Nutritional Quality of the “Algarrobo” Neltuma pallida Fruit and Its Relationship with Soil Properties and Vegetation Indices in the Dry Forests of Northern Peru

Abstract

The dry forests of northern Peru are home to extensive populations of algarrobo (Neltuma spp.). Its fruit serves as feed for goats and is used in various agro-industrial products. However, the nutritional quality can be influenced by the physicochemical properties of the soil and vegetation coverage. The objective of this study was to understand and predict the concentration of protein and ether extracts of carob and evaluate its relationship with soil properties and vegetation indices. Principal component analysis (PCA) and correlation analyses were conducted. The prediction of protein and ether extract was carried out using the Eureqa-Formulize software 1.24.0. In the PCA, protein showed a positive relationship with ash and ether extract but a negative relationship with moisture. Likewise, moderate correlations were observed between protein and ash content (0.51). Protein also showed positive correlations with pH (r = 0.19), BI (r = 0.22), and NDSI (r = 0.22). Additionally, the ether extract exhibited correlations with sand content (r = 0.22), Ca²⁺ (r = −0.26), Cu (r = −0.20), S5 (r = 0.26), and Si (r = 0.24). Protein predictions showed moderate performance (CC = 0.73 and R² = 0.53), as did ether extracts (CC = 0.68 and R² = 0.46). These findings contribute to a better understanding of the factors that influence the nutritional quality of carob and can be used for the development of sustainable management strategies in the dry forests of northern Peru.

1. Introduction

The tropical dry forest represents 50% of the forested areas in Central America and 22% in South America [1]. In Peru, individuals of the genus Neltuma, known as carob trees, are found in the dry forests of La Libertad, Lambayeque, Piura, and Tumbes [2,3] and play a fundamental role in the ecological balance of this biome [4]. Carob trees contribute to the control of water erosion, improve soil fertility, favor the survival of local fauna, and regulate the climate [5]. They also play key roles in soil bioremediation and constitute a valuable timber and fodder resource for local communities [5,6].

Carob fruits, known as “algarroba”, are an essential food resource for humans and animals in arid and semi-arid regions [7]. In 2023 and 2024 in Peru, the estimated production of algarroba reached 445 and 341 tons, respectively [8]. The algarroba is essential for feeding cattle and goats in the dry forest [9], particularly during prolonged periods of drought. Due to its importance in the local diet and economy, optimizing carob production and improving its nutritional quality is crucial to meet the growing demand of an expanding population [10,11]. However, studies evaluating the relationship between soil composition and the nutritional quality of carob are limited, leaving a knowledge gap on how edaphic conditions can influence the concentration of the proteins and fats of this fruit. Soil properties have been shown to significantly affect the quality of various crops, such as blueberries, tomatoes, cotton, sugarcane, and citrus, through their impact on nutrient uptake and storage of biochemical compounds in fruits [12,13,14,15,16,17].

The analysis of soil properties and environmental variables has an important impact on fruit quality, but there are limitations due to the high costs of the analyses and the difficulty of covering large areas of land [18]. In this context, new techniques such as remote sensing and vegetative index analysis have emerged as efficient alternatives to assess and manage soil quality as well as its impact on crops [19,20]. Although remote sensing methods have certain limitations, such as capturing soil heterogeneity and depth, as well as spatial spectral resolution, in addition to calibration and validation requirements [21,22], their advantages make them indispensable tools for agriculture and crop management.

Protein estimation by remote sensing is essential for managing crop nutritional quality [23]. Satellite imagery has been used to predict protein concentrations in rice and wheat crops [24,25]. Like the previous applications, satellite images have also been used to estimate soil parameters such as salinity, organic carbon, and pH, using vegetation indices [21,26]. It is established that the variation in reflectance between the visible, near infrared (NIR), and shortwave infrared (SWIR) bands is most used to determine specific soil characteristics, such as salinity. In the case of carob, remote sensing studies have been developed to analyze the change in land use coverage, using vegetation indices [27], as well as to model the current and future potential distribution of the Neltuma habitat [28].

Despite the importance of carob in dry forest ecosystems, few studies integrate the analysis of the nutritional quality of the fruit with soil characterization and vegetation indices. In this context, the present study aims to evaluate the influence of soil physicochemical properties on the composition and nutritional quality of carob, through chemical analysis and data obtained from satellite images, with which we seek to predict the content of protein and ether extract. The results are expected to provide key information for designing sustainable management strategies for carob, promoting the preservation of the dry forest, enhancing ecosystem services, and improving the well-being of local communities that depend on this resource.

2. Materials and Methods

2.1. Study Area

The study was conducted in five provinces of the Piura region, Talara, Sullana, Piura, Morropón, and Sechura, in February and March 2024, located at the following geographic coordinates: −4.956° latitude, −80.53° longitude. In the productive areas surrounding the forests, crops such as rice, corn, bananas, and fruit trees like mangoes are grown. This area is characterized by a mean annual precipitation of 49 mm and a maximum daily temperature of 34.5 °C, recorded in March [9]. Like other regions on the northern coast of Peru, Piura is also affected by the El Niño-Southern Oscillation (ENSO), with an accumulated rainfall of 2500 mm [29]. This precipitation refills the aquifers and is vital for the development of dry forest and carob trees. The selection of sampling points (SPs) focused on areas with carob presence and goat grazing activities. These areas were identified based on the studies of [30] for the distribution of carob and [31] for the location of localities with a high presence of goats. Once the goat paddocks were located, a 4 km buffer was generated around them, which was intersected with the distribution of carob trees. In the areas resulting from the intersection, random SP were established using the QGIS 3.34.7 tool “Determination of random points”. Although these points served as an initial reference for sampling, several of them proved to be inaccessible, so it was decided to sample trees near the access roads. The selected trees had to be in the fruiting phase, with pods in the ripening stage, to guarantee fruit collection. A total of 71 SP were evaluated; soil and fruit samples were collected from each of them. The distribution of the SP is shown in Figure 1.

