Next Article in Journal
Opportunity or Challenge? Research on the Influence of Digital Finance on Digital Transformation of Agribusiness
Previous Article in Journal
Green Innovation as a Mediator between Green Human Resource Management Practices and Sustainable Performance in Palestinian Manufacturing Industries
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 15
Issue 2
10.3390/su15021066
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Mapping Soil Parameters with Environmental Covariates and Land Cover Projection in Tropical Rainforest, Hangadi Watershed, Ethiopia

by
Berhanu Tamiru
1,*,
Teshome Soromessa
1,
Bikila Warkineh
2 and
Gudina Legese
1
1
Center for Environmental Sciences, Addis Ababa University, Addis Ababa P.O. Box 1176, Ethiopia
2
Department of Plant Biology and Biodiversity Management, Addis Ababa University, Addis Ababa P.O. Box 1176, Ethiopia
*
Author to whom correspondence should be addressed.
Sustainability 2023, 15(2), 1066; https://doi.org/10.3390/su15021066
Submission received: 4 September 2022 / Revised: 28 September 2022 / Accepted: 10 October 2022 / Published: 6 January 2023
Browse Figures
Review Reports Versions Notes

Abstract

:
Machine learning and geostatistics are efficient techniques for investigating the geographic distribution of soil properties. This study’s objectives were to assess soil fertility status, map the spatial variability of selected soil parameters and compare random forest with ordinary kriging. Soil samples were collected to analyze parameters: pH, cation exchange capacity (CEC) and organic carbon (OC) using systematic sampling. Some environmental covariates were used in the machine learning process: a digital elevation model (DEM) collected from USGS distributing shuttle radar topography mission data and a LULC map generated from a 30-year time series (1988–2018) of Landsat 8. Georeferenced samples were sent to Batu Soil Research Laboratory. pH, CEC and OC were mapped and status was determined using random forest and ordinary kriging. Random forest was more accurate with low mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (high R2). In random forest, pH varied between 5.03 and 5.76 and ordinary kriging pH ranged from 4.96 to 5.76. pH was greater in cultivated land. CEC and OC were higher in the forest. The higher pH in cultivated land was due to grass coverage and minimal tillage. The addition of organic matter and CEC to a forest may result in higher OC. Environmental covariates (topographic, bands, NDVI and LULC) were used to predict the gradients of pH, OC and CEC. For pH, OC and CEC, DEM was the most important predictor. CEC was high in low landscape, but low in high landscape positions. Low OC requires composting, fallow and organic fertilizers. Future research should include the remaining predictors: physiochemical and lithological data to improve the performance of random forest.

1. Introduction

The dynamic physicochemical and biological system of soil fertility provides essential nutrients in the right amounts and ratios, and it also effectively converts nutrients into plant growth. For the production of food, industrial resources, and averting climate change, fertile soil is necessary. Soil fertility is impacted by the rapid change of natural ecosystems into agroecosystems. It supplies nutrients to maintain the soil’s biodiversity, quality, and water-holding ability. Reduced soil fertility makes it difficult to feed an expanding population. The depletion of macro and micronutrients, acidity, topsoil removal, and salt accumulation are factors contributing to land degradation [1]. Deforestation, overgrazing and continuous cultivation have triggered soil fertility losses at the rate of 130 tones/ha for cultivated fields. This is caused by wrongful land management practices and human interference resulting in changes in land use and land cover [2].
Multiple linear regression (MLR) [2], partial least squares regression (plsr) [1], kriging, and regression-kriging [3] are statistical methods that have been used in several studies to map parameters. In fact, the connection between soil properties and predictor variables is complicated and non-linear, contrary to the assumptions of multiple linear regressions [4]. In order to map soil properties with higher accuracy than conventional geostatistical models, several studies have employed machine learning approaches such as Cubist [5], support vector machine [6] and random forest methods [7]. Machine learning techniques—random forest (RF), geostatistics and ordinary kriging (OK)—are considered to be effective for examining the spatial distribution of soil properties [8]. Digital elevation models, covariates, and Landsat 5 Thematic Mapper satellite images are variables used in machine learning RF approaches. Geostatistical analyses make use of geographic coordinates to anticipate soil qualities of unsampled areas and compare them to actual sampled values. Geostatistical sample density ought to be spatially dependent. Otherwise, classic statistics are unpredictable. To fill the gap, it is crucial to assess the advantages of both tools (RF and OK) [9,10]. Compared with other machine learning algorithms and conventional statistics, random forest is less sensitive to noise in training data [11,12]. Additionally, it is very flexible and not restricted to linear relations, unlike, for example, regression kriging and linear regression [13]. Wide research regions with uneven topography and climate have an impact on RF residue interpolation [14]. OK is an unbiased predictor leveraging data from sample points to estimate soil parameters from unsampled sites, minimizing error variability and costs. For its simplicity and availability in GIS, geostatistical interpolation is often used. Disadvantages include the fact that the model excludes environmental factors (elevation, climate and vegetation) influencing soil nutrients, which limits the simulation of landscape soil characteristics [8].
The random forest package’s random forest function was utilized to fit a random forest model. The covariate values at the sample locations were saved in a distinct data frame, whereas the response variables (soil properties) were kept in a separate vector. All variables that were stored in the covariate data were used by default by the model frame. A digital elevation model (DEM), retrieved from the USGS/Earth Explorer, which distributes shuttle radar topography mission data (STRM v.3, 30 m), and an LULC map produced from a 30-year time series (1988–2018) of Landsat 8 were some of the environmental covariates utilized in the machine learning process.
It is crucial to predict land use and land cover using time series data to provide practitioners with knowledge for managing soil fertility. Soil fertility is frequently used as a proxy for human interference on land change processes for a variety of suitability measures. In a Markov model, the transition probability matrix of the LULC change from period one to period two is used to forecast the future state of the system solely based on the proximately preceding state [15]. Cellular automata and the Markov chain model can be coupled with GIS and remote sensing to enable the spatiotemporal modeling of LULC change (CAMCM). In order to estimate the LULC change in different regions by computing the states between different land uses and the transition rate, multispectral satellite imagery and the CA–Markov chain model were utilized [13]. Due to its flexibility and capability, the model integrating Markov chain and cellular automata has frequently been used [16]. Cellular automata were integrated to solve the spatially referred outcome weaknesses of the Markov chain model [17]. Land uses refers to human habitat use and alterations of the land due to anthropogenic activities, while land cover refers to the physical characteristics of the earth’s surface, such as water, vegetation, forests, rocks, and other physical characteristics. Forest fragmentation, biodiversity loss and accurate assessments of the world’s forest, grassland and agricultural lands are all important components of future management plans to support human needs and welfare [18,19]. Monitoring the effects of humans on the environment is made easier by valuing LULC changes. The change detection of LULC is crucial for a better understanding of the dynamics of the landscape within a period of time [20]. Additionally, change detection research has made it possible to comprehend the dynamics of human activities in both space and time [21].
This watershed has the only natural forest biodiversity in the Guji zone. The prediction of LULC change and its effects on soil fertility in the Hangadi watershed have not been studied. Nonetheless, two decades ago, local people and investors started coffee cultivation as agroforestry at the expense of the virgin forest, without considering its implications on soil fertility. To our knowledge, however, none of the existing strategies seeking to advance agroforestry explicitly presented in the results of this study protect the seedlings and saplings of the forest land in the watershed. Maps of the extent and distribution of the major soil types in the landscape, as well as their physicochemical properties, thus reflect the essential data needed to develop and disseminate site-specific recommendations for integrated sustainable development, thereby guiding the scaling up of best practices that are supported by the available scientific evidence. With this in mind, the following study objectives were set through the use of a combination of techniques:
➢
Assessing the fertility status of soils, mapping the spatial variability of selected soil fertility parameters in the Hangadi watershed, and providing soil quality information to practitioners, natural resource managers and other stakeholders.
➢
Comparing the test predictions of soil pH, CEC, and OC using random forest and ordinary kriging techniques.

