Land Use Change Monitoring and Modelling using GIS and Remote Sensing Data for Watershed Scale in Thailand Land Use Change Monitoring and Modelling using GIS and Remote Sensing Data for Watershed Scale in Thailand

Landsat 7 Enhanced Thematic Mapper (ETM), Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) images obtained in 1991, 2005 and 2014 with maps, and field survey data were used to classify land use and land cover (LULC) changes over 23 years and predict soil erosion risk locations in the Khlong Kui watershed (73,700 ha), Prachuap Khiri Khan province, Thailand. Classified images together with soil features, slope and rainfall data were used to identify potential risk areas of soil erosion. Based on field check data, the overall classification accuracy was accessed from random samples that resulted as 80% for 1991, 83% for 2005 and 86% for 2014. The study discovered that rice field and rangeland increased by 1.12 and 2.81%, respectively, deciduous forest, and on the other hand, it decreased by 8.28%. GIS analysis identified the potential risk areas of soil erosion as 46,431 ha (0.63%) at very high risk.


Introduction
Since the 1990s, global, regional and local studies of land use and land cover changes (LULCC) have greatly developed, thanks to advances in earth observation and monitor methods including remote sensing and GIS techniques. The matter of land use changes has been measured in many international and interdisciplinary researches such as remote sensing, environment and biogeography [1,2].
In Southeast Asia, including Thailand, deforestation has been occurring during the last 15 years because of an increase in agricultural crops [3]. Land use land cover change in Prachuap Khiri Khan province was reported by the Office of Agriculture Economics (OAE) in 2014 that deforestation has been occurring 6.96% while agriculture and other land use increase 34.97 and 45.44% respectively [4].
Recently, remote sensing is widely applied for monitoring changes and dynamics in land use and land cover (LULC) observation and its impact to the environment. It offers a variety of benefits in LULC study and an opportunity to assess remote area such as tropical forest, high mountains, update land and terrain information and explore historical LULC. To offer more efficiency in identifying land cover changes, remote sensing is often combined with Geographic Information System (GIS) technique. GIS technology refers to for analyzing and managing spatial and temporal data associated with their features [5]. Both technologies provide capability to collect land use characteristics and changes by integrating existing remotely sensed data and relevant environments such as tropical forests, urban areas and coastal zone and different land transformations such as deforestation, urban development and desertification [2,[6][7][8]. This study shows environmental problems such as deforestation and soil erosion in Thailand caused by human activities. The results of this study could support local governments, local residents and farmers to focus on environmental problems in their regions. The erosion risk map can be used as the potential disaster information to establish field experiments plots for warning the risk area of soil erosion.

Study area
The Khlong Kui watershed is a large watershed in the Southwestern Thailand and is located between 11° 58′16"N and 12°15′50"N and between 99°31′56″E and 99°58′30″E as mapped in Figure 1. The entire area of the watershed covers approximately 73,700 ha (460,625 rai) in Kui Buri district, Prachuap Khiri Khan province, Thailand. Khlong Kui watershed, with the main river of the watershed, named the Kui Buri River, is surrounded by three main watersheds as (1) Pran Buri; (2) Khlong Khao Daeng and (3) Khlong Saphan Yai of the Prachuap Khiri Khan coast basin, the major river basin in Thailand.
Topography: Khlong Kui watershed includes high mountain range (max. 958 m) on the West, hilly and rolling land, plain and floodplain to the cost on the East as presented in Figure 1. High mountain ranges, the major landscape of the Khlong Kui watershed, are mostly in the upper watershed and are mostly covered by forest. Forests in these areas are strictly conserved as water sources. Plains, which cover a small part of the watershed, are used for crop, orchard, and vegetable cultivations. Floodplains, the second large landscape, surround the main rivers and are mostly located in the lower watershed. These areas are generally used for rice cultivations. Deforestation and soil erosion are the major environmental problems in the watershed. These problems are more prominent in the mountain ranges and hilly and rolling lands.

Climate:
The study area has a tropical savanna climate with drying season from January to May and raining season from June to December. The annual rainfall 30-year average is 1153 mm as the highest in November and the annual average temperature is 31.4°C as the highest in April [9]. Due to the highest rainfall in November, soil erosion and land slide might be occurred in the area where nonvegetation and bared land with high slope are the types of land use in Khlong Khui watershed.

