Nutrient mass balance of koi–basil aquaponics production systems

1 Introduction

Aquaponics integrates hydroponic crop production with aquaculture, motivated by nutrient circularity. Coupling is practical because aquaculture waste is already waterborne and hydroponic production is soilless, so nutrient-bearing water transfers directly between the two. Some definitions of aquaponics specify thresholds for nutrient supply from aquaculture production to plant production, with a value of > 50% of nutrients and > 80% of all nutrients supplied to plants being sourced from aquaculture (Palm et al., 2018; Lennard, 2017; Lennard and Goddek, 2019; Palm et al., 2024). Baganz et al. provided one helpful definition that emphasizes nutrient and water coupling without being prescriptive on nutrient thresholds. They say that:

Aquaponics is a technology that couples tank-based animal aquaculture with hydroponics—involving microbiological processes -using water from aquaculture for plant nutrition and irrigation (Baganz et al., 2022).

Hao et al. (2020) identified system construction, nutrient management, and microbial community structure as areas needing deeper research to build a stronger theoretical basis for aquaponics in practice. Research on nitrogen use efficiency (NUE) in aquaponics highlights that approximately 10%–50% of the nitrogen in coupled aquaponics is ultimately incorporated into plants, depending on the cultivated plants (Paudel, 2020; Gao et al., 2024). Wongkiew et al. (2018a) showed that fruiting plants can achieve higher NUE due at least in part to nitrate depletion in the water column which they attributed to greater root surface area supporting a higher abundance of nitrifying bacteria and improved water quality. The nitrogen that does not make it to crops is thought to be incorporated into fish growth mass and fish waste sludge. However, some have proposed that denitrification to N₂ or N₂O (0.5%–2.0% of input N) is also a possible leak from what was originally provided in fish feed (Zou et al., 2016a; Endut et al., 2014). Studies of N₂O emissions in soils generally indicate that as pH declines, N₂O reductase is inhibited, which means that if denitrification proceeds, the product ratio of N₂O:N₂ will gradually tend toward higher N₂O rather than N₂ (Šimek and Cooper, 2002; Qiu et al., 2024; Liu et al., 2014; Hénault et al., 2019).

Comparisons across system architectures give mixed results. Aslanidou et al. (2023) reported 42% higher NPK and Ca use in coupled aquaponics (CAP) than in decoupled (DCAP) or hydroponic (HP) systems, yet still favored DCAP because fruiting crops have mineral demands (particularly K) that are easier to tune when loops are separated. Yang and Kim (2020) reached a similar conclusion for nitrogen across tomato, basil, and lettuce: whole-plant N use was lower in aquaponics (9–17%) than hydroponics (14–24%), but whole-system N use including fish (30–41%) exceeded hydroponics. Whether NUE is reported at plant or system level therefore drives the conclusion.

At a fundamental level, nitrogen is different from most other nutrients supplied through fish feed because nitrogen must be transformed from ammonia (NH₃) to nitrate (NO₃) to be most useful to plants. The transformation occurs through nitrifying bacteria that colonize surfaces in a designed biofilter and aquaponics tank surfaces. Conventional nitrifiers found in aquaponic systems are most effective at a neutral to slightly alkaline pH of 7.5–8.0 because as pH falls, the equilibrium shifts increasingly toward ammonium (NH₄⁺), a form that nitrifier enzymes do not use. In addition, the proton-motive force across the cell membrane is hindered by lower pH, making ATP more difficult to synthesize (Lehtovirta-Morley et al., 2016; Fumasoli et al., 2017; Frijlink et al., 1992). Yet phosphorus (P) and many micronutrients, such as Fe and Mn, are more available to plants at lower pH of 5.5–6.0. The conflict between effective nitrification at moderate pH (6.0–7.5) and other nutrient availability for plants at low pH is a reason for using separate loops for fish and plants.

Higher NUE for both phosphorus and nitrogen is needed to address the problem that the fish feed-to-inorganic fertilizer cost ratio is unfavorable for aquaponics, with a value of 2–4x higher for fish feed compared to monoammonium phosphate and 14–17x for anhydrous ammonia and triple superphosphate (Colt and Schuur, 2021).

Phosphorus use in aquaponics is underutilized because of its entrapment in fish waste sludge material. Kalvakaalva et al. (2023) found that the P use efficiency (PUE) was favorable for fish at 33% but quite low in plants at only 2.6%. Cerozi and Fitzsimmons (2017) reported a higher whole-plant PUE of 31.9%. They showed that 13.1% of the P in the aquaponics nutrient solution was unavailable, speculating that it was in suspended solids and sludge. Thus, one issue with P utilization is preventing precipitation of P that might otherwise remain dissolved and bioavailable. Maintaining pH in a mildly acid to neutral range (5.5–7.2) may help because P is in the form of a singly deprotonated acid (H₂PO₄^-). At high pH, precipitates of Ca form, while at lower pH, precipitates of Al and Fe form (Cerozi and Fitzsimmons, 2016b; Devau et al., 2009).

Because much sludge P remains in the solid phase, active mobilization has been tested. Panana et al. (2021) found direct acidification (pH 7.5 to 6.3) released more P than aerobic mineralization alone, and Guyapale et al. (2024) achieved greater release by adding a carbon source, though pH fell below 4. Chen et al. (2023) reached a system PUE of 65.0% by holding pH near 6 and adjusting C:N to favor heterotrophs, attributing the gain to Bacillus spp. solubilizing P in fish feces. These interventions were outside the scope of the present study but indicate the achievable range.

Recent work illustrates interest in aquaponics using brackish groundwater or saline water and the need for additional experimentation to identify appropriate crop–fish combinations according to salinity (Tarigan et al., 2021; Vlahos et al., 2023; Puccinelli et al., 2022, 2025; Fisk et al., 2026). Existing studies do not point to a single unifying role for salinity in aquaponics-grown crops. Bordignon et al. (2024) found that brackish water up to 6,000 mg/L total dissolved solids (TDS) did not impair leafy vegetable growth in trout/catfish systems with Swiss chard and tomatoes. Kaburagi et al. (2020) grew Swiss chard at 1,150 mg/L TDS over multiple cycles; performance was good initially, but Mn, Zn, and Fe deficiencies developed over time. Pascual et al. (2024) linked light intensity to salinity tolerance in hydroponic kale, finding optimal growth at 150–250 PPFD under moderate salinity (1,500–10,000 mg/L TDS), while kale at up to 17,000 mg/L TDS and below 150 PPFD concentrated micronutrients while growing less. Thomas et al. (2021) obtained the highest spinach yield in a tilapia system at the highest salinity tested, 9,000 mg/L TDS.

We have found few peer-reviewed studies that deal directly with aquaponics in the US Great Plains. These include work on micro-system aquaponics design, co-culture of lettuce and prawn, and Pak Choi grown in hydroponics (Adams et al., 2020; Choi et al., 2025; Plocek et al., 2024). In the current study, we examine NUE in aquaponics production in the southern US Great Plains using typical raw groundwater. We estimate the operational ease of groundwater-sourced aquaponics production in this region, defining operational ease as the knowledge and effort required to monitor and adjust water quality and quantity to optimize production capacity leading to profitability. One groundwater source was the Ogallala aquifer, and the other was a saltier, brackish, confined aquifer, the Dockum. The study’s novelty is the comparison of local groundwater sources using a multi-cycle mass balance. We used one coupled aquaponics experimental system per water source and followed each over three sequential crop cycles (three temporal replicates) to address the following objectives, posed as questions.

• What difference do water sources make for aquaponics production in terms of fish growth, crop growth, and operational ease? Based on preliminary experiments, we hypothesized that the combination of hard water and increased salinity would lower the growth rate of basil crops and have little effect on fish growth.
• How efficiently are nutrients being transferred and transformed from fish feed to plant uptake over multiple cycles and across varying water sources? We hypothesized that nutrient use efficiency for plants would increase with time as systems became more established. We further hypothesized that brackish groundwater salinity would reduce nutrient use efficiency in basil crops.

