Spectrometry for better soil fertility management in Abbay River Basin

Abstract

The aim of this study was to determine how soil bunds affected important soil characteristics based on the soil's reflectance in Aba Gerima, an Ethiopian midland of the Abbay River Basin. Soil samples were collected from 0 to 30 cm of the soil depth in 48 plots for soil texture, pH, organic carbon (OC), total nitrogen (TN), available phosphorus (av. P), and available potassium (av. K). Using a spectroradiometer, we evaluated the reflectance of soils that had been air-dried, pulverized, and sieved. To locate and forecast the change in soil parameters, we employed regression modeling. We identified and predicted soil properties using the models, which were evaluated by the coefficient of determination (R²) and root mean square error. Soils from bunded plots had less reflection. Higher amounts of clay (61%), OC (2.30%), av. P (13.72 ppm), and av. K (116.55 mg/kg) were found in bunded plots. For calibration data, R² was 0.93, 0.83, and 0.82 for clay, sand, and OC contents, respectively. Spectroradiometry could therefore complement conventional soil analysis. More study is needed to enhance spectroscopy's prediction performance.

Abbreviations
• av. K
• available potassium
• av. P
• available phosphorus
• OC
• organic carbon
• R²
• coefficient of determination
• RMSE
• root mean square error
• TN
• total nitrogen

INTRODUCTION

Soil quality is the ability of a specific kind of soil to function, within natural or managed ecosystem boundaries, to sustain plant and animal productivity, maintain or improve water and air quality, and support human health and habitation (Karlen et al., 1997). A soil quality is fitness for soil use (Letey et al., 2003). Soil quality also refers to the ability of a particular kind of soil to function, within natural or managed ecosystem boundaries, to sustain plant and animal productivity, maintain or improve water and air quality, and support human health and habitation (Soil Science Society of America, 1995). Assessments of soil quality frequently recenter on identifying a “minimum data set” of soil properties that have the influential impact on soil quality.

Knowing and quantifying the regional variation of soil characteristics that are important for crop development is necessary for understanding soil quality (De Marins et al., 2018). The soil qualities are widely distributed in Ethiopia (Kassawmar et al., 2018). Sustainable land management (SLM) techniques, such as soil bunding, change essential soil properties and thereby soil functions (Berihun et al., 2020). SLM practices improve soil quality, reduce soil deterioration, and increase crop outputs (Reichert et al., 2022). Therefore, understanding the state and qualities of soil's main nutrients requires the characterization and assessment of spatial variations of soil properties.

Characterizing and evaluating geographic variation in soil parameters is necessary to understand soil quality (De Marins et al., 2018). For location-based soil and crop management, information on soil fertility is essential (Rosemary et al., 2017). To quantify soil fertility, people frequently look at crop yield and production, plant indications, and the texture and color of the soil (Karltun et al., 2013). The majority of soil suggestions are not site specific, and soil fertility improvement frequently comes with general advice. Utilizing updated soil data can prevent nutrient loss and low crop production.

In Ethiopia, a large number of soil samples are collected and examined utilizing pricy and cumbersome traditional analysis techniques to identify soil characteristics essential to managing soil and crops. The nation's use of conventional procedures, however, is severely constrained by the high cost of chemicals and the ineffectiveness of conventional laboratories. Thus, soil research has increasingly used visible-near infrared (VNIR) reflectance spectroscopy (Stenberg et al., 2010). This is because it solves the drawbacks of wet chemistry, such as being non-destructive, requiring no chemicals, analyzing data fast, and being applicable in the field (Liang, 2004). Furthermore, it just needs a small sample of soil and can identify numerous soil characteristics in a single scan (Rossel et al., 2006).

Remote sensing approaches, including airborne satellites, might be useful for dividing the landscape into more or less homogeneous soil-landscape units. Besides, soil parameters can be determined by analyzing remotely sensed data using empirical or physically based approaches (Ben-Dor et al., 2008). It has also been employed for soil survey, mapping, and quantitative soil property assessment (Ben-Dor & Banin, 1994, 1995). In general, by eliminating the need for large, time-consuming, and expensive field surveys, remote sensing techniques make it easier to map inaccessible locations.

