Machine Learning Models to Develop Land Suitability Map for Coffee Cultivation by Integrating CHIRPS and SRTM DEM

doi:10.21203/rs.3.rs-5316099/v1

Download PDF

Research Article

Machine Learning Models to Develop Land Suitability Map for Coffee Cultivation by Integrating CHIRPS and SRTM DEM

https://doi.org/10.21203/rs.3.rs-5316099/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Kodagu region is a major coffee exporter, with production concentrated in three taluks, including the Somwarpet Taluk. Coffee yields have decreased due to unfavorable factors such as climate change, disease and insect outbreaks, landslides and inadequate land-use planning in turn affecting the family income. Thus, the goal of this research is to identify suitable land for cultivation of coffee based on Food and Agriculture Organization (FAO) land suitability assessment methodology for Somwarpet Taluk of Kodagu District. For this purpose, six soil chemical properties (potential of hydrogen, electrical conductivity, organic carbon, sulphur, iron, potassium and nitrogen), two topographic data (elevation and slope) and one climatic condition (rainfall) was considered to map land suitability for coffee crops. After determining land suitability classes for coffee cultivation, the study area was then mapped using machine learning (ML) methods such as random forest (RF), Naive Bayes (NB), K-Nearest Neigbhor (KNN), Extreme Gradient Boosting Tree (XgBoost) and Decision Tree (DT). The prediction of land suitability classes by ML model showed a significant variation. For example, in case of RF model, results showed the 94% of higher accuracy when compared to the XgBoost (93.5), DT (92%), NB (75%) and KNN (50%) models.

The area of S1 (highly suitable) classified through RF, XgBoost, DT, NB

and KNN was 8.66%, 8.75%, 8.57%, 19.17% and 28% respectively. Similarly, the S2 (moderately suitable) class area via RF, XgBoost, DT, NB and KNN was 84.17%, 82.18%, 81.33 %, 69.61% and 44%, respectively. Conversely, the area of S3 (marginally suitable) classified through RF, XgBoost, DT, NB and KNN was 6.64%, 7.64%, 8.5%, 10.52% and 27.8%. Correspondingly, the N (unsuitable) class area via RF outperformed the land suitability class for XgBoost, DT, NB and KNN by 0.53%, 1.43%, 1.6%, 0.7% and 0.2%. The

sulphur and pH were the major limiting factor affecting the land suitability to map coffee cultivation. Thus, the methodologies developed in this study area can be very useful tool to ensure food security and carry out an effective assessment of land suitability in coffee crop growth and production for Somwarpet Taluk of Kodagu District, Karnataka State.

Machine Learning (ML) Models

Principal Component Analysis (PCA)

SRTM DEM

Geographic Information System (GIS)

CHIRPS

Land Suitability Mapping

Coffee Cultivation

Coffee is a woody perennial evergreen dicotyledonous plant that doesn’t shade its leaves for the entire year and it is a member of the Rubiaceae family (Thanuja & Singh, 2017; Weldon, 2016). Now a days, only two main species are used to cultivate coffee crop (Arabica coffee, or Coffea arabica) accounting to about 75–80% of the coffee that are produced globally. Coffea canephora, often known as Robusta coffee, makes up about 20% of all coffee and has a flavor which is distinct from Arabica coffee (Thanuja & Singh, 2017). Both these species share the same characteristics, namely a vertical main trunk and primary, secondary, and tertiary branches that are typically plagiotrophic in nature. If they are left unpruned, they can grow up to ten meters in height, however upkeep is always necessary for convenient harvesting (Abdisa, 2020; Weldon, 2016).

The main plantation crop grown in India’s southern states is coffee, with a tiny amount of coffee produced in the northern states of India. Where Kerala, Tamil Nadu, and Karnataka are India’s three largest coffee-growing states. This first spot belongs to Karnataka, which produces 71.03 percent of all coffee in India. The only three districts in Karnataka where coffee is grown are Kodagu, Chikmagalur, and Hassan. Approximately 45.66%, 38.99%, and 15.45% of the state’s total area and 54.06, 34.10, and 11.84 percent of its total coffee production are accounted for by these districts (Thanuja & Singh, 2017).

India is an agrarian country, and the production of coffee is important to the national economy. Since, India’s economy is totally dependent on agriculture. India is now the world’s sixth-largest producer of coffee in terms of productivity. India is the nation most impacted by climate change, especially in the southern regions where coffee is grown, according to the 2015 Global Climate Risk Index. Thus, weather, topographic and soil properties play a crucial role in the establishment and development of coffee plantations, leading to significant intra-seasonal production fluctuation. As a result, there are notable differences in coffee production performance amongst the nation’s various geographic areas. Therefore, according to India’s annual economic report, farmers make 20–25% less money as a result of poor weather and soil related problems (Chengappa and Devika, 2016).

The earth’s average temperature has been rising in recent years due to climate uncertainty brought on by pollution, solid waste management, population development, surface and global warming, and ground water hydrology. These changes have resulted in drastic alterations to the ecosystem. Today’s impoverished farmers have significant obstacles in growing coffee crops because of the frequent occurrence of droughts, heavy rains, floods, and landslides (Chengappa and Devika, 2016). Additionally, the unavailability of labor (small holders), high capital income to develop road infrastructure, transportation facilities, production facilities, and open residential land are the challenging issues faced by the farmers as most of the coffee location are situated in mountainous regions (Pagiu et al., 2020). In upcoming days, the land use for coffee farming may be less likely due to limited land availability and government regulations i.e., government are pushing farmers to convert from less productive annual crops to perennial grasslands, plantations, etc. in order to transition to organic farming (ThiTuyen et al., 2019). Thus, these factors may have a bigger influence on coffee cultivation, and this risk also affects the ongoing production of food (Mishra et al., 2016; Sharma et al., 2021).

Further it leads to other major issues that need to be resolved as soon as

possible, like food scarcity, deforestation brought on by urbanization, deteriorating agricultural land, etc. The most important step in achieving this is to complete the land suitability process. Therefore, by taking into account land assessments with comparable specific land abilities and comparing them with land use needs, the land suitability technique gives agronomists a better explanation of all such concerns. Land suitability is the process of determining the potential of land types by knowing the relationship between the circum- stances of uses and the land to which they are placed. A new model and analysis are essential with use of large number of datasets as input variables for this kind of operation.

Up to date, many researchers have applied methods to regulate the weights of factors for production of coffee crop to find out suitable land like Geographic Information System (GIS) approaches (Muliasari et al., 2022; Salas Lopez et al., 2020; Karim et al., 2020; Rono and Mundia, 2016; Nurfadila et al., 2020; Hidayat et al., 2020; Rahmatika et al., 2022; Hidayat et al., 2020; Abdisa, 2020; Ndabasanze, 2018; Bich and Phuong, 2023; Fachruddin et al., 2020; Nugraha, Prayitno, and Khoiriyah, 2021; Pravitasari et al., 2023; Chairani et al., 2017; Zhang et al., 2021; Eshete, 2022; Gross, 2014; Musli- hah et al., 2020; Ranjitkar et al., 2016; Nzeyimana et al., 2014; Neupane and Pangali Sharma, 2022; Pravitasari et al., 2023; Bindumathi, 2014; Nuary et al., 2022; Mighty, 2015; Rahmawaty et al., 2021; Barus et al., 2015; Hartono et al., 2018;Dermawan et al., 2018;Silaban et al., 2016; Weldon, 2016; Salima et al., 2012; Santoso and Ayu, 2022; Purba et al., 2019; Chairuddin et al., 2023; Tram et al., 2023; Pham et al., 2021; Gruter et al., 2022; Zhang et al., 2021; Ochoa et al., 2017; Gomes et al., 2020; D’haeze et al., 2005; Chemura et al., 2015; Raharja, 2016; Auliansyah, 2019; Liem & Duyen, 2020; Azsari et al., 2022), Fuzzy (Nurfadila et al, 2020; Neswati et al., 2023; Ranjitkar et al., 2016; Bindumathi, 2014; Zhang et al., 2021), AHP (Salas Lopez et al., 2020; Rono and Mundia, 2016; Abdisa, 2020; Zhang et al., 2021; Neupane and Pangali Sharma, 2022; Bindumathi, 2014; Mighty, 2015; Weldon, 2016; Pham et al., 2021; Zhang et al., 2021), parametric method (Marbun et al, 2019; Lop- ulisa et al., 2020; Juita et al., 2021; Nurdin et al., 2022; Pagiu et al., 2020; Juita et al., 2020), Non parametric methods like chi square test, ANOVA, Wilcoxon, Kolmogorov Smirnov tests (Ryder, 1994), MaxEnt (Zhang et al., 2021; Nuary et al., 2022; Purba et al., 2019; Gruter et al., 2022; Gomes et al., 2020; Chemura et al., 2015) and Weighted Overlay Analysis (Muliasari et al., 2022; Salas Lopez et al., 2020; Rono and Mundia, 2016; Rahmatika et al., 2022; Abdisa, 2020; Ndabasanze, 2018; Marbun et al., 2023; Chairani et al., 2017; Nzeyimana et al., 2014).