2 Experimental aquaponics mass balance studies

A high-level analysis of studies that involved experimental aquaponics and mass balance provides additional context. The Supporting Information (SI) provides a detailed list of these studies, which cover the last 10 years. Table 1 is a summary of conditions used. Study trends provide insight into what investigators learned from mass balance, how they designed experiments, and why they conducted an aquaponics mass balance. The system configuration is more often coupled (18) (Aslanidou et al., 2023; Cerozi and Fitzsimmons, 2016a, 2017; Endut et al., 2014; Fang et al., 2018, 2017, 2017; Gao et al., 2024; Jaeger et al., 2019; León-Cañedo et al., 2023; Maucieri et al., 2020; Outa et al., 2025; Ru et al., 2017; Wongkiew et al., 2018a, 2018; Zou et al., 2016a, 2016; Groenveld et al., 2020) than decoupled (4) (Aslanidou et al., 2023; Kalvakaalva et al., 2023; Yang and Kim, 2019, 2020), with more recent studies in the “other” category (Chen et al., 2023; Goda et al., 2025; Legarda et al., 2019; Panana et al., 2021; Strauch et al., 2018) designed around recovering nutrients from sludges and using multi-tropic configurations. The numbers of studies according to aquaponic system size are similar between those that are small (100–450 L, 12 out of 26), which would be laboratory or bench scale, and medium-sized (1,000–4,000 L, 11 out of 26), which we would consider to be at a pilot scale. We found few mass balance studies that examined aspects of mass balance at the commercial scale.

Table 1. Trends in size, purpose, configuration, and operation for mass balance studies.

A total of 26 studies (2014–25) are used in the analysis (Aslanidou et al., 2023; Cerozi and Fitzsimmons, 2016a, 2017; Chen et al., 2023; Colt and Schuur, 2021; Endut et al., 2014; Fang et al., 2018, 2017, 2017; Gao et al., 2024; Goda et al., 2025; Groenveld et al., 2020; Jaeger et al., 2019; Kalvakaalva et al., 2023; Legarda et al., 2019; León-Cañedo et al., 2023; Maucieri et al., 2020; Outa et al., 2025; Panana et al., 2021; Ru et al., 2017; Strauch et al., 2018; Wongkiew et al., 2018a, 2018; Yang and Kim, 2019, 2020; Zou et al., 2016a, 2016). All numbers in parentheses indicate the number of studies that had the corresponding characteristic.

The fish and crops used are of interest because the selection and density of each provide insight into what researchers wanted to know and what they consider to be typical and relevant to mass balance studies. Most studies used non-fruiting plants, either leafy greens (18) or herbs (4). Only one study in this set used a fruiting crop (tomato), and two used cereals. The most common planting type was deep water culture (DWC), we suspect, because it is easy to capture nutrient mass balance without losses in media, soil, or gravel. The aquatic organisms were primarily finfish, with only carp and tilapia reported, and a small number of studies using crustaceans, algae, and bacteria as study organisms. The range of stocking densities is large (1.25–45.0 kg/m³). Considering study purpose, we find that approximately half the studies were directly interested in mass balance for nutrient utilization (10), with a smaller number using mass balance to understand questions of environmental impact, such as mercury accumulation or nitrous oxide emissions (León-Cañedo et al., 2023; Wongkiew et al., 2018b), or system configuration questions, such as those concerning co-culture, bioflocs, or CAP vs. DCAP (Aslanidou et al., 2023; Gao et al., 2024; Goda et al., 2025; Outa et al., 2025).

It is important to examine the scales used and the reasons for their use in a study’s design. Colt et al. (2022) emphasize the value of careful and independent replicates in aquaponics studies to answer research questions with higher certainty. In this group, the majority (19) were conducted as independent replicates with a single aquatic organism rearing system tied to a single plant growth system. As experimental systems increase in size, it is more difficult to conduct replication, and so we observed a semi-replicate system used in four studies, in which the aquatic organism rearing tank was shared with multiple plant growth systems. In these four studies, the system configuration used was decoupled, which means that the plant replicate experiments shared a common source water, as most decoupled systems are only run in “flow forward” mode (Gao et al., 2024; Kalvakaalva et al., 2023; Yang and Kim, 2019, 2020). What is common across nearly all of these studies is the temporal replication method used. Most studies (19) used a single crop cycle. There can be differences in nutrient use in aquaponics as a system ages and moves past startup. In our work, we endeavored to achieve both independent spatial conditions (multiple experimental systems) and temporal replication (multiple cycles) to obtain a more robust conclusion that would be relevant to the design of larger-scale experiments.

3 Methods

3.1 Experimental and aquaponics system design

The region for our work is the Texas Panhandle, and it is of interest to understand how local water resources might be used in aquaponics. There are two major aquifers that are part of public and private water supplies: the very large and well-studied Ogallala aquifer, an unconfined aquifer that has been the traditional water supply source for the region, and a confined aquifer, the Dockum. The Dockum is frequently described as brackish, and its salinity varies over its extent. Details on water quality differences according to timing and location in this region of the United States for both aquifers are provided in earlier work (Howell, 2021). We collected raw groundwater from municipal supply wells in Hereford, TX, to examine the effects of source water quality compared with those of a reverse osmosis (RO) water supply provided to our laboratory from a source within the research building. We constructed three aquaponics bench-scale growth systems, and we operated them in the same way over three basil crop growth cycles except for the difference in source water.

Figure 1 provides the details of the experimental design, with additional technical specifications available in Supporting Information (SI). One purpose of the experiment is to understand the effects of source water quality differences; we had to ensure that everything except the water quality would be identical. The nominal system volume was the same (250 L), ranging from 241 to 250 L depending on the exact placement of pipe inlets and outlets. We used 2.4 and 3.0 N HCl for pH adjustments as pH consistently trended upward rather than downward in our regional groundwater. Previous experience has shown that these water sources are deficient in iron, and Dockum groundwater is deficient in Ca and Mg. Using water quality testing beforehand, we knew we needed to add CaCl₂, MgCl₂, and chelated Fe. For the basil crop, we grew basil from seed to the seedling stage (roots clearly extending from the coconut coir) in net pots (diameter 5 cm) with coconut coir under T5 grow lights. We transferred them at a planting density of 55.8 plants/m² which did not change throughout the course of the experiment. We grew each crop under artificial lighting for 12 h/day, with Mylar reflective sheets blocking outside light. We wanted photosynthetic photon flux density (PPFD) to be at 100 µmol/m²-s, which required us to raise the lights as the plants grew to keep this intensity at the canopy level. A single crop cycle was intended to last approximately 30 days. We wanted to achieve three temporal replicates across crop cycles, and this meant that the experiments were to run for approximately 90 days. The choice of three temporal replicates represented a balance among difficulty, statistical clarity, and what we thought would demonstrate meaningful changes over time for monitoring. Based on previous experience with koi, we decided to monitor fish mass and fish feeding rates in automatic feeders to keep the fish feed rate consistent across all replicates at 2% of fish mass per day. We obtained fish at small sizes from Koi Fish USA (Pennsylvania, USA) at lengths of 10–15 cm and placed them in tanks to achieve an initial stocking rate of 3.0–3.5 kg/m³. We selected this stocking rate as so to achieve a final stocking rate of 15–20 kg/m³ at the end of 90–120 days of growth. We found that this final rate was a fair balance between what is typical in aquaponics experiments in general and the feed-to-area ratio (FAR) we could achieve in our laboratory space. The density also fell within the typical values seen in similar studies as provided in Table 1.

