Identifying key factors controlling potential soil respiration in agricultural fields

Soil respiration is one of the main soil health indicators and is influenced by several factors in agricultural fields. Identifying key factors that control soil respiration is desirable for informed soil management decisions and for promoting and scaling up soil health. This study aimed to (i) quantify the relationships between potential soil respiration and selected soil properties, crops, and slope positions, and (ii) identify key factors controlling these relationships using a neural network model. Ninety soil samples from 0‐ to 5‐ and 5‐ to 20‐cm soil depth were collected from footslope, backslope, and summit in three fields planted with soybean (Glycine max L. Merr.), alfalfa (Medicago sativa L.), and corn (Zea mays L.). The model provided great accuracy (coefficient of determination: 0.96; root‐mean square error: 7.8; and mean absolute deviation: 3.8) and explained nearly 96% of variations in soil respiration across soil depth, crop, and slope positions. Soil depth, ammoniacal nitrogen (NH4‐N), crop types, slope position, and silt content were identified as the top five factors influencing potential soil respiration at the field level. Potential soil respiration was more sensitive to potassium, phosphorus, pH, cation exchange capacity, and mean weight diameter and less sensitive to NH4‐N, nitrate nitrogen, soil organic matter, and clay content. It increased with pH, electrical conductivity, mean weight diameter, potential nitrogen mineralization, and potassium, and it decreased with increasing silt content. Soil from 0 to 5 cm under soybean or at the summit slope position exhibited a higher respiration. Using a small dataset, this pilot study accurately predicted potential soil respiration in agricultural fields and identified key drivers controlling it. The results from this study highlight the complexity of using potential soil respiration as a standalone test for evaluating soil health. This does not diminish the usefulness of potential soil respiration as a soil health indicator to support agricultural management decisions and as a reference in future soil health studies. However, it emphasizes the importance of considering multiple factors when interpreting the significance of soil biological indicators for soil health assessments.

decisions and as a reference in future soil health studies.However, it emphasizes the importance of considering multiple factors when interpreting the significance of soil biological indicators for soil health assessments.

INTRODUCTION
Soil respiration is an energy-yielding process that measures the amount of carbon dioxide (CO 2 ) released as a result of soil organic matter (SOM) decomposition by soil microbes, as well as respiration from plant roots and soil fauna (Anderson, 1983; USDA-NRCS, 2014), and it is a primary path of CO 2 enrichment in the atmosphere (Schlesinger & Andrews, 2000).Soil respiration is considered one of the major biological indicators of soil health, representing living organisms and their activities (Schloter et al., 2003;Sciarappa et al., 2016;Soil Health Institute, 2018), and reflects carbon (C) and nutrient cycling in soils, without which many essential elements would not be available to plants (Prosser, 2007).Soil respiration can be measured in situ (Rochette & Hutchinson, 2005) or in the laboratory under a controlled environment (Anderson, 1983).The most reliable and repeatable methodology to measure soil respiration, which is comparable to the C mineralization potential of soils, is long-term incubation studies under laboratory conditions (Zibilske, 1994).Most recently, the C mineralization potential of soils has been identified as one of the top three soil health indicators to be measured for soil health assessments (Bagnall et al., 2023).
The relationships between soil properties like soil respiration and its influencing factors are complex, and machine learning algorithms can be used to quantify such relationships (Adhikari et al., 2022).Identifying the drivers of soil respiration is crucial for food security, ecosystem service delivery, climate change mitigation, and agricultural sustainability (Doran & Zeiss, 2000;Kibblewhite et al., 2008;Lal, 2016), and it also has implications for agricultural management decisions.However, there is not much literature about the drivers of soil respiration and their interrelationship from agricultural fields where farm decisions are made, and the present study aims to fill this gap.Using laboratory data and field observations coupled with machine learning, this study will quantify the relationship between soil respiration and its controlling factors at field level, which will assist in soil health promotion and agronomic decisions.Our primary objectives were (i) to quantify the relationship between potential soil respiration over a 28-day incubation period and soil properties, crop type, and slope position in agricultural fields, and (ii) to identify the key controlling factors of respiration.

