen et al. (2001) incorporated the percentage of bare soil into their analysis of post-wildre sedimentation, nding that this driver had a strong correlation with sediment generation in addition to burn severity. Factors involved in wildre prevention, suppression, and mitigation of effects (i.e., mechanical thinning, mulching, and chemical re-retardants) are also less commonly incorporated, though could provide insights important to re management efforts. Understanding how each driver impacts hydrology and water quality independently and jointly can also assist in the creation of catchment-scale predictive models, which typically incorporate multiple drivers as model parameters. Benavides-Solorio and MacDonald (2005), for example, used re severity, percent bare soil, rainfall erosivity, soil water repellency, and soil texture as model parameters to predict post-wildre sedimentation.
Responses systematically tested over ranges of drivers allows for an understanding of the shape of the response function (e.g., monotonic, parabolic, etc.). Studies typically use only $2-3 increments of burn severity and rainfall intensity due to logistical and time constraints. 83,90,121,124 However, a higher number of increments proved to be important in Hohner et al. (2019). Here, soil samples were heated in a muffle furnace at ve temperature increments ranging from 150-550 C, nding that water extractable organic carbon and nitrogen had a roughly negative parabolic relationship with temperature, peaking around 250-350 C. 26 We recommend that for a given number of total samples, future studies carefully consider the tradeoffs between the number of increments and the number of replicates.

Analyses at multiple spatial scales
A limitation of single-scale wildre and rainfall simulation experiments is the lack of consideration for how properties and processes at one scale may effect water quality and supply responses at larger scales. 67 Incorporating multiple physical scales in simulation experiments can provide insight into upscaling operators which can inform catchment-scale predictions. This is particularly important in sedimentation analyses, as geomorphic and erosional processes vary greatly from the laboratory-scale to the catchment-scale. 27 Post-re sedimentation mechanisms such as streambed erosion may be entirely missing, even from plot-scale analyses. However, understanding how mechanisms (e.g., rill erosion, streambed erosion, etc.) are introduced and change at increasing scales can allow for indirect estimation and inference about catchment-scale response.
Simulating multiple physical scales is challenging in a laboratory setting due to xed equipment size. 29,89,90,106,122 Multi-scale analysis is also uncommon in plot-scale studies. However, Ferreira et al. (2005) analyzed sediment and runoff in post-re plots on a microplot-(<1 m 2 ), plot-(16 m 2 ), and catchmentscale (<1.5 km 2 ), allowing for comparison of results across varying scales. We recommend that future studies consider analyzing post-re responses at more than one scale, if feasible with their study design and logistical and nancial limitations.

Uncertainty quantication
Uncertainty is most commonly estimated in experimental systems by testing multiple replicate samples, or uniform samples tested under the same conditions. 29,80 Quantifying uncertainty can be useful in differentiating the water quality responses of different drivers and can help inform upscaling of results to the catchment scale. 141,142 High spatial and temporal variability, albeit representative of natural systems, can introduce additional uncertainty due to difficulties in the attribution of responses to specic drivers. Thus, uncertainty analysis and mitigation efforts must consider both replicate uncertainty, as well as the role of natural variability on attribution uncertainty. In Keesstra et al. (2014), for example, soil samples were homogenized-reducing uncertainty from spatially variable soil structure and vegetation, but decreasing the samples' semblance of a natural environment. We recommend opting for greater numbers of replicate samples to quantify uncertainty, while weighing incorporation of multiple drivers at different increments and scales-which may constrain the feasible number of replicates across each study dimension.

Conclusion
This review provides a synthesis of knowledge on wildre and rainfall simulation techniques used to understand the impacts of wildre on water quality and supply. Wildre and rainfall simulation techniques offer solutions to logistical challenges faced in the collection of in situ data, including potentially dangerous post-re environments, expensive eldwork expeditions, and lack of control data. However, each technique has unique strengths and weaknesses. Plot-scale analyses are oen able to capture a higher spatial variability more representative of natural settings than laboratory-scale analyses, as well as simulate larger-scale hydrologic processes (i.e., erosion). Yet, attribution of responses to specic drivers is oen difficult due to high variability of conditions within and across plots.
Laboratory-scale analyses can more precisely control factors in the simulated system, limiting variability and allowing for drivers to be tested at a range of increments. This allows for a more direct attribution of the role of each driver on system responses, independently and jointly across ranges of values. Laboratory-scale experiments also have the benet of more precise measurements (e.g., using thermocouples to measure temperature proles) and control over drivers (e.g., muffle furnaces which can be set to exact temperatures for exact durations), which assists in the quantitative analysis of results. The downsides of laboratoryscale experiments are that they are less representative of a natural wildre system, due to limited spatial variability and scalemeaning only small-scale hydrologic processes can be analyzed.
Common design considerations across these studies include representation of natural processes, incorporation of multiple key drivers, analysis at multiple spatial scales, and uncertainty quantication. As studies are limited by time, resources, and logistical constraints, prioritization of these design considerations in future studies must be made based on scale, scope, and subject matter. Representation of natural processes can increase variability, and therefore increase uncertainty in results. Similarly, increased complexities in the study design, such as incorporation of multiple drivers and spatial scales, can decrease the amount of replicate samples at each condition, thereby limiting a robust quantication of uncertainty. Thus, future studies must weigh which design considerations are important for each aspect of their experiment, focusing resources on realistic representation of the key drivers or constituents of interest.
This review seeks to support the advancement of knowledge in the eld of wildre impacts on water quality and supply. These ndings may be informative for future practitioners, as well as for water management efforts in mitigation and adaptation strategies for wildre impacts. As wildres continue to represent an increasing threat to water quality and supply, developing advanced techniques to provide further understanding of wildre effects will become increasingly essential.

Conflicts of interest
There are no conicts to declare.