We chose to use koi fish for this study because the High Plains has little experience with food fish in aquaculture. Because koi fish are sold at a higher price and can be shipped in small quantities easily, this type of fish is more amenable to the adoption of aquaponics in our region. Similarly, basil is a crop that is relatively easy to grow and has good revenue potential for local markets because it is already in demand in this region.

Table 2 shows the details of the volume-based design of each aquaponics replicate. More than half of the volume is in fish-rearing and sedimentation tanks. The target flow rate was 1 L/min, which yielded a system hydraulic residence time of 4.1 h or 5.5 exchanges/day.

Table 2. Nominal size and volume information for aquaponics water units.

Physical systems varied slightly from these values. The design flow rate used throughout all experiments was 1 L/min.

3.2 Operations and management.

3.2.1 Luminosity

We used four T5 fluorescent light bulbs for plant light on a frame that measured 4 ft × 2 ft (1.22 × 0.61 m, Hydrofarm Agrobrite FLT24). We hung the lights on a PVC frame that allowed height adjustment and used Mylar blankets to reflect light for more even coverage. We measured the photosynthetic photon flux density (PPFD) with a Phantom PHOTOBIO Advanced Quantum PAR Meter. We measured within the photosynthetically active region (PAR) at five points throughout the grow bed and adjusted the height of the grow light so that the center point had PPFD ranging from 100 to 200 µmol/m²-s. The spatially averaged PPFD for each water source across all measurements was (mean ± SD) 91.6 ± 29.3, 89.8 ± 30.2, and 95.7 ± 32.8 µmol/m²-s for RO, Ogallala, and Dockum water sources, respectively. The photoperiod that we chose was 12 h, which yielded an intended daily light integral (DLI, across all nine water-source–cycle combinations) of 3.23–5.10 mol/m²-day. A PPFD of 100 µmol/m²-s can limit basil yield. It is at the lower end of what is commonly used (Pennisi et al., 2020; Dou et al., 2018). To determine bed luminosity at a point in time, we averaged across five points in the bed (corners and center). There was considerable variation in these spatial measurements, with a coefficient of variation (CV) of 41%–58% for the combinations, but the variation was greater when Mylar reflective sheets were not used. We took these measurements 5–6 times per cycle, which corresponded to one measurement every 5–6 days. We provided additional details on PPFD measurements in the SI.

3.2.2 Water changes and flows

We measured flows using a time–volume method commonly used for manual flow-rate measurements. We logged system volume using the depth of water in each system’s sump, and for every measurement (approximately every 3 days), we recorded the volume and added water when the level was low. Frequently, the hydraulic conditions around the submersible pump caused flow rates to decrease below 1 L/min. We had to adjust them often. The mean flow rate that we were able to achieve over the three experiments was 0.87, 0.89, and 0.90 L/min for the RO, Ogallala, and Dockum water sources, respectively. We tracked the amount of water we added along with new depths after each addition. We interpreted the water loss as evapotranspiration (ET) using the total grow-bed area (0.43 m²).

3.3 Measurements and monitoring

3.3.1 Water quality

We measured typical water quality parameters, including electrical conductivity (EC, µS/cm), water temperature (°C), and pH using a handheld Hanna brand probe (no. HI98129 Meter) which we calibrated weekly for EC and pH. We conducted these measurements approximately once every 3 days in each system, obtaining two measurements each time, one from the grow bed and one from the fish tank.

3.3.2 Nutrients

We measured nutrients in two ways. The first was a full water ion profile that we obtained from Servitech Labs, Amarillo, TX, USA. The constituents measured included nutrient speciation for nitrogen; major, secondary, and minor nutrients; and carbonate species. Due to time and expense, we could only obtain these measurements 1–2 times for each basil growth cycle. In addition to these measurements, we used HACH reagent kits with a handheld spectrophotometer to measure nitrate-nitrogen (NO₃-N, mg/L, HACH 10020 NitraVer X), ammonia nitrogen (NH₄-N, mg/L, HACH 10023 AmVer Salicylate), and phosphate (PO₄, mg/L, HACH 8114, Reactive P) at intervals of every 3 days. We used method blanks and sample duplicates to ensure precision and the absence of outside influences in every analytical batch. We did not filter water samples. All reported nutrient concentrations in water therefore may include suspended particulate matter alongside the dissolved phase.

3.3.3 Room conditions

We placed portable temperature and relative humidity (RH) loggers (HOBO model Onset MX1101) in the room to record at 1 min intervals. We also placed a large round stainless steel pan (diameter 35.6 cm) in the room and measured evaporation over time by emptying the pan periodically to weigh the water. We used these pan evaporation data to distinguish water loss from open-water surface evaporation in the aquaponics systems from that due to plant transpiration.

3.4 Mass balance and nutrient use efficiency

Equations 1 and provide the overall balance for any given nutrient over a production cycle irrespective of its speciation. The first equation shows the accumulation (or depletion) term in system water from time 1 (t₁, production cycle start) to time 2 (t₂, production cycle end), a period of approximately 30 days. The mass accumulation terms can be explained by inputs (fish feed, miff) and sink terms (fish waste, mifw, fish mass gain, mifg, growth of basil crop, mibc) as indicated in Equation 2. There are no water-based system outputs because we drained no water. There is the potential for gas-based emissions (most likely N species, denoted as migas). We did not measure gas flux from the experimental systems and effectively neglected this mass transport pathway.

(1) macc,iw=mt2,iw−mt1,iw

(2) macc,iw=miff−mifw−mifg−mibc−migas

The individual terms on the right side of Equation 2 require further explanation regarding how we calculated them for individual inputs and sinks in Equations 3–5

(3) miff=mffCiff

(4) mifw=mfwCifw

(5) mibc=mtopCitop+mrootCiroot

Because every term in the mass balance except those that we neglected is specified, the final form of the mass balance is fully specified. We solved these mass balance equations for a residual for each element. We explain the presence of a quantifiable residual (> 25% of fish feed inputs) as either measurement error or the presence of an unmonitored process. The final residual (r_i) that we calculated to examine mass balance closure is presented in Equation 6 as raw mass and in Equation 7 as the fractional residual (fir). A positive mass residual means that mass of element i in the fish feed was unaccounted for in sinks. A negative mass residual indicates that more mass was found in the sink compartments than was added in the fish feed during a production cycle.

(6) ri=−macc,iw+miff−mifw−mifg−mibc

(7) fir=rimiff

3.5 Statistical analysis and data presentation

We conducted data visualization and statistics using MS Excel (MS 365, Version 2509) or Minitab (version 21.4.3). We provided most summary statistics as means with standard errors, standard deviations, or 95% confidence limits to show variation. We presented water-source-based data with three temporal replicates (same water source, different growth cycle). There was no physical replication of water sources—that is, no two systems were run in the same configuration with the same water source. We did not have the resources available for this level of replication at the aquaponics scale that we chose (V_sys 250 L). Differences between treatments were based on one-way or multiway ANOVA conducted as standard ANOVA or Welch’s ANOVA (depending on equality of variances among treatments). In multiway ANOVA, we kept a treatment variable or its interaction only if the probability of a Type I error was less than 5% (p< 0.05). We evaluated post-hoc comparisons when there were differences between treatments using Tukey’s honest significant difference (HSD) or Games–Howell, depending on equality of variances. Residuals from each ANOVA model were assessed for normality using the Ryan–Joiner test and visual inspection of normal quantile plots, and homogeneity of variance was evaluated using Levene’s test with median centering.