2. Study Area Description and Methodology

2.1. Study Area Description

The research took place in Ethiopia’s Oromia Region, Guji zone, Odo-Shakiso district, Hangadi watershed (Figure 1). The Borena zone borders the zone on the south, the southern nations, nationalities and peoples’ region of the west, the Bale zone on the north, and the Somale region on the east. The Dawa River in the south, west, northeast, north, and east separates the Arero, Bule Hora, Uraga, Bore, Adola, Wadera and Liben districts, respectively, from the district in question [22]. The studied watershed is located 550 km south of Addis Ababa at 38°46′0″ E and 5°54′0″ N. The district’s population is 268,630 people (148,724 men and 119,906 women) with 15% lowland, 20% midland and 65% highland, comprising an area of 4165.62 km2 with a density of 59.3 people per km2 [23].
The National Meteorological Services Agency (NMSA) provided climatic data (1987–2017) that show the region has a bimodal rainfall pattern of roughly 49.3 and 34.2% during the summer (March–May) and autumn (September–November) seasons, respectively, with a mean annual rainfall of 87 mm. Similar to this, the average monthly temperature during the past 30 years (1987–2017) has ranged from 24.7 °C to 26.8 °C, with a mean annual temperature of 20 °C (Figure 2).
The highest mean sand and silt contents were found in agroforestry, forest and cultivated land areas (62.1%, 61.6%, and 27%), respectively. The lowest concentrations of sand (53.4%) and silt (11.93%) were found in the forest and agroforestry areas. It has also been demonstrated that slope and altitude significantly affect land use types. The greatest mean altitude was found in the forest land (2122), followed by agroforestry (2098) and cultivated land (1889 masl). The highest and lowest mean altitudes were found in cultivated land (1889 masl), agroforestry (2098) and forest land usage (2122 masl). However, the mean slope values for cultivated land (22.9), agroforestry (24.6), and forests (17) were the greatest and lowest, respectively as the watershed classified into three land uses (Table 1).

2.2. Methodology

2.2.1. Soil Sample Analysis and Classification

To collect soil samples for parameter analysis, systematic sampling was performed (Figure 3). Independent plots were created along line transects at 300 m intervals, 1km apart, following the initial sample plot. A total of 100 composite samples were collected from 5 subsamples—4 from the corners and 1 from the center—and labeled in polythene bags and sent to the Batu Research Soil Laboratory. Georeferenced sample locations were recorded using GIS. Prior to chemical analysis, samples were dried at room temperature, crushed, mixed and sieved through a 2 mm sieve. pH, cation exchange capacity and organic carbon (CEC and OC) were determined using the ammonium acetate technique (1MNH4OAC) of water suspension (soil to water ratio 1:2.5 on a pH meter), [24,25].
Results of pH, CEC and OC were separated into two datasets at random. Then, 75% of the data was used to train models and 25% (out of 100 = 25) of the data was used to evaluate RF and OK accuracy at validation points by comparing the observed and projected pH, CEC and OC values. The training dataset was used to interpolate and produce spatial distribution maps. RF and OK procedures were used to replicate the training soil property map of spatial distribution. The coefficient of determination (R2) [25] was determined to evaluate the accuracy of the interpolation techniques. Mean error (ME) and (MAE) [26,27] were used for error measurement. Random forest and ordinary kriging in R software and ArcGIS 10.2 were used to generate spatial distribution maps of soil attributes.
RF is a non-parametric machine learning approach that uses a lot of decision trees and bagging. Through a set of splitting criteria, the outcomes of each individual decision tree are combined to produce a single prediction. In comparison with other machine learning techniques and conventional statistics that were not used during the tree building process, the regression prediction error, out-of-bag (OOB) data were used to rank the relative significance of predictors with greater accuracy [28]. The number of trees (ntree) and the number of variables available for selection in each split (mtry) are two important parameters in the RF method [29]. The RF model was implemented using the R statistical package software [30]. Because the parameter of mtry is sensitive to the RF prediction performance, we selected a train function (an interactive approach) in the R Caret Package to determine the best mtry value in terms of the smallest OOB mean square error. The final RF algorithm was applied with model settings of n = 1000, node size = 5, and an mtry value of integer values [30].
One of the often-used kriging variants, ordinary kriging (OK) assumes a geostatistical model and derives the best interpolation from it. One of the most popular kriging approaches, conventional kriging, utilizes a geostatistical method to estimate the values of unsampled points [31]. Using it, pH, CEC and OC were mapped. The ordinary kriging framework, which employs semi-variogram analysis to evaluate spatial variability, and R software were used to determine OK, according to [32]. The spatial prediction of the unmeasured point Z*(X0) was equal to the line sum of the previously known measured values (i.e., observed values). Researchers, including [33,34,35,36], and others, provide the following formula as an elegant and straightforward way to describe OK:
Z * X 0 = ∑ i = 1 n λ i z X i
where Z* (X0) is the predicted value at the unmeasured location X0; Z (Xi) is the measured value at position Xi; λi is the weighting coefficient from the measured position to X0; and n is the number of positions within the neighborhood searching. To represent the spatial continuity of the data and display the spatial link between the pairs of points, a fitted model based on the input data distribution is required [37,38]. In this study, the framework developed by [39] was used to calculate the OK method using R software. In its basic form, the OK approach calculates the value of a function (response) from an unknown sample. It is the process of summarizing the modeled relationship between independent variables and target values into a regression equation. Using the linear relationship independent variables, the regression equation predicts the target variables. The form of the regression equation is as follows:
Y = β0 + β1(X1) + β2(X2) + β3(X3) + … + (Xn) + €