2.2. Analysis and Sampling

2.2.1. Soil Sampling

Soil samples were collected at a depth of 30 cm and were located at the outer limit of the canopy projection on the soil. The distance between the trunk and the subsample sampling sites varied according to the tree’s canopy size, ranging from 8 to 12 m. For each tree, four subsamples were taken at the four cardinal points (north, south, east, and west), which were homogenized to form a composite sample representative of the zone with root influence. The samples were stored in airtight polyethylene bags to prevent contamination and moisture loss and were transported under controlled conditions to the Soil, Water, and Fertilizer Laboratory (LABSAF) of the National Institute for Agrarian Innovation (INIA), Peru, for analysis. The physicochemical parameters evaluated are detailed in

Table 1. Soil physicochemical properties and methods used in the laboratory.

Parameters Assessed	Method and/or Instrument	Remark	Source
pH	Digital potentiometer	soil suspension: water (1:2.5)	[32]
Electrical conductivity (EC)	Digital conductivity meter	Same suspension of pH analysis	[33]
Organic matter (OM)	Walkley-Black wet oxidation		[34]
Texture	Bouyucos method		[35]
Cation-exchange capacity (CEC)	Ammonium acetate	pH 7	[35]
Nutrients available	Flamometry: exchangeable potassium Bray II: phosphorus available Kjeldahl: total nitrogen		[35] [35] [36]
Heavy metals	Inductively coupled plasma mass spectrometry		[37]

.

2.2.2. Evaluation of the Carob Nutritional Quality

Mature plant material was collected from the same trees from which the soil samples were taken. Carob pods were placed in paper bags until a minimum weight of 100 g was reached, then transferred to the laboratory for drying. The samples were then taken to the Food Analysis Laboratory of the Nutritional Research Institute, where they underwent grinding and chemical analysis to determine ether extract and protein concentrations. In addition, moisture and ash parameters were evaluated. Moisture was determined according to NTP 208.017:2021: cocoa and chocolate using the gravimetric method [38]. For the esther extract, NTP 208.016:2021: cocoa and chocolate used the Soxhlet extraction method [39]. Crude protein was determined according to AOAC 2001.11 Edition 2023: Protein Crude assay in Animal Feed, Forage (Plant Tissue), Grain, and Oil Seeds. Block Digestion Method Using Copper Catalyst and Steam Distillation into Boric Acid [40]. For ash AOAC 930.05, Ash of plants [41].

2.3. Multispectral Data Collection and Processing

The vegetation indices (VIs) were calculated from Landsat-8 sensor satellite images, using the Python programming language (version 3.8, Python Software Foundation) in the Jupyter Lab environment, with the Google Earth Engine extension [42]. The relation of VIs is detailed in

Table 2. Vegetation indices calculated from Lansat-8 satellite images.