The literature revealed that traditional Multi-criteria Decision Analysis (MCDA) approaches to mapping land suitability for different agricultural crops are limited by a number of issues. These include: (i) AHP and fuzzy methods did not perform well in controlling land feature weights and computing LS values (Pramanik, M. K., 2016; Mistri & Sengupta, 2019; Reza et al., 2020; Zhang et al., 2015); (ii) AHP is limited to five to nine covariate or covariate groups and must closely monitor the upper limit of 0.10 CR, as advised by Saaty’s method, so that the inclusion of extra significant criteria results in further complex processing (Radocaj et al., 2021; Zhang et al., 2015); (iii) To evaluate the relative importance of the criteria discretely, the AHP method requires information from experts or multiple scientists and further if a variety of datasets are available, then using scientific data or objectives are required while deriving the pairwise comparison matrix (Zhang et al., 2015; Talukdar et al., 2022); (iv) Precaution should be taken when utilizing the AHP as any inaccurate parameter assessment may affect the weights and scores that are allocated (Hassan et al., 2020; Ramamurthy et al., 2020). Consequently, MCDA and AHP 9approaches are time consuming and cost-effective procedures in ground surveying and sampling without covering the spatio-temporal properties. Thus, Machine Learning (ML) models offer a potential remedy for the GIS-based MCDA in land suitability’s aforementioned shortcomings to map suitable zones in large spatio-temporal scales. Also, ML provides more complex computational efficiency and accuracy than traditional methodologies. Thus, through several research studies, it was suggested that ML algorithms outperform other modelling techniques for non-linear correlations to a variety of elements and environmental characteristics (Radocaj et al., 2021).

According to the previous researchers, a very few studies have been conducted or tested using machine learning like Bayesian network (Lara-Estrada et al., 2017; Lara-Estrada et al., 2021; Estrada, 2019), PCA (Irawan et al., 2022), MDA (Irawan et al., 2022), Correlation test (Irawan et al., 2022), Regression (Lopez-Carmona et al., 2023; Neswati et al., 2023; Marbun et al., 2023), Decision tree (Trangmar et al., 1995), ALES model (D’haeze et al., 29005; Trangmar et al., 1995) and Ecological niche modelling (Ranjitkar et al., 2016) in terms of land suitability and land capability for production of coffee crop. From the literature, it was noticed that an inadequate study on machine learning models to identify the suitable zones for coffee cultivation in the Malnad region of Kodagu district was carried out (Bindumathi, 2014). These limitations so prompted this work to suggest a new contribution to the literature. This study provides information on the coffee suitability zones and may improve regional land use laws.

The main objective of this work is to compare GIS based machine learning techniques to map the suitability of land in Somwarpet Taluk of Kodagu district for cultivation of coffee. To build decisional maps based on machine learning, a set of eight soil chemical properties (N, pH, Fe, S, K, EC and OC), one climatic factor (rainfall) and two topographic parameters (slope and elevation) were selected as the spatial input. Therefore, by taking into account of such parameters, this methodology can aid systematic assessment of coffee cultivation for the actual status of land suitability in Somwarpet Taluk of Kodagu district. Thus, this proposed approach is unique and a helpful tool for decision makers as it combines soil macronutrients data, topography and climatic data with use of soil micronutrients contents. The results of the study will help stakeholders and decision-makers to accomplish sustainable development objectives and keep raising regional coffee production output in Somwarpet Taluk of Kodagu district.

2.1. Description of Experimental Region

The research was carried out for the year 2023. Study area is comprised of Somwarpet Taluk, Kodagu District, Karnataka state, India, which is located between 12° 35’ 47.7348’ N latitude and 75° 50’ 41.7084’ E longitude as shown in Fig. 1. There are 158 villages and 40gram panchayats in this taluk. Figure 1 depicts the location map of Somwarpet Taluk study area. Somwarpet Taluk enjoys a tropical climate with mild to medium humidity levels due to its proximity to the coast. Rainfall, high humidity, and a moder- ate summer are some of its well-known, enjoyable, and healthful attributes.

This taluk depends on the rainfall for all agricultural needs. In contrast, there was an average of 2105 mm of annual rainfall during the 2013–2022 period. The taluk is classified as piedmont and hilly zones based on geomorphology with between 70 and 80 percent of the total land falling into these groups. The terrain of the taluk is diverse, consisting of piedmont zones, plains, rivers, streams, tanks, and reservoirs. In plain land, the master slo9pe runs from northeast to southwest. The entire topography elevation of the taluk fluctuates, with the west at 1200 meters above sea level and the east at 800 meters above sea level. The drainage infrastructure in this taluk is quite advanced. The basic drainage pattern used in the study area is dendritic to sub-dendritic. Pre-Cambrian igneous and metamorphic rocks that are either surface-exposed or have a thin layer of transported and residual soils on top of them make up the majority of the study area’s geology. The soil types of the taluk are clayey sketal, clayey, and lateritic. It can retain moisture well and is productive. Since these soils are fertile, they usually yield well. The southwestern region has lateralitic soil, which is perfect for paddy, sugarcane, arecanuts, and plantation crops like plantains and cardamom. It is further distinguished by elevated concentrations of iron and aluminum. The textures of the soil vary, ranging from fine to coarse. The soil in valleys and on intermediate slopes has a lot more loam than the soil on higher slopes, which has considerably coarser dirt (Official website of Central Ground Water Board, 2023). Coffee and spice crops like cardamom, peppers, oranges, ginger, and vegetables are the main crops farmed in the region. The main crop grown in the area is coffee. Coffee is grown in accordance with the unique environmental circumstances of the area. The main crop grown in the Somwarpet Taluk is arabica coffee. Conversely, the 158 villages that comprise Somwarpet Taluk were planted with 24050 hectares of robusta and 6870 hectares of arabica, with a total bearing area of 22400 and 5890 hectares. Over a five-year period, the Somwarpet Taluk produced an average of 16141 and 6749 metric tons of arabica and robusta coffee (i.e., total 22890 metric tons) and with average yielding of total 813 kg/ha (Official website of Kodagu District, 2023).

2.2. Soil Data

A total of 1384 soil samples, or 87 soil samples from each of the 16 blocks in Somwarpet Taluk, were gathered from the Soil Testing Center in Kudige. Soil sample collection was carried out for 16 blocks of Somwarpet Taluk of Kodagu District i.e., Abhimata, Beluru Basavanhalli, Besuru, Byadagotta, Doddakundha, Gonimaruru, Kalkandhuru, Mulluru, Nandhigundha, Thalthare Shetalli, Hebbale, Heggadalli, Heruru, Kodagarahalli, Thorenuru, and Ulluguli. A homogenous distribution of the sample locations within the corresponding blocks was taken into consideration during this selection. All these soil samples were subjected to a chemical analysis at the soil testing facility located in Kudige, Kodagu district for 2023 year.

2.3. Climate Data

With a 0.05° high resolution, the Climate Hazards group Infrared Precipitation with Stations (CHIRPS) environmental record presents a new quasi- global dataset of daily, pentadal and monthly rainfall for the 50°S and 50°N regions, covering the period from 1981. CHIRPS generates gridded rain- fall time series for seasonal drought monitoring and trend analysis by fusing satellite photos at a resolution of 0.05° with in-situ station data. As of February 12, 2015, CHIRPS data for version 2.0 is complete and available to the general public via the website https://www.chc.ucsb.edu/data/chirps. It was automated to help the US Agency for International Development’s Famine Early Warning Systems Network (FEWS NET). CHIRPS data aggregates ranging from six hours to three months are readily available to end users. Nearly every dataset has a spatial resolution of 0.05° × 0.05° degrees. CHIRPS data are accessible in NetCDF, GeoTiff and Esri BIL formats. The measuring units are in the form of millimeters (mm) per time period, or mm per pentad, month and day (Funk et al., 2015).

2.4. Topographic Data

Digital elevation maps (DEMs) with a spatial resolution of one arc (30 m) were obtained for the Shuttle Radar Topographic Mission (SRTM) database using the https://earthexplorer.usgs.govwebsite.

2.5. Methodology

The research is conducted to improve the performance results by updating land suitability for cultivation of coffee in Somwarpet Taluk of Kodagu district using Machine Learning classification processes based on climate, soil and topographic factors as shown in Fig. 2.

2.6. Land Suitability Assessment

2.6.1. Processing of Soil Data

The soil data that was obtained from the Kudige Soil Testing Center was in the form excel spreadsheets. To make sure that none of the data that represented the soil data had any obvious errors or flaws, they were meticulously reviewed on an Excel spreadsheet. The latitudes and longitudes of soil sample locations, along with other spatial data, were integrated with relevant soil data in an MS Excel file that included the potential of hydrogen (pH), Sulphur (S), Iron (Fe), Nitrogen (N), Electrical Conductivity (EC), Potassium (K) and Organic Carbon (OC). To enable further processing, the results of the Excel file data were exported to the ArcGIS 10.8 edition of the program.

Next, in order to map the spatial reference with the research area boundary, XY data are displayed using X field as longitude and Y field as latitude with Geographic Coordinate Systems (GCS) for global datum of World Geodetic System 1984 (WGS 1984). Later, a geostatistical interpolation technique known as Interpolation Distance Weighting (IDW) was utilized to interpolate point data into continuous surface for Ph, N, OC, Fe, EC, K and S soil data using spatial analyst tools in ArcGIS. This allowed for the generation of a raster image from the shown XY data. Finally, using the ”Extract by Mask” feature found in the ArcGIS spatial analysis tool, the resultant image was extracted in accordance with research area boundary shapefile projections of WGS 1984 for UTM Zone of 43N. Additionally, using the ”Resample” function from the data management tool’s raster option, which contains a raster processing tool, all of the soil data were resampled using the bilinear resampling technique to produce output cell sizes with a spatial resolution of 30 meters as shown in Figs. 3, 4, 5 and 6.

