GIS-based analysis of landslides susceptibility mapping: a case study of Lushoto district, north-eastern Tanzania

Landslides are becoming increasingly widespread, claiming tens of thousands of fatalities, hundreds of thousands of injuries, and billions of dollars in economic losses each year. Thus, studies for geographically locating landslides, vulnerable areas have been increasingly relevant in recent decades. This research is aimed at integrating Geographical Information Systems (GIS) and Remote Sensing (RS) techniques to delineate landslides susceptibility areas of Lushoto district, Tanzania. RS assisted in providing remote datasets including; Digital Elevation Models (DEMs), Landsat 8 OLI imageries, and past spatially distributed landslides coordinate with the use of a handheld Global Position System (GPS) receiver, while various GIS analysis techniques were used in the preparation and analysis of landslides influencing factors hence, generating landslides susceptibility areas index values. However, rainfall, slope angle, elevation, soil type, lithology, proximity to roads, rivers, faults, and Normalized Difference Vegetation Index (NDVI) factors were found to have a direct influence on the occurrence of landslides in the study area. These factors were evaluated, weighted, and ranked using Analytical Hierarchy Process (AHP) technique in which a 0.086 (8.6%) Consistency Ratio (CR) was attained (highly accepted). Findings reveal that rainfall (29.97%), slopes’ angle (21.72%), elevation (15.68%), and soil types (11.77%) were found to have high influence on the occurrence of landslides, while proximity to faults (8.35%), lithology (4.94%), proximity to roads (3.41%), rivers (2.48%), and NDVI (1.69%) had very low influences, respectively. The overall results, obtained through Weighted Linear Combination (WLC) analysis techniques indicate that about 97669.65 Hectares (ha) of land are under very low levels of landslides susceptibility, which accounts for 24.03% of the total study area. Low susceptibility levels had 123105.84 ha (30.28%), moderate landslides susceptibility areas were found to have 140264.79 ha (34.50%), while high and very high susceptibility areas were found to cover about 45423.43 ha (11.17%) and 57.78 ha (0.01%), respectively. Furthermore, 81% overall model accuracy was obtained as computed from the Area Under the Curve (AUC) using Receiver Operating Characteristic (ROC) curve.