Vegetation Index	Equation	Source
Brightness Index (BI)	${\left({\displaystyle \frac{{Red}^2+{Green}^2+{Blue}^2}{3}}\right)}^{0.5}$	[45]
Coloration Index (COI)	${\displaystyle \frac{Red-Green}{Red+Green}}$	[46]
Difference Vegetation Index (DVI)	$NIR-Red$	[47]
Enhanced Vegetation Index (EVI)	${\displaystyle \frac{NIR-Red}{\left(NIR+Red+0.5\right)\times (1+0.5)}}$	[48]
Greenness Index (GI)	${\displaystyle \frac{2Green-Red-Blue}{2Green+Red+Blue}}$	[49]
Grain Size Index (GSI)	${\displaystyle \frac{Red-Blue}{Red+Blue+Green}}$	[50]
Global Vegetation Index (GVI)	(−0.29 $\times$ Green) − (0.56 $\times$ Red) + (0.6 $\times$ IR) + (0.49 $\times$ NIR)	[51]
Infrared Percentage Vegetation Index (IPVI)	${\displaystyle \frac{NIR}{(NIR+Red)}}$	[52]
Normalized Burn Ratio (NBR)	${\displaystyle \frac{NIR-SWIR2}{NIR+SWIR2}}$	[53]
Normalized Difference Salinity Index (NDSI)	${\displaystyle \frac{Red-NIR}{Red+NIR}}$	[54]
Normalized Difference Vegetation Index (NDVI)	${\displaystyle \frac{NIR-Red}{NIR+Red}}$	[55]
Normalized Difference Water Index (NDWI)	${\displaystyle \frac{Green-NIR}{Green+NIR}}$	[56]
Salinity Index (S1)	${\displaystyle \frac{Blue}{Red}}$	[57]
Saturation Index (SI)	${\displaystyle \frac{Red-Blue}{Red+Blue}}$	[46]
Redness index (RI)	${\displaystyle \frac{{Red}^2}{Blue\times {Green}^3}}$
Salinity Index (S2)	${\displaystyle \frac{Blue-Red}{Blue+Red}}$
Salinity Index (S3)	${\displaystyle \frac{Green\times Red}{Blue}}$	[58]
Salinity Index (S4)	${\displaystyle \frac{Blue\times Red}{Green}}$	[45]
Salinity Index (S5)	${\displaystyle \frac{Red\times NIR}{Green}}$
Salinity Index (S6)	${\left(Blue\times Red\right)}^{0.5}$
Salinity Index (S7)	${\left(Green\times Red\right)}^{0.5}$	[59]
Soil Adjusted Vegetation Index (SAVI)	${\displaystyle \frac{\left(NIR-Red\right)\left(1+L\right)}{(NIR+Red+L)}}$	[60]
Saturation Index (SI)	${\displaystyle \frac{Red-Blue}{Red+Blue}}$	[46]
Transformed Soil Adjusted Vegetation Index (TSAVI)	${\displaystyle \frac{a\left(NIR-aRed-b\right)}{Red+a\left(NIR-b\right)+x\left(1+a^2\right)}}$ Donde a = 0.33, b = 0.5, x = 1.5	[61]
Wetness Index (WI)	${\displaystyle \frac{NIR-SWIR1}{NIR+SWIR1}}$	[56]
Redness index (RI)	${\displaystyle \frac{{Red}^2}{Blue\times {Green}^3}}$	[46]
Ratio Vegetation Index (RVI)	${\displaystyle \frac{NIR}{RED}}$	[62]
Ferric Iron Index (FI)	${\displaystyle \frac{Red}{Blue}}$	[63]
Clay Index (CI)	${\displaystyle \frac{SWIR1}{SWIR2}}$	[64]

. For their compilation, the Python libraries Geemap [43], os, and Geopandas [44] were also used. To generate the orthomosaic covering the entire study area, images acquired on three different dates were merged: March 6, April 7, and May 2. These dates were chosen due to their low cloud cover (≤20%), as in previous periods the cloud cover in the area was high (>20%), which affected the quality of the data. Once the VIs were obtained, the values were extracted within a 90 m radius influence area, using the sampling point (SP) as the centroid. This strategy minimized the influence of outliers associated with the spectral variability in a single pixel. Finally, the values obtained were matched to the protein and ether extract concentration in the carob and used for building the predictive models.

2.4. Generation of Protein and Ether Extract Prediction Equations

The generation of equations to estimate the protein and ether extract content of carob pods was carried out using the Eureqa-Formulize software 1.24.0 (Noutonian, Newton, MA, USA). This software allows the generation of equations and hidden mathematical relationships [65]. The input variables were the soil properties and VI detailed in Table 2, with protein and ether extract concentrations as output variables. In the preprocessing of the input data before starting the generation of the equations, the outlier removal and the variable normalization of those that did not match normality were carried out. For each of the predicted variables, multiple equations were generated in order to select the one with the best performance. The selection of the model was based on those that presented the highest values of determination coefficient (R²) and those that showed the best results according to other indicators of fit quality [66].

2.5. Statistical Analysis

Data were analyzed using R software (version 4.2). Statistical analyses were performed to evaluate differences in soil physicochemical properties, carob nutritional quality, and VI. Pearson correlation analyses were also performed to analyze the relationship between variables, as well as the use of principal component analysis (PCA). In the case of Pearson, the Bonferroni correction was used. The performance of the generated equations was evaluated by comparing the simulated values with the observed values using the following indicators: coefficient of determination (R²), mean absolute error (MAE), mean square error (MSE), and correlation coefficient (CC). The equations were chosen on the basis of their indicators’ performance. The equations used are as follows: 1, 2, 3, and 4.

$$R^2={\displaystyle \frac{{\sum }_{i=1}^n\left(T_i-\overline{T} \right) \left(t_i-\overline{t}\right)}{\sqrt{{\sum }_{i=1}^n{\left(T_i-\overline{T}\right)}^2{\sum }_{i=1}^n{\left(t_i-\overline{t}\right)}^2}}}$$

(1)

$$MAE={\displaystyle \frac{1}{n}}\sum _{i=1}^n\left|T_i-t_i\right|$$

(2)

$$MSE={\displaystyle \frac{1}{n}}\sum _{i=1}^n{(T_i-t_i)}^2$$

(3)

$$CC=\sqrt{R^2}$$

(4)

where n is the number, T_i is the original measured values, t_i is the predicted values in the stablished model, and bar shows the average value of the variable referred to.

3. Results

3.1. Analysis of Soil Physicochemical Properties, Fruit Quality, and Vegetation Indices

3.1.1. Soil Physicochemical Properties

Table 3. Statistical analysis of the soil physicochemical properties.

