Soil moisture index estimation from Landsat 8 images for prediction and monitoring landslide occurrences in Ulu Kelang, Selangor, Malaysia

Received Nov 9, 2019 Revised Sep 11, 2020 Accepted Oct 13, 2020 Soil moisture is one of the contributing factors that accelerates soil erosion and landslide events due to the increase in pore pressure which eventually reduces the soil strength. For landslide prediction and monitoring purposes, large-scale measurement involves estimating the soil moisture. However, estimation of soil moisture usually involves point-based measurements at a particular site and time, which is difficult to capture the spatial and temporal soil moisture dynamics. This paper presents the estimation of the SMI using Landsat 8 images for prediction and monitoring of landslide events in Ulu Kelang, Selangor. The selected SMI map for dry, moist, and wet seasons are obtained from climatology rainfall analysis over 20-year periods (19982017). SMI is assessed based on remote sensing data which are land surface temperature (LST) and normalized difference vegetation index (NDVI) using GIS software. Overall results indicated that rainfall distribution is high during inter-monsoon (IM), followed by northeast monsoon (NEM) and southwest monsoon (SWM) season. High rainfall distribution is a direct contributor towards SMI condition. Results from simulation show that April 2017 is known to have the highest SMI estimation season and selected to be the best SMI mapping parameter to be applied for prediction and monitoring of landslide events.


INTRODUCTION
Rainfall has been known as the main factor for most of the landslide events in the region of high seasonal rainfalls [1][2][3][4]. Any given rainfall event will cause an increase of pore water pressure within the soil on the area. Whenever a slope filled with water, the fluid pressure will provide a block of flexibility which reduces the resistance of movement and eventually causing the slope to fail. This condition can get even worse due to the soil conditions of the affected area [5][6][7]. It is important to identify the soil moisture of the area before considering rainfall as the factor that causing the landslide event, especially on deep-seated landslides and terrains with complex hydrological [8,9]. Commonly, soil moisture measurement can be estimated by using three methods: in situ measurements, hydrological model, and remote sensing. In situ measurements were the best method among the three which can provide the highest accuracy, but it only offers point-based measurement and limited due to high dependencies of installation and maintenance cost. The second method is to obtain variations of continuous soil moisture is through the land surface or hydrological model. Nevertheless, model-based approaches tend to deal with the issue of time drifts, large numbers of reliable data inputs and computationally intensive for large monitoring areas. Therefore, remote sensing and GIS techniques proved to be reliable alternative to soil moisture estimation on a global scale [10][11][12][13][14].
Remote sensing techniques offer a continuous estimation of soil moisture for a large area. For this case-study, soil moisture estimations are referring to the near-surface soil moisture (NSSM), which characterizes the first 5 cm or less of the topsoil profile. In recent years, remote sensing techniques have been advanced and varied their estimation to make more effective tool for monitoring soil moisture index (SMI) and other related variables such as land surface temperature (LST) and the normalized difference vegetation index (NDVI) [15][16][17]. LST calculation is defined from thermal emission, while NDVI is estimated based on portions of the electromagnetic spectrum, namely red and near-infrared (NIR) surface reflectance. These methods are sometimes known as optical and thermal infrared of remote sensing. Previous studies were conducted based on passive or active microwave data to estimate the soil water substance in the surface soil layer within 0 to 10 cm [18,19].
Other usage of these optical and/or thermal data is to indirectly identify soil moisture condition by referring the changes of biophysical factors, such as surface energy balance and vegetation cover, which were affected by the availability of soil water. Results from various studies show a potential in monitoring both root zone and surface of soil moisture by using thermal and/or optical derived from vegetation. Over the years, various vegetation indices have been used to estimate soil moisture and the response of vegetation to the spatial and temporal variations [20][21][22]. The aim of this study is to estimate the soil moisture index from Landsat 8 images based on dry, moist, and wet seasons. The analysis of SMI condition season is conducted to provide the best SMI mapping parameter in predicting and monitoring of landslide occurrences in Ulu Kelang, Selangor.

RESEARCH METHOD 2.1 Study area and data set
This study was carried out at Ulu Kelang, Selangor which is located at the latitude of 3°12'30''N and longitude of 101° 45' 28'' E with a 5 km distance from Kuala Lumpur city center as shown in Figure 1. Ulu Kelang is a residential area which is known as one of Malaysia's most landslide-prone areas. The average of annual rainfall in Ulu Kelang area is about 2,440 mm. The rainfall distribution mainly based on two monsoon seasons, known as the southwest monsoon (SWM), beginning from May to August and the northeast monsoon (NEM) which is from November to February. In Ulu Kelang, the soil moisture is mostly influenced by rainfall distribution [23][24][25][26]. The ground-based measurement rainfall data is used to analyze the climatology of rainfall in Ulu Kelang, Selangor. The rain-gauge data was provided by jabatan pengairan dan saliran (JPS), Ampang. Two rain-gauge stations (JPS Ampang and Genting Kelang) were acquired to represent the rainfall distribution for 20-year periods from 1998 to 2017. The soil moisture index was analyzed from Landsat 8 images were downloaded using USGS Earth Explorer website. Landsat 8 images were selected based on the climatology of rainfall and the availability of cloud free satellite images. The SMI map was produced by using ArcGIS 10.2.2. The selected images are listed in Table 1.  Figure 2 shows the methodology of the SMI analysis based on Landsat 8 images which were selected from climatology of rainfall analysis for dry (low moisture), moist (medium moisture) and wet (high moisture) seasons. The Landsat 8 images were selected in February, April and June of the year 2017. SMI is calculated based on the combination of the NDVI and LST calculation using (1) [27,28].