Physical and chemical soil research can be replaced by laboratory soil spectrometry in the visible, near-infrared, and short-wave infrared (Vis-NIR-SWIR, 350–2500 nm) range (Gholizadeh et al., 2013). Contents of soil organic carbon (OC) (Vašát et al., 2017), clay (Tümsavaş et al., 2018), total nitrogen (TN), and soil texture (Vasava et al., 2019) could be predicted from reflectance. SLM practices, such as soil bunds, are crucial to ensure environmental sustainability (Bastos Lima, 2021). Bund is an engineering measure of soil conservation used for creating obstruction across the path of surface runoff to reduce the velocity of flowing water. It retains the running off water in the watershed and thus helps to control soil erosion. Bunds are simply embankment-like structures, constructed across the land slope. Different types of bunds are used for erosion control and moisture conservation in the watersheds. Bunds increase the fertility of the soil by minimizing runoff and soil erosion (Amare et al., 2013).

Low agricultural yield is a serious problem in the Aba Gerima basin due to soil erosion and diminished soil productivity. According to Amare et al. (2013), spectroradiometry was effective at quantifying soil carbon in Ethiopia's various agroecologies and soil types. However, there are no conclusions about the effects of soil bunding on soil properties in the catchment, elsewhere in Ethiopia, or Africa.

Detailed soil knowledge is needed to address the soil fertility issue. Therefore, in order to use and manage different soil types, it is crucial to grasp their properties and qualities. Furthermore, it is difficult to transmit soil and crop technologies to other regions because the majority of the current soil knowledge and expertise are retained by farmers and agricultural consultants. Soil color indicators were used in this work to develop regression models for soil properties in the watershed. We hypothesized that the soil reflectance-based color indices would provide acceptably accurate estimates of soil properties under different soil management alternatives. Therefore, the goals of this work were to determine how soil bundling affected soil texture, pH, OC, TN, available phosphorus (av. P), and available potassium (av. K) based on spectroradiometric data.

Core Ideas

• Soil bunds improved clay, silt, organic carbon, and potassium contents of soils.
• Bunded soils recorded less reflectance.
• Low soil reflectance indices show increased soil quality.
• Spectroscopy could better characterize soils as compared to convectional one.

METHODOLOGY Study area

The research was done in Aba Gerima, a midland in the Abbay River Basin of Ethiopia (Figure 1). The site is classified as a midland area (Peel et al., 2007), it is situated between 1900 and 2000 m above sea level. The research site receives annual rainfall of 1076–1953 mm, monthly maximum temperature of 27.0°C, and monthly minimum temperature of 12.6°C, according to data from 1994 to 2021 at adjacent meteorological stations (Figure 2). The year's rainfall falls between June and August).

FIGURE 1. Locations of research site and soil plots: (a) Ethiopia, (b) Aba Gerima altitudes, (c) Aba Gerima soil units, and (d) Aba Gerima land-use/land-cover types.

FIGURE 2. Cumulative rainfall and mean temperature per month from January to December at research site.

Methods Methods for soil samples collection

According to data from Sentinel-2 images, which were retrieved from https://sentinels.Copernicus.eu/, cultivated soils were sampled in 2019 (Figure 1d). In a digital elevation model with a resolution of 0.5 m, the site, which has 426 ha, was divided into three topographic classes: gentle (2%–5%), moderately (5%–10%), and strong (10%–15%) sloped. No soil bunding (WB, n = 24) and with soil bunds (SB, n = 24) were the two land management scenarios. The bunds were 5 years old and had a width of 0.8 m and a height of 0.5 m. Each plot was deliberately scattered (Figure 1d) following the slope classes. Five soil samples from each plot were randomly collected with an Edelman auger near the surface (0- to 30-cm depth).

Laboratory procedures

Soil samples (n = 48) were crushed, sieved to 2 mm, and air-dried. Later, they were examined at Amhara Design and Supervision Works Enterprise for soil texture, pH, OC, TN, av. P, and av. K. The hydrometer method was used to determine the soil's texture (sand, silt, and clay) following the destruction of organic matter (OM) and soil dispersion (Bouyoucos, 1962). The proportion of sand was computed as the difference between the amounts of silt and clay, which were measured using hydrometer. Finally, we determined soil textural classes using the USDA system's textural triangle (Rowell, 1997; Soil, 1996).