Properties	Units	Mean	Maximum	Minimum	Standard Deviation (±)
Electrical conductivity (EC)	mS cm−1	85.93	966.40	3.71	184.81
pH	-	7.00	8.10	4.40	0.56
Potassium available (Kavai)	ppm	130.08	600.42	40.31	110.35
Phosphorus available (Pavai)		6.91	48.70	0.00	8.00
Organic matter (OM)	%	4.65	12.99	0.86	2.38
Total nitrogen (TN)		0.01	0.03	0.00	0.01
Sand		84.68	98.44	38.29	7.31
Clay		8.60	17.56	1.14	3.75
Silt		6.72	44.14	0.28	5.36
Total carbon (TC)		3.72	9.30	0.81	1.46
Carbon organic (OC)		2.66	7.53	0.50	1.44
Calcium ion (Ca2+)	cmol+ kg−1	4.36	28.47	0.88	5.49
Magnesium ion (Mg2+)		0.81	5.33	0.00	0.99
Sodium ion (Na+)		0.50	8.02	0.00	1.56
Potassium ion (K+)		0.37	3.10	0.10	0.43

presents the average values of the soil physicochemical properties at the evaluated points. The electrical conductivity (EC) showed an average value of 85.93 mS cm⁻¹, with a range between 3.71 and 966.40 mS cm⁻¹, indicating variability in soil salinity. The pH ranged from 4.40 to 8.10, with a mean value of 7.00, reflecting acidic to slightly alkaline conditions. The organic matter (OM) content ranged from 0.86% to 12.99%, with an average of 4.65%, while total nitrogen (TN) presented low values, with an average of 0.01%. As for available nutrients, potassium (K_avai) and phosphorus (Pavai) presented average values of 130.08 ppm and 6.91 ppm, respectively, with a high variability in their concentration. The soil texture was mainly composed of a sand fraction, with an average content of 84.68%, while clay and silt showed average values of 8.60% and 6.72%, respectively. Regarding the cation exchange capacity, calcium (Ca²⁺) was the predominant cation, with a mean of 4.36 cmol⁺/kg, followed by magnesium (Mg²⁺) (0.81 cmol⁺ kg⁻¹), sodium (Na⁺) (0.50 cmol⁺ kg⁻¹), and exchangeable potassium (K⁺) (0.37 cmol⁺ kg⁻¹). Finally, the total carbon (TC) and organic carbon (OC) contents showed mean values of 3.72% and 2.66%, respectively, with moderate variability.

3.1.2. Presence of Metals in Soil

The metals present in the soil (

Table 4. Statistical analysis of metals present in the soil.

Properties (ppm)	Mean	Maximum	Minimum	Standard Deviation (±)
Arsenic (As)	41.59	59.65	29.23	7.66
Beryllium (Be)	0.38	1.00	0.00	0.19
Cadmium (Cd)	0.92	2.70	0.30	0.56
Calcium (Ca)	5532.15	20,615.35	2876.40	3056.64
Cobalt (Co)	6.14	20.20	1.89	4.32
Copper (Cu)	16.46	58.05	4.23	11.96
Strontium (Sr)	78.42	213.99	36.88	35.89
Molybdenum (Mo)	0.76	1.88	0.32	0.37
Nickel (Ni)	9.64	37.17	3.26	7.29
Lead (Pb)	9.12	33.18	3.05	5.19
Selenium (Se)	4.64	14.07	0.00	2.77
Thallium (Tl)	0.16	0.48	0.04	0.11
Vanadium (V)	65.29	139.90	43.32	22.04
Iron (Fe)	9046.52	22,008.29	4050.02	4644.70
Potassium (K)	1591.56	4667.54	498.19	770.42
Magnesium (Mg)	3130.53	8352.83	1023.90	1536.74
Sodium (Na)	700.93	6819.66	78.93	1324.26
Mercury (Hg)	0.05	0.82	0.00	0.15
Barium (Ba)	60.65	275.93	10.28	52.87
Zinc (Zn)	65.63	144.00	29.65	29.30
Antimony (Sb)	0.31	0.57	0.18	0.08
Manganese (Mn)	275.88	913.85	98.08	172.81
Chromium (Cr)	19.06	41.13	11.38	6.67
Aluminum (Al)	6716.76	14,559.98	2961.31	2948.43

) show that iron (Fe), calcium (Ca), and aluminum (Al) were the most abundant elements, with averages of 9046.52 ppm, 5532.15, and 6716.76 ppm, respectively. Additionally, there was a significant dispersion in their values, as reflected in the high standard deviation values. In contrast, mercury (Hg) and thallium (Tl) presented low concentrations, with averages of 0.05 and 0.16 ppm, respectively, and lower variability. Arsenic (As) had a mean of 41.59 ppm and a standard deviation of 7.66 ppm, showing moderate dispersion, while cobalt (Co), copper (Cu), and zinc (Zn) had mean values of 6.14, 16.46, and 65.63 ppm, respectively. Manganese (Mn) and potassium (K) with averages of 3130.53 and 1591.56 ppm, respectively, showed high variability, as indicated by the standard deviations of 172.81 and 770.42 ppm, respectively. Other metals include beryllium (Be), cadmium (Cd), lead (Pb), selenium (Se), nickel (Ni), strontium (Sr), and vanadium (V). The average values found in this study are Sr = 78.42 > V = 65.29 > Ni = 9.64 > Pb = 9.12 > Se = 4.64 > Cd = 0.92 > Be = 0.38 ppm. Molybdenum (Mo), sodium (Na), barium (Ba), antimony (Sb), manganese (Mn), and chlorine had average concentrations of 0.76, 700.93, 60.65, 0.31, 275.88, and 19.06 ppm, respectively, with mixed standard deviations.