4 Results

4.1 Production narrative

Table 3 provides the periods of growth that we achieved over three independent cycles. We intended there to be three cycles, each approximately 30 days, but the conditions of the experiment, the time required to harvest, and the need to avoid excessively disrupting fish water conditions led to the three durations shown: 38.0, 32.2, and 41.4 days of basil production. The production thus ran from the middle of summer into the middle of fall of 2023. As is the case in indoor spaces, we could not maintain constant temperature and humidity. The timing of the indoor growth period reflected some seasonality in the Texas Panhandle. For ease of comparison, all temporal data hereafter are displayed in terms of either experiment days within a cycle or days across the total continuous experiment as provided in Equations 8, 9. In the equations, tcyclen is the number of days on a current cycle n, tcaln is the calendar date-time within the cycle, tcal,on is the calendar date-time at the start of the cycle, t_exp is the number of days since the start of the entire experiment, t_exp is the current calendar date-time, and t_cal,o is the calendar date-time at which the entire experimental series began.

Table 3. Basil crop growth cycle periods and duration.

All time periods are mm/dd/yyyy hh:mm on a 24-hour clock.

(8) tcyclen=tcaln−tcal,on

(9) texp=tcal−tcal,o

4.2 Water resources

4.2.1 Balance and flow

We measured the total volume in the aquaponics systems on average every 3.5 days, which resulted in a total of 33 volume measurements over all three production cycles. Figure 2 is a summary of the temporal cumulative ET for each system according to its water source and experimental cycle. The series is informative because ET should relate to overall water demand and water use efficiency and serve as an indicator of basil crop growth. The trend reveals that the rates of ET over the course of the experiments are not substantially different from one water source to the next. The RO- and Ogallala-sourced experiments are close to one another, although it seems clear that during cycle 1, the Ogallala aquifer aquaponics system used more water than the RO experiment. In contrast to these, the brackish Dockum aquifer aquaponics system lagged in water demand, and the gap only increased over time. The difference in ET at the end of cycle 1 between Dockum and the fresher water sources was 14–20 cm, while this gap increased to 35–40 cm. The pan evaporation data also indicated the relative contributions of evaporation from the aquaponics tanks in the room (evaporation, E) and water loss attributable to crops (transpiration, T). The mean evaporation rate during cycles 1 and 2 was 0.19 cm/day (Table 4) while it decreased during cycle 3 to 0.15 cm/day. We examined T as a fraction of ET over the experiment, and it was notable that the transpiration fraction, based on cumulative T (cm), did not vary much over an individual cycle but certainly varied depending on water source, with RO, Ogallala, and Dockum showing transpiration fractions of 0.86, 0.87, and 0.82, respectively. Thus, in this indoor environment, 82%–87% of the overall water loss was due to plant transpiration.

Table 4. Summary of evapotranspiration water loss in aquaponics water sources per cycle.

4.2.2 Water quality

Table 5 and Figure 3 presents temporal trends in nutrient species and aggregate water quality parameters. Because these were newly established systems, early nitrogen dynamics were prominent: ammonia-N peaked during initial biofilter colonization before transitioning to predominantly NO₃-N. By the end of cycle 1, nitrate had stabilized, consistent with approximate equilibration between fish excretion and plant demand. Phosphorus behaved differently: PO₄-P rose continuously in the RO and Ogallala systems across all three cycles, from ∼0.5 mg/L in cycle 1 to 10–18 mg/L by cycle 3. The Dockum system diverged beginning around day 20 of cycle 1, remaining below the other two until late in cycle 3 when concentrations reconverged. The reason for lower PO₄-P in Dockum may be PO₄ precipitation due to the slightly higher pH.

Table 5. Summary statistics (mean ± 95% confidence) according to water source and cycle.

Capital letters (A, B, C) indicate differences within a parameter for the entire experimental period (all). Lowercase letters (a, b, c) indicate differences among groups based on cycle and water source. The significance threshold for Welch’s ANOVA was p = 0.05. The cycle × water source interaction was significant only for EC (mS/cm).

**Indicates statistically significant differences when all cycles are grouped together. These are differences by water source across all cycles. *Indicates two-factor differences observed by cycle and water source.

Water temperature did not differ significantly (p > 0.05, standard one-way ANOVA) among sources, reflecting the shared indoor environment. Despite periodic manual HCl additions, a statistically significant pH ordering persisted throughout: Dockum (7.57) > Ogallala (7.37) > RO (7.11), driven by the higher natural alkalinity of the groundwater sources. By the final 20–30 days of the experiment, we were able to keep pH within the target range of 7.0–7.5. EC remained stable in the RO and Ogallala systems (1.0–1.8 mS/cm across all cycles) but rose continuously in the Dockum system, reaching an overall mean of 2.53 ± 0.46 mS/cm—significantly higher than both other sources (p< 0.05, Welch’s one-way ANOVA). This rise reflects the Dockum water’s elevated Na (430 mg/L vs. 99 mg/L Ogallala and< 0.5 mg/L RO), SO₄, and Cl, none of which were diluted or removed during the experiment.

Table 6 provides detailed nutrient concentrations. All three source waters entered the experiment with negligible NO₃-N and PO₄-P; therefore, all dissolved inorganic nitrogen and phosphorus observed during production therefore originated from fish. The RO system, beginning from a near-zero nutrient baseline, effectively isolates fish-derived additions: the dominant contributions were P, K, SO₄-S, and Mg. The Ogallala source contributed substantial background Ca, Mg, and SO₄-S relative to RO, while the Dockum added high SO₄-S (76 mg/L in source water), Na, and Cl. Across all three water sources, P and K accumulated monotonically throughout the experiment, while NO₃-N stabilized after the initial cycle.

Table 6. Nutrient concentrations (all in mg/L) for aquaponics (n = 5) and groundwater (n = 1).

Statistics shown are means ± 95% confidence limits using a t distribution. In any instance in which a compound was below the detection limit, we substituted the detection limit for that value.

4.3 Plant production

4.3.1 Plant growth and water use

During cycle 2 (see Figure 4), final dry masses per plant ranged from 14.0 to 16.2 g, corresponding to estimated fresh (wet) masses of 140–162 g at 90% moisture, at a planting density of 55.8 plants/m². Area-based dry productivity in that cycle ranged from 32.6 to 37.7 g/m². Individual basil plant dry masses at harvest are summarized in Figure 5, with grow-bed totals and water use efficiency (on a whole-plant dry basis and a plant-top wet basis) in Figure 6.

A mixed-effects ANOVA on individual plant total, top (shoots), and root dry mass showed statistically significant effects of cycle, water source, and their interaction (p< 0.05, Welch’s ANOVA, Games–Howell pairwise comparisons). Each combination of water source and cycle comprised 24 individual specimens. Post-hoc analysis revealed that the three lowest shoot dry mass yields were for Dockum in cycles 1 and 3 and Ogallala in cycle 2 (indicated as group c in Figure 5B). The lowest root dry mass was for Ogallala in cycle 1 and Dockum in cycle 3. Thus, the lowest shoot yield in cycle 3 (6.57 g) was significantly lower at the level of individual plant specimens than the plant-top yield in the RO-fed water source in cycle 3. The variation in plant-top mass in the Ogallala water source in cycle 3 was large, making it indistinguishable from Dockum in cycle 3.

Grow-bed-level totals (one value per water source per cycle) showed trends as well. The RO system gained mass continuously across all three cycles, doubling from cycle 1 to cycle 2 and adding a further 26% in cycle 3. The Ogallala system increased threefold from cycle 1 to 2 and then stabilized. The Dockum system increased 2.6-fold from cycle 1 to 2 but then declined in cycle 3, returning to near cycle-1 totals. Top-mass fraction (green vegetative material as a proportion of total plant mass) was lower in cycle 1 across all sources, with a modest increase by cycle 3 in the RO and Dockum (from 0.516–0.604 in cycle 1 to 0.666–0.716 in cycle 3).