Data use
In this study, Landsat 7 Enhanced Thematic Mapper (ETM) images in 1991, 2005 and Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) image in 2014 were used for land use and land cover (LULC) classification of the Khlong Kui watershed. A digital form of the watershed boundary was utilized. Field survey, topographic maps and LULC thematic maps were used for classification accuracy assessment. Digital Elevation Model (DEM), soil series digital maps and rainfall data were used as ancillary data to identify potential risk areas of soil erosion.   Rainfall data were composed from The Tropical Rainfall Measuring Mission (TRMM) is a joint mission between National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA) designed to monitor and study tropical rainfall. The rainfall measuring instruments on the TRMM satellite include the Precipitation Radar (PR), an electronic scanning radar operating at 13.8 GHz; TRMM Microwave Image (TMI), a nine-channel passive microwave radiometer; and Visible and Infrared Scanner (VIRS), a five-channel visible/ infrared radiometer. The purpose updated algorithm is to produce the best-estimate precipitation rate (in mm/h) and root-mean-square (RMS) precipitation-error estimates from TRMM and other data sources [10]. Vertical hydrometeor profiles and surface rainfall means are computed monthly with the grid size as 0.5°×0.5°. The amounts of annual rainfall in the study area and its six categories with higher amount of rainfall were ranked with the higher scores ( Table 1). Soil series maps of six provinces that were collected during field work between 1999 and 2002 were created by Land Development Department, Thailand in the shape file format. They came with soil series' soil materials properties in the Excel format. The attributes of soil series' soil materials properties were in the Excel format which was standardized with type of lithology prepared by FAO (in 2006). The soil materials were graded into six classes based on their resistant to water as provided in Table 1.
Slope is shown as the percentage of slope gradient that was calculated from Triangulated Irregular Networks (TIN) come from Digital Elevation Model (DEM) by using the Spatial Analyst Surface. The slope gradient structures were classified into six classes in accordance with the slope gradient classes [11] and the slope classes for water erosion [12]. The classes were classified from 1 to 6 as presented in Table 1.
Land use/land cover in 2014 classified from Landsat 8 OLI-TIRS images that were reclassified into six LULC types based on the crop management factor values provided by Land Development Department, Thailand (in 2000). The ranking scores of LULC are described in Table 1.

Methods
This study was accomplished using three major procedures: image classification and analysis, modeling LULC changes in 23 years (during 1991-2014) and identification of potential risk areas of soil erosion in the Khlong Kui watershed described as shown in Figure 2.

Satellite image geometric correction
The geometric correction process geometrically converts the image coordinates from (x, y) into Universal Transverse Mercator (UTM) Zone 47P map projection coordinate by using eight ground control points (GCPs). For the purpose of and use change and soil erosion analysis, all the satellite image and maps must be registered in the same pixel size and map projection with precise overlaying together. The second-order polynomial transformation and cubic convolution are used for image registration. In this study, the GCPs have been collected during 10-20 April, 2014 by using GPS-GLONASS L1 receiver brand ASTECH model Promark 100.

Satellite image enhancement
With respect original multispectral data set, the color distortion of pan-sharpening technique is significant limitation as shown in Figure 3. The statistics analysis was used to evaluate the digital value and characteristic of original data before pan-sharpening transform with enhanced data after pan-sharpening transform.

Image classification and analysis
The Landsat satellite images described in the previous section were used to investigate LULC in the Khlong Kui watershed, Thailand in 1991, 2005 and 2014. The images were analyzed with the image processing software GEOMATICA Ver. 2013, a widely used image processing software package, which is often used to perform LULC classification of remotely sensed data.

Image classification for land use and land cover map
Digital image classification is the process of recognizing pixels, which are given in multispectral bands of a satellite image. The process generates clusters of pixels with similar digital values into the same informational categories [8]. The classification performed by automated (unsupervised) or semiautomated (supervised) approaches are widely used in many LULC studies [1,6,[13][14][15][16].
In this study, supervised method was used to classify LULC in the Khlong Kui watershed. Supervised classification employs samples of pixels that are already known informational categories to classify unknown pixels on an image. The class names were assigned into 12   online Google map for navigating to the sample location by using 3G internet connection and GPS-GLONASS L1 receiver brand ASTECH model Promark 100. These samples were then applied for image classification accuracy assessment by generating classification confusion matrices and accuracy report.