Introduction
Globally, landslide events are progressively increasing resulting in adverse impacts on people's lives, the economy, and the environment at large (Collins and Jibson 2015;Kirschbaum et al. 2015). Landslides are events where the mass of rocks, soil, or debris falls down a slope (Li and Chen 2020;Silalahi et al. 2019). They include sliding and falling of earth's materials due to the effects of gravity (Das 2011;Motamedi 2013) as a result of internal factors mainly including the geological and geomorphological configuration of the terrain as well as external factors triggered by human activities imposing pressure on the terrain, hence altering the surface slope, which leads to instability of the area causing slope failure (Agliardi et al. 2012;Gupta et al. 2021). On the other hand, landslides are a result of earthquakes and rainfall (Gupta et al. 2021). Effects of landslides events globally are critical Díaz et al. 2020): However, in recent times, the impacts are more likely to increase due to the growth in population and increase in human activities as people continue to expand their settlements and other activities thus, infringing and encircling vulnerable high grounds that are prone to failure (Tegeje 2017). Landslides not only affect human production and life but sometimes has a great influence on human society Froude and Petley 2018;Haque et al. 2019) and the environment (Chalise et al. 2019;Turner 2018). It is estimated that worldwide from 1998 to 2017, landslides accounted for 5.2% of natural hazards (CRED, 2018). However, effects due to landslides caused global total annual damage of about 18 billion Euros ) and approximately 66,438 deaths by 2020 worldwide (Guha-Sapir et al. 2020).
In Africa, landslides occur almost in every region but are more frequently reported in Sierra Leone, the Democratic Republic of Congo, Nigeria (Igwe 2018), Cameroon (Nama 2020), Ivory Coast, Kenya, Uganda, Rwanda, Ethiopia, and Tanzania (Knapen et al. 2006;Nyssen et al. 2003). However, in most highlands of East Africa such as Ruwenzori in Uganda, Elgon in Uganda/Kenya, Aberdare Ranges in Kenya, and Bukavu in the Democratic Republic of Congo landslides are more common (Mugagga et al. 2012) with haphazard impact on the environment and society. In Tanzania, various types of landslides have been widely documented from different regions of the country in a lapse of time. This includes large quaternary debris avalanches reported on Mt. Meru and Kilimanjaro (Delcamp et al. 2013), Ol Doinyo Lengai, and Kerimasi Mountains (Kervyn et al. 2008), Uluguru Mountains (Westerberg and Christiansson 1999), and Rungwe Mountains (Fontijn et al. 2012). Nevertheless, mudflow has been reported in different time frames at Usambara Mountains, Tanga region, and in Rondo & Makonde Plateaus in the southern regions of Tanzania. Rock falls and rock topple are the most common landslide events in Lushoto district, northeastern Tanzania, reported in various time frames (Tegeje 2017). Nevertheless, in 1998, landslide hazards destroyed 60 hectares of land with intense impacts to the society (Tegeje 2017). On the other hand, during the rainy season, most landslide events in the area leads to human death and injuries, loss of properties including farm products, swiping away soil nutrients by surface runoff, destruction of houses, roads, and occasionally electrical cutoff among others. Short-term migration of families and their livestock to secure safe places are a common phenomenon during this period. Severe rainfall, changes in drainage patterns, farming on slopes, and deforestation as a result of settlement growth and the nature of the topography are the agents of landslides in the area, resulting in catastrophic corollaries on society and the environment. However, they have not been effectively and scientifically studied within the area, and their spatial distributions are still undefined.
In recent years, several landslide predictive models have been employed to widely map the landslide's susceptibility zones based on the analysis, scale, modelling approaches, and causative evaluation factors (Brenning 2005;Reichenbach et al. 2018). These models consist of a qualitative approach, which usually returns a landslide susceptibility zonation in terms of weighted indices and relative ranks (e.g., low, medium, high); adopted for local scale and site-specific studies, and quantitative approaches which estimate landslide occurrence as a numerical value (Roccati et al. 2021). Currently, GIS and RS technologies have been extensively employed and proven effective for landslide susceptibility mapping (LSM) and assessment worldwide (Dou et al. 2019). These technologies are advantageous in managing a large volume of datasets both in terms of data volume and the geographical scale, as well as analyse dynamics of landslides and determining susceptibility zonation representing an important requisite for proper management of land and risk mitigation on the earth's surface (Lai and Tsai 2019;Turconi et al. 2019). Often, these technologies combine various datasets collected from innovative techniques such as satellite remote sensing and light detection and ranging (LiDAR) images (Roccati et al. 2021). These technologies are widely integrated with numerous statistical models and approaches extensively used in LSM as engaged in various studies globally including frequency ratio (FR) (Abdo 2022;Melese et al. 2022), index of entropy Singh et al. 2021), analytical hierarchy process (AHP) (Jam et al. 2021;Moragues et al. 2021), analytical network process (ANP) (Hamzeh and Amiri 2020; Swetha and Gopinath 2020), information value (Sharma et al. 2020;Wubalem and Meten 2020), and fuzzy logic (Baharvand et al. 2020;Bahrami et al. 2020) among others, and a hybrid of the aforementioned techniques can be used to overcome biases (Gheshlaghi and Feizizadeh 2021). However, for regional and small-scale landslide mapping, statistical methods are the most widely frequently adopted techniques (Shano et al. 2021). As a result, the current study has employed AHP a multicriteria decision analysis (MCDA), which allows the evaluation of consistency in the assignment of causative weights, hence lessening the subjectivity factor in the treatment of variables. The technique is widely used for analysing complex spatial-related challenges such as site selection, groundwater assessment, ecological studies, flood assessment, and landslides susceptibility analysis among others (Biswas et al. 2020;(Makonyo and Msabi 2021a, b;Morandi et al. 2020;Msabi and Makonyo 2021). The method applies semi-quantitative approaches and the judgement is analysed through weighting of factors by pairwise comparison matrix (PCM) (Saaty 1980).
LSM is a complex process that requires expertise from different disciplines include geology, environmental, and hydrogeology among others (Ayalew et al. 2004). However, various researchers have integrated various causative factors from different disciplines including slope angle, geomorphology, slope, lithology (Moragues et al. 2021), relief, surface roughness, distance to streams (Bostjančić et al. 2021), slope angle, slope aspect, rainfall, lithology, altitude, land use (Jam et al. 2021), lithology, elevation, slope, aspect, curvature, land cover, proximity from drainage networks, lineament, and roads, (Ozioko and Igwe 2020) but in different combination. However, nine influencing factors slope angle, aspects, elevation, lithology, proximity to rivers, proximity to geological faults, roads, rainfall, and land use/land cover (LULC) were employed in this study. Therefore, the current study is aimed at identifying landslides susceptibility zones at Lushoto district, northeastern Tanzania based on GIS and MCDA.

Description of the study area
Lushoto district lies on the toes of Usambara Mountain making it one of the eight districts of Tanga region in mainland Tanzania, Lushoto town, being its capital city covering a total area of about 4125. 17 km². The district is bordered by the Republic of Kenya on the northeastern side, same district of Kilimanjaro region in the northwest, Korogwe district on the southern side, Bumbuli district on the southeastern and Mkinga with Muheza district on the further eastern side (Fig. 1).

Climate
Lushoto district is situated in the tropical climatic region receiving an average amount of annual precipitation of about 999.9 mm (39.37 inches). Heavy (rainy season) rainfall is experienced in March, April, May, October, and November in which May is the wettest month with an average of 278 mm (10.94 inches) of precipitation. However, the driest season is experienced in February with an average of 33 mm (1.3 inches) of precipitation. However, the region has an average annual temperature of 28 °C, February being the hottest month of the year and drops to 23.8 °C in July the coldest month. The study area's average annual relative humidity is almost 70%. The portion of the study Fig. 1 Lushoto district administrative boundary area covering the Piedmont plain is located in a dry-warm area, while the portion covering the plateau and escarpment is located in a humid cold agroecologic zone (Massawe 2011).

Geomorphology
The geomorphology of Lushoto district is categorized by steeply rugged terrain with a lot of mountain peaks separated by deep valleys. Thus, higher peaks within the study area range between 1800 and 2500 m above the mean sea level. Topographically, a high area including the Mabughai-Mlomboza plateau symbolizes a down-faulted block bounded by a sharp fault escarpment. However, Usambara Mountain is characterized by a series of uplifted blocks parted by many faults. The tectonic movement causes bauxitization within the area causing a lowering of bauxitized blocks with fewer effects on erosion (Braslow and Cordingley 2016).