2.2.2. LULC Change Prediction for the Year 2048

The CA–Markov model, an integrated model with cellular automata and Markov chain, offers a strong method for spatial and temporal dynamic modelling [39]. It has been implemented in order to predict the characteristics and trends of LULC change over time and provide future scenarios of land use and land cover to help the design of policy responses (Figure 3). IDRISI Selva uses the CA–Markov model to display the transition probability and transition matrix for the chosen region of interest. It provides the likelihood that a pixel will change from one LULC class to another within the same time period and predicts upcoming changes using the state of each class at the time of the forecast and the probability of the transition [40]. With the help of the CA–Markov model in IDRISI software version 17, future LULC changes from the research site were predicted. The procedures that followed throughout this were as follows:
  • Using land use maps from 1988 and 2018, the IDRISI software’s CA–Markov model was utilized to construct a transaction probability and area matrix. The forecast map was produced using a 30-year run cycle 5 × 5 contiguity filter.
  • The Markov chain was used to compute the transition probability matrix.
  • The transition probability was used to create the LULC atlas.
  • In the end, the LULC change for the year 2048 was projected using the transition probability images and base map (Figure 3).
The CA model is shown in the following Equation (2) [41,42,43]:
S (t, t + 1) = f(S(t), N)
where S (t + 1) is the system status at the time of (t, t + 1), functioned by the state probability of any time (N).
The Markov chain model is often used in LULC monitoring, ecological modeling, simulation changes, LULC trends, and to predict the stability of future land development in the study region [43]. Equation (1) describes the computation of the prediction of LULC changes (Markov chain model), which pronounces the LULC change from one time to another to predict future change [44].
S(t, t + 1) = Pij × S(t)
where S (t) is the system status at the time of t; S (t + 1) is the system status at the time of t + 1; and Pij is the transition probability matrix in a state which is calculated as follows [44].
P1, 1 P1, 2 P1, N
P1, 1 P2, 2 P2, N
=Pij=  · · · · · · · · ·
PN, 1 PN, 2 PN, N
0 ≤ Pij ≤ 1,
where P is the transition probability; Pij stands for the probability of transforming from the present state i to another state j in succeeding time; and PN is the state probability of any time. The high transition has probabilities near (1) and the low transition will have a probability near (0) [44]. Markov chain concludes precisely how much land would be estimated to change from the latest date to the predicted date. The transition probability file is the result of this process, which is a matrix that registers the probability that each land use/land cover class will change to every other class [45].

3. Results

3.1. Particle size and Topography

Land cover types differed significantly in texture, altitude and slope (<0.01). Agroforestry, forest, and cultivated land areas had the greatest mean sand and silt content (62.1%, 61.6% and 27%, respectively). However, forest and agroforestry areas had the lowest levels of sand (53.4%) and silt (11.93%), respectively. Altitude and slope were also shown to have significant differences among land uses. The highest mean altitudes were recorded in forest land use (2122) and agroforestry (2098), and the lowest in cultivated land (1889 masl). However, the highest mean slope values were identified in agroforestry (24.6), followed by cultivated land (22.9) and forest (17). The clay texture and bulk density of the three land uses did not differ significantly (Table 2).
pH and OC were significant differences among three land cover types. Forest and cultivated land use areas had the lowest (5.2) and highest (5.6) pH means, respectively. For cultivated and forest land use, the mean OC value of studied watershed ranged from 2 to 2.4%. The quantity of OC found on cultivated land was less than in forests (Table 2). This might reflect differences in organic matter management, the forest oxidation of organic matter, or the limited addition of organic inputs (compost or manure).

3.2. Predictors’ Importance Obtained from Random Forest

One of the advantages of the RF model is the ability to rank the relative relevance of covariate variables. DEM was the most important predictor variable for pH, CEC and OC in the Hangadi watershed. The relative importance of variables employed as predictors in the RF model was shown in (Figure 4, Figure 5 and Figure 6). Both topographic- and satellite-derived predictors had a high relative usage, showing that several variables had a smaller influence on the prediction of pH. The most used predictor variables in the RF model to predict pH were the digital elevation model (DEM), B1, B4, B2, B7, NDVI, B5, B6, LULC 2018, B3 and slope. The topographic predictors were more important than the satellite-derived indices in predicting CEC. The most important predictors to predict CEC were DEM, slope, B3, B2, B1, B4, NDVI, B6, B7, B5 and LULC 2018. The predictor variables of OC listed according to their importance were DEM, B1, slope, B4, B7, NDVI, B2, B5, B6, B3 and LULC 2018.

3.3. Semi-Variogram Model Parameters Derived from Ordinary Kriging

Semi-variograms were used to compute the spatial structures of soil pH, CEC and OC. The nugget/sill ratio (spatial dependency) of pH and OC is zero (0), suggesting that pH and OC content have a strong spatial correlation in the investigated region (Table 3). The nugget/sill ratio of 0.2 in CEC is more than zero (0), and observations are separated by the minimum distance.