Methodology
Where, for a given NDVI, LSTmax and LSTmin are the maximum and minimum of surface temperature and the land surface temperature is LST, the surface temperature of a pixel for a given NDVI derived from remote sensing. The calculation of LST is based on the (2).
NDVI is define as the ratio of reflectivity differences between NIR and the Red band to their sum. NDVI is calculated using (6): Finally, the SMI analysis was obtained using the raster calculator in ArcGIS 10.2.2. The SMI maps provide a value between 0 and 1, which is represents the relative amount of soil moisture within the area, 0 indicates the lowest soil moisture and 1 indicates the highest soil moisture on a specific day.

RESULTS AND ANALYSIS 3.1. Climatology of rainfall analysis
Rainfall climatology during the southwest monsoon (SWM) occurs between May-August whereas inter-monsoon (IM) happens in March-April and September-October, while northeast monsoon (NEM) is in November-February. The rainfall data were taken in Ulu Kelang, Selangor using ground-based measurement as in Figure 3. The monsoonal rainfall fluctuation in Figure 3 is based on average monthly rainfall and rainy days over 20 years (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017). During the SWM season, the highest rainfall was recorded is in the month of May which reached up to 531.2 mm with 16 rainy days. The following month was in declining trends as the Kelang area received higher rainfall as compared to WSM season. The highest rainfall distribution was recorded in November, which is at 703.1 mm with 22 rainy days. However, the rainfall pattern for the NEM season is similar to the SWM season whereby the rainfall distribution decreases in the following months: December, January and February. In general, rainfall received during NEM and IM seasons was the highest contributor to its annual rainfall distribution which the rainy months are in November and April. While the dry months are in January and February which happened at the end of NEM season.

Soil moisture index (SMI) analysis
Based on the climatology of rainfall analysis, the months of February, April and June were selected to represent the SMI map as the parameter for landslide events in Ulu Kelang, Selangor. The estimation of normalized difference vegetation index (NDVI) and land surface temperature (LST) are based on essential data for obtaining SMI calculation. The NDVI values range from -1 to 1 where the negative value of vegetation indicates a poor vegetative cover, while the positive value indicates a dense and good vegetative cover. Figures 4(a) The minimum SMI for all selected months is between 0.000001 to 0.000006 and the maximum SMI is between 0.999997 to 1.00000. This is in line with SMI indicator which in the range of 0 to 1. The SMI map is classified into three classes, SMI less than 0.3 (<0.3, dry area), SMI between 0.3 and 0.5 (0.3-0.5, moist area) and SMI more than 0.5 (> 0.5, wet area). The brown spot areas known as dry areas, which indicate the SMI value close to zero was highly affected by water deficit. While the yellow spot areas are identified as moist areas. The SMI value near to 1 is represented by green spot areas which known as the wet area or forest cover and have the highest moisture as compared to the rest of land cover. Figure 7 shows the total area of SMI classes in February, April and June for the year of 2017. From the analysis, there is a large dry area which is approximately 41% (6.5349 km 2 ) in February compared to moist (28%) and wet area (31%). While in April, 17% (2.709 km 2 ) was identified as dry area, 32% ) as the moist area and 51% (8.1882 km 2 ) as the wet area. In June, all SMI classes were within the same percentage where dry area covers 31% (5.1921 km 2 ), moist area at 33% (4.9419 km 2 ) and wet area at 36% (5.9013 km 2 ). Based on the analysis conducted on the selected months, April was found to be the most wet/rainy month compared to February and June. This is due to 83% of the area in April was covered by moist and wet areas, while in February it was only at 51% and June at 69%.
Based on these analyses, April is found to be the most wet/rainy season which aligned to the IM season periods of March-April and September-October and also the beginning of NEM in November and December. While June is represented as the moist season for SWM. The month of February is identified as the dry season which indicates the end of NEM season (January and February). However, only one SMI map will be selected to represent the SMI parameter for landslide events. The best selected parameter is based on the SMI map in April as most of the landslide events occur in high SMI value during the rainy season.

CONCLUSION
Soil moisture was the key parameter to monitor and predict the rainfall-landslide occurrences, especially in hilly areas. The main objective is to generate SMI map estimation, derived from Landsat 8 images by considering the monsoonal season which is influenced by the rainfall distribution. Based on the analysis, results indicated that rainfall distribution is high during IM season followed by NEM season. The month of February, April and June 2017 were selected to present the soil moisture condition for dry, moist, and wet seasons. The highest SMI estimation seasons (wet season) was selected as the SMI parameter in the prediction of landslide occurrences analysis for Ulu Kelang, Selangor.