Land use/land cover (LULC)
The study area is characterized by both natural and planted vegetation cover especially forest plantations. The natural forest in the area is composed of mainly Camphor (Ocotea usambarensis) with Podo (podocarpus usambarensis and Podocurpus pensiculi), Lansthus cirumilee, and other shrubs. However, the associated species are Parinari excels Pygeum africanum, Ficalhoa laurifolie, Polycas spp. Macaranga kilindscharica, and Crysophylum spp. Olea hochstetteri and Cassipourea spp. The main species of forest plantation are the Cedar (Juniperus proceria) Cypress (Cupresuus lusstanica), Pinus petula, and Pinus radiata (Kamugisha et al. 2007). Furthermore, the mainland uses in the area include food & cash crops agriculture, commercial agriculture, protected forest reserves, and pastures (Kamugisha et al. 2007).

Soil
The study area's soils are primarily clays, sandy clays, and sandy clay loams, ranging in colour from red to grey-brown to black, with low pH values and high iron, manganese, and magnesium content. Thus, the dominant soils on the slopes have an argic horizon that is often heavily eroded and thus acidic (Acrisols, Acric Ferralsols, Lixisols, and Alisols). Furthermore, younger, less eroded soils (Luvisols and Lixisols) can be found. The valley bottoms have dominant soils influenced by a fluctuating groundwater table (Fluvisols and Gleysols), whereas the hilltops have superficial soils with an abundance of rock and petro-plinthic outcrops rich in iron (Neerinckx et al. 2010).

Population size, growth, and structure
Since 2002, the population of Lushoto District Council has grown from 279,096 to 492,441 in 2012 (NBS, 2012) as the 2022 district population census results still await. Tanga District, which is primarily the city, has the highest population density in Tanga Region at 144.54 people per square kilometre. The fertile agricultural land for the 1 3 production of cash and food crops contributes to the former's population density. In general, topography, social service availability, and climate have all influenced the population distribution pattern in Lushoto district.

Data types and sources
Various datasets including structural maps, soil, and geological maps were acquired from the Geological Survey of Tanzania (GST), while rainfall datasets were acquired from the Centre for Environmental Data Analysis (CEDA) archive ranging from 1901 to 2021 with 30 m spatial resolution (https:// cruda ta. uea. ac. uk/ cru/ data/ hrg/). Faults were directly generated from the structural map and road datasets were downloaded from open street maps (https:// www. opens treet map. org) in the form of shapefiles. Roads covering the study area were extracted and proximity values from this dataset were calculated for further analysis. Four Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) tiles datasets with 30-m spatial resolution each covering the study area were acquired from the USGS earth explorer website (https:// earth explo rer. usgs. gov/), which were later merged and analysed to generate slopes and elevation datasets as well as drainage systems, which were further processed to generate proximity values used the analysis. Furthermore, Landsat 8 OLI satellite imageries with 30-m spatial resolution of April 2022 were acquired from (https:// earth explo rer. usgs. gov/) in which band 5 (near-infrared) and band 4 (red band) were used to generate the normalized difference vegetation index (NDVI) values of the study area. Finally, a handheld Global Positioning System (GPS) survey was undertaken in the study area in which landslide spatial points (coordinates) including the rock falls, mudflow, and the rock topples types of landslides spatial points were collected.

Landslide influencing factors
LSM is a complex process that requires various landslide influencing factors from different disciplines, which contribute differently to landslides occurrence. These factors closely relate to the environmental settings of an area including the geomorphological, geological, hydrological, and topographical characteristics (Kawabata and Bandibas 2009). The combination and diversification of data from different fields for landslide evaluation optimize the findings (Moragues et al. 2021). In the current study, the diversity of conventional thematic maps of geological factors (proximity to faults, lithology & soil types), topographical factors (slope angles & elevation), anthropogenic factors (proximity to roads), geomorphological factors (rivers and NDVI), and finally metrological factors (rainfall) were employed in which among these factors have been used in various landslides studies worldwide (Berhane et al. 2020;Das and Lepcha 2019;Hong et al. 2018;Panchal and Shrivastava 2020). However, the rasterization followed by resolution merging was undertaken and finally, these thematic maps were converted into Universal Transverse Mercator (UTM) zone 37 S projected coordinate system in the ArcGIS environment. Thus, to achieve the objective of this study, the following landslides influencing factors were employed.

Rainfalls
Rainfall is one of the most landslide influential factors in different parts of the world (Jennifer et al. 2021;Youssef 2015). It is considered an active factor; thus, it directly influences the stability of the surface slope in a given area (Fig. 2a). Furthermore, both heavy, short, and prolonged rainfall controls the surface runoff and energizes soil pore water pressure, thus triggering the weakening of the soil and compromising terrain stability (Jennifer et al. 2021). Various scholars have employed rainfall datasets in landslide studies (Gheshlaghi and Feizizadeh 2021;Hong et al. 2018;Jennifer et al. 2021) and found it the most significant landslide influential aspect.

Slope angle
The slope is the degree of change in height over a given area with lower slope values representing flat surfaces and high values showing elevated areas (Rahaman and Aruchamy 2017). Slope angle is among the most influential factor in landslides assessments Nguyen et al. 2019); thus, it directly impacts the shear forces (Lee and Min 2001), soil water content, soil formation, soil stability (Berhane et al. 2020;Riaz et al. 2018), development of plants roots and evapotranspiration (Faber 2003;Wang et al. 2011). A slope gradient triggers the initiation of mass movement due to gravity effects acting on the slope acting as a sliding plane (Torizin 2011). However, mass removal is highly accelerated when a threshold of the surface angle of inclination reaches 30° (Tofelde et al. 2017) (Fig. 2b).