2.6.2. Processing of Climate Data

For a period of 43 years (1981–2023), global annual precipitation data was taken from CHIRPS: Rainfall Estimates from Rain Gauge and Satellite Observations in Geo Tiff format. Then, the rainfall raster data was imported into ArcGIS for additional processing. Next, the file was saved to a specified location once the loaded data were exported into the current data frame of spatial reference. The exported data are then projected to WGS 1984 for the UTM Zone of 43N, and they are then exported once more to the full length of a tiff file’s raster dataset. The ”Composite Bands” function from the data management tool of the raster option, which contains a raster processing tool, is then used to composite the raster dataset. Additionally, the rainfall data is resampled using the Bilinear resampling technique with the ”Resample” function to produce an output cell size of 30m spatial resolution. The resulting image is then clipped using the ”Clip” function to study area boundary shapefile projections of WGS 1984 for UTM Zone of 43N by choosing the use input features for clipping geometry option from the data management tool of raster option containing raster processing tool.

Furthermore, an overlay statistics of sum value using the ”Cell Statistics”

function is used from spatial analyst tools of local tool in ArcGIS software in order to compute a function of cell values from all inputs at that loca tion to value of each and every location on the output composited raster data. Next, a conversion tool is used to transform the cell statistics data into raster to point data. Interpolation Distance Weighting (IDW), a geo-statistical interpolation approach employing spatial analyst tools in ArcGIS, was thereafter used to produce raster images from the point data in order to offer representative rainfall values throughout the whole research area. Finally, using the ”Extract by Mask” feature found in the ArcGIS spatial analysis tool, the resultant image was extracted in accordance with research area boundary shapefile projections of WGS 1984 for UTM Zone of 43N as shown in Fig. 7.

2.6.3. Processing of Topographic Data

The SRTM DEM data were exported as raster Geo Tiff files to ArcGIS

10.8 software. Using the ”Mosaic” option from the Image analysis tool, the supplied DEM data are mosaiced to create a single raster dataset from several raster datasets. As a result, in ArcGIS 10.8 software, the elevation data was produced utilizing the ”Hill Shade” option from the Arc Toolbox properties of the surface tool as shown in Fig. 5. Subsequently, the raster image yielded a continuous surface as the result. The ”Clip” feature in the data management application was then used to clip the resultant image based on the study area border shapefile. Using the ”Slope” option from the Arc Toolbox properties of surface tool in ArcGIS 10.8 software, a slope was calculated from the elevation data after it had been converted to a raster image as shown in Fig. 8a and b.

2.6.4. Determining Land Suitability Index Using Principal Component Analysis

In this research, a qualitative land assessment was conducted to establish land suitability classes for coffee cultivation. Soil pH, electrical conductivity, nitrogen, iron, sulphur, organic carbon, potassium and rainfall as well as DEM data (elevation and slope), were utilized to create land suitability indices. The pixel values of each variable are then extracted to each sample point using the ArcGIS spatial analysis tool named as multi-value extraction to points function to create a complete sample set of data. The selected factors were graded according to Table 1. Where, Table 1 displays the growth requirement tables that affecting the cultivation of coffee crop acquired through various literature reviews and expert opinions from coffee board centre.

After the selection of auxiliary variables, principal component analysis (PCA) technique was applied to compute land suitability indices for a coffee cultivation based on the reference Table 1. Principal component analysis (PCA) is a multivariate statistical approach that reduces numerous variables to a few composite variables using the concept of dimensionality reduction. During PCA, as many properties of the original sample as feasible are preserved (Wei and Zhou, 2023). PCA method reduces correlations between variables, resulting in more stable and interpretable composite indices. In- stead of using arbitrary or expert weights, PCA provides an objective approach for weighting variables in the index based on the amount of variation explained by each component; in contrast, the researcher (Irawan et al., 2022) used PCA and Weighting index to determine soil fertility index by selecting indicators with the best sensitivity. package is applied with the help of R 4.3.2 software to perform PCA approach. This approach first handles any missing values and then eradicate the columns with constant values (i.e., Zero variance) as shown in the Eq. (1):

$$\:Var\left({x}_{j}\right)=0\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(1\right)$$

Then standardization of data is performed to scale each feature to have zero mean and unit variance as per the Eq. (2):

$$\:{X}_{scaled,\:j}=\:\frac{{x}_{j}-{\mu\:}_{j}}{{\sigma\:}_{j}}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(2\right)$$

Where $\:{X}_{scaled,\:j}$: represents original data matrix, $\:{\mu\:}_{j}$: indicates mean of column j, $\:{\sigma\:}_{j}$: indicates standard deviation of column j.

Table 1

Depicts the growth requirement thresholds and degrees for coffee cultivation. (LSF: Land Suitability Factor; pH: Potential of Hydrogen; N: Nitrogen; OC: Organic Carbon; Fe: Iron; S: Sulphur; K: Potassium; EC: Electrical Conductivity; S1: Highly Suitable; S2: Moderately Suitable; S3: Marginally Suitable; N: Unsuitable).
LSF	LSF Range	LSF Class	Adapted From
Soil Conditions
pH	< 4.0 and > 8.5	N	Salas Lopez et al., 2020; Estrada, 2019; Neupane et al., 2022; Bindumathi, 2014
	4.0-4.4 and 7.6–8.5	S3
	4.5–4.9 and 7.1–7.5	S2
	5–7.0	S1
OC (%)	< 0.6	S3	Bindumathi, 2014
	0.6–1	S2
	> 1	S1
	--	N
N (Kg/ha)	< 280	S3	Expert judgement
	280–560	S2
	> 560	S1
	--	N
Fe (ppm)	< 16	S3	Expert judgement
	16–72	S2
	> 72	S1
	--	N
S (ppm)	< 16	S3	Expert judgement
	16–32	S2
	> 32	S1
	--	N
K (Kg/ha)	< 141	S3	Expert judgement
	141–336	S2
	> 336	S1
	--	N
EC (ds/m)	< 1.0	S1	Bindumathi, 2014
	1.0–1.5	S2
	1.5–2.0	S3
	> 2.0	N
Topographic Factor
Slope (degree)	< 15	S1	Hidayat et al., 2020; Hidayat et al., 2020; Neupane et al., 2022; Bindumathi, 2014
	15–35	S2
	35–50	S3
	> 50	N
Elevation (m)	< 1000 and > 2000	N	Hidayat et al., 2020; Hidayat et al., 2020; Neupane et al., 2022; Muliasari et al., 2022; Expert judgement
	1000–1500	S1
	1500–1750	S2
	1750–2000	S3
Climatic Conditions
Annual Rainfall (mm)	< 950	N	Salas Lopez et al., 2020; Estrada, 2019; Neupane et al., 2022; Bindumathi, 2014; Expert judgement
	950–1000	S3
	1000–1250	S2
	> 1250	S1

As a consequence, the final land suitability index is generated using a combination of soil, meteorological and topographic factors. Thus, computation of the land suitability index applies the PCA technique in the following Eq. (3):

$$\:PCA:\:{X}_{scaled}=w*\:P\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(3\right)$$

Where $\:{X}_{scaled}$: represents standardized data matrix, $\:w$: matrix of eigen vectors, $\:P$: indicate matrix of principal component scores. After performing Principal Component Analysis (PCA) on the needed components, the resultant principal component scores were utilized to indicate land suitability. These scores were calculated using a weighted combination of the original constituent factors, with higher values of the composite index indicating higher suitability for coffee growing in Somwarpet Taluk and lower values indicating poor suitability. As a result, this approach comprises preprocessing and scaling the data using PCA to reduce dimensionality while retaining the most significant variation and levelling the findings to a consistent scale for comprehension.

2.6.5. Determination of Land Suitability Classes

Land suitability classes were established by generating a land index for coffee crop cultivation. Further they were categorized into four land suitability classifications according to FAO guidelines as highly suitable (S1), moderately suitable (S2), marginally suitable (S3) and unsuitable (N) (Bindumathi, 2014). The normalizing method improves the speed of convergence of the loss function, prevents gradient explosions and increases computational correct- ness. Whenever predicting suitable coffee cultivation areas, it is essential for the model’s input to include a range of data from multiple variables. Therefore, data normalization must be performed to reduce the impact of different dimensions on prediction outcomes and increase the model’s accuracy and efficiency (Wei and Zhou, 2023). In this study, the min-max approach was employed to normalize the data from 0–100 to make the data simpler to read and compare the results. The land suitability classification for the S1, S2, S3 and N classes were ranged between 75–100, 50–75, 25–50 and 0–25, respectively as per land indices (Sys et al., 1991). The unsuitable (N) land class was assumed to have major constraints that could never or only partially be overcome by various improvement approaches. Following the determination of land suitability classes, it was randomly divided into 75% training data and 25% testing data to extract the variable relevance of each variable using a machine learning model.

2.7. Machine Learning Techniques

2.7.1. K- Nearest Neighbor

One of the most straightforward machine learning algorithms based on the supervised learning approach is the k-nearest neighbor method, also known as k-NN or KNN. It groups the new example in the category to which it most closely resembles, based on the assumption that the new and current data are similar. It is possible to apply the KNN approach to both regression and classification. The KNN algorithm process are as follows: an accurate estimate of K value is selected (i.e., odd number is chosen) and the precision increases with the higher value of K. In next stage, computation of Euclidean distance between the test data point and every other data point is performed (Ismaili et al., 2023). The caret package as “knn” method has been applied in R 4.3.2 version to develop KNN model.