Soil pH was determined potentiometrically with a digital pH meter in a 1:2.5 (soil: water) supernatant suspension. We poured 10 g of air-dried soil and 25 mL of purified water into 100-mL beakers, stirred it for 1 min with a glass rod, and allowed it to equilibrate for 1 h before we measured the pH of the supernatant. The wet digestion method was used to measure soil OC levels (Walkley & Black, 1934). Soil OC content was determined by the wet digestion method, which entails digesting the OC with potassium dichromate in a sulfuric acid solution. The Kjeldahl method was used to estimate TN. The Kjeldahl procedure was followed for the determination of TN that follows oxidizing of the same with concentrated sulfuric acid and converting the nitrogen in the organic compounds into ammonium sulfate during the oxidation (Bremner & Mulvaney, 1982). av. P was determined by Bray II method in which av. P was determined by shaking the soil sample with extracting soil solution of 0.3 N ammonium fluoride in 0.1 N hydrochloric acid as described in Bray and Kurtz (1945). Then, a spectrophotometer was used to measure the av. P (Murphy & Riley, 1962). By using Morgan's solution to extract a soil sample and a flame photometer to measure the amount of av. K, the amount of K was determined (Morgan).

Collecting and pre-processing soil spectra

A spectroradiometer was used to measure the soil reflectance in the visible, near-infrared, and short-wave infrared (350–2500 nm) spectrum. The spectroradiometer was recalibrated against a white panel (Labsphere Inc.) every 10 min. The soil reflectance data in the Vis-NIR-SWIR (λ350–2500 nm) range were collected with a Fieldspec 4 spectroradiometer (Analytical Spectral Devices [ASD] Inc.). The field of view was set at 8°, and the distance between the trigger of the spectroradiometer's refiber optic cable and the soil specimen was held at 15 cm for all observations. Direct sunlight was used to assess reflection between 10:30 and 11:00. A white Spectralon (LabsphereR) was used for calibration. View Spec Pro v. 6.2 and Remote Sensing 3 v. 6.4 software were used to record and pre-process the reflectance. Indices of soils were also calculated as shown in Table 1. The present study performed regression analysis to investigate the association between soil parameters and soil reflectance-based indices, including brightness index, redness index, and hue index as independent variables. It is common practice to calibrate reflectance indices to forecast soil attributes based on soil color using models (Bachofer et al., 2015). Soil color often varies due to soil humus and soil water content (Schulze et al., 1993). However, soil OC, clay, and sand contents were related weakly with color indices at a 95% confidence level.

TABLE 1 Spectral color indices of soils in the study area.

Index	Formula	Index property	References
Brightness index (BI)	$$\frac{{\sqrt {{R^2} + {G^2} + {B^2}} }}{{\sqrt 3 }}$$	Average soil reflectance magnitude	Mathieu and Pouget (1998)
Hue index (HI)	$$\frac{{2 \times R - G - B}}{{{\mathrm{G}} - {\mathrm{B}}}}$$	Dominant wavelength, primary colors	Mathieu and Pouget (1998)
Redness index (RI)	$$\frac{{{R^2}}}{{B \times {G^3}}}$$	Hematite content	Mathieu and Pouget (1998)

Abbreviations: R, red; G, green; B, blue.

Data analysis

The data were examined using Pearson's linear correlation (r) analysis at p < 0.05. Statistical Analysis Software 9.4 was used to analyze variance with the Duncan method and simple linear regression. For calibration (31 soil samples) and validation, we computed the coefficient of determination (R²; Moriasi et al., 2007), root mean square error (RMSE), and Nash–Sutcliffe efficiency (NSE; Nash & Sutcliffe, 1970) as: [Image Omitted. See PDF][Image Omitted. See PDF][Image Omitted. See PDF]where n is the number of samples, ŷ is the predicted value of a soil property, $$\bar{y}$$ is the mean measured value of a soil property, and y is the observed value of a soil property.