Study site
The study was conducted in three adjacent fields located in the eastern corn belt plain in northeast Indiana (Figure 1a).The fields were conventionally plowed and were in an 8-year rotation of wheat (Triticum aestivum L.), corn (Zea mays L.), soybean (Glycine max L. Merr.), oat (Avena sativa L.), and alfalfa (Medicago sativa L.).The slopes of the three fields varied from 6% to 20%, with some closed depressions (Smith & Livingston, 2013).The soils are derived from compacted glacial till with the predominate soil series being the Glynwood silt loam (Fine, illitic, mesic, aquic Hapludalfs) and the Blount silt loam (Fine, illitic, mesic, aeric Epiaqualfs) (Soil Survey Staff, 2014).

Soil sample collection and laboratory analysis
Forty-five soil sampling locations (15 in each field) were selected based on major field slope positions (summit [SU], backslope [BS], and footslope [FS]).At each slope position, five sampling sites were randomly selected for a composite soil sample (Figure 1b).Soil samples were collected from 0to 5-cm and 5-to 20-cm soil depths and analyzed for aggregate stability expressed as mean weight diameter (MWD), particle size distribution (% sand, silt, and clay), soil pH, electrical conductivity (EC), SOM, nitrate nitrogen (NO 3 -N), ammoniacal nitrogen (NH 4 -N), total inorganic nitrogen (T-Inorg-N), available phosphorus (P), potassium (K), cation exchange capacity (CEC), potential nitrogen mineralization (Pot-N-Min), and soil respiration.The MWD was determined by wet sieving following Kemper and Rosenau (1986), percent sand, silt, and clay using the Malvern Pipette Method (Malvern Worcestershire), and soil pH and EC (1:2 soil-water ratio) (Thomas, 1996) with an Accumet Research AR50 Dual Channel pH and Conductivity Meter.Loss on ignition method was used to determine SOM (Nelson & Sommers, 1996) and NO 3 -N and NH 4 -N (1:10 soil-2 M KCl ratio) (Bremner & Keeney, 1966) using the Lachat Quikchem 8500 Series 2 Flow Injection System (Lachat Instruments).Samples were extracted with Mehlich III solution (Ross & Ketterings, 1995) to determine P and base cations, while CEC was determined by summation.Similarly, Pot-N-Min was determined aerobically over a 28-day period, as described in Drinkwater et al. (1997).
Soil respiration was determined using a 28-day incubation method by measuring CO 2 release (Broadbent & Nakashima, 1974).Samples were placed in sealed glass jars along with a base trap containing KOH, which were changed seven times (days 1, 3, 7, 10, 14, 21, and 28) throughout incubation.The base traps were acidified with 4 N hydrochloric acid, and CO 2 was measured (mg CO 2 -C kg −1 ) with a Varian CP-3800 gas chromatograph (Varian Inc.).

Data analysis and modeling
Data analyses included descriptive statistics, correlation among variables, and predictive modeling using the JMP software (JMP Pro, Version 17), while terrain analysis, geomorphons, and classifying slope positions were performed using ArcGIS and GRASS GIS.All measured soil properties together with soil depth, crop type, and slope position were considered as independent variables or predictors that could influence potential soil respiration in the study area.To avoid model overfitting from possible multi-collinearity among predictors, the linear relationship between soil respiration and measured soil properties was assessed with Pearson's correlation coefficient (r), and highly correlated predictors (r > ±0.72) were excluded from the modeling.A feed-forward artificial neural network (ANN)-based deep learning model with two hidden layers was employed to predict potential soil respiration using the selected predictors.The ANN is a form of artificial intelligence analogous to the biological nervous system, and it can be used to solve complex and nonlinear problems as in the soil systems (Levine et al., 1996;Zhao et al., 2009).A typical ANN model architecture consists of an input layer, one or more hidden layers, and one output layer where the predictions are stored.Hidden layers are located in between input and output layers where mathematical functions are applied generating different weights and biases, and an activation function to produce an output.In the present study, the model parameter learning rate was set to 0.1, the penalty method set to be absolute, and the fitting was accomplished on transformed predictors with

Core Ideas
• Potential soil respiration is influenced by soil depth, nutrient levels, and slope positions.• A neural network model accurately predicted and identified key controllers of potential soil respiration.• Potential soil respiration was highly sensitive to changes in mean weight diameter, CEC, potassium, and phosphorus.
robust fit to minimize the impact of potential outliers in the data.The model was iterated 100 times, and the prediction performance was evaluated using the coefficient of determination (R 2 ), and root mean square error (RMSE) calculated on a 10-fold cross-validation set with each fold repeated five times (Adhikari et al., 2014) and with mean absolute deviation (MAD), which is the average absolute difference between each prediction from its mean (Equation 1).The contribution of each predictor was calculated from variable importance standardized as percent relative importance (RI) to identify key drivers of potential soil respiration.Furthermore, the response of potential soil respiration change due to changes in soil physicochemical properties was also evaluated using sensitivity analysis.
where X i is the measured value at the ith location, µ is the mean, |X i -µ| is the absolute deviation, and N is the sample size.