Water use efficiency (WUE) was low in cycle 1 across all water sources (28.8–48.1 g dry whole-plant/1,000 L; 185–302 g wet top mass/1,000 L) and increased for RO and Ogallala in cycles 2–3. The Dockum system showed persistently low WUE in cycles 1 and 3, with only partial recovery in cycle 2. WUE ranged from 361 to 615 g wet top mass per 1,000 L across all cycles and water sources. Because water consumed per cycle was similar regardless of water source or treatment (CV = 2.4%, mean = 241 ± 5.8 L), WUE tracked grow-bed mass gain closely.

4.3.2 Plant tissue analysis

Leaf/stem and root tissue compositions are presented in Table 7, with published reference ranges for basil leaves in Table 8. No nutrients fell below the recommended leaf ranges in any treatment. Nutrients consistently exceeding recommended concentrations (denoted [+]) were K in all three water sources (5.0%–5.7% vs. recommended 1.55%–2.05%), Mg in all three sources (1.18%–1.89% vs. 0.60%–1.00%), Fe in all three sources (205–336 mg/kg vs. 75–200 mg/kg), and Zn in all three sources (150–278 mg/kg vs. 30–70 mg/kg). Phosphorus was elevated in the RO and Ogallala treatments but within range in the Dockum.

Table 7. Nutrient element concentration in basil (a) leaf and stem tissue (top portion of plant) and (b) roots summarized over three growth cycles.

Each set represents n = 5 samples. There was one sample for cycles 1–2 and three samples in cycle 3. All values are given as means ± 95% confidence limits using the t distribution. Nutrients with letters indicate a significant difference shown using one-way ANOVA for water source (p< 0.05), with differences between groups indicated by different letters using Tukey’s honest significant difference (HSD).

Table 8. Recommended concentrations of nutrients in healthy basil leaves (Bryson et al., 2014).

Among water source differences in leaf tissue, Mg was significantly higher in the Ogallala treatment (1.89% ± 0.50%) than in RO or Dockum (p< 0.05). Boron was elevated in the Dockum treatment (54.2 ± 3.33 mg/kg) relative to RO and Ogallala (39–41 mg/kg), consistent with the higher B concentration in Dockum groundwater. Leaf Na showed the expected highest value in the Dockum treatment (20,900 ± 3,260 mg/kg), but RO plants accumulated significantly more Na than Ogallala plants (12,300 vs. 8,330 mg/kg), a result not predictable from the respective EC values alone.

Observed leaf NPK mass fractions averaged 3%–5% N, ∼1% P, and 5%–6% K across all treatments, compared to a published optimum of approximately 6.2:1:2.2. The principal deviation was excess K, which exceeded the recommended range in all treatments and yielded N:K mass ratios of 0.5–0.8, well below the recommended value of ∼3. N concentrations (3.7%–4.1%) were near or slightly below the recommended minimum of 4.0% but were not associated with visible deficiency symptoms.

Root tissue had lower concentrations of most macronutrients and secondary nutrients than leaf/stem tissue. The exceptions were Fe, Cu, and Mn, which had higher concentrations in roots across all treatments (e.g., Fe: 2,450–2,880 mg/kg roots vs. 205–336 mg/kg shoots; Mn: 92–211 mg/kg roots vs. 32–65 mg/kg shoots). Boron in roots followed the same water-source ranking observed in leaves, with Dockum significantly elevated (52.8 ± 6.88 mg/kg) relative to RO and Ogallala (34–40 mg/kg). The Na ranking in roots (Dockum > RO > Ogallala) was numerically consistent with leaf tissue, though greater inter-cycle variability in root samples reduced statistical significance.

4.4 Fish production

4.4.1 Fish growth

We summarized fish growth data and feed metrics in Figure 7 (additional details in SI). The numbers of individual fish were 8, 9, and 5 for RO, Ogallala, and Dockum, respectively, from experiment day 15 onward. One fish died in the Dockum system around that time. There was a small pH upset on day 3 that killed four fish. Because this occurred early in the experiment, we moved fish between systems to equalize fish mass among systems. This shifting of animals explains the drop in fish mass during the first 5 days.

Fish experienced an initial mass decline upon transfer to the aquaponics systems, recovering to starting mass by days 20–25 of cycle 1. Growth rates during this initial period averaged 1.37–1.97 g/day. Following acclimatization, growth rates increased substantially: 3.76–6.15 g/day in cycle 2 and 4.47–5.36 g/day in cycle 3. Total mass gain over the full experiment was similar across the RO (435 g), Ogallala (434 g), and Dockum (394 g), though the Dockum system accumulated progressively less mass relative to the other two beginning in cycle 2. We could not statistically examine differences in fish growth by water source because there was only one replicate for each source. However, visual examination of fish mass suggested a difference in growth rate for the Dockum water source. In cycles 2–3 (38–115 d, n = 13 time steps), the Dockum growth rate was 4.33 ± 1.01 g/day (mean ± SE), while Ogallala and RO were 7.02 ± 2.74 and 4.99 ± 1.61 g/day, respectively.

Feed-to-area ratio (FAR) increased across cycles in proportion to growing fish biomass, rising from<10 g/m²-day in cycle 1 to 14–20 g/m²-day in cycle 2 and 19–24 g/m²-day in cycle 3. Because we fed at a target of 2% of current live mass per day, FAR necessarily tracked fish mass over time. Achieving a consistent 2%/day delivery was difficult with the automatic rotary feeder, as indicated by the scatter in the values.

Feed conversion ratio (FCR) was not statistically different across water sources (p > 0.05, n = 3 cycles/water source). Numerically, FCR was lowest (most efficient) in cycle 2 (1.29–1.56 g feed/g fish across sources), consistent with the steeper growth observed in that cycle. The Ogallala system had the highest FCR across all cycles (overall mean 1.99 g feed/g fish) compared to Dockum (1.71) and RO (1.68), though the small sample size precludes strong conclusions about water-source effects on feed efficiency.

4.4.2 Fish waste

Fish waste generation is summarized in Table 9, with elemental composition in Figure 8. Across all cycles and water sources, we recovered approximately 10% of dry feed mass as solid fish waste in the gravity sedimentation basin. The Dockum system produced less waste than the other two sources in all cycles. FCR was comparable in Dockum and RO, whereas it was higher in Ogallala. Conversion efficiency cannot explain the difference in waste generation. It is possible that poor settling of fish waste in the brackish Dockum solution was due to a high quantity of monovalent Na⁺ relative to multivalent Ca²⁺ and Mg²⁺. Na+ is known to be a poor flocculator. The mean sodium adsorption ratios (SARs) in aquaponics water for RO, Ogallala, and Dockum were 15.1, 7.17, and 28.7, respectively. Higher SAR promotes deflocculation but is counteracted by higher ionic strength (greater EC). Thus, differences in flocculation are a possible explanation that would require further examination using jar settling tests.

Table 9. Fish waste generation during experiments.

Note that all values for the first cycle were based on estimates using the relationship between fish mass and feed mass from the other two cycles. The %/day of animal mass is based on the average mass over the period it represents.

Indicates the fish waste was estimated using multiple linear regression relationship from other data based on the following equation: waste (g) = 0.0707fish mass (g)+0.0486*feed mass (g)-11.22 g.

We measured elemental composition of fish waste for cycles 2 and 3 (cycle 1 waste was not analyzed). Total P exceeded total N as a dry-mass fraction in all samples, consistent with the greater solubility of nitrogen compounds and their preferential loss to the aqueous phase. N content in the fish waste increased from cycle 2 to cycle 3 by approximately 0.75% (∼25% relative increase) across all three water sources, while total P remained constant at a mean of 4.32%. Among secondary nutrients, Ca was the dominant constituent at a mean of 7.41%, exceeding Mg and S (both near or below 1%), and accounted for much of the ash fraction (28.4–36.6% of dry mass).