Elevation
This is the height above the mean sea level, which has direct influences on biological and natural aspects (Kavzoglu et al. 2014), and it is one of the most landslide-influencing factors ). This factor normally influences most geological and geomorphological aspects of the earth's surface (Ayalew and Yamagishi 2005). However, stable surfaces are represented by low altitudes, while higher grounds are usually unstable (Devkota et al. 2013) (Fig. 2c).

Soil types
The soil type of a given area influences landslide events, particle size, shape, and pore characteristics of the soil sway slope stability (Fig. 2d). However, soil characteristics can be useful in assessing landslides (Das 2011;Rossi and Reichenbach 2016) and controlling the type of landslides in the area. Soil also determines water infiltration in a given area, water movement, and the capacity of soil to hold water (Silalahi et al. 2019). Soil such as clay and silt have large pore spaces, hence holding a higher volume of water, therefore, more prone to landslides (Lepore et al. 2012). However, soil with a thin depth is more prone to landslides (Das and Lepcha 2019).

Proximity to geological faults
Geological faults are gaps between two different lithological units and describe the zones of weakness (Fig. 2e). They induce fractures, which can trigger landslide events due to shaking of the tectonic plates as instability increases nearer to the lineaments and conversely (Regmi et al. 2016). However, active faults accelerate the process due to the weakness of rocks as a result of vibration waves that occur frequently (Okuwaki et al. 2021).

Lithology
Lithology is the main factor that directly influences landslide development in a given area (Abedini et al. 2019;Tian et al. 2019). The properties and resistance of each material from a given lithology control the distortion characteristics and its stability (Meena et al. 2019;Moreiras 2009) as shown in Fig. 2f. However, it determines the degree of weathering, hence distinguishing one rock from another, influencing slope stability in a given set of the environment (Sarkar and Kanungo 2017). Various researchers have employed lithology in landslide studies (Dang et al. 2019;Panchal and Shrivastava 2020). However, the nature of the rock type has to be understood to determine the rock that permits the storage of water in pore spaces acting as landslide agents (Jennifer et al. 2021).

Proximity to roads
Proximity to roads has a significant association with the occurrence of landslides, which can be a result of cutting the toe's slope for road construction purposes, hence disturbing the geological and slope formation of a given area, which eventually leads to slope failure (Kavzoglu et al. 2014). This proximity accounts for one of the significant factors for slope instability, hence resulting in slope failure (Jennifer et al. 2021) (Fig. 2g). This factor is widely used in landslide studies as one of the causative parameters for landslide occurrence as it greatly influences the area due to moving automobiles (Chen, Pradhan, et al., 2019;Moragues et al. 2021; Ozioko and Igwe 2020).

Proximity to rivers
Rivers and groundwater channels play a key role in landslide occurrence, as they scrape the channels' escarpments and the slope bases (Fig. 2h). However, water from rivers soaks part of the material forming the slope, hence triggering landslides (Demir et al. 2013). Proximity to river datasets has been widely employed in landslide studies (Gheshlaghi and Feizizadeh 2021;Moragues et al. 2021;Shano et al. 2021).

NDVI
Landslide events are more related to the density of vegetation in a given area (Meng et al. 2016). Thus, the presence of vegetation forces the soaking of water underground hence a well-saturated soil can trigger landslides (Fig. 2i). NDVI parameter can well determine the presence of vegetation coverage (Hong et al. 2016). This dataset helps to determine the distribution of vegetation in the area, which triggers the stability of soil particles (Löbmann et al. 2020).

3 3 Methodology
Landslides influencing factors employed were standardized using a 1-5 suitability scale, weighted, and ranked by using the heuristic AHP technique. Each factor was assigned a weight according to its relative influence in LSM. The higher the factor's weight the higher the influence (Saaty 1980). Furthermore, the WLC method was employed to calculate the final LSM of the study area as employed in Fig. 3 below.

Normalization and factor's weight evaluation
The method undergoes three evaluation phases, including the construction of hierarchies, determining priorities, and evaluation, which is done utilizing the consistency index (CI) (Saaty 1980). AHP process involves the following logical steps; (a) Itemization of the influencing factors into components, (b) systematize the factors into a hierarchical order, (c) assigning a numerical value to determine the relative importance of each factor, (d) construction of a comparison matrix, (e) computing the eigenvectors (final score), and (f) ranking the alternatives (Malczewski 1999;Saaty and Vargas 2001). This method is useful in landslide mapping as the inconsistency in the judgement is easily determined (Kayastha et al. 2013). The factors were standardized based on 1 to 9 Saaty's assessment scale of relative importance ( Table 1).

Pairwise comparison matrix (PCM) of the influencing factors
The PCM is constructed where factors are listed based on their relative influence on landslide occurrence. The diagonal values should have equal influence (equal to 1). The PCM of the matrix (μ) is computed by Eq. 1 and as shown in Table 2 below. Strongly more significant 7 Very strong more significant 9 Extremely more significant 2,4,6,8 Intermediate values between adjacent values Furthermore, these criteria were normalized (standardized) to check for any redundancy and the weights were computed (Table 3).

Calculation of the principal eigenvalue( w )
This factors (λ w )helps to access and measure the consistency of the evaluated weights (Saaty 1980). The calculated weights are accepted if only the λ w is equal to or greater than the number of involved factors and the consistency ratio (CR) is less than 10% (0.1) (Saaty 1980). In the current study, λ w was found to be 10.04, which is greater than the number of factors (9) ( Table 4).