3.4. Mapping Soil pH, CEC and OC in Hangadi Watershed

Forests, agroforestry and cultivated land were identified as land uses in 2018 using satellite data and field observations. In the studied watershed, no land without a personal certificate was found. Agroforestry was the most common land use/cover type in the studied watershed in the year, accounting for 1575.4 ha (44.8%), followed by forest land 988.4 ha (28.1%), and cultivated land 955.1 ha (27.1%) (Table 1) [22].
In random forest, soil pH ranges from 5.0 to 5.8, while in ordinary kriging, it ranges from 4.96 to 5.76. A higher pH was observed in cultivated and agroforestry land uses, whereas a lower pH was reported in the studied watershed’s forest site. The borders between terrain types in Oklahoma can be used to identify pH map changes. Yellow stripes on the boundary between land uses of CEC and OC illustrated these changes.
Random forest interpolation of CEC values ranges from 22.4 to 33.7, while ordinary kriging CEC values vary from 19.4 to 35.50. In forest and agroforestry land uses, the highest CECs were 33.7 and 35.5, respectively, interpolated with random forest and ordinary kriging. As seen in Figure 7, CEC values declined from forest and agroforestry to cultivated land in the watershed. When using random forest, the OC contents range from 1.7 to 2.9. This ranged from 1.5 to 3 when interpolated with ordinary kriging. The study area’s agroforestry and cultivated use types had lower percentages of OC, but the forest part of the watershed had a larger percentage of OC (7 and 8). According to the focal group discussion, one of the major factors for the low concentration of OC across the study region is long-term cultivation. The land use pattern, which may be the dominating factor of OC in the research area, may be the cause of OC variability. Low to high OC values were seen as a continuous reddish and greenish band across the land use types’ boundaries. The forest land use band corresponds to the highest terrain, where virgin vegetation is abundant and farming operations are difficult to reach. This shows that the input for OC comes from a greater net primary production. The remaining areas of lower elevation include agroforestry and cultivated land areas, which have comparatively middle-ranging (yellowish areas) and lower (reddish areas), soil organic carbon levels.

3.5. Model Accuracy

The accuracy of the predictive model was evaluated using validation indicators such as MSE, MAE, RMSE and R2. The results of both random forest and ordinary kriging are indicated in Table 4. Random forest was a more accurate prediction with low MSE, RMSE, MAE and high R2, compared with ordinary kriging, with high MSE, RMSE and MAE, and low R2 in modeling pH, CEC and OC in the studied watershed (Table 4).

3.6. LULC Change Prediction for the Year 2048

For 2018–2048, Table 5 illustrates a probabilistic matrix of land use and cover conversion for the Hangadi watershed. Forests have a 52.13% chance of remaining as forests, a 35.01% risk of being converted to agroforestry, and a 12.87% risk of being converted to cultivated land. The CA–Markov model was used to estimate the future LULC scenarios for 2048 using the 1988 map as a base map. To forecast future changes in LULCs for 2048, Markov and cellular automata were utilized. According to the 2048 LULC, the net percentage estimates for forests, agroforestry, and cultivated land are 539.97, 1730.42 and 1249.54 ha, respectively (Table 6, Figure 4).

4. Discussion

4.1. Soil Bulk Density and Topography

The insignificant difference in bulk density in the studied watershed implies that the studied soil has normal pore space and no aeration constraints. It also demonstrates that bulk density does not exceed the crucial limit of 1.63 g.cm−3, reducing compaction and drainage issues while promoting biological activity and infiltration [46].

4.2. Selected Soil Chemical Properties

The soils from the forest land use area in the studied watershed were classified as having strong acidic range, whereas the soils of other land uses were classified as moderately acidic [47]. Agroforestry and cultivated land occur on steep slopes in the Hangadi watershed, whereas forest land use is in the lower terrain (Table 2). The study implied that slope has a negative impact on soil organic matter [48]. A steep slope may cause increased soil erosion, resulting in a loss of soil organic matter.
According to focus group discussions, elevated soil pH in cultivated land could be linked to potassium supplementation through shifting cultivation; this is in line with the findings of [49,50,51], that all found a significant increase in soil pH under cultivated land. Furthermore, when interpolated using RF and OK, the soil pH distributions in forest, agroforestry and cultivated land (upper, middle and lower) are 5.0 to 5.76 and 4.96 to 5.76, respectively. Buffer zones with grasses for animal feed surrounding cultivated land might be attributed to a higher pH on the upper slope (Table 1). According to [51], higher altitude correlates with lower pH. Furthermore, farmers in the watershed practice minimal tillage, and thus pH is unlikely affected. However, the results contradict those of [52,53,54], who found that cultivated land had lower pH values than agroforestry and other land uses due to cation depletion and removal. According to [55]’s ratings, the watershed’s soils fall into the strongly acidic to moderately acidic range, and the very strongly acidic to acidic range, respectively (Figure 6 and Figure 8). Differences in parental material, topographic position, land use type, the removal of basic cations by crop harvests and prevailing weather conditions may all contribute to pH variation among land uses. pH values between 5.5 and 8.0 are considered ideal for plant growth, as suggested by [56]. Because 5.0 is more acidic than 5.5, the pH of the soils in this study may not be suitable for plant development. The current pH could imply that monitoring the effects of various land use schemes on soil quality is difficult.
According to [56], the contents of OC play a vital role in assessing land quality. When mapped using both random forest and ordinary kriging, the OC content of the watershed is very low to low, according to [57]. However, according to [57]’s assessments, it is moderate to high (2007). Higher slope locations have a very low SOC, whereas lower slope positions have a relatively high OC. A similar result was reported by [58], who conducted research at Shenkola watershed and reported that the lower landscape position had the highest value of OC, and the lowest value was found in the upper landscape position. On one hand, a higher value of OC in forestland might be due to a lower rate of organic carbon turnover for the lowest possible amount of soil disturbance and the addition of OM in forestland. The lower value of OC in cultivated land is due to the high oxidation of organic matter and the removal of crop residues, which coincides with reports of [59,60,61].