We found that micronutrients B, Cu, Fe, and Mn were strongly concentrated in fish waste relative to their concentrations in feed and water. Sodium partitioned substantially into fish waste sludge across all water sources, with Dockum waste containing the highest Na mass fraction consistent with its elevated source-water Na. Iron ranged from 2,430 to 9,230 mg/kg in fish waste across water sources and cycles, with lower concentrations found in cycle 3; however, because total sludge mass was greater in cycle 3, the mass of Fe sequestered may have been comparable across cycles. The progressive concentration of micronutrients in fish waste and their accumulation in the sedimentation basin has implications for long-term nutrient availability to crops, discussed further below.

4.5 Nutrient mass balance

4.5.1 Closing the mass balance

We provided the average residuals with error bars showing the 95% confidence limits across all water source–cycle combinations for each nutrient in Table 10. The fractional residual is the mass residual divided by the mass in the fish feed quantity over a cycle. The raw signed and absolute residuals are also provided. We achieved the lowest fractional residuals with nitrogen (abs fraction mean 18.1%), calcium (43.3%), and boron (92.1%). The worst residual was for sodium (332%).

Table 10. Mass balance residual analysis statistics by element on a water source-cycle basis.

Ranks are based on the elements with the lowest absolute fractional residuals. Fractional residuals are based on the mass residual divided by the amount of each element in the fish feed during a production cycle. Residuals labeled as “in balance” are those meeting the criterion of an absolute fractional residual less than or equal to 25%.

We also wanted to examine the trends in mass residuals. To do this, we examined the nine mass balance residuals for each element according to the absolute fractional residual (|fir|). We used a threshold value of 25%. If the absolute fractional residual is less than or equal to 25%, we called this case “in balance”. Otherwise, we classified the case according to the sign of the signed fractional residual, either positive or negative. Using this process, the table shows that the mass closure was in balance for N approximately half the time (44%), and the rest of the time it was negative. For the other macronutrients (P, K), the mass balance always had a negative residual. The interpretation of the residual sign is that a negative residual means that nutrient accumulation in sinks (fish growth, plant growth, fish waste) or in water exceeded the measured nutrient inputs from fish feed. Thus, for P and K, this means that P and K were either overreported in terms of the amounts utilized by the production system or under-reported in the fish feed. If the residual is positive, this means that more of an element was found in the fish feed than in sinks or as accumulation in the water compartment. This occurred in the case of most micronutrients. Errors in the mass balance can occur from incorrect measurements of media quantity (water volumes, fish mass, fish waste), element concentrations in compartments, or both. It is not clear how to determine which case might be occurring for each nutrient here. It was not possible to close the mass balance in these experiments for many of the nutrients except N, Cu, and Mg. Regardless, for the remainder of the analysis, we analyzed the mass in compartments relative to one another rather than in an absolute sense.

4.5.2 Nutrient use efficiency

The overall mean NUE values for fish and basil are provided in Table 11. We examined differences among source waters for each basil plant NUE, but we found no statistical differences (p > 0.05, n = 3 cycle/water source), and so we present only temporal trends. The NUE values for N, P, and Ca in the fish were sometimes > 100% due to large negative mass residuals in these nutrients. We sampled tissue from only one koi fish to determine the elemental composition of the animal. If we had sampled one animal for each system, the mass residuals and thus the NUE for these elements may have been more clearly< 100%. Despite this limitation, there were some trends in the fish NUE. The highest NUE values were for N, P, Ca, and S, which would seem to correlate well with the production of animal tissue. Considering the differences in water quality, the Ogallala source water had higher Ca and Mg, and the Dockum water had far more Na. Out of all nutrients, the Dockum water source had the greatest value nominally. Concerning all nutrient as a group, the Dockum water source had the greatest value nominal NUE compared to the other water sources. The lowest fish growth rate and the lowest fish mass were observed in the Dockum, and this may explain a higher nominal NUE for fish.

Table 11. Nutrient use efficiency of fish and plants within each source-water type and across all experiments.

Each value is a mean over three cropping cycles.

The plant NUE was lower than fish NUE for nearly every nutrient except Mg, B, and Fe. The highest plant NUE values were, in descending order, B > Fe > Mg > K. The lowest plant NUE values were for N, P, and S. Considering that N and P are macronutrients, it is discouraging that their NUE values are so low, but these low values are consistent with other CAP experiments. The ranges provided in the literature review in Table 1 for N, P, and K were 15%–30%, 2%–10%, and 12%–28%, respectively. Our experimental values (all cycles) ranged 2.19%–9.48%, 2.06%–9.14%, and 10.2%–47.5%, respectively. In this respect, plant N use efficiency is low, P use efficiency is typical, and K use efficiency is high. The low NO₃-N concentration (7–14 mg/L on average, Table 6) may explain the low N use efficiency. Looking at temporal trends, the NUE increased for several nutrients over time in the RO water source (N, P, S, and Fe), while K decreased. There were no consistent temporal trends in the Ogallala and Dockum water sources. Instead, cycle 2 frequently had the highest plant NUE. This high cycle 2 NUE was especially pronounced in the Dockum experiments for K, Mg, B, Fe, Zn, and Na. The pH was held consistent over time and across all three experiments. However, the salinity, expressed as EC (mS/cm) in Figure 3, showed nearly a doubling in the Dockum compared to the other water sources. This increase put stress on the basil crops, driving lower plant nutrient efficiency in the third cycle.

4.5.3 Nutrient media distribution

Figure 9 provides nutrient distributions in one of four media—fish tissue, fish waste, plant root, and plant top—by water source and by cycle. Rather than accounting for all the nutrients in this analysis, the fractions used here consider only where nutrients end up in solid-phase sinks. For macronutrients (NPK), the largest sink for N and P was clearly fish tissue. There was a substantial amount of P in fish sludge, as has been observed by others (Lobanov et al., 2021; Yang and Kim, 2020; Cerozi and Fitzsimmons, 2017). For K, 30%–50% was present in the plant, with the majority in the top rather than the roots. The remainder was in fish tissue, with only a small portion in the fish waste. For the secondary nutrients, Ca and S had the largest fractions in fish tissue, with sizable fractions in fish waste. In contrast, Mg was distributed as 30%–50% in the basil plant, 20%–30% in fish waste, and 20%–40% in fish tissue. These fractions did not change substantially by cycle except for K and Mg. For the RO system, K was much higher in the basil plants in cycle 1 than in cycles 2–3, and Mg showed higher capture in cycle 2 in the Dockum experiment compared with lower amounts in cycles 1 and 3.

Looking at the micronutrients, a substantial portion of B (30%–80%) ended up in the fish sludge, while Cu, Fe, and Mn were consistently 70%–80% in the fish sludge. Zn showed a more even distribution between fish sludge and fish tissue, while Na was distributed as 30%–50% in fish tissue, 10%–50% in sludge, and 10%–30% in plant tissue. The basil crops did not appear to be deficient in the nutrients that were heavily bound up in fish sludge. Nevertheless, the fraction in the sludge remained relatively consistent, while the concentrations of many micronutrients increased by cycle 3, as shown in Figure 8. This indicates that these sludge-bound micronutrients may never become available for plant use if this production continues unchanged.

5 Discussion

It is important to know the difference that the water source makes to fish and crop production because alterations to the water source require treatment technology, which increases costs. This study considered fresh water (< 1,000 mg/L TDS, Ogallala aquifer 1.07 mS/cm EC, 696 mg/L TDS) and brackish water (1,000–10,000 mg/L TDS, Dockum aquifer 1.33 mS/cm EC, 865 mg/L TDS). Dockum aquifer source water did not reach > 1,000 mg/L TDS at our source, but it is broadly considered to be brackish based on samples we tested (Howell, 2021). We saw that salinity in the brackish-water coupled system increased gradually over time, while the fresh and RO systems reached a relatively constant EC of approximately 1–1.5 mS/cm, compared with 3.5 mS/cm in the brackish system. Avdouli et al. (2021) studied basil in cascade hydroponics and showed that it has an upper threshold of 5 mS/cm for salinity stress. Thus, salinity alone should not have reduced basil yield as much as we observed. Interestingly, the freshwater Ogallala source water did not behave very differently in salinity buildup to RO water.