Computation of the consistency index (CI) of the model
Computation of CI is performed in the AHP method to evaluate the judgement undertaken in assigning weights. CI is given by the ratio of λ w and the numbers (n) of factors evaluated and is given by (Eq. 2).

Computation of the consistency ratio (CR)
In the computation of CR, Thomas Saaty's Random Index (RI) values are used (Table 5).
These RI values give the index values based on the number of factors evaluated in the model (Saaty 1997). The CR is computed by (Eq. 3).

Weighted linear combination (WLC) analysis
The aforementioned reclassified landslides influencing factors employed in the current study: rainfall, slope angle, elevation, soil types, proximity to faults, lithology, proximity to roads, proximity to rivers, and NDVI with their respective influencing weights ( r i ) were integrated and analysed by the WLC technique (Eq. 4) to produce LSM of Lushoto district using Arc-GIS Pro software. This technique employs a mathematical algorithm by summing and multiplying influencing factors' weights with their conforming reclassified thematic layers. Thus, using the raster calculator toolset available in the map algebra toolset of the spatial analyst tools each reclassified factor was imported one by one and multiplied by each respective computed weight, and finally, the additional algorithm was employed to sum these products, hence (3) CR = CI RI Therefore, CR = 0.129 1.5 = 0.086%(Accepted) generating the final landslides susceptibility index of the study area. Thus, the rainfall reclassified raster dataset was added to a raster calculator and multiplied by its relative weight computed from AHP, and then the slope angle reclassified raster dataset was again added and multiplied by its weight and added until the last influencing factor, which was NDVI, and finally, LSM was generated (Eq. 4). The above mathematical algorithm enables the production of the LSM as a single landslides index of the study area, which is ranked from 1 (very low landslides susceptibility) to 5 (very high susceptibility). This number indicates landslides susceptibility levels of the area, which are finally renamed into; very low, low, moderate, high, and very high landslides susceptibility levels. Various colour coding schemes have been proposed for disaster mapping; however, from a landslides perspective, "Red" colour has been proposed for highly susceptible landslides areas, "Orange" for high susceptible areas, "Yellowish" for moderate, "Quetzal Green" for low, and "Peacock Green" for very low landslides susceptible areas (Singh 2009). Therefore, very high susceptible areas represented by 5 were assigned a red colour, high shown by 4 were assigned orange colour, moderate areas shown by 3 were assigned a yellow colour, and low and very low susceptible areas represented by 2 and 1, respectively, were shown by Quetzal green and Peacock green colours, respectively.

Mudflow
Mudflow-type landslides are observed in the southernmost parts of Lushoto district and mainly in Usambara ward. About two (2) mudflow type of landslides events were observed in the lowland areas of the district, which mostly receive high-intensity of surface runoff during the rainy season where surface rainwater is accumulated from the highlands of Usambara Mountain flowing to low-lying areas eventually resulting in a mudflow (Fig. 4). These types of landslides are typically associated with significant property destruction and loss of life because they typically transport unconsolidated materials, mud, and logs, which are normally agents of destruction as they travel a long distance while causing additional destruction along the way (Mücher 2009).

Rock falls
Results show that these types of landslides are common within the study areas with severe effects on society and the community at large. Figure 4 indicates that most wards of Lushoto district are more pounded by rock falls mainly as a result of human intervention, shaking of the earth's crust, and rainfall-related processes that weaken the surface slopes and changes in surface water level within the study area. The findings also clearly show that approximately nine (9) rock fall types of landslide events occurred in the study area, with the majority of these events occurring in elevated (plateau) areas of the district, as most of these areas have high slope angles that are easily weakened and affected by earth's movement. These results greatly agree with other scholars on landslide mapping and management including Jennifer et al. (2021), who found that most rock fall types of landslides are common in mountainous areas or most elevated areas as these areas easily experience weathering and hence are facile weakened.

Factors influencing the occurrence of landslides in Lushoto district
The identified landslides influencing factors were reclassified on a scale of 1-5 using the proper user-defined interval algorithm in the ArcGIS 10.8 environment based on the identified values in each used dataset. Each factor was reclassified into; very low, low, moderate, high, and very high susceptibility levels. Based on the analysis, the following susceptibility levels with respect to their percentage area coverage were identified in each employed influencing factor;

Rainfall levels of landslides influence
High rainfall intensity is the most influential factor in the occurrence of landslides in many parts of the world (Youssef 2015). It plays a vital role in loosening the surface landmass by increasing the soil pore pressure. The rainfall dataset was categorized into 1000-1100 mm, 1100-1200 mm, 1200-1300 mm, 1300 mm-1400 m, and above 1400 mm, which was finally reclassified into very low, low, moderate, high, and very high influences on landslides occurrence (Fig. 5a). The analysis reveals that rainfall highly contributes to landslides occurrence within the study area with 29.97% (Table 6) as also obtained in other studies (Jennifer et al. 2021;Michael and Samanta 2016;Youssef 2015). Thus, about 22.66% of the total areas, which is equal to 926.325 ha have very high influences on the occurrence of landslides, thus, receive very high rainfalls. On the other hand, 19.89% of the area, which is approximately equal to 813.32 ha is under a region of high landslides influences and thus receives high rainfalls, 30.34% of the total study area, which is equivalent to 1240.42 ha is under moderate areas of influences receiving moderate rainfall while low and very low areas of landslides influence cover about 15.99% and 11.12%, respectively, equivalent to about 653.74 ha and 454.66 ha of land, respectively. Thus, about 50.23% of the total study area has a higher likelihood of being pounded by landslide events and is mostly in high areas. These findings are similar to other studies around the world, thus according to Youssef (2015), rainfall contributed to around 46.15% of the likelihood of landslide occurrence in Ar-Rayth area, Jizan, Kingdom of Saudi Arabia. Jennifer et al. (2021), found that rainfall is an important factor to consider in landslide susceptibility mapping Nilgiris District, Tamil Nadu, India in which rainfall influenced approximately 30% of the landslide occurrence within the study area. Furthermore, Michael and Samanta (2016), found that approximately 80% of landslide occurrences were due to rainfall thus, rainfall acts as the common triggering agent for landslide events occurrence in most areas worldwide.