Accuracy assessment
Accuracy assessment is an essential requirement of image classification, and it can be resulted by the confusion matrix. Confusion matrices quantitatively compare the relationship between the classified images and the reference data which contains field survey, high resolution digital map and/or thematic maps. After the confusion matrix is generated, overall accuracy, producer's and user's accuracies, omission and commission errors, and Kappa statistics [1,[15][16][17][18] can be written as shown in Eq. 1.
where N is the total number of sites in the matrix, r is the number of rows in the matrix, x ii is the number in row i and column i, and x i+ is the total for column i, x +i is the total for row i, [1].

Identification of the potential risk areas of soil erosion
According to the literature, soil erosion of a land surface is caused by various factors. These factors include topography (e.g., slope orientation, steep and length), soil cover (e.g., trees, grasses, water, bare soil and paved surface), soil character (e.g., soil mass, soil components and soil materials), and climate (e.g., rainfall amount and intensity, temperature and wind) [7,12,19]. In this study, we chose four different factors based on data availability to identify the potential risk areas of soil erosion in the Khlong Kui watershed. These factors are (1) slope, (2) LULC, (3) soil parent material and (4) rainfall. To construct the model, we executed two processes: (1) variable ranking and layer creation and (2) model development.

Variable ranking and layer creation
The factors were categorized into six thresholds based on a review of the literature. The threshold categories were ranked from 1 as lower risk of soil erosion to 6 as higher risk of soil erosion as shown in Table 1.

Model development
The model was constructed using Multi-Criteria Modeling (MCM). MCM is an influential efficient technique for managing different types of ecological modeling for decision-making and environmental planning [20,21]. Weights for the erosion factors were derived from pairwise comparison by ranking the importance of each factor and comparing them with another as shown in Table 2.
Slope was ranked (as 7) as the most important factor because steep slope areas usually have high potential for soil erosion. However, different types of vegetation cover can prevent erosion; hence, LULC was ranked (as 5) as the second most important factor. Soil material and rainfall were ranked (as 3 and 1) as the third and the fourth important factors. The Fuzzy Logic method (IDRISI Software Ver.17) calculates the weights which were obtained by the relative importance matrix as 0.3496 for slope; 0.2496 for LULC; 0.1496 for soil material and 0.0496 for rainfall as displayed in Table 2.
The weights were then used to create two equations using the attribute calculator tool from software QGIS Ver. 2.6 as given in Eq.
The final risk scores of each model were standardized in percentage of potential risk by Eq. 3: where X is the final risk score, Min is the least score and Max is the highest score [12]. Finally, the percentages of potential risk of erosion were divided into five classes: very low (<20), low (20-40), moderate (40-60), high (60-80) and very high (>80) potential risk of soil erosion.

LULC classification
The water and reservoir, deciduous forest, evergreen forest, rice field, urban and village land categories presented good classification performance during the study period. Based on the ground truth and field check data of 100 samples for LULC types, the classification assessment with the confusion matrices were generated for evaluating the overall, producer and user accuracy of each LULC types. As we can see in the confusion matrix as shown in Table 3 water and reservoir archived 100% for both producer and user accuracy because the signature of water is sufficient difference from vegetation and other land cover types. Deciduous forest and evergreen forest are also classified with high accuracy as 96 and 95% for producer accuracy and 100 and 95% for user accuracy, respectively. The cropland (91.67% producer accuracy, 73.33% user accuracy) and orchard (71.43% producer accuracy and 83.333 user accuracy) categories had moderate classification performance. Rangeland (62% producer accuracy) had lower accuracy as it was mixed with barren land and urban village type. The wetland category had poor classification performance except in the 2005 classified image where it had high accuracy performance. The uncertainty of classification among forests, agricultural lands and wetlands occurred due to similar spectral reflectance of green vegetation. This confusion usually occurs when using moderate spatial resolution images such as Landsat satellite images to classify areas that have heterogeneous LULC [22].   Table 3. LULC classification 2014 Feb 02: confusion matrix.
Land Use -Assessing the Past, Envisioning the Future  in Table 3. These overall accuracies are decreased for old dated data (1991 and 2005) due to many changes in forest and agricultural land use in comparison with those identified during ground truth period. The LULC types are presented in Figure 4, and the statistics of area is calculated in Table 4.