4.3. Mapping and Covariate Variables

In this study, regression was performed using the supervised learning technique of random forest. It develops an ensemble model that predicts the target value using basic decision rules discovered from data attributes. The random forest approach creates decision trees during the training phase. This technique integrates the predictions made by various trees to produce a single result [62]. Even when variables were highly linked with one another, the RF model calculated the relative significance of anticipated variables, preventing the periodically important ones from being deleted. The stronger capacity of the RF model to deal with the non-linear and hierarchical correlations between the predictors and target soil parameters might be attributed to its superior performance in minimizing susceptibility to overfitting [63].
In its basic form, the OK approach calculates the value of a function (response) from an unknown sample [63]. It is the process of summarizing the modeled relationship between independent variables and target values into a regression equation. Using the linear relationship between independent variables, the regression equation predicts the target variable. The form of the regression equation is as follows:
Y = β0 + β1(x1) + β2(x2) + β3(x3) + … + (xn) + ε
In this case, the dependent variable is Y, the independent variables are x, β is the coefficient which reflects the relationship of each independent variable to the dependent variable, and ε the portion of dependent variable that cannot be explained.
Figure 7 and Figure 8 exhibit the two maps of the predicted pH, CEC and OC contents using random forest and ordinary kriging, respectively. The spatial differences between the maps produced by random forest and ordinary kriging, despite their visual similarity, may be due to the predictor variables that random forest utilized to predict pH, CEC and OC contents. The random forest model’s prediction accuracy can be increased by applying the predictor variables [63]. As seen in Table 4, random forest had higher R2 and lower RMSE values than ordinary kriging, making it a more accurate technique of prediction of pH, CEC and OC in the studied watershed.
The influence of auxiliary data collected from the DEM demonstrates the importance of topography as the main influencing factor of pH, CEC and OC. Many studies have found that when topography is interpolated using random forest, it has an influence on soil properties and formation, for example, refs, and prediction accuracy may be increased by incorporating auxiliary data. DEM has a considerable influence on SOM, according to the random forest model. The next major predictor factors for SOM were Band 1, slope, Band 4, Band 7, NDVI, Band 2, Band 5, Band 6 and LULC 2018. The satellite image bands: Band 1 (430–450 nm), Band 2 (450–510 nm), Band 3 (530–590 nm), Band 4 Red (640–690 nm), Band 5 near infrared (850–880 nm), Band 6 SWIR1 (1570–1650 nm) and Band 7 SWIR2 (2110–2290 nm) were acquired on 11 October 2021.
CEC refers to a soil’s capacity to retain nutrients and prevent leaching. According to [63], samples obtained in the watershed were classified as high to medium and high to moderate, respectively, when interpolated with RF and OK, respectively. CEC concentration varied with topography, according to the finding. CEC was high in the lower slopes, whereas it was low on the top slopes. CEC variation was also reported along a slope gradient. DEM, slope, Band 2, Band 3, Band 1, NDVI, Band 4, Band 6, Band 5, Band 7 and LULC 2018 were important and included in the final set of operators of cation exchange capacity in the watershed when interpolated with random forest.

5. Conclusions

In this study, the pH, CEC and OC contents based on environmental and georeferenced soil and topographic data were predicted. Random forest and ordinary kriging models were used for the prediction results, and showed that random forest achieved a better performance. Since random forest yielded more accurate predictions, it can be concluded that the use of environmental covariates is necessary for predicting PH, CEC and OC contents in the studied watershed.
The SOC, pH and CEC of forest, agroforestry and cultivated land in the Hangadi watershed were mapped in this study. The random forest approach was used to model the relationship between pH, CEC, OC, and their associated variables (DEM, Band (1–7), NDVI, slope, and LULC, 2018). The outcomes were predicted using random forest and ordinary kriging models, and it was discovered that random forest outperformed ordinary kriging. Because random forest performed better than ordinary kriging, it may be inferred that environmental factors are required for predicting pH, CEC and OC in areas with gradients.
The findings presented in this research from maps using RF and OK displayed a higher pH at the upper slope position, which could be attributed to the grass coverage of buffer zones used for livestock feed. Moreover, farmers practice minimum tillage to avoid affecting pH. Contrary to pH, OC is rated as very low in higher slopes and high in lower slopes. The higher value of OC in forest areas with lower organic carbon turnover rates and the addition of OM to forest land indicate that land use types significantly affect pH and OC. Similarly, lower landscape positions had a high CEC, but upper landscapes had a low CEC. The study found that the expansion of cultivated lands depleted OC. Low OC levels require the application of organic manures, fallow cropping systems, organic fertilizer applications, and residual cropping to increase OC.

Author Contributions

Conceptualization, B.T. and T.S.; data curation, B.T. and T.S.; formal analysis, B.T.; funding acquisition, B.W.; investigation, B.T.; methodology, B.T.; resources, B.T. and G.L.; software, T.S. and B.T.; supervision, T.S.; B.W. and G.L.; validation, B.T.; writing—original draft, B.T. and writing—review and editing, T.S.; B.W.; and G.L. All authors have read and agreed to the published version of the manuscript.

Funding

The research was founded by Addis Ababa University, College of Natural and Computational Science and Moredocofe S.C.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data used to support the findings of this study are included within this article without restriction.

Acknowledgments

Authors deeply acknowledge Addis Ababa University and Moredocofe S.C. for financial support.

Conflicts of Interest

The author declares no competing interests with respect to research, authorship and/or publication of the article.