Soil respiration measurements
Average potential soil respiration for the study area was 517 ± 178 mg CO 2 -Ckg −1 with a CV of around 35% (Figure 1c).It was higher at 0-to 5-cm depth (588 mg CO 2 -C kg −1 ) compared to 5-to 20-cm depth (446 mg CO 2 -C kg −1 ).The highest respiration was measured from soils under alfalfa (542 mg CO 2 -C kg −1 ), followed by soybean (535 mg CO 2 -C kg −1 ), and corn field had the lowest respiration (472 mg CO 2 -C kg −1 ).Soil respiration under soybean had the maximum CV (38.5%).The footslope had the highest potential respiration (530 mg CO 2 -C kg −1 ), followed by the backslope, and the summit had the least potential respiration (493 mg CO 2 -C kg −1 ).There was a significant difference in potential respiration by soil depth (p <0.0001), but no significant difference was observed between crop types (p >0.29) and landscape positions (p >0.42).

Key factors affecting soil respiration
Among 14 measured soil properties, Pot-N-Min had the highest positive correlation (r = 0.41) with potential soil respiration followed by soil pH (r = 0.37), MWD (0.32), CEC and K had the strongest negative correlation (−0.18), and clay and P had the weakest correlation of all (Figure 1d).Sand (0.12) and SOM (0.16) had positive, and silt had a negative (−0.13) correlation with potential soil respiration.Sand content was negatively correlated with all measured properties except for soil respiration and pH, whereas SOM, EC, and MWD were positively correlated with all the measured properties except for percent sand content.To reduce the effect of multi-collinearity among predictors, Sand and T-Inorg-N were removed as they were highly and significantly correlated, respectively, with Silt (r = −0.76),Clay (−0.81),NO 3 -N (r = 1.0), and EC (r = 0.72) (Figure 1d).
Figure 2 shows model performance by crop types, slope position, soil depth, and RI of the variables in predicting potential soil respiration.The ANN model was robust and was able to predict potential soil respiration with high accuracy (R 2 = 0.96; RMSE = 7.8; and MAD = 3.8 mg CO 2 -C/kg).Potential soil respiration from 0-to 5-cm soil depth, under soybean crop, and at the summit position were better predicted compared to 5-to 20-cm depth, or other crops and slope positions.Overall, the model explained nearly 96% of variations in potential soil respiration by crop types, >81% by soil depth, and up to 94% by slope positions (Figure 2a).The model identified soil depth, NH 4 -N, and crop type as the top three influencing predictors (>11% RI), followed by slope position, K, and Pot-N-Min.Soil pH, NO 3 -N, and P had 6%-9% RI, whereas EC had the lowest RI of 3% among all predictors.The MWD, clay, CEC, and SOM had less influence than soil pH.The model also indicated that silt and K affected potential soil respiration with their corresponding RI of about 9%.Thus, the model results predicted that soil nutrient contents were a major influence on potential soil respiration.As reported by Liang et al. (2021), increasing nutrient levels increased microbial activity in soils and had a positive response to fertilization (Chen et al., 2017;Liang et al., 2019;Liu et al., 2016).However, Jong et al. (1974) reported decreased respiration due to mineral nitrogen addition, and we also observed a negative correlation between soil respiration and NO 3 -N and NH 4 -N.Similarly, changes in soil pH, C:N ratio, available P, and differences in local topography impacted microbial diversity, activities, and respiration (Gurmessa et al., 2021;Liu et al., 2007;Xu et al., 2020).In a separate study, Jong (1981) reported the influence of slope position on soil respiration, as also observed in the present study.
This research found potential soil respiration responded to pH, CEC, Pot-N-Min, MWD, and soil K and P. Soil respiration increased with MWD, EC, pH, K, P, and Pot-N-Min, but it decreased with silt content.Although NH 4 -N had a higher RI than other predictors (except for soil depth), soil respiration was less sensitive to changes in NH 4 -N than it was for NO 3 -N and SOM.The soils on these fields have relatively high SOM varying from 1.5% to 12%, which can partially explain the lack of soil respiration sensitivity to NH 4 -N changes.However, findings from other research indicate inconsistencies in the relationship between NH 4 -N and microbial respiration (Jackson & Caldwell, 1993;Salonius & Mahendrappa, 1975;Thirukkumaran & Parkinson, 2000).Increased NH 4 -N in soil has been found to suppress soil microbial respiration perhaps due to changes in soil pH, which can reduce the availability of soil carbon (Thirukkumaran & Parkinson, 2000), while in other instances, additions of NH 4 -N have increased respiration rates (Davies et al., 2017).High spatial and temporal heterogeneity of soil microbial respiration, as found by Scott-Denton et al. (2003), driven by climate and soil microenvironment further adds to the complexity of modeling potential microbial respiration response to changes in NH 4 -N.We observed an increased response on respiration for 0-to 5-cm soil depth, summit slope position, and soybean crop compared to 5-to 20-cm depth and other slope positions and crop types (Figure 2c).Soil respiration has been recognized very early as an indicator of soil fertility (Birch, 1960;Gainey, 1919).More recently, soil respiration has been identified as one of the major soil health indicators (Haney et al., 2018;Soil Health Institute, 2018) and suggested as a helpful indicator for monitoring short-term soil health changes (1-4 years) following the adoption of conservation practices (Crookston et al., 2023) and other soil management practices such as crop diversity, tillage, and application of soil amendments (Williams et al., 2020).Most of the proposed methods utilize short-term (24 h to a few days) soil incubation (Haney et al., 2018;Williams et al., 2020).However, the dynamic spatial and temporal nature of soil respiration found in our study and pointed out by other research (Kuzyakov & Blagodatskaya, 2015) highlights the importance of considering multiple factors when interpreting the significance of soil biological indicators or soil health and functioning.