Slope's angle levels of landslides influence
A higher slope angle influences the occurrence of landslides as a result of slope failure due to the instability of the underlying surface (Tofelde et al. 2017). Thus, from the analysis slope angle ranging from 0 to 7° were reclassified as having very low levels of influence,  (Fig. 5b), with the overall influence of 21.72% (Table 6). Very low slope angle influence lies in most northeastern and western parts of the study area covering about 171660.7 ha equal to 42% of the total study area. Low landslides influences occupy approximately 96575.1 ha of land, which is equivalent to 23.6%, while moderate, high, and very high slope angle influences occupy 74756.7 ha, 50732.3 ha, and 15127.9 ha, respectively, dominating central and southern parts of the study area, which is equal to 18.3%, 12.4%, and 3.7%, respectively. Very high slope angle influences are mostly observed in the outermost part of the central region, southern and most parts of the middle east of the study area. Furthermore, the results show that from a slope's angle perspective, very high landslide levels of influence cover a very small area of the study area, which is mostly observed in the outermost part of the central region, southern and most parts of the middle east of the study area. However, high slope angle influences are observed in the central and southern parts of the study area with few patterns indicating a downward orientation. Generally, this indicates that very high and high slope angle influences are most concentrated in the southern region of Lushoto district. Furthermore, moderate, low, and very low slope angle levels are seen in the northern parts of the area. The finding from this study agrees with other scholars elsewhere, Hong et al. (2018), indicated that the slopes of an area determine the occurrence of the landslide as their stability and failure decide these events. Furthermore, Jennifer et al. (2021) indicated that slope is commonly affected by most water permeability, thus controlling slope stability hence is of vital importance in LSM.

Elevation levels of landslides influence
Very high elevations were considered to have very high landslide influence levels (1700-2300)m as articulated by various researchers worldwide Iqbal et al. 2021), while low-lying regions ranging from 215 to 600 m were reclassified as having very low levels of landslides influences. Elevations ranging from 600 to 1000 m, 1000 to 1300 m, and 1300 to 1700 m were considered as low, moderate, and high levels of landslides influences, respectively (Fig. 5c). From the analysis, very low elevation levels of influence covers about 191061.1 ha, which is approximate 46.7% while very high elevation influences lie in an area equal to 51705.47 ha, which is equivalent to 12.6%. Low, moderate, and high levels of influence cover an area equal to 58209.71 ha, 49341.61 ha, and 58534.51 ha, which are equivalent to 14.2%, 12.1%, and 14.3%, respectively (Table 6). Furthermore, results indicate that very high elevation levels of influences are more concentrated in the eastern part; thus, these areas have a high likelihood of facing landslides in the study area. However, moderate levels of influence are revealed in the outer ward region of the central area. Generally, based on this factor, landslides are more likely to occur in the central to southern parts of the study area. Furthermore, very low elevation levels of influences are revealed in the northeastern parts of Lushoto district covering the lowest-lying areas as an indication of a very low likelihood of landslides occurrence based on this criterion. Thus, these findings indicate that elevation has a significant role in landslide occurrence as indicated in other studies globally (Chen, Shahabi, et al., 2019;Iqbal et al. 2021).

Soil-type levels of landslides influence
Soil types within the study area were reclassified based on their influences on water retention. Eutric leptosols and Umbric acrisols soil types were reclassified as having very high and high levels of landslides influences, respectively, while Haplic luvisols and Chromic luvisols were considered to have moderate and low influences, respectively, within the area (Fig. 5d), contributing to almost 11.77% influence of landslides occurrence ( Table 6). The analysis indicates that low levels of influence cover an area equal to 186646.3 ha equivalent to 45.7% of the total area, moderate lies at about 241.6068 ha, which is almost 0.1% while high and very high cover an area equal to 110486.6 ha, and 111462.1 ha, respectively, equivalent to 27.3% and 23.0% of the total study area, respectively. These results indicate that very low influences based on the soil type are revealed in the southwest part of the district. While higher influences are shown in the central region protruding from the south to the northwest part of the study area. This shows that the likelihood of landslides occurrence with respect to soil type is in the southwestern part of the area. Through a thorough literature review, scholars indicate that soil type that permits water lodging are more prone to landslides as also found in this study (Jennifer et al. 2021; Swetha and Gopinath 2020).