3.1.3. Nutritional Quality of Carob

The humidity content (

Table 5. Statistical summary of the carob fruit nutritional quality.

Properties	Mean	Maximum	Minimum	Standard Deviation (±)
Humidity (%)	12.11	13.61	10.38	0.74
Ash (g 100 g−1)	3.46	4.98	2.59	0.38
Ether extract (g 100 g−1)	1.10	1.27	0.90	0.08
Protein (g 100 g−1)	10.15	18.12	7.21	1.82

) presented an average of 12.11%, with a standard deviation of ±0.74 g 100 g⁻¹, which indicates a moderate variability. The ash content, representative of the mineral fraction, had an average of 3.46 g 100 g⁻¹, with a standard deviation of ±0.38 g 100 g⁻¹. On the other hand, the ether extract content was the most stable parameter, with an average of 1.10 g 100 g⁻¹, with a standard deviation of ±0.08 g 100 g⁻¹. In contrast, protein showed the greatest variability, with an average of 10.15 g 100 g⁻¹ and a standard deviation of ±1.82 g 100 g⁻¹, suggesting differences in concentration at the different sampling points.

3.1.4. Vegetation Index Analysis

Figure 2 illustrates the box plots of the vegetation indices employed in this study. The RI index has the widest and most dispersed range of values, with a median close to 200 and outliers. Similarly, the RVI index also exhibits high dispersion, albeit on a smaller scale of values. The indices with a more compact distribution and centered around the median were GI (−0.26 ± 0.02), GVI (0.18 ± 0.02), S2 (−0.26 ± 0.03), S5 (0.30 ± 0.02), FI (1.75 ± 0.10), CI (1.53 ± 0.19), and SI (0.27 ± 0.04). On the other hand, the NDVI (0.39 ± 1.55), IPVI (0.69 ± 0.07), and NBR (0.24 ± 0.15) indices registered low values. However, their dispersion suggests variability in carob tree cover, possibly due to heterogeneity in canopy density or the presence of rainfall in some places.

3.2. Relationship of Soil Physicochemical Properties and Vegetation Indices with the Protein and Ether Extract Content of Carob Fruit

3.2.1. Principal Component Analysis

Principal component analysis (Figure 3) reveals that proteins are positively correlated with the amount of ash and ether extract fat concentration but negatively correlated with moisture. Concerning soil physicochemical properties (Figure 3a), ether extract is associated with sand and total nitrogen (TN). On the other hand, the metals that present a greater relationship with ether extract and protein are selenium (Se) and antimony (Sb). Regarding vegetative indices, S1 and S2 show a positive correlation, while GVI and RI show a negative association with the ether extract concentration in carob. The first principal component (Dim1) in Figure 3a explains 19.6% of the variance, while the second principal component (Dim2) explains 14.5%, making a total of 34.1%. In the analysis of metals, the variance explained by the principal component (Dim1) is 55.4%, and that of the second component (Dim2) is 8.2%, totaling 63.6% (Figure 3b). In the case of vegetation indices, the explanation of the variance by the principal component (Dim1) is 58.3%, and that of the secondary component is 14%, making a total of 72.3% (Figure 3c), which is the highest value.

3.2.2. Correlation of Proteins and Ether Extract with Soil Physicochemical Properties and Vegetation Indices

Figure 4 shows moderate positive correlations between protein and ash (CC =0.51) and ether extract (CC = 0.33), as well as a moderate negative correlation between moisture and ash (CC = −0.33). Other weak correlations also occur between ether extract and sand (CC = 0.22), antimony (CC = −0.28), calcium (CC = −0.26), and S5 (CC = −0.26). Protein shows weak correlations with pH (CC = 0.19) and NDSI (CC = 0.22). These results suggest that the chemical composition of carob is influenced by multiple factors, such as the percentage of sand, the pH, and the concentrations of some elements, such as Ca²⁺, Sb, and Cu, with some associations being more significant than others.

3.3. Predictive Modeling and Key Factor Analysis

The equations were obtained using the Eureqa-Formulize software (version 1.24.0), with carob fruit protein and ether extract fat as output variables and soil properties and vegetation indices as input variables. The selected equations presented the lowest margin of error and met the criteria for a good fit, as indicated by the statistical indicators shown in Figure 5.

Table 6. Protein and ether extract concentration prediction equations from soil properties and vegetation indices.