References

  1. Amare, T.; Birru, Y.; Hans, H. Effects of Guie on Soil Organic Carbon and Other Properties: A Traditional Soil Fertility Management Practice in the Central Highlands. J. Agric. Sci. 2013, 5, 236. [Google Scholar] [CrossRef] [Green Version]
  2. Meersmans, J.; De Ridder, F.; Canters, F.; De Baets, S.; Van Molle, M. A multiple regression approach to assess the spatial distribution of Soil Organic Carbon (SOC) at the regional scale (Flanders). Geoderma 2008, 143, 1–13. [Google Scholar] [CrossRef]
  3. Yigini, Y.; Panagos, P. Assessment of soil organic carbon stocks under future climate and land cover changes in Europe. Sci. Total Environ. 2016, 557–558, 838–850. [Google Scholar] [CrossRef] [PubMed]
  4. Wang, B.; Waters, C.; Orgill, S.; Cowie, A.; Clark, A.; Liu, D.; Simpson, M.; McGowen, I.; Sides, T. Estimating soil organic carbon stocks using different modelling techniques in the semi-arid rangelands of eastern. Ecol. Indic. 2018, 88, 425–438. [Google Scholar] [CrossRef]
  5. Lacoste, M.; Minasny, B.; McBratney, A.; Michot, D.; Viaud, V.; Walter, C. High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape. Geoderma 2014, 213, 296–311. [Google Scholar] [CrossRef]
  6. Kuang, B.; Tekin, Y.; Mouazen, A. Comparison between artificial neural network and partial least squares for on-line visible and near infrared spectroscopy measurement of soil organic carbon, pH and clay content. Soil Tillage Res. 2015, 146, 243–252. [Google Scholar] [CrossRef]
  7. Subburayalu, S.; Slater, B. Soil series mapping by knowledge discovery from an Ohio County soil map. Soil Sci. Soc. Am. J. 2013, 77, 1254–1268. [Google Scholar] [CrossRef]
  8. Behera, S.; Mathur, R.; Shukla, A.; Suresh, K.; Prakash, C. Spatial Variability of Soil Properties and Delineation of Soil Management Zones of Oil Palm Plantations Grown in a Hot and Humid Tropical Region of Southern India. Catena 2018, 165, 251–259. [Google Scholar] [CrossRef]
  9. Gallo, B.; Demattê, J.; Rizzo, R.; Safanelli, J.; Mendes, W.; Lepsch, I.; Lacerda, M. Multi-Temporal Satellite Images on Topsoil Attribute Quantification and the Relationship with Soil Classes and Geology. Remote Sens. 2018, 10, 1571. [Google Scholar] [CrossRef]
  10. Castro-Franco, M.; Córdoba, M.; Balzarini, M.; Costa, J. A pedometric technique to delimitate soil-specific zones at field scale. Geoderma 2018, 322. [Google Scholar] [CrossRef]
  11. Zhao, Z.; Yang, Q.; Sun, D.; Ding, X.; Meng, F. Extended Model Prediction of High-Resolution Soil Organic Matter Over a Large Area Using Limited Number of Field Samples. Comput. Electron. Agric. 2021, 169, 105172. [Google Scholar] [CrossRef]
  12. Hoover, C.; Ducey, M.; Colter, R.; Yamasaki, M. Evaluation of alternative approaches for landscape-scale biomass estimation in a mixed-species northern forest. Ecol. Manag. 2018, 409, 552–563. [Google Scholar] [CrossRef]
  13. Hashimoto, H.; Wang, W.; Melton, F.; Moreno, A.; Ganguly, S.; Michaelis, A.; Nemani, R. High-resolution mapping of daily climate variables by aggregating multiple spatial data sets with the random forest algorithm over the conterminous. Int. J. Climatol. 2019, 39, 2964–2983. [Google Scholar] [CrossRef]
  14. Chatterjee, S.; Santra, P.; Majumdar, K.; Ghosh, D.; Das, I.; Sanyal, S. Geostatistical approach for management of soil nutrients with special emphasis on different forms of potassium considering their spatial variation in intensive cropping system of West Bengal. Environ. Monit. Assess. 2015, 187, 1–17. [Google Scholar] [CrossRef]
  15. Li, J.; Heap, A. Spatial interpolation methods applied in the environmental sciences: A review. Environ. Model. Softw. 2014, 53, 173–189. [Google Scholar] [CrossRef]
  16. Singh, S.; Mustak, S.; Srivastava, P.; Szabo, S.; Islam, T. Predicting spatial and decadal LULC changes through cellular automata Markov chain models using earth observation datasets and geo-information. Environ. Process. 2015, 2, 61–78. [Google Scholar] [CrossRef] [Green Version]
  17. Azizi, A.; Malakmohamadi, B.; Jafari, H. Land use and land cover spatiotemporal dynamic pattern and predicting changes using integrated CA-Markov model. Glob. J. Environ. Sci. Manag. 2016, 2, 223–234. [Google Scholar]
  18. Menne, M.; Durre, I.; Vose, R.; Gleason, B.; Houston, T. An Overview of the Global Historical Climatology Network-Daily Database. J. Atmos. Ocean. Technol. 2012, 29, 897–910. [Google Scholar] [CrossRef]
  19. Huffman, G.; Bolvin, D.; Nelkin, E. Integrated Multi-Satellite Retrievals for GPM (IMERG), Late Run, Version V06A. 2014. Available online: http//jsimpson.pps.eosdis.nasa.gov/data/imerg/gis/ (accessed on 31 March 2019).
  20. Mayunga, S. Monitoring of Land Use/Cover Change Using Remote Sensing and GIS techniques: A case study of Loliondo Game Controlled Area. Trends J. Sci. Res. 2018, 3, 18–32. [Google Scholar]
  21. Akyürek, D.; Koç, Ö.; Akbaba, M.; Sunar, F. Land use/land cover change detection using multitemporal satellite dataset: A case study in Istanbul new airport. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2018, 42, 17–22. [Google Scholar] [CrossRef] [Green Version]
  22. Berhanu, T.; Teshome, S.; Bikila, W.; Gudina, L.; Merga, B. Woody species composition and community types of Hangadi watershed, Guji zone. BMC Ecol. Evol. 2021, 21, 225. [Google Scholar] [CrossRef]
  23. Central Statistical Agency, CSA. Population Projection of Ethiopia for Regions at District Level from 2014–2017. Addis Ababa; 2013. Available online: http://www.csa.gov.et (accessed on 3 January 2019).
  24. Walkley, A.; Black, I. An Examination of the Degtjareff Method for Determining Soil Organic Matter, and a Proposed Modification of the Chromoc Acid titration method. Soil Sci. 1934, 37, 29–38. [Google Scholar] [CrossRef]
  25. Henok, K.; Dondeyne, S.; Poesen, J.; Frankl, A.; Nyssen, J. Impact of deforestation on soil fertility, soil carbon and nitrogen stocks: The case of the Gacheb catchment in the White Nile Basin, Ethiopia. Agric. Ecosyst. Environ. 2017, 247, 273–282. [Google Scholar] [CrossRef] [Green Version]
  26. Chen, D.; Chang, N.; Xiao, J.; Zhou, Q.; Wu, W. Mapping Dynamics of Soil Organic Matter in Croplands with MODIS Data and Machine Learning Algorithms. Sci. Total Environ. 2019, 669, 844–855. [Google Scholar] [CrossRef] [PubMed]