Proximity to geological faults levels of landslides influence
Areas close to faults are more prone to failure, hence resulting in a higher likelihood of landslides occurrence (Michael and Samanta 2016). Thus, proximity distance from faults was calculated and reclassified based on other works of the literature (Moragues et al. 2021). In this study, proximity analysis of the geological faults was performed, and proximity values were reclassified; less than 200 m as having very high influences of landslides occurrences, 200-500 m as high while 500-800 m, 800-1200 m, and greater than 1200 m as moderate, low, and very low influences, respectively (Fig. 5e), with overall 8.35% of landslides influence (Table 6). The results disclose that very low proximity to faults levels of influence lie in an area equal to 355881.8 ha, which is equivalent to 87% of the total study area. Low and moderate levels cover an area equal to 17348.98 ha and 13138.01 ha equivalent to 4.2% and 3.2%, respectively. High and very high proximity to fault levels of influences lie in an area equal to 13379.05 ha and 9094.365 ha covering about 3.3% and 2.2% of the total area, respectively. The findings reveal that areas very close to fault are more likely to trigger landslides due to the effects of tectonic plates shaking hence considered zones of weakness. High and moderate levels of influence are also shown very close to faults, while very low levels of influence dominate the area. On the other hand, very high levels of influence are observed very close to faults that cover the low-lying areas of the district. These findings agree with other scholars globally, Michael and Samanta (2016), argued that proximity to faults influences its stability, hence increasing the susceptibility of landslides occurrence.

Lithology levels of landslide influence
The area comprises mainly two lithological units in which Neoarchean rocks with Neoproterozoic overprint, which was reclassified as having a very high influence on the occurrence of landslides and Neoproterozoic granulite complexes as having a higher influence on the occurrence of landslides events as adopted from other studies (Jennifer et al. 2021) (Fig. 5f). The analysis indicates that lithology contributes to almost 4.94% of landslide influence with very low lithological levels of influences occupying about 103434.8 ha, which is equal to 25. 3% of the total study area, while high influences occupy most parts of the area covering an area equal to 305398.6 ha, which is equivalent to 74.7% of the study area (Table 6). The lithology of the area highly influences the movement of the earth's materials within the area. Thus, the results indicate that most of the study area is covered by Neoproterozoic overprint, which protrudes from the southern part of the study area to the northern region. Thus, based on this factor likelihood of landslides occurrence is very moderate in the central region and very low in the southwestern and northeastern areas of the study area. These results strongly agree with other scholars who found that the geology of an area strongly controls landslide occurrence as it also determines water permeability and lodging in a given area (Singh and Chakrapani 2015). However, Shano et al. (2021) pointed out that when the lithology of an area is loose or fragile the likelihood of landslides occurrence increases, and these types of rocks are commonly extremely weathered and fractured.

Proximity to road levels of landslides influence
Proximity distances were calculated from the existing roads and reclassified based on the nature of the study area and other pieces of the literature (Ozioko and Igwe 2020;Shano et al. 2021). Distance less than 200 m was considered to have a very high influence on landslides occurrence, while a distance ranging from 200 to 500 m, 500-800 m, 800-1200 m, and greater than 1200 m were considered to have high, moderate, low, and very low levels of influences (Fig. 5g) and contributes about 3.41% of landslides influence ( Table 6). The findings reveal that very high levels of influences cover an area equal to 36531.99 ha, which is about 8.9% of the total study area. High and moderate levels cover about 45328.76 ha and 39257.98 ha equals 11.1% and 9.6%, respectively, while low and very low occupies about 45491.72 ha and 242232.2 ha equivalent to 11.1% and 59.2%, respectively. These results portray significant information on the anthropogenic disturbances with respect to landslide occurrence in the study area. Furthermore, results indicate that very high levels of landslide influences are significant near roads due to disturbances and shaking related to human activities including construction and the passing of heavy trucks, which results in the loosening of uppermost soil. High and moderate influence areas are also shown at a few distances from the road networks. Very low landslide influences dominate most parts of the area as an indication that these areas are not interacted by road networks hence free from human disturbances such as road maintenance, construction and passage of heavy equipment. Thus, the analysis revealed that areas very close to roads are more likely to be affected by various anthropogenic activities such as road construction and vibration due to heavy passing trucks as also suggested by other scholars globally (Chen, Sun, et al., 2019).

Proximity to river levels of landslides influence
The analysis shows that areas very close to river channels are more prone to failure due to water movement and erosion. In the current study, proximity distances were calculated in which less than 200 m were considered to have very high landslide influences as compared to further distances. Thus, proximity distances ranging from 200 to 500 m, 500-800 m, 800-1200 m, and greater than 1200 m were considered as having high, moderate, low, and very low levels of landslides occurrence, respectively (Fig. 5h), with overall of 2.48% of landslides influences (Table 6). Further analysis was undertaken to find the percentage coverage of each level of influence, results indicated that very high and higher levels of landslide influences cover an area equal to 42264.23 ha and 57201.03 ha, which are equal to 9.8% and 13.3% of the total study area. Moderate and low levels lie at about 52015.48 ha and 61439.38 ha, which are almost 12.1% and 14.3%, respectively, while low influences cover about 217505.1 ha equivalent to 50.5% of the total study area. Streams play a vital role in landslide occurrence as they are agents of soil erosion and destructors of riverbanks (Moragues et al. 2021). The results reveal that the likelihood of landslides occurrence is very high in areas very close to river channels. However, the likelihood decreases significantly away from streams as the effects due to water flow are low as you move away from these streams. Moderate and low proximity to rivers susceptibility levels is very minimum away from channels and the probability of occurrence based on this factor becomes very minimal in areas with no river channels.