Model	ID	Equation
Protein	A	$11.20+{\displaystyle \frac{-7.59}{CE}}+{\displaystyle \frac{7.72}{(12.08-CE\times CT)}}+{\displaystyle \frac{-9.63}{137.52-21.99\ast CE)}}-0.008\times CE-0.01\times CT\times P_{avai}$
	B	$11.52+2.63\times Ca+1.66\times pH-{\displaystyle \frac{192.01}{K_{avai}}}+{\displaystyle \frac{1.41}{K_{avai}-83.33}}+0.79\times {10}^{-9}\times Ca\times K_{avai}^3-0.37\times Ca\times pH$
	C	$4.8+{\displaystyle \frac{-9.45}{Ash-5.26}}+{\displaystyle \frac{0.23}{0.103\times H\times Ash-5.25}}+{\displaystyle \frac{0.0678}{0.148\times H\times Ash-5.27}}+{\displaystyle \frac{-0.0366}{0.148\times H\times {Ash}^2-23.3}}$
	D	$10.26+{\displaystyle \frac{-0.0001}{SAVI-0.33}}+{\displaystyle \frac{-0.00272}{0.32683-EVI}}+{\displaystyle \frac{-0.00692}{0.31115-EVI}}+{\displaystyle \frac{0.0003}{0.05419+{SAVI}^2-0.469\times SAVI}}-1.8892\times EVI$
	E	$133.72+WI+82.13\times \mathrm{S}4+15.107\times CI+15.288\times CI\times {S3}^2\times COS\left(194.54\times S5\right)-19.24-69.90\times CI\times WI-319.98$
	F	$15.75+{\displaystyle \frac{0.015}{0.46-NDVI}}-{\displaystyle \frac{0.0039}{NDVI-0.161}}+{\displaystyle \frac{0.044}{0.1212+NDVI\times NDWI}}+{\displaystyle \frac{0.49-0.0058\times Sand}{BI-0.153}}-0.0688\times Sand$
Ether extract	A’	$1.096+0.083\times K+{\displaystyle \frac{-0.019}{K\times {Ca}^4}}+{\displaystyle \frac{0.0025}{Ca-1.07-Ca\times K}}+{\displaystyle \frac{-0.0085}{0.226\times Sand-19.28-K}}-0.0022\times Ca\times Silt$
	B’	$1.14+{\displaystyle \frac{-0.01}{0.226\times Sand-19.44}}+{\displaystyle \frac{0.0026}{Ca-1.23-Ca\times K}}+{\displaystyle \frac{0.0045\times Silt-0.0089}{Ca-1.277}}-0.0125\times Silt$
	C’	$1.0769+{\displaystyle \frac{-0.0054}{Cu-0.0437\times Mn}}+{\displaystyle \frac{0.083\times Cu-0.62}{Co\times Cu-36.607}}+{\displaystyle \frac{292.7457}{{Mn}^2-3577.773-Co\times Cu\times Mn}}$
	D’	$1.096+{\displaystyle \frac{0.0065-0.02\times \mathrm{S}5}{28.075\times SI+26.61\times \mathrm{S}2-0.44}}+{\displaystyle \frac{2.334\times SI-0.63}{18108.53+23.89\times \mathrm{S}1+198866.95\times {S5}^2-120065.24\times S5}}+{\displaystyle \frac{3.066\times \mathrm{S}5-0.926}{1191.97+13159.0147\times {S5}^2-26.606\times \mathrm{S}2-7944.71\times S5}}$
	E’	$1.096+0.083\times COI\times RVI\times {FI}^3+{\displaystyle \frac{-0.0006}{3.415\times COI-18.14\times {BI}^2}}-{\displaystyle \frac{0.0000219}{0.009+BI\times COI-1.99\times {BI}^2}}-0.744\times COI$
	F’	$1.05+{\displaystyle \frac{-0323}{{Ca}^2}}+{\displaystyle \frac{0.015\times Sand\times SI}{Ca}}+{\displaystyle \frac{-0.047\times NDVI}{3.11-0.0147\times Sand\times NDVI\times {Ca}^2}}-SI\times {NDVI}^3$

where EC, electrical conductivity; Ca²⁺, calcium ion; K_avai, potassium available; P_avai, phosphorus available; H, humidity; TC, total carbon; Mn, manganese; Cu, cupper; Co, cobalt; S1, S2, S3, S4, S5, salinity index; NDVI, Normalized Difference Vegetation Index; COI, Coloration Index; BI, Brightness Index; RVI, Ratio Vegetation Index; FI, Ferric Iron Index; NDWI, Normalized Difference Water Index; CI, Clay Index.

presents the details of each selected equation. In the nonlinear multiple regression analysis, although the coefficient of determination (R²) reflects the goodness of the model fit, the other statistical indicators showed values within acceptable ranges, which supports the validity of the obtained fit.

The prediction equations for protein content (Figure 5) in the algarroba obtained a coefficient of determination (R²) in the range of 0.31 to 0.53, indicating a moderate predictive capacity. The correlation coefficient (CC) ranged from 0.53 to 0.73, while the mean squared error (MSE) ranged from 0.99 to 1.45, and the mean absolute error (MAE) ranged from 0.76 to 0.90. Among the equations evaluated, the one that presented the best performance was equation F, taking into account variables such as sand content, as well as the NDVI, NDWI, and BI indices. Equation F is better than E by 3.77% and 41.5% than equation A, taking into account R².

The prediction equations for the ether extract content (Figure 6) in carob yielded coefficients of determination (R²) ranging from 0.37 to 0.46, indicating a moderate predictive capacity. The correlation coefficient (CC) ranged from 0.61 to 0.73, while the mean square error (MSE) was 0.004 and the mean absolute error (MAE) was 0.05 for all cases. Among the equations evaluated, those that showed the best performance were equations 6-B’ and 6-E’, considering variables such as silt, K, and calcium (Ca) content, as well as the RVI, FI, BI, and COI indices. Equations (B’) and (E’) are 2.17% higher than F’ and 19.56% higher than equation C’, considering R².