CONCLUSIONS
On the basis of this study, the following results were obtained: • Potential soil respiration was positively correlated with Pot-N-Min, pH, and MWD and negatively correlated with CEC, K, and silt content.• The ANN model accurately predicted potential soil respiration (R 2 : 0.96; RMSE: 7.8, and MAD: 3.8) and explained nearly 96% of the measured variability across soil depth, crop type, and slope position.• Soil depth was a key variable influencing potential soil respiration, followed by NH 4 -N and crop types, and EC had the lowest importance among all predictors.• Potential soil respiration was highly responsive to K, P, pH, CEC, and EC but was less responsive to SOM, NO 3 -N, NH 4 -N, and clay content.• The response of potential soil respiration to soil depth and crop type was in the order 0-5 cm > 5-20 cm, and SO > AL > CO.

F
Soil sample locations in the study area (a and b), histogram and box-plot of potential soil respiration measurements (c), and its correlation with predictors with corresponding p-value [−log 10 (p-value)] (d).Res., potential soil respiration; CEC, cation exchange capacity; EC, electrical conductivity; MWD, mean weight diameter; NH 4 -N, ammoniacal nitrogen; N-Min, potential nitrogen mineralization; NO 3 -N, nitrate nitrogen; SOM, soil organic matter; T-In-N, total inorganic nitrogen.

F
Measured versus predicted soil respiration from different slope positions, crop type, and soil depth (a), the relative importance of predicting variables (b), and a profiler graph showing the response of respiration due to changes in predictor values (c).Dotted lines represent a reference value on either axis; the purple triangle is a sensitivity indicator showing the partial derivation of the profile function at its current value.AL, alfalfa; BS, backslope; CEC, cation exchange capacity; CO, corn; EC, electrical conductivity; FS, footslope; MWD, mean weight diameter; NO 3 -N, nitrate nitrogen; NO 4 -N, ammoniacal nitrogen; Pot-N-Min, potential nitrogen mineralization; RMSE, root-mean square error; SO, soybean; SOM, soil organic matter; SU, summit.
Conceptualization; formal analysis; investigation; methodology; software; validation; visualization; writing-original draft.Kelsey R. Anderson: Data curation; formal analysis; writing-review and editing.Douglas R. Smith: Conceptualization; funding acquisition; project administration; resources; supervision; writing-review and editing.Phillip R. Owens: Conceptualization; writing-review and editing.Philip A. Moore Jr.: Writing-review and editing.Zamir Libohova: Writing-review and editing.A C K N O W L E D G M E N T SUSDA is an equal opportunity provider and employer, and mention of trade names in this study does not imply endorsement by USDA.