Chloride (Cl^-) and sodium (Na⁺) were primary drivers of salinity increase over most other constituents. Chloride was higher in the Dockum source AP water than in the Ogallala (120 vs. 37 mg/L in our regional characterization), and by the end of the experiment, the system concentrations were ordered as Dockum (568 mg/L) > Ogallala (284 mg/L) > RO (220 mg/L). Two caveats attach to this interpretation. First, we did not measure chloride content in plant, fish, or sludge compartments; our expectation that chloride is removed only minimally through these sinks rests on its high solubility and weak biological sequestration (McCarter et al., 2019; Svensson et al., 2012). Second, chloride was not supplied by the source water alone—our pH and nutrient amendments (HCl, CaCl₂, MgCl₂) also added chloride, and because the Dockum source was deficient in Ca and Mg, these chloride-bearing salts were added to the Dockum treatment while no such salts were added to the other two treatments. A calculation of the amount of Cl^- added to each system from additions shows that the added amounts were 40%–50% of the final value found in the water at the end of the experiments (see SI for more details). Na⁺ accumulated to a much higher concentration in the brackish groundwater system (final concentration 500 mg/L) than in the fresh (175 mg/L) and RO (178 mg/L) systems. The higher source concentration (430 mg/L in brackish vs. 99 mg/L in fresh water) supports the argument that the brackish source did drive the rise in salinity over time. The increase in salinity over time was therefore partly operational (nutrient addition) and partly a source-water property (naturally high Na⁺ and Cl^-). With those caveats, the mechanism remains directionally consistent: chloride that is neither taken up nor precipitated will accumulate in any fully coupled system until water is drained to remove it. In addition, high sodium, while not conserved in the water column, was not absorbed by system sinks (fish, fish sludge, and plants). A system merely using brackish water, rather than one designed for halophytes, will therefore require more frequent water-waste events. Because we operated at 0% water waste, we did not test this directly, but such events would be expected to lower overall nutrient use efficiency—a cost that, without additional capital investment in water treatment, appears unavoidable for salinity control.

The brackish water affected basil crop growth more than fish growth, reducing basil yield and visual quality. There were statistically significant effects of water source, cycle, or their interaction (p< 0.05) on both the total plant dry mass and the top (harvested portion) dry mass. At the level of whole-grow-bed totals, which we did not evaluate statistically (small sample size, n = 1 per water source), the RO and Ogallala water sources showed either increasing or stable yield across cycles, and the Dockum system increased from cycle 1 to cycle 2 before declining in cycle 3 in both yield and visual appearance. Converging lines of evidence—the whole-bed totals, the lower water use efficiency, the lagging evapotranspiration, and the statistical significance of growth cycle with water source—point in the same direction. The basil plants were independent of previous cycles because each cycle began with starter plants grown in the same way; a cycle-3 decline must therefore reflect accumulated system conditions rather than carryover in the plants. In this last cycle, salinity and Cl^- and Na⁺ concentrations were highest, so the rising salinity background of brackish groundwater contributed to reduced basil performance. We cannot, however, attribute this to source-water salinity alone: Dockum water also had a significantly higher pH (7.61, above the 7.5 threshold at which nutrient availability begins to decline), so salinity and pH are confounded across the treatments. The amount of water consumed was also lower in the brackish water, which we attribute to lower basil growth. The fish, by contrast, showed no significant differences in growth rate by water source (p > 0.05, n = 3 cycles), although the Dockum fish had numerically lower mass and began to lag in cycles 2 and 3.

One leaf tissue finding that did not follow the expected source-water salinity pattern is Na accumulation. Dockum-grown plants had the highest leaf Na (20,900 mg/kg), as expected. Less expected was that RO-grown plants accumulated significantly more leaf Na than Ogallala-grown plants (12,300 vs. 8,330 mg/kg), despite the RO source containing essentially no Na and the Ogallala source containing 99 mg/L. Because Na concentrations in the water column were broadly similar between these two treatments across cycles, the differential uptake may reflect competition from other ions: the higher Ca and Mg background in the Ogallala system may have partially suppressed Na uptake at the root, though we did not measure ionic competition directly. This finding warrants attention in future work.

A valuable use of mass balance in a study where all system components are actively measured is the assessment of nutrient use and its efficiency. We measured low plant nutrient use efficiency for the major nutrients N and P compared with that in other studies. The NUE was highest in the second cycle for the brackish groundwater and then diminished with increasing salinity. The plant NUE was higher for micronutrients and secondary nutrients of importance (B, Mg, and Fe). The nutrient use was related to two factors. One was how efficiently the plant could take up the nutrient from the water in its dissolved form. The other was whether the nutrient was present in the dissolved phase at all. Several nutrients (N, P, Ca, and S) efficiently partitioned to fish tissue, while others (B, Cu, Fe, Mn, and Zn) preferentially concentrated in fish waste or accumulated in the water column. These two phases frequently held > 50% of the nutrients in sink (non-water) compartments. Nutrients held in fish tissue are generally beneficial because fish are desirable as food or, in this case, for their aesthetic value. The areas in which there are opportunities for increased circularity are nutrient buildup in the water column and in fish waste. Phosphorus built up in all systems over time based on water concentrations, and it remained constant at approximately 4 mass% in fish waste. Nitrogen remained at approximately 3–4 mass% in fish sludge across all cycles, while NO₃-N reached a relatively steady concentration of 10–15 mg/L in the water in cycle 2, which continued into cycle 3 except in the brackish-water system, where it dipped to 6–9 mg/L. Taken together, this means that plants were exposed to a relatively constant concentration of NO₃-N once the systems became established (after cycle 1) and a rising concentration of PO₄-P. There was more PO₄-P for plants to take up, but they did not do so. The rate of NO₃-N uptake seems constant given that the concentration in water did not change greatly. We did not measure any gaseous emissions. The primary gaseous emissions we would expect would be N₂O and N₂ from denitrification. Zou et al. (2016a) have noted denitrification losses indicating more complete denitrification to N₂ than to N₂O at higher pH. Because the residuals for our nitrogen mass balances were negative for every cycle (average -33.7%), this indicates that a gaseous loss of nitrogen was possible. Denitrification requires a lower redox potential, pointing to the possibility that it could occur in areas of our systems where there was locally lower dissolved oxygen (Zhang and Furman, 2021). Places where lower DO occurred would be areas of high chemical oxygen demand (COD), such as the vicinity of some biofilms and settled fish solids, and flow dead zones where compressed air was not introduced as efficiently as in well-mixed areas (Timmons et al., 1998). Future work involving tracers may help to characterize such issues and reduce gaseous nitrogen loss (Recsetar et al., 2020).

Beyond overall NUE values, the pattern of nutrient accumulation in plant tissue suggests that K availability was disproportionately high relative to N across all three water sources. Observed leaf NPK mass fractions averaged 3%–5% N, ∼1% P, and 5%–6% K—compared with a published optimum of approximately 6.2:1:2.2—yielding N:K mass ratios of approximately 0.5–0.8 versus a recommended value of ∼3. K rose monotonically in the water column throughout the experiment and exceeded the recommended leaf range in all three treatments. That plants did not show visible signs of K excess or pronounced N deficiency may suggest tolerance within this range, but it raises the potential hypothesis of whether the nutrient dynamics of fish-fed aquaponics systems—where K contributions from feed and waste are not balanced against crop demand by any deliberate amendment—structurally tend toward K excess relative to N. This pattern would be worth monitoring in longer-duration experiments or DCAP configurations.