4. Discussion

4.1. Soil Physicochemical Characterization

The analyzed soils exhibit significant variability in their physicochemical properties, which may impact the productivity of the dry forest ecosystem in northern Peru. The carob Neltuma spp. thrives in arid, semi-arid, tropical, and subtropical regions, in saline and alkaline soils [67], as reported in this study, due to the lack of rainfall. Low levels of available phosphorus are present, which can limit the growth and development of carob, affecting root formation and the efficiency of water and nutrient absorption [68]. The average organic matter content indicates the good capacity of the soil to retain moisture and nutrients, being 30% higher than that reported by [69]. Soil texture is dominated by the sandy fraction, with a maximum of 98.44% and a minimum of 38.29%, similar to those reported by [70], who found 96.4% sand. Neltuma species are resistant to water scarcity and tolerant to salinity [71]. These characteristics allow it to grow in the semi-arid zone of Peru, with minimal soil nutrient requirements. The physicochemical properties of the soil, such as the percentage of sand and the pH, had the highest correlations with the concentration of ether extract and protein, respectively. In the case of pH, it aligns with the study by [72] that examines the variation in nutritional quality of grapefruit. The texture of the soil and the content of microelements modify the nutritional quality of the seed oil of an edible halophyte [73]. Nitrogen (N) is also an essential element of amino acids (proteins); its deficiency reduces leaf growth and pod productivity [74]. Like nitrogen, potassium (K) is a primary macronutrient for plants, and it regulates key enzymatic functions involved in metabolism and protein synthesis [75]. Together with N and K, the other macronutrient phosphorus (P) had lower correlations than those found by [76] between available phosphorus and protein content in potato crops.

4.2. Potential for Heavy Metal Pollution in the Soil Where the Carob Tree Grows

The concentrations of arsenic (As), nickel (Ni), lead (Pb), and cadmium (Cd) in the soil samples were within the permissible limits established by the World Health Organization (WHO). Similar results were reported by [77], with the following values found in soils: Ni = 466.256 > Pb = 12.009 > As = 5.486 > Cd = 0.394 mg kg⁻¹, reversing the order of Ni, As, and Pb. However, Cr concentrations exceeded the limits allowed by the WHO, as they were present in 84.4% of the samples. The presence of trees favors the immobilization of cadmium in soils, as indicated by [78], who found higher cadmium contents in soils with trees than in soils without trees; this is due to the density of the trees. The high presence of heavy metals in soils causes intoxication and, in agricultural soils, poses a serious risk to plant life, human health, and the global food supply, with negative implications for both health and crop yields [79]. Therefore, it is necessary to study the concentration of metals in carob to rule out that these are translocated from the soil to the fruit. The highest correlations found in this study are between Sb and ash and between Cu and ether extract; the other metals do not present correlations greater than 0.20. [80] found that there is a strong mobility of metals from the soil to fresh fruits, especially in areas with low CEC values. However, it has also been demonstrated that the highest concentration of heavy metals is found in the root system, rather than in the aerial components of the plant [81].

4.3. Proximal Composition of Carob

The results obtained for ether extract and protein are 20% and 30% higher than those reported by [9]; as for ash content, the values are similar. Regarding the ash content, no significant differences were observed. When comparing the ether extract content with total lipids, it is observed that ether extract is lower by 43.3%, 60.6%, and 50.5% compared to Prosopis alba, Prosopis chilensis (yellow), and P. chilensis (brown) flour, respectively [82]. Compared to the previous study, the protein content was 28.8% and 6.8% higher than P. alba and P. chilensis (yellow) but 5.9% lower compared to P. chilensis (brown). These high values of protein and ether extract content have led to the flour of various species of Neltuma being used as an alternative to improve the nutritional quality of certain products [83]. Knowing the concentrations from soil characteristics or spectral indices will enable the accurate implementation and development of these products at a lower cost. Additionally, the use of vegetation indices facilitates the observation of spatial variations, which can optimize decision-making in the management and production of carob trees. It can help in the prediction of yields [84], the detection of stress disputes [85] and, in this case, the prediction of protein and ether extract.

4.4. Carob Vegetation Indices

In general, the values of the spectral indices are low, which aligns with the findings of [29] for the visible spectrum. Meanwhile, the near-infrared values tend to be higher, depending on the time of year. These values can also be affected by wet or dry periods. In denser forests, NDVI values can exceed 0.5, although they are typically around 0.4 on average, as reported by [27,29]. The same study indicates that EVI values range from 0.14 to 0.20, although during wet periods, they can reach values of up to 0.5. In this study, the average value recorded is 0.27, which is close to the expected range, considering that these values only represent those of the carob tree. The use of vegetation indices obtained from satellite imagery has been widely applied in different studies, including the early prediction of protein and oil in soybean cultivation with an RMSE of 1.804 and 1.04, respectively [86].

4.5. Forecasting of the Protein and Ether Extract Content from Soil Physicochemical Properties

The values obtained for R² in the equations generated are similar to those reported by [72], who achieved an R² of 0.50 and an MAE of 0.0003 in the prediction of the nutritional quality of grapefruit using a random forest model. In that study [72], the most important variables were pH, calcium, potassium, magnesium ions, and soil organic matter, coinciding in part with the most relevant variables in this study (pH, calcium, and potassium ions). Although studies on the prediction of protein content regarding soil properties are scarce, some research relates the yield to soil quality indices. For example, with regard to apple trees, [87] reported an R² of 0.56. Likewise, [76] estimated the soluble protein content in potatoes with an R² of 0.32, a value lower than the maximum values obtained in this study (R² = 0.53). However, both models highlight the importance of variables such as potassium, available phosphorus, and pH. In addition, [17] employed artificial neural network models to predict peach fruit quality through the calcium, potassium, nitrogen, and manganese content in the soil.