LULC changes and analysis (1991-2014)
The gains and losses are shown in Figure 4, and the cross tabulation of the changes between 1991 and 2014 ( Table 5) is reliable with the previous two periods (1991-2005 and 2005-2014). Although there were gains in evergreen forests (6.32%) from croplands, orchards, barren land and deciduous forests, the great loss of deciduous forests (−8.28%) occurred due to conversion to evergreen forests, rangeland, barren land and croplands. Moreover, Table 5 shows that the increase of evergreen forests, rice fields and croplands (3.01%) was mainly from deciduous forests. Most of the orchard losses (−9.40%) were converted to evergreen forests, barren land and croplands.
The major loss of coastal forest (−0.01%) was due to conversion to deciduous forests. Although the minor changes among LULC types could have followed by a result of agricultural activities such as shifting cultivation, crop rotation and infrastructure development and some of these changes could be added to the error of classification caused by similar spectral reflectance or mixed pixels from the various characteristic of LULC in the region.

Identification of the potential risk areas of soil erosion
Fuzzy Logic presented a major weighting factor in development of the model because mountains are a major landscape of the watershed. The results from Fuzzy Logic seem to be more  conventional based on the topography of the watershed with very high risk (0.63%), high risk (32.00%), moderate risk (32.40%), low risk (31.21%) and very low risk (3.76%) as shown in Figure 5. The clusters of very high risk were consistent with the northern, central, eastern regions of the watershed as also presented in Figure 5. They were mainly located in mountainsides or hillsides, which are usually steep slope and boundaries between forests and highland crops.
In general, most areas of the Khlong Kui watershed had high potential risk of soil erosion due to the combination of mountainous topography and agricultural activities. High rainfall in high mountain area generated more areas of higher risk while low rainfall in low and flat area generated areas of lower risk.

Conclusions
This study used remote sensing and GIS techniques to assess land use and land cover (LULC) and its dynamics of change with identify the potential risk areas of soil erosion in the Khlong Kui watershed in 1991, 2005 and 2014. The Khlong Kui watershed was selected as the study area because this watershed has been experiencing deforestation and soil degradation due to the development of agricultural lands and urban areas. Moreover, the topography of the watershed, which includes mountains, hills and slopping lands, makes the Khlong Kui watershed an interesting region to examine potential risk areas of soil erosion. The key findings of the research are as follows:

Image classification and analysis
The major LULC of the Khlong Kui watershed are forests and agricultural lands. The study monitored an increase in orchards, croplands, evergreen forests, rice field and urban areas while a decrease in deciduous forests and wetlands in the watershed in 1991, 2005 and 2014. The overall accuracy assessment of the image classification was satisfactory in all three different years of satellite data acquisition.

LULC changes and dynamics
Deciduous forest, evergreen forest and orchards types were major drivers of land use and land cover changes. An increase of range land, croplands and evergreen forests were mainly derived from deciduous forests. The development of range land, barren land and crop land was related to an increase in infrastructure of the Khlong Kui watershed. There is a high probability of change from deciduous forests, wetlands and orchards to rice fields and croplands in 2014.

Potential risk areas of soil erosion
High-risk areas of soil erosion were primarily located in the northern and eastern regions of the watershed which are also with mountain ranges and hilly areas. High rainfall in high mountain area generated more areas of very high risk at 0.63% of the watershed. The change from forests to agricultural lands in the northwestern and northeastern regions of the watershed led to higher risk areas of soil erosion in the last 9 years.

Recommendation for further research
Due to limitation of research, financial budget and time, land use change and soil erosion model have lacked sample questionnaire for validation process. It is recommended for further research works that develop an additional surveying method to improve the soil erosion model to archive more accurate and creditable result.