Plant tissue micronutrient data warrant additional discussion. Fe, Zn, and Cu exceeded recommended leaf concentration ranges in all treatments, and Fe, Cu, and Mn were more concentrated in root tissue than in shoots across all water sources (e.g., Fe: 2,450–2,880 mg/kg roots vs. 205–336 mg/kg in shoots). This pattern of root accumulation with limited shoot translocation is characteristic of hydroponic systems operating at elevated pH, where Fe³⁺ and other polyvalent cations tend to form hydroxide or carbonate precipitates at or near the root surface. Despite measured leaf Fe being above recommended ranges, we found visual chlorosis consistent with Fe deficiency in Ogallala-grown plants during cycle 2. We interpret this as a functional Fe deficiency: Fe was reaching the root zone and accumulating in root tissue but could not move effectively to the shoot. This pattern—chlorosis with adequate or elevated total leaf Fe—has been reported elsewhere and reflects Fe that is present but physiologically inactive rather than absent (Langenfeld and Bugbee, 2025; Wang et al., 2020; Kasozi et al., 2019). Root Fe accumulation alongside shoot chlorosis suggests that Fe was immobilized in the root apoplast and inefficiently loaded into the xylem, with the residual Fe that reached the leaf remaining unavailable. Contributing mechanisms reported in the literature include bicarbonate-induced apoplastic pH buffering that inhibits Fe(III) reduction and precipitation of Fe-phosphate in the root apoplast under elevated P (Shi et al., 2018; Rediske and Biddulph, 1953; Liu et al., 2023). The subsequent addition of iron chelate in cycle 3 increased water-column Fe accumulation substantially, suggesting that chelated forms were more plant-available than the inorganic species present previously. Taken together, this indicates that total measured tissue Fe—or even water-column Fe concentration—may be a misleading indicator of Fe nutritional status in coupled aquaponics systems where pH is chronically above 7.0. Future work should consider measuring Fe speciation or using chelated micronutrient additions from the outset in systems like these, particularly when using groundwater sources that are naturally alkaline.

Even if more PO₄-P had been released from fish sludge, as is often suggested, the basil would not have used it. Given that fish sludge—its removal, handling, degradation, and mineralization—is an area of substantial operational control in aquaponics, it is worth examining how it might be used more beneficially within the aquaponics growth system rather than removing it for beneficial use or disposal elsewhere. In our study, it is not clear that different fish waste management practices and nutrient solubilization from fish waste would have made a significant difference in overall productivity. There are three areas, however, where this might be explored. One is better alignment of the total amount of fish waste with the crop-growing area. Our FAR ranged from 7 to 24 g/m²-day, which is a typical or even low range (Colt et al., 2022). Greater plant area would lower the FAR below what is typical. What would differ from other studies if the FAR were lowered in this way is that more nutrients would be reclaimed from fish waste due to an increase in plant receptors. Another way to make use of excess P in this system might be to harvest basil more frequently. We selected 30 days as a typical value. A commercial operation might be able to harvest more frequently and still obtain a quality product. The second area of further exploration is nutrient ratios. We did not attempt to create any recommended or strategic balance of nutrients. Such a balance should be easier to achieve in a DCAP system. Yet even in this case, if the nutrients are to come from aquatic animal rearing, then ratios are largely determined by the native ratios in the fish-rearing tank. We could strategically mix the separate crop loop in a DCAP with a fish-rearing stream to adjust nutrient ratios. Yet it is much easier to use these two streams (fish and crop loops) to control pH and EC. It may still be valuable to recover nutrients from fish waste to return to either CAP or DCAP, but the best management strategy for obtaining a benefit that would justify the additional design and cost is unclear.

A third possibility for P utilization is to examine P precipitation in suspended particles rather than fish sludge. Others have reported that P precipitation (P combined with Ca and Mg) is a real sink that might be considered (Cerozi and Fitzsimmons, 2016b; Yang and Kim, 2020). Some P found in fish sludge may consist of settled solid precipitates. The fate of P precipitate depends on its particle size and the mixing conditions in the aquaponics system. If suspended P, perhaps as colloids, had been present within our water samples, we would not have known this because we analyzed whole (unfiltered) water. The rising P in the water column we report may therefore represent finely suspended particles. Management of pH to prevent this precipitation and keep P dissolved and bioavailable might therefore have increased its plant utilization. We could not find any aquaponics studies to date that have examined the fate of P as colloidal suspended material using direct measurements. Such examinations may provide additional insight into the overall mass balance of P which will AID AP nutrient circularity in general.

6 Conclusion

In this study, we examined important aspects of water balance, nutrient use, and plant–fish behavior according to the type of water source used. The water balance indicated that consumptive water use was linked to plant growth, with water use efficiencies of 361–615 g wet/1,000 L and a daily aerial ET of 0.86–1.6 cm/day. Water source type produced no strong effect on fish but had a statistically significant effect on basil. Converging measures—whole-bed yield, WUE, ET, and visual quality—validate the reduction in basil performance in the brackish water source shown by replication with multiple plant specimens. WUE was lower in the brackish groundwater source, reflecting salt-stressed growth, of which the Na-driven nutrient antagonism is one expression. Na was far higher in the aquaponics water and basil plant tissue in the brackish-sourced water. The nutrient use efficiency did not differ by water source for any nutrient, but NUE did frequently show noticeable differences over time. We conclude that NUE changed over time and that conducting a single-cycle experiment would thus have been misleading about its dynamics. Possible reasons for changes in NUE over time include system establishment, rising salinity, and changing room temperature. The source water, through salinity, the associated accumulation of Cl- and Na+, and the rise in pH, appears to influence the quantity of water used and the yield and quality of basil. Any means of increasing basil yield or quality under high salinity, such as blending or strategic aquaponic water draining, would make the use of marginal-quality brackish water for aquaponics more feasible.

The nutrient mass balance in this study was difficult to close within ± 25%, and the fractional partitioning of nutrients differed substantially among nutrients. The nutrient use efficiency for macronutrients (N, P, and K) was high in fish but low in plants, except for K. Improved transfer of N and P from fish feed to plants would be beneficial given the cost of chemical fertilizer and the circularity benefits frequently attributed to aquaponics. Losses of N to fish sludge were not high, whereas losses of P were. Elemental analysis of dried sludge showed that little N but a substantial fraction of P was retained in settled solids. Higher utilization of P may be possible, but pursuing this depends on the cost–benefit of sludge P mobilization techniques. A decreased fish-to-plant ratio (i.e., more plants) may help increase P utilization, or further examination of dissolved-phase nutrient ratios or P precipitate colloids may help address this.

These data show that koi and basil crops can grow together in an indoor environment in the Texas Panhandle using local water supplies, including water sources that are brackish or moderately saline. These sources may require treatment or blending. Growing herbs or other high-value crops in this region would use water efficiently and reduce groundwater withdrawals relative to commodity crop production. Growers could pair koi fish and other freshwater fish with high-value leafy greens or herbs to provide a coupled agricultural output. Future research should examine the marketability of high-value koi and another high-value crop at scale. Both coupled and decoupled aquaponics are reasonable options for implementing this approach, but larger-scale demonstrations will be necessary to validate our findings.

Correction note

This article has been corrected with minor changes. These changes do not impact the scientific content of the article.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. We used Claude Opus 4.8 to aid in editing the manuscript and checking for internal consistency.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/faquc.2026.1914289/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Isabela Pinheiro, Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research (AWI), Germany

Reviewed by: Nawroz Kareem, Keele University, United Kingdom

Radhakrishnan Subramanian, United Arab Emirates University, United Arab Emirates