4.6. Prediction of Protein and Ether Extract Content from Vegetation Indices

Although no specific studies have been found on the prediction of protein content in Neltuma, previous research has shown that the near-infrared region allows information on compounds such as phospholipids and phosphoproteins in plants [88]. Within this spectrum, vegetation indices such as NDVI and NDWI are found, which, in the present research, were representative in the generated equations. Previous studies have reported R² values greater than 0.60 for predicting the protein content using NDVI [89,90,91]. These results are supported by [92], who used multispectral images obtained with unmanned aerial vehicles. Other indices, such as RVI, have also been used in predicting protein content [93]. However, information on estimating lipid concentration in fruit trees from soil properties and vegetation indices remains limited. In this context, research has shown that multispectral reflectance in the range of 460 to 1680 nm allows for the prediction of lipid content in pastures, with R² values between 0.60 and 0.80 [94]. In this spectral range, the vegetation indices are used to generate models of ether extract content in the algae sheath in the present research. Likewise, these remote sensing techniques have been applied to estimate the carbon-nitrogen ratio in grasslands [95].

The prediction of the ether extract and protein content in carob fruit from soil physicochemical properties and vegetation indices aims to support the more accurate estimation of these nutritional components. The predictive equations developed in this study incorporate multiple variables and terms. However, the complexity of these models could be refined with a larger sample size, which would likely reduce the number of variables required and improve the R² and CC values, thereby enhancing model performance. It is essential to note that carob production occurs within a natural system, with minimal human intervention in management practices. Future research could explore the impact of specific management practices on soil quality and carob productivity. Additionally, incorporating functional biodiversity analyses would be valuable to provide a deeper understanding of the interactions among soil properties, vegetation, soil biota, and the ecosystem services provided by dry forests.

The findings of this study contribute to the sustainability of dry forests by demonstrating that the nutritional quality of carob, specifically their protein and ether extract content, is influenced by the specific physicochemical properties of the soil. This finding highlights the importance of integrated ecosystem management, where soil conservation is recognized as an essential component for maintaining the productivity and ecological functionality of these forests. The assessment of protein and ether extract concentrations supports the valorization of carob as a strategic food resource for local communities and livestock systems, thereby strengthening food security and promoting the sustainable use of native species. Furthermore, the application of vegetation indices derived from remote sensing enables the generation of spatial information regarding the nutritional quality of carob fruit. This approach offers new possibilities for monitoring the impacts of climate change [96] on carob yield, estimating the ecosystem’s energy contribution, and determining the livestock carrying capacity within forested areas. Collectively, these findings position the study as a valuable tool for supporting evidence-based ecological restoration strategies [97] and guiding the sustainable management of natural resources in dry forest ecosystems.

5. Conclusions

The results of this study provide insight into the influence of soil physicochemical properties, metal concentrations, and vegetation indices on the protein and ether extract concentrations of the carob tree (Neltuma) fruit in the dry forest of the northern coast of Peru. The soil characterization results indicate a texture with an average of 84.68% and a standard deviation of ±7.31%. The protein and ether extract content of carob is 10.15 g 100 g⁻¹ ± 1.82 g 100 g⁻¹ and 1.10 g 100 g⁻¹ ± 0.08 g 100 g⁻¹, constituting an important source for feeding goats and sheep. In the case of metals in the soil, the Cr content exceeds the levels suggested by the WHO. This is an issue that should be addressed in future studies to rule out its presence in carob. There are correlations between the ether extract and the percentage of sand (0.22), Ca+2 (−0.26), Cu (−0.20), S5 (0.26), and Si (0.24). In the case of the protein, correlations are observed with pH (r = 0.19), BI (r = 0.22), and NDSI (r = 0.22).

In the development of predictive equations for protein concentration, Equation F demonstrated the best performance, with an R² of 0.53 between the observed and predicted values, representing a 41.5% improvement over the lowest-performing equation, which had an R² of 0.31. The variables contributing to this model were NDVI, NDWI, BI, and sand content. For ether extract prediction, two equations, B’ and E’, were identified as the most effective, both achieving an R² of 0.46, with an MSE of 0.004 and an MAE of 0.05. The key predictive variables in these models included sand, potassium, calcium ions, and the vegetation indices (VIs): COI, RVI, BI, and FI. In both protein and ether extract models, sand content and the BI vegetation index consistently emerged as significant predictors. These equations should be considered as initial baselines, and their predictive accuracy could be substantially improved by increasing the sample size and incorporating sampling points from other regions. Additionally, the application of more advanced modelling approaches, such as machine learning algorithms, may further enhance the predictive accuracy. Monitoring the protein and ether extract concentrations in Neltuma pallida (carob) fruit using soil physicochemical properties and vegetation indices offers a more accessible approach and contributes to its sustainable utilization. These findings aim to strengthen the integrated management of the dry forest ecosystem and to enhance the valorization of carob as a vital food resource for local communities.