RESULTS AND DISCUSSION Effects of soil bunding on modifications to soil characteristics

In the three slope classes, there were significant differences at p < 0.05 in soil texture, OC, and av. P concentrations between bunded and non-bunded plots (Table 2). The Aba Gerima catchment's soil pH ranged from severely acidic to moderately acidic. Clay and OM, which hold on to more basic cations, may be to blame for the higher soil pH values in bunded plots. According to Tellen and Yerima (2018), improper use of fertilizers and leaching of basic cations may be to blame for the lower pH values at non-bunded plots (Osman et al., 2013). As a result, acidity issues may damage the research area's soils.

TABLE 2 Effect of bunding on soil properties in Aba Gerima.

Treatments		pH	Sand (%)	Silt (%)	Clay (%)	OC (%)	Av. P (ppm)	Av. K (mg kg−1)
Bunded plots (n = 24)	S1B	5.62a	18.75c	20.25c	61.00b	2.30b	13.72b	116.55a
	S2B	5.72a	22.38ac	31.00ab	46.63ab	1.51ab	10.23ab	97.35a
	S3B	5.60a	18.88c	25.75ac	55.38b	1.57ab	11.03b	106.71a
Non-bunded plots (n = 24)	S1W	5.54a	33.25ab	31.25ab	35.50ac	1.22ac	10.41b	106.73a
	S2W	5.63a	35.50b	33.25b	31.25c	1.54ab	13.03b	101.30a
	S3W	5.52a	36.75b	27.50ab	35.75ac	0.90c	6.37a	94.61a
Mean		5.61	27.58	28.17	44.25	1.45	10.5	103.88
Significance		ns	**	***	***	***	*	ns

Note: S1, 2%–5%; S2, 5%–10%; S3, 10%–15%. n = 48.

Abbreviations: av. K, available potassium; av. P, available phosphorus; B, soil bund; OC, organic carbon; W, without soil bund; ns, not significant.

*, **, and ***

represent significant correlation coefficients at p < 0.05, 0.01, and 0.001 levels, respectively.

The Aba Gerima's soils vary from clayey to sandy loam (USDA, 1996). Bunding's ability to reduce soil erosion may be the reason silt and clay particles predominate in bunded plots in all three slope classes (S1B, S2B, and S3B; Table 2). On the other hand, sand can predominate in non-bunded plots on steep slopes (S2W and S3W) as a result of the selective erosion of smaller soil particles (Reichert & Norton, 2013). Similarly, higher silt and clay and lower sand contents were reported in bunded soils (Guadie et al., 2020). Values of av. P and av. K were lesser than values founded in the watershed (Ebabu et al., 2020) and the Uwite Catchment, Ethiopia.

Improper farming, clearance of crop residue, and animal dung, an increased mineralization owing to a higher temperature, leaching, and increased erosion due to rainfall are all factors that contribute to low levels of OC and av. K (Iticha & Takele, 2019). However, the buildup of fine soil particles and nutrients may be the cause of the highest amounts of clay (61%), OC (2.30%), av. P (13.72 ppm), and av. K (116.55 mg/kg) in bunded plots on mild slopes (S1B). Bunding enhanced clay and silt buildup, soil OC, P, and K levels are consistent with claims that bunding improves soil fertility (Ebabu et al., 2019).

The decreasing soil quality in the Aba Gerima watershed was explained by the pH, OC, av. P, and av. K values are below threshold levels. To improve the qualities of Ethiopian soils, it is crucial to increase soil OM values via the use of SLM techniques and biomass addition (Reichert et al., 2021).

Modeling of soil properties

Soil OC, clay, and sand contents showed almost no significant relation with color indices at a 95% confidence level. Therefore, the model was carried out to validate the performance of prediction of the soil parameters using soil reflectance. For clay, the R² was high for both calibration (0.93; RMSE = 4.51) and validation (0.96; RMSE = 3.97; Figure 3), better than values reported by Camargo et al. (2015); 0.70, Zhang et al. (2017); 0.64, and Silva et al. (2021); 0.66, and R² = 0.73 with RMSE = 5.40 (Vasava et al., 2019), R² = 0.83 with RMSE = 0.34 (Yang et al., 2020), R² = 0.62 with RMSE = 2.06 (Haghi et al., 2021), and R² ranging from 0.71 to 84 (Zhao et al., 2021).

FIGURE 3. Scatter plots of results of (upper) calibration and (lower) validation models for contents percentage of clay, organic carbon (OC), and sand. NSE, Nash–Sutcliffe efficiency; R2, coefficient of determination; RMSE, root mean square error.

For sand, we achieved R² = 0.83 with RMSE = 5.36 for calibration and R² = 0.83 with RMSE = 6.23 for validation (Figure 3). Similar values were reported: R² = 0.80 with RMSE = 3.28 (Vasava et al., 2019), R² = 0.81 with RMSE = 3.84 (Tümsavaş et al., 2018), R² = 0.90 with RMSE = 11.66 (Haghi et al., 2021), and R² = 0.56–0.71 (Zhao et al., 2021). Our performance of sand values was more precise than R² = 0.76 with RMSE = 0.92 (Yang et al., 2020) and R² = 0.77 (Rodríguez-Pérez et al., 2021).

For soil OC, we achieved R² = 0.82 with RMSE = 0.29 for calibration and R² of 0.83 with RMSE = 0.35 for validation (Figure 3) was excellent based on the R² threshold values of Saeys et al. (2005). The accuracy of the OC model is consistent with previous reports: R² = 0.84–0.93 (Leone et al., 2012), R² = 0.63–0.90 with RMSE = 6.40–0.78 (Kuang & Mouazen, 2012), R² = 0.91 (Yang et al., 2012), R² = 0.85 with RMSE = 3.77 (Amin et al., 2020), R² = 0.57–0.7 (Gomez et al., 2013), R² = 0.77–0.83 (Gras et al., 2014; Wijevardane et al., 2016), and R² = 0.764 with RMSE = 0.344 for validation (Feyziyev et al., 2016).

CONCLUSIONS

In the Aba Gerima watershed in Ethiopia, soil bunds improved the soil's clay, silt, OC, and K values. The differences in soil characteristics (texture, OC, and av. P) between bunded and non-bunded plots in various slopes were substantial. However, soils on lower slopes had higher pH values and higher concentrations of clay, silt, sand, OC, P, and K than soils on higher slopes. These results imply that for sustainable soil management in the research area, site-specific data could guide soil management interventions like soil bunding. Bunded plots' soils with less reflection increased soil quality. Soil OC, clay, and sand content were poorly explained with reflectance indices. Thus, soil quality can be indicated with its reflectance and improved with soil management under the study catchment conditions.

For highly accurate prediction of the clay, sand, and OC contents, empirical equations were created. Our findings indicate spectroscopy is a rapid, non-destructive method for soil characterization, evaluation, and monitoring. The results have suggestions for the research site's spatial administration and soil property monitoring. The creation of sustainable, site-specific soil management methods will benefit land owners and governments with the aid of spectroradiometrically based soil prediction models. However, more study is needed to determine which parts of the spectrum could enhance the spectroscopy's predictive capabilities, help soil management interventions, and investigate their impacts on other soil metrics.

AUTHOR CONTRIBUTIONS

Gizachew Ayalew Tiruneh: Conceptualization; data curation; formal analysis; investigation. Derege Tsegaye Meshesha: Conceptualization; data curation; formal analysis. Enyew Adgo: Conceptualization; data curation; formal analysis. Tiringo Yilak Alemayehu: Data curation; formal analysis; investigation. Genetu Fekadu: Conceptualization; data curation; formal analysis; investigation. Temesgen Mulualem: Conceptualization; data curation; investigation. Simeneh Demissie: Conceptualization; data curation; formal analysis; investigation; methodology. Kefyialew Tilahun: Conceptualization; data curation; formal analysis; methodology. José Miguel Reichert: Conceptualization; data curation; investigation; methodology.

ACKNOWLEDGMENTS

We thank Agerselam Gualie and Melkamu Wudu for their support during our field and laboratory work. The authors also benefited from reviewers’ and the editors’ comments.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.