2.7.2. Decision Tree

A Decision Tree (DT) is a supervised learning approach that can do classification and regression tasks, but is mostly utilized for classification. A decision tree, like a tree, consists of root nodes, branch nodes and leaf nodes. Each node represents a characteristic or attribute, whereas a branch represents a decision or rule and a leaf represents the outcomes. Decision tree algorithms are used to separate features. At each node, splitting is tested to determine the best fit for the relevant classes (Sheth et al., 2022). In this study, the decision tree works by recursively separating a dataset into subgroups based on the value of input characteristics like soil, climate and topographic conditions. Each split is intended to optimize homogeneity in the subsequent subsets, resulting in a tree structure with leaves representing the suitability class. The decision Tree was applied in this work using the rpart package as ”rpart” method using R 4.3.2.

2.7.3. Random Forest

Random Forest (RF) is a machine learning approach that utilizes several decision/classification trees where each tree’s value is determined by selecting a random vector’s values uniformly and independently across all of the trees in the forest. As a forest’s tree population rises, overfitting approaches its limit. The relative strengths and correlations between the trees in a forest of tree classifiers affect the forest’s generalization error. Error rates are obtained by splitting each node using a random selection of features. In order to regulate the inaccuracy, estimation is made on model’s response to increase in number of variables based on the analysis and as well as through their importance (Ismaili et al., 2023). In this study, a fundamental concept of bootstrap sampling approach is applied to select k samples at random from the original dataset and then bring them back in as a new training set. Later, the classification trees are then combined to form a random forest of k decision trees and the affiliation of additional samples is determined by a majority vote depending on the results of each classification tree. The algorithm’s goal is to enhance decision trees by combining many decision trees, each of which is built from an autonomously generated sample set (Hayat et al., 2024). In this study, Random Forest was implemented using the ”rf” technique from the caret package in R 4.3.2.

2.7.4. Extreme Gradient Boosting Tree

Extreme Gradient Boosting (XgBoost) models combine an end-to-end tree boosting system with a scalable machine learning method by categorizing or forecasting the desired models from a bigger learning database. It is an algorithm based on decision trees and its primary optimization method is gradient boosting. It is a member of the boosting family of algorithms, which are commonly referred to as ”boosting techniques” since they optimize sequential processes by using misclassification information from a previously formed tree to improve the subsequent tree. Parameter tuning is the most challenging aspect of modelling as decision tree algorithms are prone to over-fitting. To produce a model with low bias and low variance, it essential to find ideal method for hyper-tuning the parameters. This prevents the model from getting overly precise on the training dataset and also makes it possible for the model to get more precise on test datasets. When using multispectral data, the XgBoost model is an excellent option for categorizing soil and land crops that have almost identical spectral signals because of these characteristics (Ismaili et al., 202-3). To implement XgBoost model in this work, the caret package as ” xgbTree ” technique in R 4.3.2 was used.

2.7.5. Na¨ıve Bayes

Naive Bayes (NB) is a commonly used probabilistic statistical classification approach with a simple and efficient algorithm that achieves excellent accuracy. The algorithm uses Bayesian theory to assign weights to each class in training data depending on desired feature conditions (Xing et al., 2022). During this study, NB was implemented using the caret package as ”nb” technique using R 4.3.2 and calculated the likelihood of all the attributes for each suitability class based on the training data. It then forecast the class with the greatest likelihood.

2.7.6. Model Hyperparameters

To detect hyperparameters of each model, we verified the hyper tune for every model. The outcomes are as follows: XgBoost has a (number of boosting iterations) as 100 iterations, (maximum tree depth) of 3, 6 and 9, (Learning rate) for 0.01, 0.1 and 0.3, (minimum loss reduction), (Subsample ratio of columns), (Minimum sum of instance weight needed in a child) and (subsample ratio of the training data); In case of RF, there are two hyper-parameters that must be set by user: (number of trees) and (the number of variables in each tree). The value of parameter was set to 100 and the value of the parameter was set to 10; Whereas for DT, it produces a grid of complexity parameter () values (ranging from 0.01 to 0.1) for adjusting a decision tree model. The parameter regulates tree pruning and determines model complexity. The objective is to identify the ideal value that balances model accuracy and generalizability while limiting overfitting via pruning. This grid of values will be examined during cross-validation to discover the optimal ; In NB model, it generates a hyperparameter grid to test various values for: (Laplace smoothing) - Adds a smoothing constant to prevent zero probability for unobserved data- Indicates whether to utilize kernel density estimation for continuous variables rather than assuming a Gaussian distribution, control the kernel’s bandwidth when, hence altering the smoothness of kernel density estimation; For KNN, created a grid of odd values for the k hyperparameter, which ranges from 3 to 15 by increasing to 2, for use in a KNN model. The odd numbers serve to eliminate ties in the voting process, making it easier to assign a class label to fresh data points. This grid will be used to test various values of k throughout the cross-validation procedure to discover the optimal number of neighbors for the model.

2.8. Evaluation of Machine Learning Performance

2.8.1. Statistical Measures

Using the land suitability modelling process, a number of statistically based metrics like overall accuracy method was employed, to accurate model by combining suitable and unsuitable areas for coffee cultivation. To assess ML model, four categories of potential outcomes such as true positive (TP), false positive (FP), true negative (TN), and false negative (FN) are considered in Eq. (4). Here, in this study 0.5 cutoff value is used to determine the suitable and unsuitable areas of coffee cultivation in Somwarpet region of Kodagu District. It is then determined by comparing each suitability truth pixel with the corresponding suitability pixel on the resulting categorized map. The terms ”TP” and ”FP” denote the places that are identified as a suitable and unsuitable based on land suitability assessments. Not suitable locations are categorized as suitable and unsuitable places, respectively, by FN and TN (Dang et al., 2019). The following list includes all of the equations that were used to determine these parameters.

$$\:\varvec{O}\varvec{v}\varvec{e}\varvec{r}\varvec{a}\varvec{l}\varvec{l}\:\varvec{A}\varvec{c}\varvec{c}\varvec{u}\varvec{r}\varvec{a}\varvec{c}\varvec{y}=\:\frac{\varvec{T}\varvec{N}+\varvec{T}\varvec{P}}{\varvec{T}\varvec{P}+\varvec{F}\varvec{P}+\varvec{T}\varvec{N}+\varvec{T}\varvec{P}}\:\:\:\:\:\:\:\:\:\:\:\left(4\right)$$

2.8.2. ROC Curve and AUC

The Receiver Operating Characteristic (ROC) curve is regarded as a common metric for evaluating the results of using machine learning models. The ROC curve is created by combining different cut-off limits with sensitivity and specificity as shown in the Eqs. (5 and 6). An area under the curve (AUC) value of 0.5 indicates an imprecise model, whereas a value of close to 1 indicates a flawless model (Dang et al., 2019). The ROC curve was used in this work to validate the best-selected model.

$$\:\varvec{S}\varvec{e}\varvec{n}\varvec{s}\varvec{i}\varvec{t}\varvec{i}\varvec{v}\varvec{i}\varvec{t}\varvec{y}\:=\:\frac{\varvec{T}\varvec{P}}{\varvec{T}\varvec{P}+\varvec{F}\varvec{N}}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(5\right)$$

$$\:\varvec{S}\varvec{p}\varvec{e}\varvec{c}\varvec{i}\varvec{f}\varvec{i}\varvec{c}\varvec{i}\varvec{t}\varvec{y}\:=\:\frac{\varvec{T}\varvec{N}}{\varvec{F}\varvec{P}+\varvec{T}\varvec{N}}\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(6\right)$$

3.1. Variable Importance of the LSF Affecting Coffee Cultivation

In studies of the land suitability for cultivation of coffee, other crops (Bindumathi, 2014; Reza et al., 2020; Zhang et al., 2015; Ramamurthy et al., 2020) or agriculture in general (ThiTuyen et al., 2019; Pramanik, 2016; Mistri & Sengupta, 2019; Radocaj et al., 2021; Talukdar et al., 2022; Hassan et al., 2020); it is common to select climatological, topographical and soil parameters. Differentiating from the previous studies (Lopulisa et al., 2020; Juita et al., 2021), this research included two sub-criteria of soil micro nutrients (Iron and Sulphur) for mapping land suitability of coffee crop. Furthermore, only few previous researchers (Hayat et al., 2024; Xing et al., 2022; Gandhi, 2022; Ozalp and Akinci, 2023; Wei and Zhou, 2023; Sahu et al., 2023; Sarkar et al., 2021; Taghizadeh-Mehrjard et al., 2020; Yang et al., 2023; Talukdar et al., 2022) have used ML techniques to rank the variable importance and prioritize sub-criteria for various other crops for land suitability analysis. Thus, this study employed a variable analysis approach to categorize the most significant factors by utilizing RF, KNN, XgBoost, DT and NB techniques to evaluate the ability of soil, climate and topographic covariates between suitable and unsuitable areas for coffee crop as shown in Fig. 6.

The RF model applied to calculate the relative influence of each variable within the modelling using the Gini importance mechanism; but for DT, XgBoost, NB and KNN models, the weights represent the relevance of the auxiliary data. In terms of development of coffee cultivation in this region, our study showed that the OC (26.22%), Fe (23.7%), slope (12.1%), N (11.67%), S (8.88%), K (8.25%), elevation (7.15%), pH (1.23%), rainfall (0.45%), followed by EC (0.35%) are the most important variables to map land suitability for coffee crop using XgBoost model. However, RF and DT models indicated that Fe (25.6% and 15.98%, respectively), N (23.33% and 16.6%, respectively), OC (14.94% and 16.51%, respectively), slope (13.18% and 9.45%, respectively), elevation (8.56% and 9.6%, respectively), K (6.47% and 9.53%, respectively), S (3.29% and 6.87%, respectively), pH (2.04% and 1.14%, respectively), EC (1.42% and 7.97%, respectively) and rainfall (1.17% and 6.35%, respectively) are the most relevant factors. In accordance with the KNN and NB models, they had an insignificant difference in predicting the most important factors for coffee land suitability classes with elevation (11.7% and 11.65%), Fe (11.6% and 11.62%), slope (11% and 10.83%), S (10.7% and 10.74%), K (10.5% and 10.5%), OC (10% and 10.1%), N (10% and 10.02%), pH (9.4% and 9.35%), EC (8.7% and 8.76%) and rainfall (6.4% and 6.43%).