  27. Tziachris, P.; Aschonitis, V.; Chatzistathis, T.; Papadopoulou, M. Assessment of spatial hybrid methods for predicting soil organic matter using DEM derivatives and soil parameters. Catena 2019, 174, 206–216. [Google Scholar] [CrossRef]
  28. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef] [Green Version]
  29. Houborg, R.; McCabe, M. A hybrid training approach for leaf area index estimation via Cubist and random forests machine-learning. ISPRS J. Photogramm. Remote Sens. 2018, 135, 173–188. [Google Scholar] [CrossRef]
  30. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2019. [Google Scholar]
  31. Kumar, P.; Sajjad, H.; Tripathy, B.; Ahmed, R.; Mandal, V. Prediction of spatial soil organic carbon distribution using sentinel-2A and field inventory data in Sariska Tiger Reserve. Nat. Hazards 2018, 90, 693–704. [Google Scholar] [CrossRef]
  32. Omuto, C.; Vargas, R. Re-tooling of Regression Kriging in R for Improved Digital Mapping of Soil Properties. Geosci. J. 2015, 19, 157–165. [Google Scholar] [CrossRef]
  33. Isaaks, E.; Mohan, S. An Introduction to Applied Geostatistics; Oxford Unviversity Press: New York, NY, USA, 1989. [Google Scholar]
  34. Wang, Y.; Zhang, X. A dynamic modeling approach to simulating socioeconomic effects on landscape changes. Ecol. Model. 2001, 140, 141–162. [Google Scholar] [CrossRef]
  35. Cressie, N. Statistics for Spatial Data, Revised Edition; John Wiley & Sons, Inc.: New York, NY, USA, 1993. [Google Scholar]
  36. Hengl, T.; Nikolić, M.; MacMillan, R. Mapping efficiency and information content. Int. J. Appl. Earth Obs. Geoinf. 2013, 22, 127–138. [Google Scholar] [CrossRef]
  37. Abdulrahman, A.; Ameen, S. Predicting Land use and land cover spatiotemporal changes utilizing CA-Markov model in Duhok district between 1999 and 2033. Acad. J. Nawroz Univ. 2020, 9, 71–80. [Google Scholar] [CrossRef]
  38. Regmi, R.; Saha, S.; Balla, M. Geospatial analysis of land use land cover change modeling at Phewa lake watershed of Nepal by using cellular automata Markov model. Int. J. Curr. Eng. Technol. 2014, 4, 2617–2627. [Google Scholar]
  39. Zhilong, Z.; Xue, W.; Yili, Z.; Jungang, G. Assessment of changes in the value of ecosystem services in the Koshi River Basin, central high Himalayas based on land cover changes and the CA-Markov model. J. Resour. Ecol. 2017, 8, 67–77. [Google Scholar] [CrossRef]
  40. Behera, D.; Mukun, D.; Borate, S.; Panda, S.; Behera, P.; Roy, P. Modelling and analyzing the watershed dynamics using Cellular Automata (CA)– Markov model–A geo-information-based approach. J. Earth Syst. Sci. 2012, 121, 1011–1024. [Google Scholar] [CrossRef] [Green Version]
  41. Ghosh, P.; Mukhopadhyay, A.; Chanda. Application of cellular automata and Markov-chain model in geospatial environmental modeling—A review. Remote Sens. Appl. Soc. Environ. 2017, 5, 64–77. [Google Scholar] [CrossRef]
  42. Camara, M. Integrating cellular automata Markov model to simulate future land use change of a tropical basin. Glob. J. Environ. Sci. Manag. 2020, 6, 403–414. [Google Scholar]
  43. Subedi, P.; Subedi, K.; Thapa, B. Application of a hybrid cellular automaton-Markov (CA-Markov) model in land-use change prediction: A case study of saddle creek drainage basin, Florida. Appl. Ecol. Environ. Sci. 2013, 16, 126–132. [Google Scholar] [CrossRef] [Green Version]
  44. Kumar, S.; Radhakrishnan, N.; Mathew, S. Land use change modelling using a Markov model and remote sensing. Geomat. Nat. Hazards Risk 2014, 5, 145–156. [Google Scholar] [CrossRef]
  45. Mishra, V.; Rai, P.; Mohan, K. Prediction of land use changes based on land change modeler (LCM) using remote sensing: A case study of Muzaffarpur (Bihar). J. Geogr. Inst. Jovan Cvijic SASA 2014, 64, 111–127. [Google Scholar] [CrossRef]
  46. USDA/NRCS. (United States Department of Agriculture/ Natural Resources Conservation Service). Soil Health- Bulk Density/Moisture/Aeration- Guide for Educators. 2014. Available online: https://www.academia.edu/5114008/Guides_for_Educators (accessed on 5 March 2021).
  47. Tekalign, T. Soil, Plant, Water, Fertilizer, Animal Manure and Compost Analysis; Working Document No. 13; International Livestock Research Center for Africa: Addis Ababa, Ethiopia, 1991. [Google Scholar]
  48. Majoro, F.; Niyonzima, A.; Mukamwambali, C.; Mugabushaka, A.; Twagirimana, A. Digital Soil Nutrients Mapping in Sebeya Catchment Agricultural Land. Am. J. Innov. Res. Appl. Sci. Rwanda. 2020, 11, 26–34. [Google Scholar]
  49. Yimer, F.; Ledin, S.; Abdelkadir, A. Changes in Soil Organic Carbon and Total Nitrogen Contents in Three Adjacent Land Use Types in the Bale Mountains, South-Eastern Highlands. For. Ecol. Manag. 2007, 242, 337–342. [Google Scholar] [CrossRef]
  50. Kizilkaya, R.; Dengiz, O. Variation of Land Use/Cover Effects on Some Soil Physico-Chemical Characteristics and Soil Enzyme Activity. Zemdirb. Agric. 2010, 97, 15–22. [Google Scholar]
  51. Weldemariam, S.; Eyasu, E.; Girmay, G. The Effects of Land Use and Landscape Position on Soil Physicochemical Properties in a Semiarid Watershed, Northern. Appl. Environ. Soil Sci. 2020. [Google Scholar] [CrossRef]
  52. Emiru, N.; Gebrekidan, H. Effect of land use changes and soil depth on soil organic matter, total nitrogen and available phosphorus contents of soils in Senbat watershed, western. J. Agric. Biol. Sci. 2013, 8, 206–212. [Google Scholar]
  53. Alemayehu, K.; Sheleme, B. Effects of the Different Land-Use Systems on Selected Soil Properties in South Ethiopia. J. Soil Sci. Environ. Manag. 2013, 4, 100–107. [Google Scholar]
  54. Belayneh, B.; Eyasu, E.; Getachew, A. Effects of landscape positions on soil physicochemical properties at shenkella watershed, south central. Environ. Syst. Res. 2021, 10, 14. [Google Scholar] [CrossRef]
  55. Hazelton, P.; Murphy, B. Interpreting Soil Test Results: What Do All Numbers Mean? 2nd ed.; CSIRO Publishing: Collingwood, Australia, 2007. [Google Scholar] [CrossRef]
  56. Gazey, C.; Davis, S. Soil Acidity: A Guide for West Australia Farmers and Consultants; Department of Agriculture and Food: Western Australia: Perth, Australia, 2009. [Google Scholar]
  57. Landon, J. Booker Tropical Soil Manual: A Handbook for Soil Survey and Agricultural Land Evaluation in the Tropics and Subtropics; Longman Scientific and Technical: Essex, NY, USA, 1991; p. 474. [Google Scholar]