NDVI levels of landslides influence
NDVI ranging from (−0.04) to 0.017 were considered to have very high influences on the occurrence of landslides in the study area and 0.17-0.23 as high, while 0.23-0.31, 0.31-0.39, and 0.39-0.64 NDVI values were considered as moderate, low, and very low levels of landslides influences within the area (Fig. 5i), which generally NDVI contributed to almost 1.69% of landslides influences. The results indicate that very low levels of landslides influence cover about 184362.5 ha, which is equivalent to 42.8% of the total study area. Low and moderate influences cover an approximate area equal to 103216.9 ha and 59638.89 ha, which are almost equivalent to 24.0% and 13.9%, while high and very high influences have areas equal to 50629.32 ha and 32592.85 ha, which are equivalent to 11.8% and 7.65%, respectively (Table 6). Vegetated areas accelerate the interception of rainfall, while plant roots play a great role in improving soil stability by removing water from the soil. Areas with very little or void vegetation are more likely to be affected by soil erosion and hence forming gullies resulting in shallow landslides (Löbmann et al. 2020). Furthermore, the results reveal that very high NDVI levels of influences are concentrated in the southernmost part of the study area and some parts of the central. High and moderate influences prevail in the central and some parts of the southern region of the study area. Thus, the likelihood of landslides occurrence based on this factor is higher in these regions as compared to other parts of the study area. Moreover, the low-NDVI levels of landslide influences are observed in the central part protruding from the southeastern to western parts of the study areas thus, indicating that the likelihood of landslides occurrence based on this factor is medium. Finally, very low NDVI susceptibility level is very low in the most northern parts of the area and some parts of the western edge indicating that the probability of landslides occurrence in these areas is very low as compared to other parts of the study area also found true in other studies elsewhere globally (Löbmann et al. 2020).

Landslide susceptibility mapping (LSM) of Lushoto district
From the WLC analysis, the landslide susceptibility areas' map was generated based on the identified landslides influencing factors with their respective weights in the ArcGIS-Pro environment. This technique generates a single index of probability ranging from the lowest rank defined to the highest. In this study, the lowest rank was 1 (very low), and the highest was 5 (very high). These numbers indicate landslide susceptibility levels of the study area, which were later renamed based on their meaning as earlier defined; Very low, low, moderate, high, and very high susceptibility levels instead of 1,2,3,4, and 5, respectively ( Table 7) and (Fig. 6). Further analyses were carried out to determine the susceptibility area within the study area in hectares (ha). By the use of the pixel-based area calculation method, various susceptibility areas were computed, and results revealed that about 97669.65 ha of the land is covered by very low landslides susceptibility levels, which accounts for about 24.03% of the total study area. However, low susceptibility levels were found to cover about 123105.84 ha of land, which is equivalent to 30.28%. Moderate landslides susceptibility areas were found to have 140264.79 ha of land, which accounted for about 34.50% of the total area. High and very high susceptibility areas were found to cover about 45423.43 ha and 57.78 ha, respectively, which are equivalent to 11.17% and 0.01%, respectively (Table 7). However, these indicate that most southern and central mountainous regions of the study areas are more susceptible to landslides. High landslides susceptibility is observed in the central and western and protrudes to the southeastern parts of the study area. Furthermore, the results portray that low susceptibility areas are concentrated in the outer central parts of the low-lying slopes of Usambara Mountain, while very low landslide susceptibility is observed in the northeastern parts of the district with small patches of susceptibility shown within the areas indicating that the probability of landslides occurrence is very low. Thus, the government, stakeholders, and environmentalists should take agent measures, such as demarcating and zoning these high to very high landslide susceptibility areas to limit further human interventions that may result in property loss and the death of living organisms within the study area, to ensure human prosperity.

Validation of LSM by past landslides events
Eleven past landslide spatial sites collected by handheld GPS in the field were overlaid with LSM results, and about eight of them exactly matched with high to very high landslide susceptibility areas of the study area, demonstrating the high accuracy of the obtained results. The remaining geographical sites also correctly matched moderate and low susceptibility levels as shown in Fig. 7 below.

Validation of LSM results by receiver operating characteristic (ROC) curve
ROC is widely used in assessing the efficiency (validity) and performance of the model employed in suitability mapping using true positive values of the known events in a given area (Kayastha et al. 2013;Msabi and Makonyo 2021). The area under the curve (AUC) is determined from ROC in which the cumulative percent of the area under different landslides susceptibility levels and the past landslides events are used. From the analysis, AUC was found to be 0.811, indicating that the model is very good by 81.1% because the curve approaches the genuine positive rate very closely (Desalegn et al. 2022; Dhamija and Joshi 2022) (Fig. 8).

Conclusion
The current research work has demonstrated the ability of geospatial techniques in the assessment and management of landslides. Findings from the analysis conclude that; among others rainfall, slopes' angle, elevation, soil type, lithology, proximity to roads, faults, rivers, and NDVI plays a vital role in the assessment of landslides in a given area with the same characteristics as the current study area. Also, it is concluded that; based on the nature of the study area, rainfall, slope angle, elevation, and soil type landslides factors have higher influences on the occurrence of landslides within the area, while NDVI, proximity to roads and rivers have low influences. Furthermore, mudflows and rockfall types of landslides are more common within the area and are more concentrated in the central, western, and southern parts of the area. Finally, GIS, RS, and statistical techniques such as AHP play a great role in the assessment and management of landslides events and, they are more advantageous in terms of being cost-effective and less time-consuming while giving Fig. 7 Results validation by past landslides more reliable results and, generating datasets to be stored in various databases for easy storage and retrieval for the future use and planning. As a result, it is recommended that the government and other stakeholders intervene in the formulation of policies and measures on human-induced activities in disaster-prone areas, and their delineation, to limit and avoid the risk of economic loss and death.