According to XgBoost and RF model, iron, organic carbon, slope and nitrogen are the most important contributing factors for LSM; as contrast to NB and KNN models, where elevation, sulphur, iron, slope and potassium plays a significant role for mapping land suitability for coffee production in Somwarpet taluk of Kodagu District. While potential of hydrogen plays a least role in case of DT model; but to XgBoost, NB, KNN and RF models, rainfall is considered as a fewer important factor in the land suitability characterization for coffee cultivation. Similar results have been reported by the researcher but using AHP and fuzzy method to rank and prioritize sub- criteria. In relation to climatological sub-criterion, the author (Ochoa et al., 2017; Mighty, 2015; Rono and Mundia, 2016) highlighted that rainfall is the most essential criterion for their region. In fact, coffee crops are susceptible to climate, particularly the temperature at which they grow. Therefore, the suitability models generated in this type of study are based on current environmental circumstances and must be revised regularly due to climate change’s impact on climatological sub-criteria (Salas Lopez et al., 2020); conversely, the researcher (Ndabasanze, 2018) suggested that by adapting the contour farming practices can prevent the extreme events especially to avoid run off i.e., land sliding or heavy precipitation related to variation in climatic conditions.

With regard to soil sub-criteria, soil pH is the most significant criteria

(Salas Lopez et al., 2020) and this also agrees (Ochoa et al., 2017). Conversely, soil pH is the least important for growing coffee crop (Rono and Mundia, 2016); in a similar way, the study (Nurfadila et al., 2020) indicated that pH followed by organic carbon was least important factor. In fact, soil pH plays a crucial role in plant nutrient digestion, as nutrient availability changes with acidity levels; more prominently, some minerals can be harmful to plants at high acidity levels (Salas Lopez et al., 2020). The researcher indorsed that by adding sulphur nutrient content and liming, it is possible to reduce the acidic levels ; conversely, salinity issues can be reduced by managing the water system i.e., by increasing the drainage (Juita et al., 2021; Rahmatika et al., 2022); whereas, the study discovered that including cover crops, organic matter, fertilizing, composting, making terraces or planting the parallel to contour and drainage can increase the low pH value, organic carbon, N and K nutrients (Irawan et al., 2022; Rahmatika et al., 2022; Hartono et al., 2018; Silaban et al., 2016).

In contrast to the work (Rono and Mundia, 2016), elevation is the most influencing factor of the topographic sub-criteria; conversely, the study (Ochoa et al., 2017) consider slope to be the most important; whereas, in case of the findings (Nurfadila et al., 2020), imply that the slope weight factor for evaluating land suitability was the least since high slope area was evaluated and people continued to farm diverse commodities for livelihoods; however, (Mighty, 2015) consider that there is no consequence difference between slope and elevation; additionally in the study (Salas Lopez et al., 2020), elevation was the most important followed by the slope factor. Also, coffee cultivated at higher elevations exhibit superior organoleptic quality. Therefore, altitudinal factors have led to coffee plantation movement to higher areas, resulting in increased deforestation in the main coffee plantation areas. In terms of planting the coffee crop without the precise land suitability class i.e., if the coffee is planted on the land where slope < 15% without proper soil fertility, poorly managed shades, soil and water conservation practices may reduce the coffee production in the region (Marbun et al., 2019).

3.2. Spatial Land Suitability Mapping

The land suitability models for mapping coffee production were built using the five machine learning techniques namely K-NN, NB, RF, DT and XgBoost. The land suitable area indices generated by the PCA approach were distinguished into four classes according to FAO as: highly suitable (S1), moderately suitable (S2), marginally suitable (S3) and unsuitable (N) areas for coffee cultivation. The land suitability maps generated by the five machine learning models are displayed in Figs. 10, 11 and 12. The LSM produced by the RF model noticed that 84.17% (83895.75 ha) of the study region has moderate land suitability for coffee production. While the remaining 8.66% (8637.12 ha), 6.64% (6614.37 ha) and 0.53% (527.85ha) of the research area covers into marginal, highly and unsuitable land suitability class, respectively.

Based on the outcomes of DT model, the study area has 81.33% (81063.72 ha), that falls into the S2 land suitability class, followed by 8.57% (8546.4 ha), 8.5% (8472.42 ha) and 1.6% (1592.55 ha) of the area are in S1, S3 and N land suitability class. According to the findings for the KNN model, the research area has a 44% (43743.42 ha) and a 28% (28011.6 ha) area in the S2 and S1areas, respectively, but it falls under S3 and N areas with 27.8% (27720.27 ha) and 0.2% (199.8 ha). In case of the NB model, the results showed that the research area has a 69.61% (69389.28 ha), 19.17% (19105.2 ha), 10.52% (10481.31 ha) area in the moderate, high and marginal land, followed by the 0.7% (699.3 ha) of unsuitable area. As per the XgBoost model, the findings indicated that 82.18% (81910.44 ha) of the research area falls into the moderate suitability class and the remaining 8.75% (8727.48 ha), 7.64% (7612.56 ha) and 1.43% (1424.61 ha) falls into the high, marginal and unsuitable areas.

The researcher (Bindumathi, 2014) used integrated fuzzy-AHP and GIS

approaches for assessing land suitability of coffee cultivation in three different zones of Karnataka region. Out of 71658 ha of total land, the Hassan, Holenarasipura, Chennarayapatana and Ariskere Taluk had 7.88%, 3.04%, 1.47% and 0.24% of highly suitable coffee plantation land for zone one; in contrast, a total 154316 ha, 72297 ha and 4970 ha of land was classified into moderate, marginal and unsuitable areas. Within zone two, a total 121465 ha of land is classified as a highly suitable, in that Belur, Alur, Chikkamagalur and Arkalgud Taluks recorded as 42614 ha, 41800 ha, 22652 ha and 14399 ha. Except for Alur Taluk, all other Taluk i.e., Chikkamagalur (2.42%), Belur (1.99%) and Arkalgud Taluk (0.88%) was classified into moderate suitable land. Their results for zone three indicated that 56200 ha, 31272 ha and 20,800 ha of land is highly suitable for coffee cultivation in Sakleshpura, Mudigere and Somwarpet Taluk; conversely, 4528 ha (4.01%) of Mudigere Taluk was identified as moderately suitable for coffee cultivation in Hemavathi Watershed.

3.3. Validation of the RF Model

ROC curves and macro-AUC values of the DT, NB, XgBoost, KNN and RF models were used to validate the findings of this research area as shown in the Figs. 13, 14 and 15. It was discovered that the information gathered in

the field survey and the predicted outcomes with good similarity. The overall accuracy provided for predicting the coffee land suitability class using RF and XgBoost methods (94% and 93.5%, respectively) were high when compared to the values of DT, NB and KNN methods (92%, 75% and 50%, respectively). The RF, XgBoost, DT, NB and KNN models have predictive rates of 1, 0.99, 0.96, 0.9 and 0.85, respectively. The macro-AUCs showed that the models for the LSM have very good to exceptional prediction accuracy, with the RF being the most reliable and accurate model for mapping land suitability of coffee production in Somwarpet taluk of Kodagu District.

From a statistical standpoint, the RF model succeeded well in terms of prediction accuracy. The advantages of RF model are to model nonlinear connections using both categorical and continuous predictor variables, along with minimal bias and variance (Gandhi, 2022; Taghizadeh-Mehrjard et al., 2020). Thus, the RF model is a powerful tool for predicting land suitability classes due to its robustness to noise in predictors, lack of overfitting, low bias and variance and ability to identify important auxiliary data (Taghizadeh-Mehrjard et al., 2020). As a result, the RF model is suggested as the best model for predicting the land suitability class of coffee cultivation in Somwarpet Taluk, Kodagu region. This outcome is consistent with the findings of previous researchers who have proved the reliable performance of RF (Taghizadeh-Mehrjard et al., 2020; Hayat et al., 2024; Gandhi, 2022; Ozalp and Akinci, 2023; Xing et al., 2022; Sahu et al., 2023).

In the study (Taghizadeh-Mehrjard et al., 2020), applied RF and SVM to predict land suitability classes for rainfed wheat and barley. They showed that Rf performed better than SVM. Conversely, the work carried out by (Hayat et al., 2024) observed high performance of random forest than support vector machine to model soil and land suitability for rainfed olive and maize crop in Khyber Pakhtunkhwa province; in a similar way, the researcher (Ozalp and Akinci, 2023) findings showed that RF model is best suitable for cultivation of olive in Artvin Province, Turkey. In contrast, the author (Gandhi, 2022) attained best performance results using two regression-based RF algorithm (i.e., 2018 and 2080 scenarios) to analyse the suitability for maize production in Canada and they reported that the out of bag root mean square error (OOB MSE) with variance values in validation data achieved were 0.004 (98.3%) and 0.005 (98.2%).

With regards to the study (Xing et al., 2022), they proposed a ML-based tea cultivation suitability assessment model by associating random forest (RF), adaptive boosting (AdaBoost), gradient boosting decision tree (GBDT), extreme gradient boosting (XgBoost), multilayer perceptron (MLP), Gaussian Naive Bayes (GNB). Their finding highlighted that Rf model provided highest accuracy to evaluate weight factor and map land suitability at Town and Village scales of China. Another research found that the random forest (RF) model was more accurate than the logistic regression model, with AUCs of 85.2% and 83.3%, respectively for cultivation of tea crop in Darjeeling Himalaya, India (Sahu et al., 2023).

This study estimated the land suitability for coffee cultivation in Somwarpet Taluk of Kodagu region, by utilizing the FAO land suitability assessment framework, principal component analysis (PCA) technique and ML mapping approach. The iron, nitrogen, organic carbon, potassium, sulphur, potential of hydrogen, electrical conductivity, elevation, slope, and rainfall map were the most significant auxiliary data for estimating land suitability class for growing coffee crop. Based on overall accuracy findings, the RF model outperformed the XgBoost, DT, NB, and KNN techniques in predicting land suitability classes for coffee growing. RF algorithm is considered as a powerful modelling approach for predicting land suitability classes due to its several benefits over other statistical modelling methods.

The results reported that area of S1 (highly suitable) classified through RF, XgBoost, DT, NB and KNN was 8.66%, 8.75%, 8.57%, 19.17% and 28%

respectively. Correspondingly, the S2 (moderately suitable) class area via RF outperformed the land suitability class for XgBoost, DT, NB and KNN by approximately 84.17%, 82.18%, 81.33%, 69.61% and 44%, respectively. Areas of N (unsuitable) and S3 (marginal suitable) were 0.53% and 6.64% for RF, followed by 1.43% and 7.64%, 1.6% and 8.5%, 0.7% and 10.52%, 0.2% and 27.8% for XgBoost, DT, NB and KNN methods. After comparing five ML approaches, it was discovered that greater regions of S2, S1 and S3 classes and smaller areas of N class were used to estimate land suitability maps for coffee growing.

In general, the research area’s results indicated low adaptability or that it was unsuitable for coffee land due to pH and Sulphur nutrient content. These restrictions are the primary reasons why the actual yield of coffee growing in this research region is lower than the crop’s potential output. To improve the suitability of the study area for coffee croplands, it is essential to increase the production, cost-benefit analysis, land use changes, crop population benefits, socio-economic factors and land improvement operations such as liming, reclamation, balanced fertilization, composting, increasing soil organic matter through farmyard manure or green manure, land cover crop cultivation, supplementary irrigation, terracing and contour farming practices are required.

As the parameters for assessing the feasibility of a region varies based on the study area and available geographical data. Therefore, it is recommended to examine all elements with all soil micro nutrients, photoperiod and percentage of shades to increase precision when evaluating land suitability for coffee cultivation. This research methodology integrates GIS with data driven machine learning techniques to analyze land suitability for coffee growing in Somwarpet Taluk of Kodagu region; in a similar way, it can be applied to other crops with nutritional, economic and environmental importance, by adjusting for local realities. Furthermore, these techniques can be influenced the field by identifying best suitable coffee crop growing areas; thus, preventing over-exploitation of soil resources and promoting sustainable agriculture practices.

Author Contribution

G. S. conceptualized and designed the study with the guidance of A . L . G. S. was responsible for data collection, preprocessing, analysis, developement of the models and interpretation of the results. The manuscript was written by G. S. with input from the author A . L . A . L . and G. A. reviewed and provided critical revisions. All authors actively participated in the manuscript review process and approved the dataset.

Acknowledgement

I would like to thank Mr. K. P. Veeranna, Mr. K. P. Nandish and Ms. C. T. Ankita from the Soil Sample Training and Testing Centre in Kudige, Somwarpet Taluk, Kodagu District, for their important assistance and direction in providing soil information during this project. I am particularly grateful to Dr. K. Chandrappa and Dr. S. A. Nadaf of the Coffee Research Sub-Station in Chethalli, Kodagu District, for their excellent advice and support in making this study feasible. Their contributions have been critical to the effective completion of this study.

Abdisa, T., 2020. Land suitability analysis for Arabica coffee (Coffea Arabica) production using geospatial and multi criteria evaluation in Jardega Jarte District, Western Ethiopia. Jimma University, Jimma.
Abdisa, T. (2020). Land suitability analysis for Arabica coffee (Coffea arabica) production using geospatial and multi-criteria evaluation in Jardega Jarte District, Western Ethiopia. (M.Sc. Thesis). Jimma University, College of Social Science and Humanities.
Akbar, M., Karim, A., & Sugianto. (2019). Investigation of utilizing coffee commodities toward land suitability: Case study: Mane Village, Aceh Province. International Journal of Multicultural and Multireligious Under- standing, 6(3), 608-613.
Auliansyah, G., Fachruddin, F., & Yunus, Y. (2019). Evaluasi Kesesuaian Lahan Pada Tanaman Kopi Arabika (Coffea arabica L.) Organik Menggunakan Sistem Informasi Geografis (SIG) di Kecamatan Pegasing Kabupaten Aceh Tengah (Evaluation of Land Suitability in Organic Arabica Coffee (Coffea arabica L.) Using Geographical Information Systems (GIS) in Pegasing District, Aceh Tengah), Jurnal Ilmiah Mahasiswa Pertanian, 4(2), 339-348.
Azsari, R., M., Mulyani, C., & Iswahyudi. (2022). Evaluasi Kesesuaian Lahan Untuk Tanaman Kopi Robusta (Coffea Canephora) di Desa Punti Payong Kecamatan Ranto Peureulak Kabupaten Aceh Timur (Evaluate the Suitability of Land for Coffee Plants Robusta (Coffea Canephora) in Punti Payong Village Ranto Peureulak District, East Aceh District), Jurnal Penelitian, 9(2), 61-70.
Barus, B., Razali, R., & Sitanggang, G. (2015). Evaluation of land suitability for Coffea arabica (Coffea arabica L Var Kartika Ateng) in Muara District, North Tapanuli Regency. Journal Online Agroekotechnology, 3(4), 1459-1467.
Bich, N. T., and Phuong, H. X. (2023). Application of GIS to identify land areas suitable for rubber and coffee trees to restructure agriculture in Gia Lai Province. Journal of Forestry Science and Technology, 15, 69–75.
Bindumathi, S., 2014. Assessment of land suitability for agricultural crops in Hemavathi watershed analysis through remote sensing and GIS ap- proach. University of Mysore, Mysore.
Chairani, E., Supriatna, J., Koestoer, R., & Moeliono, M. (2017). Physical land suitability for Civet Arabica Coffee: Case study of Bandung and West Bandung Regencies, Indonesia. IOP Conference Series: Earth and Environmental Science, 98, 012029.
Chairuddin, Z., Sra, N., & Khaerunnisa. (2023). Mapping the Land Suitability Rating of Arabica Coffee Crops: A Geographical Indication Factor Based Approach, International Journal of Plant & Soil Science, 35(20), 1320- 1336.
Chemura, A., Kutywayo, D., Chidoko, P., & Mahoya, C. (2015). Bio- climatic Modelling of Current and Projected Climatic Suitability of Coffee (Coffea arabica) Production in Zimbabwe, Regional Environmental Change, 16(2), 473–485.
Chengappa, P.G., Devika, C.M., 2016. Climate variability concerns for the future of coffee in India: An exploratory study. Int. J. Environ. Agric. Biotechnol. 1(4), 819-826.
Dang, K. B., Burkhard, B., Windhorst, W., & Müller, F. (2019). Application of a hybrid neural-fuzzy inference system for mapping crop suitability areas and predicting rice yields. Environmental Modelling & Software, 114, 166–180.
D’haeze, D., Deckers, J., Raes, D., Phong, T., A., & Loi, H., V. (2005). Environmental and Socio-Economic Impacts of Institutional Reforms on the Agricultural Sector of Vietnam: Land Suitability Assessment for Robusta Coffee in the Dak Gan Region, Agriculture, Ecosystems and Environment, 105, 59–76.
Dermawan, S. T., Mega, I. M., & Kusmiyarti, T. B. (2018). Evaluasi kesesuaian lahan untuk tanaman kopi robusta (Coffea canephora) di desa pajahan kecamatan pupuan kabupaten Tabanan. E-Jurnal Agroekoteknologi Tropika, 7(2), 230–241.
Eshete, G. G. (2022). Analyzing land suitability for Coffea arabica growing under current and future climate scenarios: Rift-Valley Lake Basin, Ethiopia. M.Sc. Thesis, Hawassa University, Wondo Genet College of Forestry and Natural Resources, School of Graduate Studies, Wondo Genet, Ethiopia.
Estrada, L., D., L. (2019). Exploring the Potential for Adaptation and Mitigation to Climate Change of Coffee Agroforestry Systems in Central America. Ph.D. Thesis, Faculty of Mathematics, Informatics and Natural Sciences, Department of Earth Sciences, Universitat Hamburg, Germany.
Fachruddin, F., Fadhil, R., Syafriandi, and Dahlan, D. (2020). Suitability analysis of scrubland for Arabica and Robusta coffee plants in Aceh Besar Regency. IOP Conference Series: Earth and Environmental Science, 644, 012011.
Funk, C., Peterson, P. Y., Landsfeld, M. F., Pedreros, D., Verdin, J. P., Shukla, S., Husak, G. J., Rowland, J., Harrison, L., Hoell, A., and Michaelsen, J. (2015). The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Scientific Data, 2(1).
Gandhi, V.H., 2022. Land suitability assessments for maize production in Ontario, Canada, using a weighted overlay method and random forest algorithm. The University of Guelph, Guelph.
Gross, J. J. (2014). Assessment of future agricultural land potential using GIS and regional climate projections for Hawaii Island: An application to macadamia nut and coffee. M.Sc. Thesis, University of Hawaii, Natural Resources and Environmental Management.
Gruter, R., Trachsel, T., Laube, P., & Jaisli, I. (2022). Expected Global Suitability of Coffee, Cashew, and Avocado Due to Climate Change, PLoS ONE, 17(1), e0261976.
Gomes, L., C., Bianchi, F., J., J., A., Cardoso, I., M., Fernandes, R., B., A., & Filho, F., E., I. (2020). Agroforestry Systems Can Mitigate the Impacts of Climate Change on Coffee Production: A Spatially Explicit Assessment in Brazil, Agriculture, Ecosystems and Environment, 294, 106858.
Hartono, B., Rauf, A., Elfiati, D., Harahap, F. S., & Sidabuke, S. H. (2018). Evaluasi kesesuaian lahan pertanian pada areal penggunaan lain untuk tanaman kopi arabika (Coffea Arabica L.) di kecamatan salak kabupaten pak-pak bharat. Jurnal Solum, 15(2), 66.
Hassan, I., Javed, M., Asif, M., Ahmad, S., Akhtar, S., Hussain, B., 2020. Weighted overlay based land suitability analysis of agriculture land in Azad Jammu and Kashmir using GIS and AHP. Pak. J. Agric. Sci. 57, 1509-1519.
Hayat, A., Iqbal, J., Ashworth, A.J., Owens, P.R., 2024. Assessing soil and land suitability of an olive–maize agroforestry system using machine learning algorithms. Crops. 4, 308–323.
Hidayat, E. et al. (2020). Land suitability evaluation of Arabica coffee (Coffea arabica L.) plantation in Subdistrict Aie Dingin, Lembah Gumanti, Indonesia. IOP Conference Series: Earth and Environmental Science, 583, 012005.
Hidayat, E. (2020). Land suitability evaluation of coffee in Tokunoshima Island, Japan. Journal of Applied Agricultural Science and Technology, 4(2), 146–154.
Irawan, S. et al. (2022). Study of the relationship between soil fertility and land suitability for Arabica coffee (Coffea arabica L.) development in Bandar Sub-district, Pacitan District. IOP Conference Series: Earth and Environmental Science, 1111, 012029.
Ismaili, M., Krimissa, S., Namous, M., Htitiou, A., Abdelrahman, K., Fnais, M.S., et al., 2023. Assessment of soil suitability using machine learning in arid and semi-arid regions. Agronomy. 13, 165.
Jutia, N., Ridwan, I., Jannah, R., K., L., & Makmur, A., A., P. (2020). Analisis Kesesuaian Lahan Untuk Pengembangan Kopi Robusta dengan Pendekatan Parametrik Terbaru: Land Suitability Analysis for Robusta Coffee Development with The Latest Parametric Approach, Environmental Science & Engineering, 74-82.
Juita, N., Ridwan, I., Jannah, R., and Parahyanti, A. M. (2021). Arabica coffee land suitability with a parametric approach based on square root method. IOP Conference Series: Earth and Environmental Science, 807, 022076.
Karim, A., et al., 2020. Land arrangement for citronella (Cymbopogon nardus) and Arabica coffee in the cultivation area in Gayo Lues District, Aceh Province Indonesia: A land suitability approach. Aceh Int. J. Sci. Technol. 9(3), 128–137.
Lara-Estrada, L., Rasche, L., and Schneider, A. (2017). Modeling land suitability for Coffea arabica L. in Central America. Environmental Modelling & Software, 95, 196–209.
Lara-Estrada, L., Rasche, L., and Schneider, A. (2021). Land in Central America will become less suitable for coffee cultivation under climate change. Regional Environmental Change, 21(88), 1–13.
Liem, N., D., & Duyen, T., T., M. (2020). Spatial Planning for Rubber and Coffee Development in Kontum Province, Dalat University Journal of Science, 10(2), 42-70.
Lopulisa, C., Neswati, R., and Norsyam, M. (2020). Land suitability index to estimate the land potential for Arabica coffee plantation: A case of Tompobulu District, Bantaeng Regency. IOP Conference Series: Earth and Environmental Science, 486, 012071.
Lopez-Carmona, D., Zuniga, F. B., & Bautista-Hernandez, D. A. (2023). Identification of soil properties associated with the peasant perception of the suitability of the land for growing organic coffee: The case of traditional agriculture in the “Mixteca Alta” Mountains of Oaxaca, Mexico. Agroecology and Sustainable Food Systems, 48(2), 161–182.
Marbun, P. et al. (2019). Evaluation of land suitability on Arabica coffee plantations by parametric method in Lintongnihuta District. IOP Conference Series: Earth and Environmental Science, 260, 012155.
Marbun, P. et al. (2023). Identification of soil physicochemical properties, land suitability, and their relationship to Arabica coffee yields based on plant age groups. Coffee Science, 18, 1–12.
Mighty, M. A. (2015). Site suitability and the analytic hierarchy process: How GIS analysis can improve the competitive advantage of the Jamaican coffee industry. Applied Geography, 58, 84-93.
Mishra, S., Mishra, D., Santra, G., 2016. Applications of machine learning techniques in agricultural crop production: A review paper. Indian J. Sci. Technol. 9(38).
Mistri, P., Sengupta, S., 2019. Multi-criteria decision-making approaches to agricultural land suitability classification of Malda district, Eastern India. Nat. Resour. Res. 29(3), 2237–2256.
Muliasari, A.A., Dewi, H., Surwarto, 2022. Spatial analysis for land suitability of Arabica coffee (Coffea arabica L.) in Bogor District. IOP Conf. Ser. Earth Environ. Sci. 974, 012096.
Muslihah, I. N., Karuniasa, M., & Herawati, T. (2020). The impact of climate change on Arabica suitability area and opportunities to reduce vulnerability. IOP Conference Series: Earth and Environmental Science, 575, 012078.
Ndabasanze, P. (2018). Suitability analysis of coffee growing areas in the context of climate change in Nyaruguru District, Rwanda. (Master’s Thesis). University of Rwanda, College of Science and Technology.
Neswati, R., Sappe, N. J., Baja, S., and Rukmana, D. (2023). Assessment of farmers’ preferences for growing particular crops and the correlation with land suitability. Journal of Agriculture and Environment for International Development (JAEID), 117(1), 85–116.
Neupane, K., & Pangali Sharma, T. P. (2022). Land suitability analysis for coffee production in hilly districts of Lumbini Province. The Geographic Base, 1–14.
Nuary, R. B., Setiyono, R., & Sukartiko, A. C. (2022). Land suitability modelling of agricultural geographical indication products under climate change scenarios. Proceedings of the 2nd International Conference on Smart and Innovative Agriculture (ICoSIA 2021), [Preprint].
Nurdin, N., et al. (2022). Comparison of land suitability class for endemic Coffea liberica Pinogu HP acquired using different methods and recommen- dations for land management in Pinogu Plateau, Bone Bolango Regency, Indonesia. SAINS TANAH - Journal of Soil Science and Agroclimatology, 19(1), 42-51.
Nurfadila, J. S., Baja, S., Neswati, R., and Rukmana, D. (2020). Evaluation of land suitability for coffee plants based on fuzzy logic in Enrekang District. IOP Conference Series: Earth and Environmental Science, 486, 012069.
Nugraha, A. T., Prayitno, G., and Khoiriyah, L. A. (2021). Land suit- ability and economic performance in the Pasuruan region for coffee development. International Journal of Sustainable Development and Planning, 16(2), 229–236.
Nzeyimana, I., Hartemink, A. E., & Geissen, V. (2014). GIS-based multicriteria analysis for Arabica coffee expansion in Rwanda. PLoS One, 9, e107449.
Ochoa, P., A., Chamba, Y., M., Arteaga, J., G., & Capa, E., D. (2017). Estimation of Suitable Areas for Coffee Growth Using a GIS Approach and Multicriteria Evaluation in Regions with Scarce Data, Applied Engineering in Agriculture, 33(6), 841-848.
Official Website of Central Ground Water Board (CGWB). (2023). Retrieved 2 August 2023, from https://cgwb.gov.in/https://cgwb.gov.in. Maintained by Ministry of Jal Shakti, Department of Water Resources, River Development, and Ganga Rejuvenation, Government of India.
Official Website of Kodagu District. (2023). Retrieved 23 February 2023, from http://www.kodagu.nic.in/www.kodagu.nic.in. Maintained by Kodagu District Administration and designed by the National Informatics Centre, Ministry of Electronics and Information Technology, Government of India.
Ozalp, A.Y., Akinci, H., 2023. Evaluation of land suitability for olive (Olea europaea L.) cultivation using the random forest algorithm. Agriculture. 13, 1208.
Pagiu, S., Ramlan, Belo, T., I., & Patadungan, Y., S. (2020). Land Index and Production of Arabica Coffee (Coffea arabica L.) in Smallholding Plantation of Tana Toraja District, Indonesia, International Journal of Design & Nature and Ecodynamics, 15(4), 587-592.
Pham, M., P., Vu, D., D., Tong, T., H., Thi, M., H., N., & Sandlersky, R. (2021). Integrated Use of AHP-GIS-Remote Sensing Predicting Potential Areas of Coffee Plants: A Case Study of Buffer Zone of Ta Dung Nature Park, Vietnam, E3S Web of Conferences, 285, 02022.
Pramanik, M.K., 2016. Site suitability analysis for agricultural land use of Darjeeling district using AHP and GIS techniques. Model. Earth Syst. Environ. 2(2), 56.
Pravitasari, A. E., et al. (2023). Land suitability analysis and direction for plantation commodities development in Pekalongan Regency, Central Java. IOP Conference Series: Earth and Environmental Science, 1133(1), 012053.
Purba, P., Sukartiko, A., C., & Ainuri, M. (2019). Modeling the plantation area of geographical indication product under climate change: Gayo Arabica coffee (Coffea arabica), IOP Conf. Series: Earth and Environmental Science, 365, 012021.
Radocaj, D., Jurisic, M., Gasparovic, M., Plascak, I., Antonic, O., 2021. Cropland suitability assessment using satellite-based biophysical vegetation properties and machine learning. Agronomy. 11(8), 1620.
Raharja, M., A. (2016). Analisis Dan Perancangan Sistem Informasi Geografis Kesesuaian Lahan Untuk Tanaman Kopi Di Kabupaten Buleleng (Analysis and Design of Geographical Information System of Land Suitability for Coffee Plants in Buleleng District), Jurnal Ilmiah Ilmu Komputer, 9(2), 1-8.
Rahamatika, D. E. et al. (2022). Evaluation of land suitability for Coffee (Coffea sp.) in Nawangan District, Pacitan Regency, East Java. IOP Conference Series: Earth and Environmental Science, 986, 012039.
Rahmawaty, Ginting, Y. A., Batubara, R., Carenina, G., Ginting, C. F., Ginting, R. K., Angelita, L., & Rauf, A. (2021). Land suitability assessment for Coffea arabica on land overgrown by Uncaria gambir. IOP Conference Series: Earth and Environmental Science, 886, 12121.
Ramamurthy, V., Reddy, G.P.O., Kumar, N., 2020. Assessment of land suitability for maize (Zea mays L.) in semi-arid ecosystem of Southern India using integrated AHP and GIS approach. Comput. Electron. Agric. 179,105806.
Rangmar, B., B., Giltrap, D., J., Burgham, S., J., & Savage, T., J. (1995). Land Suitability Assessments for Selected Crops in Papua New Guinea, PNGRIS Publication, 8 (AusAID: Canberra), 84.
Ranjitkar, S., Sujakhu, N. M., Merz, J., Kindt, R., Xu, J., Matin, M. A., Ali, M., & Zomer, R. J. (2016). Suitability analysis and projected climate change impact on banana and coffee production zones in Nepal. PLoS One, 11, e0163916.
Reza, A., Reza, H., Sohrabi, A., Sarmadian, F., 2020. Integrated fuzzy, AHP and GIS techniques for land suitability assessment in semi-arid regions for wheat and maize farming. Ecol. Indic. 110, 105887.
Rono, F. and Mundia, C. (2016). GIS-based suitability analysis for coffee farming in Kenya. International Journal of Geomatics and Geosciences, 6(3), 1722–1733.
Ryder, R. (1994). Land Evaluation for Steepland Agriculture in the Dominican Republic, The Geographical Journal, 160(1), 74-86.
Sahu, N., Das, P., Saini, A., Varun, A., Mallick, S.K., Nayan, R., et al., 2023. Analysis of tea plantation suitability using geostatistical and machine learning techniques: A case of Darjeeling Himalaya, India. Sustainability. 15, 10101.
Salas Lopez, R., Gomez Fernandez, D., Silva Lopez, J., Rojas Briceno, N., Oliva, M., Terrones Murga, R., et al., 2020. Land suitability for coffee (Coffea arabica) growing in Amazonas, Peru: Integrated use of AHP, GIS and RS. ISPRS Int. J. Geo-Inf. 9(11), 673.
Salima, R., Karim, A., & Sugianto. (2012). Evaluasi Kriteria Kesesuaian Lahan Kopi Arabika Gayo Di Dataran Tinggi Gayo, Jurnal Manajemen Sumberdaya Lahan, 460, 194-206.
Santoso, N., A., & Ayu, Z., D. (2022). Subsurface Studies and Land Suitability of Robusta Coffee Based on 2 Magnetic Data, pH, And Soil Temperature in Ulubelu, Indonesia, SSRN Electronic Journal [Preprint].
Sarkar, D., Saha, S., Maitra, M., Mondal, P., 2021. Site suitability for aromatic rice cultivation by integrating geo-spatial and machine learning algorithms in Kaliyaganj C.D. Block, India. Artif. Intell. Geosci. 2, 179- 191.
Sheth, V., Tripathi, U., Sharma, A., 2022. A comparative analysis of machine learning algorithms for classification purpose. In: 4th International Conference on Innovative Data Communication Technology and Application, Procedia Computer Science. 215, 422–431.
Silaban, S. H., Sitorus, B., & Marbun, P. (2016). Evaluasi kesesuaian lahan untuk tanaman kopi arabika (Coffea arabica), kentang (Solanum tubero- sum L.), kubis (Brassica oleraceae L.), dan jeruk. Jurnal Online Agroekoteknologi, 4(3), 2055–2068.
Sys, C., Van Ranst, E., Debaveye, I.J., 1991. Land evaluation. Part I: Principles in land evaluation and crop production calculations. General Administration for Development Cooperation, Agricultural Publication-No. 7, Brussels.
Taghizadeh-Mehrjardi, R., Nabiollahi, K., Rasoli, L., Kerry, R., Scholten, T., 2020. Land suitability assessment and agricultural production sustain- ability using machine learning models. Agronomy. 10(4), 573.
Talukdar, S., Naikoo, M., Mallick, J., Praveen, B., Shahfahad, Sharma, P., 2022. Coupling geographic information system integrated fuzzy logic analytical hierarchy process with global and machine learning-based sensitivity analysis for agricultural suitability mapping. Agric. Syst. 196, 103343.
Thanuja, P., Singh, N.P., 2017. An economic analysis of marketing and processing of coffee in Kodagu district of Karnataka. Int. J. Agric. Sci. Res. 7(4), 227-232.
ThiTuyen, T., Yen, H., ThiThuy, H., ThiTrangThanh, N., Quoc, N., Prakash, I., et al., 2019. Agricultural land suitability analysis for Yen Khe Hills (NgheAn, Vietnam) using analytic hierarchy process (AHP) combined with geographic information systems (GIS). Indian J. Ecol. 46.
Tram, N., N., B., Sieng, N., T., & Khoi, D., N. (2023). Mapping agro-climatic zone for coffee crop in the Srepok River Basin, IOP Conf. Series: Earth and Environmental Science, 1170, 012003.
Wei, G., Zhou, R., 2023. Comparison of machine learning and deep learning models for evaluating suitable areas for premium teas in Yunnan, China. PLoS One. 18(2).
Weldon, M. (2016). Application of GIS in Selecting Areas Favourable for Coffee Farming: Case Study, Kericho County. M.Sc Thesis, University of Nairobi, Department of Geospatial and Space Technology, School of Engineering.
Xing, W., Zhou, C., Li, J., Wang, W., He, J., Tu, Y., et al., 2022. Suit- ability evaluation of tea cultivation using machine learning technique at town and village scales. Agronomy. 12, 2010.
Yang, H., Ma, W., Liu, T., Li, W., 2023. Assessing farmland suitability for agricultural machinery in land consolidation schemes in hilly terrain in China: A machine learning approach. Front. Plant Sci. 14, 1084886.
Zhang, J., Su, Y., Wu, J., Liang, H., 2015. GIS-based land suitability assessment for tobacco production using AHP and fuzzy set in Shandong Province of China. Comput. Electron. Agric. 114, 202–211.
Zhang, S., Liu, X., Li, R., Wang, X., Cheng, C., Yang, O., & Kong, H. (2021). AHP-GIS and MaxEnt for delineation of potential distribution of Arabica coffee plantation under future climate in Yunnan, China. Ecological Indicators, 132, 108339.
Zhang, S., Liu, X., Wang, X., Gao, Y., & Yang, Q. (2021). Evaluation of Coffee Ecological Adaptability Using Fuzzy, AHP, and GIS in Yunnan Province, China, Arabian Journal of Geosciences, 14, 1366.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Machine Learning Models to Develop Land Suitability Map for Coffee Cultivation by Integrating CHIRPS and SRTM DEM

Status:

Version 1

Abstract

Figures

1. Introduction

2. Material and Methods

2.1. Description of Experimental Region

2.2. Soil Data

2.3. Climate Data

2.4. Topographic Data

2.5. Methodology

2.6. Land Suitability Assessment

2.6.1. Processing of Soil Data

2.6.2. Processing of Climate Data

2.6.3. Processing of Topographic Data

2.6.4. Determining Land Suitability Index Using Principal Component Analysis

2.6.5. Determination of Land Suitability Classes

2.7. Machine Learning Techniques

2.7.1. K- Nearest Neighbor

2.7.2. Decision Tree

2.7.3. Random Forest

2.7.4. Extreme Gradient Boosting Tree

2.7.5. Na¨ıve Bayes

2.7.6. Model Hyperparameters

2.8. Evaluation of Machine Learning Performance

2.8.1. Statistical Measures

2.8.2. ROC Curve and AUC

3. Results and Analysis

3.1. Variable Importance of the LSF Affecting Coffee Cultivation

3.2. Spatial Land Suitability Mapping

3.3. Validation of the RF Model

4. Conclusions

Declarations

Author Contribution

Acknowledgement

References

Additional Declarations

Status:

Version 1