  58. Worku, G.; Bantider, A.; Temesgen, H. Effects of land use/land cover change on some soil physical and chemical properties in Ameleke micro-watershed Gedeo and Borena Zones. South. J. Environ. Earth Sci. 2014, 4, 13–24. [Google Scholar]
  59. Noi, P.; Degener, J.; Kappas, M. Comparison of multiple linear regression, cubist regression, and random Forest algorithms to estimate daily air surface temperature from dynamic combinations of MODIS LST data. Remote Sens. 2017, 9, 398. [Google Scholar] [CrossRef] [Green Version]
  60. Nabiollahi, K.; Eskandari, S.; Taghizadeh-Mehrjardi, R.; Kerry, R.; Triantafilis, J. Assessing Soil Organic Carbon Stocks Under Land Use Change Scenarios Using Random Forest Models. Carbon Manag. 2019, 10, 63–77. [Google Scholar] [CrossRef]
  61. Al-Awadhi, A. Case Assessment of the Mechanisms Involved Inhu-man-Induced Land Degradation in North Eastern Kuwait. Land Degrad. Dev. 2013, 24, 2–11. [Google Scholar] [CrossRef]
  62. Kuhn, M. Caret: Classification and Regression Training. R Package Version 6.0-86. 2020. Available online: https://CRAN.R-project.org/package=caret (accessed on 2 March 2020).
  63. Nasir, J.; Arif, M.; Gyu, S.; Imam, A. A novel improved random forest for text classification using feature ranking and optimum number of trees. J. King Saud Univ.-Comput. Inf. Sci. 2022, 24, 2733–2742. [Google Scholar]
Sustainability 15 01066 g001 550
Figure 1. Showing the geographical location of the studied watershed.
Figure 1. Showing the geographical location of the studied watershed.
Sustainability 15 01066 g001
Sustainability 15 01066 g002 550
Figure 2. Rainfall and temperature diagram of Odo Shakiso district [22].
Figure 2. Rainfall and temperature diagram of Odo Shakiso district [22].
Sustainability 15 01066 g002
Sustainability 15 01066 g003 550
Figure 3. Projected LULC changes for 2048 [22].
Figure 3. Projected LULC changes for 2048 [22].
Sustainability 15 01066 g003
Sustainability 15 01066 g004 550
Figure 4. Importance of the predictors of pH derived from random forest.
Figure 4. Importance of the predictors of pH derived from random forest.
Sustainability 15 01066 g004
Sustainability 15 01066 g005 550
Figure 5. Importance of the predictors of CEC derived from random forest.
Figure 5. Importance of the predictors of CEC derived from random forest.
Sustainability 15 01066 g005
Sustainability 15 01066 g006 550
Figure 6. Importance of the predictors of OC derived from random forest.
Figure 6. Importance of the predictors of OC derived from random forest.
Sustainability 15 01066 g006
Sustainability 15 01066 g007 550
Figure 7. pH (upper left), CEC (upper right) and OC (bottom left) prediction map with ordinary kriging for 2018.
Figure 7. pH (upper left), CEC (upper right) and OC (bottom left) prediction map with ordinary kriging for 2018.
Sustainability 15 01066 g007
Sustainability 15 01066 g008 550
Figure 8. PH (upper left), CEC (upper right) and OC (bottom left) prediction map with random forest for 2018.
Figure 8. PH (upper left), CEC (upper right) and OC (bottom left) prediction map with random forest for 2018.
Sustainability 15 01066 g008
Table
Table 1. Area (ha) distribution of LULC types during 1988–2018 [22].
Table 1. Area (ha) distribution of LULC types during 1988–2018 [22].
LULC Type198820082018
Area (ha)%Area (ha)%Area (ha)%
Agroforestry113632818.9823.271575.444.8
Cultivated land517151235.9935.12955.127.1
Forest land1866531464.4641.61988.428.1
Total35191003519.431003519100
Table
Table 2. Showing selected soil physicochemical values in the watershed.
Table 2. Showing selected soil physicochemical values in the watershed.
Land UseParticle Size Distribution (%)Bulk Density (g/cm3)Altitude (m)Slope (%)pH
(pHmeter)		OM (%)CEC (cmol)
SandSiltClay
Forest61.6 a18 a21 a1.22 a2122 a17 a5.2 a4.2 a30 a
Agroforestry62.1 a17 a20 a1.22 a2098 a24.6 b5.4 b3.8 b28 a
Cultivated
Land		53.4 b27 b20 a1.23 a1889 b22.9 b5.6 c3.4 c29 a
Note: Values followed by the same letter within a column are not significantly different.
Table
Table 3. Best-fitted semi-variogram model parameters for pH, CEC and OC.
Table 3. Best-fitted semi-variogram model parameters for pH, CEC and OC.
ParametersNuggetPsillRangeNugget/Psill
pH0.00210.04390.000.00
CEC0.0020.0110010.2
OC0.000.011370.00
Table
Table 4. Comparison of mapping accuracy obtained from models.
Table 4. Comparison of mapping accuracy obtained from models.
ParameterOKRF
MSERMSEMAER2MSERMSEMAER2
OC0.130.360.150.850.020.170.020.82
pH0.020.150.540.900.010.060.010.89
CEC12.93.60.140.9210.720.950.070.91
Table
Table 5. Transitional probability matrices of land use and land cover changes.
Table 5. Transitional probability matrices of land use and land cover changes.
ForestAgroforestryCultivated Land
Forest0.52130.35010.1287
Agroforestry0.01310.61080.3762
Cultivated land0.00120.44220.5565
Table
Table 6. Area coverage of each land use/cover and percentage projection for 30 years.
Table 6. Area coverage of each land use/cover and percentage projection for 30 years.
NoLand UseArea (ha)Area (%)
1Forest539.9715.34
2Agroforestry1730.4249.16
3Cultivated land1249.5435.50
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tamiru, B.; Soromessa, T.; Warkineh, B.; Legese, G. Mapping Soil Parameters with Environmental Covariates and Land Cover Projection in Tropical Rainforest, Hangadi Watershed, Ethiopia. Sustainability 2023, 15, 1066. https://doi.org/10.3390/su15021066

AMA Style

Tamiru B, Soromessa T, Warkineh B, Legese G. Mapping Soil Parameters with Environmental Covariates and Land Cover Projection in Tropical Rainforest, Hangadi Watershed, Ethiopia. Sustainability. 2023; 15(2):1066. https://doi.org/10.3390/su15021066

Chicago/Turabian Style

Tamiru, Berhanu, Teshome Soromessa, Bikila Warkineh, and Gudina Legese. 2023. "Mapping Soil Parameters with Environmental Covariates and Land Cover Projection in Tropical Rainforest, Hangadi Watershed, Ethiopia" Sustainability 15, no. 2: 1066. https://doi.org/10.3390/su15021066

APA Style

Tamiru, B., Soromessa, T., Warkineh, B., & Legese, G. (2023). Mapping Soil Parameters with Environmental Covariates and Land Cover Projection in Tropical Rainforest, Hangadi Watershed, Ethiopia. Sustainability, 15(2), 1066. https://doi.org/10.3390/su15021066

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop