Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale

FORECASTING LAND USE CHANGE AND THE IMPACT ON WATER YIELD AT WATERSHED SCALE

YUSUF RAHADIAN

GRADUATE SCHOOL

BOGOR AGRICULTURAL UNIVERSITY B O G O R

STATEMENT OF THESIS AND INFORMATION SOURCES

Hereby I declare the thesis entitled Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale, is result of my own work under supervision of supervisory committee and it has not been submitted yet in any form in any university to obtain a degree. The researcher has full responsibility for all contents of this thesis. Sources of information derived or quoted from other researchers, whether it is published or not are mentioned in the text and listed in the bibliography at the end of this thesis.

Bogor, January 2014

Yusuf Rahadian G 051 090 061

ABSTRACT

YUSUF RAHADIAN. Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale. Supervised by SURIA DARMA TARIGAN as supervisor and IBNU SOFIAN as co-supervisor.

The focus in this study is on the impact of land-use changes from 1991 until 2030 on the water yield quantity in Upper Cisadane Watershed. This study tests a methodology, which involves coupling a land-use change model with a hydrology model. The future land-use is modelled with the CLUE-S model. Four scenarios are developed based on land use demand (population growth) and area restriction (spatial policy). Hydrology components, in case of water yield is simulated using the HEC-HMS model. Result from land use model show that trends of land use changes of the area during 1991 – 2030 depict grassland, settlement and forest area are increasing over time, whereas estate tend to decrease and water remains constant. Based on hydrology model simulation shows increasing of water yield from Year 2010 to Year 2030. The increasing values of water yield influenced by forest rehabilitation activity by the government inside forest area and the development of community forest outside the forest area. Increasing values of water yield are higher for scenario 2 and scenario 4, where government policy about restriction of land use inside forest area applied. That means government policy to prohibit land use conversion inside forest area is appropriate to apply.

ABSTRAK

YUSUF RAHADIAN. Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale. Di bawah bimbingan SURIA DARMA TARIGAN sebagai pembimbing I dan IBNU SOFIAN sebagai pembimbing II.

Fokus dari penelitian ini adalah mengenai dampak perubahan penggunaan lahan pada periode tahun 1991 – 2030 terhadap produksi air di Daerah Aliran Sungai Hulu Cisadane. Penelitian ini menguji sebuah metodologi yang merupakan kombinasi antara model perubahan penggunaan lahan dengan model hidrologi. Penggunaan lahan di masa depan dimodelkan dengan CLUE-S. Empat skenario dikembangkan berdasarkan permintaan penggunaan lahan (pertumbuhan populasi) dan pembatasan wilayah (kebijakan spasial). Komponen-komponen hidrologi, dalam hal ini produksi air, disimulasikan dengan menggunakan model HEC-HMS. Hasil dari pemodelan penggunaan lahan dalam jangka waktu 1991-2030 menunjukkan tren perubahan penggunaan lahan padang rumput, area terbangun, dan hutan bertambah, sementara area perkebunan menurun, dan badan air tetap. Hasil simulasi model hidrologi menunjukkan peningkatan produksi air pada periode 2010-2030. Peningkatan produksi air dipengaruhi oleh aktifitas rehabilitasi hutan oleh pemerintah di dalam kawasan hutan dan pembangunan hutan rakyat di luar kawasan hutan. Peningkatan produksi air lebih tinggi pada skenario 2 dan skenario 4, di mana kebijakan pemerintah mengenai larangan konversi penggunaan lahan di dalam kawasan hutan diimplementasikan. Artinya, penerapan kebijakan pemerintah untuk melarang konversi penggunaan lahan di dalam kawasan hutan sudah tepat untuk diimplementasikan.

Kata kunci: Perubahan Penggunaan Lahan, Model Penggunaan Lahan, Model Hidrologi, Produksi Air.

SUMMARY

YUSUF RAHADIAN. Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale. Supervised by SURIA DARMA TARIGAN as supervisor and IBNU SOFIAN as co-supervisor.

One of the most important issues in watershed management is the use of land, especially in the upstream catchment area. People effort to fulfill their ends meet can cause land conversion. As a result of the land conversion, the Cisadane River condition seems quite apprehensive in the last two decades. Environmental degradation in Cisadane watershed in both the upstream and downstream has a heavy impact to the availability of water resources in Cisadane River. In relation to this fact, an in-depth study on land use change in Upstream Cisadane Watersheds and its dynamics need to be conducted. Particularly, the impact of land use change on water yield. This research purpose is to forecast the future of land use changes (using CLUE-S) and assess the impact of the predicted land use changes on water yield in Upstream Cisadane Watershed (using HEC-HMS).

This research was conducted in Upstream Cisadane Watershed – Bogor. The study area is Upstream Cisadane Watershed with outlet of watershed in Empang, which located in both Bogor Regency and Bogor Municipality. Geographically, it is located between 6036’26.05” – 6047’08.49” South Latitude and 106044’30.27” – 106056’36.76” east longitude. Four scenarios are developed based on land use demand (population growth) and area restriction (spatial policy). The scenarios presented are not necessary the most realistic, but are made in such a way that they provide information on the functioning of the model.

Binomial logistic regression is used to examine the relation between land use and possible driving factor. The result of this analysis are coefficient values that shows the contribution of each driving factor to land use change. ROC (Relative Operating Characteristics) is a method to measure the goodness of the statistical model. The probability of each land use resulted from logistic regression is compared to the real land use map to calculate the equal category of each grid cell between those maps. This method will depict the capability of regression equation to represent land use characteristics. The goodness of the statistical measurement revealed that ROC values for urban water area, grassland area, estate area, settlement area and forest area were 0.903, 0.701, 0.780, 0.813 and 0.994, which indicated that the probability of land uses built from these models were capable to represent land use changes and empirical analysis by using logistic regression method was satisfactory to examine the relationship between driving factors and land use change in study area.

Through model validation process shows that overall accuracy and Kappa accuracy for land use simulation are 90.83 % and 86.00 % or categorized as fit. It indicates that the driving factors have a good capability to explain land use pattern in study area and it can be used to predict future land use pattern.

According to the change detection for the land use classification year 1991, simulated year 2000, year 2010, year 2020 and year 2030, it can be seen that settlement, forest and grassland area are increasing over time, whereas estate tend to decrease and water remains constant. Based on the findings, settlement area were significantly increase during the period 1991-2009 and causing the pressures in the study area.

Developing a model basin is an important step to assess the impact of land use changes on water yield in Upstream Cisadane Watershed using HEC-HMS. A configuration needs to be developed in the model basin to describe physical representation of a watershed based on its hydrology elements. In this research, there are seven hydrology elements available in HEC-HMS; Sub-basin, Reach, Reservoir, Junction, Diversion, Source, and Sink. The research employs 19 sub-basins, 9 reach, 9 junctions, and 1 outlet. Developing a model basin also includes calculation in 4 main sub-models, i.e.: loss model, transform model, base flow model, and routing model.

The model calibration done by adjusting the initial abstraction, curve number and impervious area values, until the results matched the field data. The process was completed manually by repeatedly adjusting the parameters, computing, and inspecting the goodness of fit between the computed and observed hydrographs. During calibration process by using half-year daily data (February 1, 2010 – July 31, 2010), the accuracy achieved 0.524 of R2. Performance of the model was objectively evaluated by using Nash-Sutcliffe Efficiency (NSE) and Relative Volume Error (RVE), in which it gave good efficiency value (NSE) of 0.67 and 42.9 % of RVE. By using three tests it can be stated that the model is satisfactory accepted.

Four hydrographs were simulated by using the same parameters in which was used during calibration and validation model. In this research, again the rainfall data of 2010 (February 1, 2010 – July 31, 2010 Period) was used as meteorological input. All these four scenarios were run using same parameters, except the curve numbers, percent impervious and initial abstraction are based on those previous land use condition. The simulated hydrograph obtain the peak flow and water yield information for four different land use scenarios. The values of each scenario peak flow are 81.00 m3/s, 81.10 m3/s, 81.00 and 81.10 m3/s for scenario 1, scenario 2, scenario 3 and scenario 4 respectively. While the value of water yield are 276,085.00 m3, 278,038.90 m3, 275,143.20 m3 and 279,178.20 m3 (all values for water yield are multiplied by 1000), for scenario 1, scenario 2, scenario 3 and scenario 4 respectively.

Based on the comparison about values of water yield between scenario-based simulation for Year 2030 and existing condition of Year 2010 data shows increasing of water yield from Year 2010 to Year 2030. The increasing values of water yield influenced by forest rehabilitation activity by the government inside forest area and the development of community forest outside the forest area. Increasing values of water yield is quite high for scenario 2 and scenario 4, where government policy about restriction of land use inside forest area applied. That means government policy to prohibit land use conversion inside forest is appropriate to apply.

Copyright © 2014, Bogor Agriculture University Copyright are protected by law,

1. It is prohibited to cite in part or the whole contents of this thesis without referring to and mentioning the sources:

a. Citation only permitted for education purpose, research, scientific writing, report writing, critical writing or reviewing scientific problem.

b. Citation does not inflict the name and honor of Bogor Agricultural University.

2. It is prohibited to republish and reproduced in part or the whole of this thesis without written permission from Bogor Agricultural University.

FORECASTING LAND USE CHANGE AND THE IMPACT ON WATER YIELD AT WATERSHED SCALE

YUSUF RAHADIAN

Thesis is a prerequisite to obtain degree Master of Science in Information Technology for Natural Resources Management Program Study

GRADUATE SCHOOL

BOGOR AGRICULTURAL UNIVERSITY B O G O R

Research Title : Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale

Student Name : Yusuf Rahadian Student ID : G 051 090 061

Study Program : Master of Science in Information Technology for Natural Resources Management

Approved by, Advisory Board

Dr. Ir. Suria Darma Tarigan, M.Sc. Supervisor

Dr. Ibnu Sofian, M.Eng. Co-Supervisor

Endorsed by,

Program Coordinator

Dr. Ir. Hartrisari Hardjomidjojo, DEA.

Dean of Graduate School

Dr. Ir. Dahrul Syah, MSc.Agr.

Date of Examination: December 12rd, 2013

ACKNOWLEDGMENT

Alhamdulillahirabbil’alamin, first of all, I would like to express my gratitude to Allah SWT., which by Allah SWT. helps, I can finally complete this study. Also, I am indebted to many individuals during the course of study, I would like to thank them through this opportunity:

 To my extraordinary supervisor: Dr. Ir. Suria Darma Tarigan, M.Sc. for his patience during the length of my research and my thesis writing. He has provided me with guidance and motivation to develop and complete this thesis. Dr. Ibnu Sofian, M.Eng., who has been so helpful and has provided me with important guides and give me inspiration for writing my thesis into its form now. I couldn’t have wished for better supervisors than those two.  To MIT coordinator, lecturers, and MIT secretariat that has provided me with

opportunity to complete my study in Information Technology for Natural Resources Management Study Program in a given time frame.

 To my friends: Iwan Ridwansyah, M.Sc., Budi Nugroho, M.Sc., Triyanto, M.Si., They are my partners that have provided me with invaluable support through my study. Also thanks to Aswin Rahadian that has been there to share his knowledge with me in mapping.

 To my parents: Ani Rochaeni, Didi Achmadi, Tuti Trihatmi, and Edris; to my brothers and sisters; Andina, Uci, Ismi, Andi and Taufiq, thanks for the endless support, patience and prayers that have helped me through this study.  Finally, to my wife, Dian Ekowati thank you so much for the love and support. You taught me how to believe that I am capable. To my two amazing children, Fathimah Azzahra and Umar Gandhi, I couldn't be more grateful to get to be your father. I've learned so much from each of you.

I wish that this thesis will give positive contribution to all peoples who read it. Thank you.

CURRICULUM VITAE

Yusuf Rahadian was born in Bandung, West Java, Indonesia on April 28th, 1981, studied Information Technology for Natural Resources Management with specialization in Geographic Information System and Modelling. He attended Forestry Management Study Program, Faculty of Forestry, Bogor Agricultural University during 2000-2005 for his bachelor degree. He has been working with several institutions e.g. Environmental Service Program of USAID Program, AidEnvironment Asia, and also have been working as independent consultant for several projects in Forestry Management related issue, policy and the use of Geographical Information System. Yusuf is currently working as an independent consultant in National Development Planning Agency (Bappenas).

CHAPTER I INTRODUCTION

1.1. Research Background

Increased use of natural resources as an impact of population growth and economic development, conflict of interest, and the absence of coordination between sectors/between areas (upstream, midstream, downstream) especially after decentralization era have worsened the damage in watershed area. It is presented by the higher occurrence of natural disaster e.g.: flood, landslide, and drought. Thus, since the last 30 years, the number of prioritized watershed in Indonesia is increasing. One of them is Cisadane Watershed that based on Forestry Minister Decree No. SK.328/Menhut-II/2009 is designated as one among 108 priority watershed in Indonesia.

Many forms of nature changes in Cisadane Watershed indicate the occurrence of natural resource degradation in its area. The width of Cisadane forest area is only 18.34%, far below the ideal number: 30% of total watershed area. This is worsened with illegal logging in forest area and forest land use change outside the area, i.e. in upstream area. To understand and describe the cause of natural resource degradation in this area, several comprehensive approaches are needed; from biophysical, social economic and community culture factors. The research on relation of land use change on hydrological condition is one way to aim for the right watershed rehabilitation direction, so that the efforts conducted are better planned and predictable.

Cisadane watershed forest area is only 18.34% (28,098.79 ha) of its total area. The land use commonly found in Cisadane Watershed is farming, both dry farming and paddy field. Total area of dry farming and mixed dry farming in Cisadane Watershed is 47,368.55 ha (30.92%), paddy field area is 29,499.34 (19.25%), settlement area is 34,194.25 ha (22.32%), and the rest 4,485.68 ha (2.93%) is bushes and open land. Analysis of land use change conducted by Balai Pengelolaan Daerah Aliran Sungai- BPDAS (The Office Manager of Watershed Area)

Citarum-for 989.7 ha and paddy field Citarum-for 201.9 ha; while the area of settlement increased Citarum-for 2,345.50 ha in between 2000-2009. In line with that, those changes have affected the hydrological condition of Cisadane Watershed.

Prasetyo and Setiawan (2006) estimate the deforestation in Halimun Salak National Park which part of it is in the area of Cisadane Watershed (21,586.1 Ha/ 25.68 %). Several causes are illegal logging, land occupation and forest fire. Those causes are all usually related to the community, both communities who live in the surrounding area of forest or inside the forest.

The use of water resources and the control over its destructive force can only be performed at its optimum if there is an adequate quantity, quality and continuity of water resources. Conservation of water resources is critical to maintain and increase the availability of water. Technically, water resource conservation efforts can be performed by controlling the surface flow, accommodate the runoff from the rain, to absorb water as much as possible and soak it into the ground.

Due to the vital role that water plays in the ecosystem, forecasting its yield for the next generation has become critical to plan and manage its use. One primary factor causes changes in water resources is the constant evolution in land use. Forecasting the spatial distribution of water yield requires hydrology modeling.

The changes of land use patterns generate many social and economic benefits. However, they also come at a cost to the natural environment. One of the major direct environmental impacts of development is the degradation of water resources and water quality (USEPA, 2001). Conversion of agricultural, forest, grass, and wetlands to urban areas usually comes with a vast increase in impervious surface, which can alter the natural hydrologic condition within a watershed. Since people started to live in settlements they have adapted the land use and their needs. The original land use has been replaced by cities, farmlands, industrial sites, roads, canals etc. Furthermore, land use change is an ongoing process (Goldewijk, 2004).

Geographic Information Systems (GIS) has increasingly become a valuable management tool, providing an effective infrastructure for managing, analyzing, and visualizing disparate datasets related to soils, topography, land-use, land cover, and climate (Liao and Tim, 1997; Miller et al., 2004). The integration of GIS with

hydrologic and/or land use models as a data pre/post-processor have simplified data management activities by enabling the easy and efficient extraction of multiple modeling parameters at the watershed scale (Putnam and Chan, 2001; Ogden et al., 2001). The improvement of land use change models combined with developments in hydrological models allows more realistic predictions of future hydrologic system. A new modeling approach is evaluated for assessing the impact of the land use changes by combining a forecasting land use change model (Conversion of Land Use and its Effects at Small regional extent/CLUE-s) with a hydrology model (HEC-HMS). The benefits of such linkages, when supported by a common GIS platform, are that planners can examine the present and future impacts land-use policies may have upon hydrologic response characteristics of a specific watershed.

1.2. Problem Definition

One of the most important issues in watershed management is the use of land, especially in the upstream catchment area. Changes in land use in upstream catchment area, will give a concrete impact on the downstream watershed. In the effort to fulfill their ends meet, especially food, and shelter, people can encourage land conversion. As a result of the land conversion, the Cisadane River condition seems quite apprehensive in the last two decades. Environmental degradation in Cisadane watershed in both the upstream and downstream has a heavy impact to the availability of water resources in Cisadane River.

Related to this fact, an in-depth study on land use change in Upstream Cisadane Watersheds and its dynamics need to be conducted. Particularly, the impact of land use change on water yield.

1.3. Research Objectives

This research purpose is to forecast the future of land use changes and assess the impact of the predicted land use changes on water yield in Upstream Cisadane Watershed.

This aim will be reached through the following objectives:

1. To forecast future land use changes in the Upstream Cisadane Watershed predicted by a land use change model (CLUE-S),

2. To assess the impact of the predicted land use changes on water yield using hydrology model (HEC-HMS).

1.4. Research Question

The present study aims to answer the following questions in order to achieve the above-mentioned objectives. Research question related to the objectives are:

1. What are the major trends of future land use change?

a. How has the land use in study area changed during the period of 1991-2000 and 1991-2000-2009?

b. What are the trends of land use change in study area during 1991-2009? c. What are the underlying driving factors that influence the land use change

in study area?

d. Which are the most important driving factors?

2. How are the impacts of land use change to water yield in the future?

1.5. Scope, Limitation and Assumption of Research

This study is intended to integrate remote sensing and GIS for forecasting land use change in Upstream Cisadane Watershed. And also assessing the impact of land use changes to water yield at Upstream Cisadane Watershed is conducted using hydrology model.

Scope, limitation and assumption of this research are:

1) Study area as a boundary of system in this research is Upstream Cisadane Watershed, with Empang as the outlet of the Watershed.

2) This study is intended to integrate remote sensing and GIS for forecasting land use change in Upstream Cisadane Watershed. Also, , assessing the impact of

land use change to water yield at Upstream Cisadane Watershed by employing hydrology model.

3) Based on the limitation of the CLUE-s Model, land use type only divided into five types, i.e.: water, grassland, estate, settlement and forest.

4) In this research, hydrology model is only employed to explore the effect of land use change to water yield, by comparing hydrograph for each land use. Thus, modeling result will not be calibrated.

1.6. Time and Study Area of Research

This research was conducted from August 2011 until December 2012. The time is allocated for collecting, preparing and processing data; and also making a final report in MIT Laboratory, Bogor - West Java. Data collection process was conducted in Upstream Cisadane Watershed – Bogor.

The study area is Upstream Cisadane Watershed with outlet of watershed in Empang, which located in both Bogor Regency and Bogor Municipality as shown in Figure 1. Geographically, it is located between 60₃₆’26.05” – ₆0₄₇’08.49” _South

Latitude and 106044’30.27” – 106056’36.76” east longitude.

The water source of Upstream Cisadane Watershed is from Gunung Gede Pangrango National Park and Halimun Salak National Park. Those river flows have been used by the community around the river-banks to support and fulfill their daily life needs, based on various kinds of practices. Based on its topography, Cisadane Watershed upstream area is a hilly area with height to 3.000 masl and 40% slope. Cisadane Watershed upstream area includes Bogor Regency and part of Bogor District; its land use is dominated with forest, dry farming, estate, settlement and open area.

1.7. Research Methodology in General.

Research methodology in general for Forecasting Land Use Change and the Impact on Water Yield at Watershed Scale is shown in Figure 2. Preparation of the data into model input consists of spatial and non-spatial data. Spatial data can be divided into raster and vector data. Topographical maps (Rupa Bumi Indonesia) that cover research area (including layers of administrative boundary, land use and road network), village potential statistics map, geological map, forest maps, critical land maps and soil map are used as vector data. Satellite imagery (ALOS-AVNIR) imagery and SRTM-DTM were used as a raster data in this research. Statistical data, including time series of population and village potential statistics of Bogor Regency and Bogor Municipality area in 2000, and 2010 was used as non-spatial data.

Figure 2. General Methodology of Research

The data processing that employs various statistics methods and spatial analysis technique in GIS has been applied to derive the data input into the development model. In addition, land use type classifications were obtained from references map and from satellite imagery in 2000 and 2010. CLUE-s model was applied to forecast future land use changes in the Upstream Cisadane Watershed. The result from this modeling process will be used as an input for the following step, i.e. hydrology model to see the effect of land use change to water yield.

1.8. Thesis Output

1) The outputs of this research are; Forecasted future land use map, based on modeling result.

2) Impact of land use change on watershed hydrology, particularly impact on water yield.

1.9. Thesis Structure

The thesis was organized into four chapters. The first chapter describes introduction chapter includes: (1) Research Background; (2) Problem Statement; (3) Research Objective; (4) Research Question; (5) Scope, Limitation and

Assumption of Research; (6) Time and Study Area of Research; (7) Research Methodology in General; (8) Thesis Output; and (9) Thesis Structure.

Generally, the result and analysis of this research activity are described in chapter 2 and chapter 3. In the second chapter, analysis of land use change and its driving forces will be discussed. Data processing that employs various spatial analysis techniques and modeling in GIS has been applied to forecast the future land use change. Meanwhile, chapter third describes the impact of the land use change into water yield.

Finally, general conclusion and recommendation are drawn in chapter fourth as last chapter of this research.

CHAPTER II

FORECASTING LAND USE CHANGE IN UPSTREAM CISADANE WATERSHED USING CLUE-S MODEL

2.1. Introduction 2.1.1. Background

Land use science can be defined as an inclusive, interdisciplinary subject that focuses on material related to the nature of land use and land cover, their changes over space and time, and the social, economic, cultural, political, decision-making, environmental, and ecological processes that produce these patterns and changes (Aspinall 2006). A variety of theories, methodologies, and technologies underpin research on land use science, and, consequently, a number of basic and applied science themes that are characteristic of land use research can be identified. These reflect the interdisciplinary and integrated analysis required to comprehend land use, as well as the role and importance of land use, land use change, and land management and policy, and the importance of land use for sustainability (Raquez et al. 2006). Land use is also considered a central part of the functioning of the Earth system as well as reflecting human interactions with the environment at scales from local to global.

Basic science questions in land use science include those that focus on (a) dynamics of change in space and time; (b) integration and feedbacks between landscape, climate, socioeconomic, and ecological systems (c) resilience, vulnerability, and adaptability of coupled natural and human systems (d) scale issues and (e) accuracy. Applied science addresses policy and management questions in land use science including (a) addressing evolving public and private land management issues and decisions; (b) interpretation and communication of scientific knowledge for adaptive management of change in land use systems; and (c) human and environmental responses to change (Aspinall and Hill 2008). The applied issues also should be set against a need for explicit management of uncertainties. This will include definition of the limits of applicability of change

participatory approaches. The need and role for spatially integrated dynamic models of coupled natural and human systems in the contexts of study and management of land use change underpin this discussion.

The rising awareness of the need for spatially explicit land-use models within the Land-Use and Land-Cover Change research community (LUCC; Lambin et al. 2000a; Turner et al. 1995) has led to the development of a wide range of land-use change models. Whereas most models were originally developed for deforestation (Kaimowitz and Angelsen 1998; Lambin 1997) more recent efforts also address other land use conversions such as urbanization and agricultural intensification (Brown et al. 2000; Engelen et al. 1995; Hilferink and Rietveld 1999; Lambin et al. 2000b).

Spatially explicit approaches are often based on cellular automata that simulate land use change as a function of land use in the neighborhood and a set of user-specified relations with driving factors (Balzter et al. 1998; Candau 2000; Engelen et al. 1995, Wu 1998). The specification of the neighborhood functions and transition rules is done either based on the user’s expert knowledge, which can be a problematic process due to a lack of quantitative understanding, or on empirical relations between land use and driving factors (Pijanowski et al. 2000; Pontius et al. 2000).

A probability surface, based on either logistic regression or neural network analysis of historic conversions, is made for future conversions. Projections of change are based on applying a cut-off value to this probability surface. Although appropriate for short-term projections, if the trend in land-use change continues, this methodology is incapable of projecting changes when the demands for different land-use types change, leading to a discontinuation of the trends. Moreover, these models are usually capable of simulating the conversion of one land-use type only (e.g. deforestation) because they do not address competition between land-use types explicitly.

The Conversion of Land Use and its Effects (CLUE) modeling framework (Veldkamp and Fresco 1996, Verburg et al. 1999a) was developed to simulate landuse change using empirically quantified relations between land use and its

driving factors in combination with dynamic modeling. In contrast to most empirical models, it is possible to simulate multiple land-use types simultaneously through the dynamic simulation of competition between land-use types.

2.1.2. Objective

The objective of this study is to forecast future land use changes in the Upstream Cisadane Watershed predicted by a land use change model (CLUE-s).

2.2. Literature Review

2.2.1. Forecasting Land Use Change

Several methods have been developed for forecasting land use change, with varying degrees of sensitivity to the influence of transportation networks. The simplest types of models for forecasting land use change are Markovian models (Brown et al. 2000; Weng et al. 2002) such as Markov chain models, which tend to treat land use change as a stochastic process. Assuming that rates of change between land use types are more or less constant from one period to the next, Markovian models project land use transitions forward to any given future date, eventually reaching an equilibrium distribution of land uses. These models tend to have a limited ability to incorporate transportation networks and other spatial features, except as states (e.g., land use types) in the model. More often, they are applied to analyses of land use change.

Cellular and agent-based models have recently gained greater acceptance as tools for simulating land use change in urban areas. Advances in computational power and data storage have facilitated the development of models that disaggregate urban space to a greater degree and can operate with individuals or land parcels as the units of analysis, rather than zones. These include micro-simulation models of urban development, as well as models based on a cellular automata framework (Jantz et al. 2005). Cellular automata models emphasize neighbor effects and dynamic interactions between agents (with land use cells as agents), while micro-simulation models treat individual households and firms as

agents and attempt to simulate their behavior in terms of location and travel choices. Micro-simulation models of land use are often coupled with transportation models and are integrated into larger urban simulation models (Waddell and Ulfarsson 2003).

Despite these methodological advances, regression models continue to be a popular method for modeling and simulating land use change. Indeed, many simulation models with a land use component use regression methods, either in the form of discrete models of land use change (Landis and Zhang 1998) or within hedonic or bid-rent frameworks for land prices (Waddel 2003). Regression models allow the identification of exogenous variables, which are thought to influence patterns of development. The variables can represent physical and social influences on development (Verburg 2004), neighborhood effects (Verburg 2004; Zhou and Kockelman 2007), or the effects of transportation and accessibility. It is these latter effects that are of the greatest interest in the current context. While regression techniques have been used previously to identify the correlates of highway network growth in terms of land use and population characteristics (Levinson 2007).

2.2.2. Modeling Land Use Change

The land-use change model, Conversion of Land-Use and its Effects at Small regional extent (CLUE-S) (Verburg 1999), is used to simulate future land-use change. The CLUE-S model is an empirical based model developed at the University of Wageningen in the Netherlands. The model attempts to identify causes of land use changes (driving forces), using a multivariate analysis on the possible contributors, to empirically derive rates of change (Verburg et al. 1999).

The CLUE-S model has been chosen for the land-use modeling in this study based on the selection criteria developed by the US Environmental Protection Agency (US EPA, 2000). The most important reasons for choosing this model were: the flexibility on the input data (driving forces), the possibility of linking the output to another model (e.g. Hydrological Model/HEC-HMS model), and free access to the model.

2.3. Methodology

Changes in land use pattern are related to a large number of biophysical and socio-economic factors. The modeling of spatially explicit changes in land use pattern requires, therefore, a large database of factors considered to be important in the case study. Therefore, the database is not similar for every application. To run the model it is minimally needed to have spatially explicit data for at least 1 year. However, to allow calibration and validation model works, it is necessary to have data of another different year. To meet this necessity, the research will employ data from 2 different years, with 6 years’ time difference, i.e. 1991 and 2009.

2.3.1. Data Preparation

2.3.1.1. Data Requirement for Land Use Change Model

For the simulation of dynamics of the spatial pattern of different land use types, data are needed for the land use distribution and a number of biophysical and socio-economic parameters that are considered as important potential drivers of the land use pattern. These drivers are most commonly variables that describe the demography, soil, geology, climate and infrastructural situation. This study only considers the biophysical aspects, while the socio-economic aspect is considered constant (Business as usual).

The data required to analyze land use change process and build scenario development were obtained from various sources. The data are derived from multispectral satellite data, extracted from digital topographic data, and from spatial processing of statistical data. According to type, data are divided in three: remotely sensed data, digital topographic data and statistical data. How the data were collected and used will be explained below.

Remotely Sensed Data

Remotely data used in this study comprises of Landsat images 1991 (30 m), and ALOS-AVNIR image 2009 (10 m). These data will be used for land use change analysis and input for trend extrapolation to calculate land use requirements year 1991-2030.

Table 1. Remotely Sensed Data Requirement for Land Use Model

Image Resolution Date Acquisition Source of Data

Landsat TM 30 m 28 July 1991 GLOVIS

ALOS AVNIR 10 m 17 July 2009 BTIC – BIOTROP

Topographic data

Topographic data sets are obtained from Agency of National Survey and Mapping (Bakosurtanal – Geospatial Information Agency/BIG, nowadays), Bogor. These data are digitized from topographic map scale 1:25.000 and are extracted for selected layers, including road, facilities and public service distribution, and central of economic. All layer data have been digitalized, infrastructure and facilities, road, river, public services and industries location. Data sets used in this study can be seen in the following table.

Table 2. Topographic Data Sets Used in Analysis Land Use Change and Scenario Development

Data Set Year and

Scale Class Source

Collection

Method Data Use Land use

map

1991, and 2009; Scale 1:50.000

5 classes (See Table 4 for detailed land use classification)

Interpretation from Landsat TM and ALOS-AVNIR images Interpretation using supervised classification

Land uses data are used for land use change analysis and trends extrapolation and land demand input in CLUE-S Road 2003; Scale

1:25.000

Classified as main road, appropriate with study area

Rupa Bumi Indonesia (RBI), 2003 Derived from topographic map (RBI)

Road map used as driving factors influencing land use change and pattern, it will be processed by using distance analysis to obtain classes of distance from road Elevation

Data

2005, Scale 30m

Continuous USGS Derived from SRTM-Level 2

Interpolation and rasterizing to obtain digital elevation data and image classification to obtain elevation class Facilities and public service distribution 2003; Scale 1:25.000

All of facilities (education, health and public service facilities) Rupa Bumi Indonesia (RBI), 2003 Derived from topographic map (RBI)

Facilities and public service are processed by using proximity analysis to obtain distance classes of facilities Central of

Economic

2003; Scale 1:25.000

Business center Rupa Bumi Indonesia, 2003

Derived from topographic map

COEs are processed by using proximity analysis Forest Designation Map 2009, Scale 1:250.000

Forest Area, Non Forest Area

Directorate General of Forest Planology, Ministry of Forestry Re-delineate in GIS environment

Forest designation map will be used as spatial policy in the model.

Socio-economic data

Statistical data in this study consist of population data in village level year 2007. Other data socio-economic are total education and health facility unit for each village in the study area year 2007).

Table 3. Statistical data used in analysis land use change and scenario development

Data Year Source Description Data Use Population 2007 Village potential

statistic –

Statistical Center Agency (BPS)

Population data for each village in the study area

Population data will be used to produce density map in sub-district level as an input of scenario modeling

2.3.1.2. Data Processing and Formats

In order to enable analysis and modeling in the CLUE modeling framework the data need to be converted into a consistent format. Therefore the data are converted to raster format. Grid size depends on the scale and resolution of the original data. All grids should have the same (equal area) projection and be geo-referenced with exactly the same extent and cell size.

All data are communicated with the CLUE-S model through ASCII files, meaning that the data are stored in a text file that contains all values of the individual grids, presenting the values for separate cells. In this format the values for the individual cells are stored in rows and columns with a header that describes the format, like number of columns and rows and the coordinates of the upstream left corner. Data can also be stored in one single column containing all the cells.

2.3.1.3. Data Normalization

The maps used as independent variables are created in raster format. To produce these maps, vector data are transformed into raster format with a spatial resolution of 50 m. This spatial resolution is appropriate with image data, which also have a resolution 50 m.

To balance the range of data values, all of variables are normalized into 0 – 1. Normalization is done by dividing the value of each grid cell with the highest value of cell data to achieve similar data range. This method is important because of the sensitivity of data transformation in logistic regression. Moreover, in the multivariate analysis such as logistic regression, the continuous independent variables should have a similar scale (Nefeslioglu et al, 2008).

2.3.2. Land Use Scenario Development by using CLUE-S Model

A development scenario of land utilization in this study is addressed to understand the dynamic of land use change. Method to implement land use scenarios in this study is based on CLUE-S model (Veldkamp, et al., 1996), which emphasizes on how scenarios can be demonstrated in study area by using simulation of land use change.

CLUE-S model is chosen in this study because of its adaptability to local scale and successful applications in tropical regions that make it appropriates to be applied in study area. Besides that, this model is based on an analysis of the spatial structure of land use that it makes the model is not bounded by the behavior of individuals or sectors of the economy. In the other hands, CLUE-S model provides many opportunities to include actor-behaviors and feedbacks from policy makers. The possibility to simulate different scenarios for various land use types at the same time makes the model powerful to support land use analysis in study area.

2.3.3.1. Land Use Classification

Land use maps are derived from classification of Landsat TM images year 1991, and ALOS-AVNIR images year 2009. In this study, the images are classified in five classes based on land use condition in the area.

Table 4. Land use classification in study area

Land Use Code Description

Water 0 River, lake

Grassland 1 Shrub, bare land

Estate 2 Irrigated & rain fed paddy field, dry land Settlement 3 Residential area, developed area

Forest 4 Forest, no matter inside or outside forest area

2.3.3.2. Location Characteristics

Location characteristics are a set of factors that affecting land use changes. The combination of factors provides preference for a specific type of land use to take place at a specific location. Preference represents the outcome of interaction

between land use and driving factors is calculated by using logistic regression. The results of regression analysis are used to create probability map for each land use type. The driving factors used in this study can be seen in Table 5.

Representation of driving factors in the CLUE-S model

In order to fit the underlying factors in logistic regression, all of determinants will be adjusted and specified. All factors that considered important as agents of land use change are provided in the following table.

Table 5. Driving Factors Included in Logistic Regression

Driving Factor Representation

of Variable Description

Nature of

Variable Source

Socio-economic Drivers Population

Density

POPDEN Person/Km2_{, village}

level Categorical, converted to floating/con tinuous Village potential statistic – Statistical Center Agency (BPS)

Bio-geographical drivers

Elevation ELEVS Interpolation and

rasterizing to obtain digital elevation data & image classification to obtain elevation class

Continuous Derived from SRTM-Level 2

Proximity Drivers

Roads ROADS Distance to road Continuous road map from Agency

of National Survey and Mapping

Public facilities PUBFAC Distance to public facilities

Continuous Infrastructure map from Agency of National Survey and Mapping Education

facilities

EDUFAC Distance to education facilities

Continuous Infrastructure map from Agency of National Survey and Mapping Health facilities HEAFAC Distance to health

facilities

Continuous Infrastructure map from Agency of National Survey and Mapping Central of

Economic

COECO Distance to central of economic

Continuous Infrastructure map from Agency of National Survey and Mapping

In this research, population density is used to represent the demographic condition. Population growth increases the level of demand for residential parcels and facilities. As population expands, cities grow spatially to accommodate more people.

Proximity is measured by direct access to roads, public facilities, education facilities, health facilities, and distance to central of economic. Proximity of each

cell is mainly measured with straight-line distance to the specific location, where the straight-line is measured by using Euclidean distance to the target location.

It is predicted that proximity factors become more dominant in this study, because of relatively smooth topography of the study area. Distance to facilities, and distance to road become important determinants in predicting land use change. Furthermore, elevation data is considered important in this area, although it has little variation. Elevation is significant for estate and settlement land use area, where it is associated with the distribution of irrigation water to agricultural crops and landscape for residential area.

Another fact is that the density of population is measured by the relative density, not an absolute density. Density is calculated based on the number population (person) divided by the village area (Km2). This is not a best-fit method to measure the density where density should be measured in urban area or named absolute density. However, the calculation of the absolute density of urban areas land use cannot be applied to logistic regression, because it will produce data that are bias, where the regression calculation applied to the same extent of areas.

2.3.3.3. Statistical Analysis

Logistic regression is divided in two types, including binomial and multinomial regression. Binomial regression, which is employed in the research, uses dichotomous value in the dependent variable, whereas the type of independent variable could be categorical or continuous. In land use analysis, logistic regression is used to examine the relation between land use and possible driving factors. The results of this analysis are coefficient values that show the contribution of each driving factor to land use change. The formula is:

� ₋ = � + � � + � � + ⋯ + ���� (Verburg 2002).

Where : is the land use change probability

��… are independent factors

To achieve the validity of land use change estimation, the model should be supported by the procedure to identify the driving factors that are statistically independent and to determine the significance of driving factors. One of methods to identify the driving factors that have significant contribution to land use pattern is the stepwise procedure. In the stepwise procedure, all of driving factors are involved in one step and eliminated according to their significance values. A driving factor that has a lower value than the significant threshold will be excluded from the analysis. In forward procedure, the analysis starts with one factor and continues to other factors respectively.

ROC (Relative Operating Characteristics) is a method to measure the goodness of the statistical model. The probability of each land use resulted from logistic regression is compared to the real land use map to calculate the equal category of each grid cell between those maps. According to Pontius and Schneider (2001), this method will depict the capability of regression equation to represent land use characteristics. The range of ROC value is between 0 – 1, where ROC value below 0,5 is categorized in low or completely random, 0,5 – 0,6 is good, 0,6

– 0.99 is very good and 1.0 is fit/perfect. The goodness of the logistic regression equation to represent land use condition indicates the suitability of driving factors as determinant of land use change.

2.3.3.4. Land Use Type Specific Conversion

Conversion setting for specific land use type is addressed to determine the temporal dynamic of the simulation by using reversibility of land use changes. This method will be implemented by using three different decision rules that represent the situation of study area:

1. Some land use types are unlikely to be converted into another land use type after first conversion.

2. Other land use types are converted more easily. Forest and grassland are more likely to be converted into another land-use type soon after their initial conversion without any restrictions.

3. Other remains land use types operate in between those settings, where the conversion will occur in specific condition. An example is grassland will be converted to estate area if estate area is more profitable.

Table 6. Land use conversion matrix for study area

Land Use Future

Water Grassland Estate Settlement Forest

P

re

sent

Water 1 0 0 0 0

Grassland 0 1 1 1 1

Estate 0 1 1 1 1

Settlement 0 0 0 1 0

Forest 0 1 1 0 1

1 likely to conversion; 0 unlikely to conversion

This method is one of specific setting to determine temporal dynamic of land use simulation. Land use with high investment will not easily be converted to other uses. Moreover, because of the differences of conversion behavior, dimensionless factor is added to each land use type. This factor represents elasticity conversion, ranging from 0 (easy conversion) to 1 (irreversible change). Water area and settlement area is an example of this rule, where the elasticity value for urban built up area is set 1 that shows urban built up area is hard to be converted to another type of land use. Grassland and forest area are more easily to be converted and the value is 0.4. Estate is set to more difficult to be converted, so the elasticity value is 0.6. The justification of the elasticity values of land uses in this study is based on field observation and local knowledge and adjustment for the model. The range of values can be seen in following table.

Table 7. Settings of conversion elasticity in the study area

Land Use Type Conversion Elasticity

Water 1

Grassland 0.4

Estate 0.6

Settlement 1

Forest 0.4

2.3.3.5. Scenario Setting

The scenarios presented are not necessary the most realistic, but are made in such a way that they provide information on the functioning of the model. A scenario for the CLUE-S model consists of a file with land requirements and a file that indicates areas where restrictions to conversion apply.

Land Requirements

Land use requirement (demand) is calculated for every land use in the whole of study area. This requirement becomes a constraint of area needed for simulation. Demand of land use will be specified in every year by using extrapolation of past land use change trends that have been quantified before (year 2001-2009).

Future land utilizations are based on two considerations:

1. Baseline.

Population growth is assumed stable and land requirements of 2009-2020 keep linear change based on the trend in the period 2001-2009. According to the baseline scenario, the demand of land use year 2009-2020 will be extrapolated from the annual change in the period 2001-2009.

2. The increasing growth of population.

The assumption is based on a slow growth of population in study area that has effects on increasing of demand of settlement and estate land use. In this scenario, the demand of urban built-up area is assumed increased twice from the demand of urban built-up area in baseline scenario.

Spatial Policies

Spatial policies (restriction) can influence the pattern of land use change. Spatial policies mostly indicate areas where land use changes are restricted through policies. Besides that, spatial policies can imply stimulation arrangements for a certain land use on a location. In this study, forest designation map is used as area restriction, where forest area is restricted to be converted into other land use area.

Figure 3. Maps of Restricted Area

Table 8. Scenarios used in spatial modeling

Scenarios Land Requirement Spatial Policies

First (Baseline)

Equal with the trend of the land use change in common

No Spatial Policy

Second Equal with the trend of the land use change in common

Forest area is restricted to be converted into other land use area

Third The increasing growth of population

No Spatial Policy

Fourth The increasing growth of population

Forest area is restricted to be converted into other land use area

2.3.3. Model Validation

An important stage in the development of any predictive change model is validation. Typically, one gauge means the understanding of the process and the power of the model by using it to predict some periods of time when the land use conditions are known. This is then used as a test for validation.

The first is called ‘validate’, and it provides a comparative analysis on the basis of the Kappa Index Analysis. Kappa is essentially a statement of proportional accuracy. Kappa is computed as:

= � ∑_{� − ∑}��= ���− ∑��=_� ��+ , �+�

�+ , �+�

� �=

Where : � = Total number of sites in the matrix,

� = Number of rows in the matrix

��� = Number in row i and column i

��+ = Total for row i, and

�+� = Total for column i.

Kappa can be used to determine if the values contained in an error matrix represent a result significantly better than random (Jensen 1996).

2.4. Result & Discussion

This section will mainly discuss the results of overall analysis, including analysis of land use changes, examination of driving factors in logistic regression and land use scenario development by using relevant variables in the CLUE-S model. The first section explains the result of identifying, quantifying and trend analysis of land use change. The second section of this chapter mainly demonstrates the use of relevant variables logistic regression analysis in development scenarios of future land use. The model used will depict the effect of land use change to the future land use. All of sections of this chapter will be directly followed by discussion to provide clear explanation about all of the findings of this research.

2.4.1. Land Use Pattern

2.4.1.1. Land Use Maps Year 1991 and 2009

According to the classification processes, there are five land uses that could be identified from all of images: water, grassland, estate, settlement and forest. Description of each land uses are explained in Table 4 and the location and distribution of land use could be seen in the figure below.

Figure 4. Land Use Map Year 1991 and Year 2009

2.4.1.2. Land Use Change Analysis

The land cover/use change patterns from one land cover/use to another are presented in the following table and the maps can be seen in Figure 5.

Table 9. Matrix of land use change year 1991-2009

Land Use 1991 2009 Change

1991 - 2009 Percentage

Annual Average Change

Water 93.50 93.50 0.00 0.00 0.00

Grassland 2209.25 2313.50 104.25 4.72 0.26

Estate 12040.25 11293.00 -747.25 -6.21 -0.34

Settlement 3055.25 3454.50 399.25 13.07 0.73

Forest 4831.25 5075.00 243.75 5.05 0.28

Based on the findings about land use changes in year 1991 and 2009 above, it can be noted that urban settlement area experienced the most significant development followed by forest and grassland. However estates tend to decrease and river is stable during the 1991-2009 period.

2.4.1.3. The Trends of Land Use Change during 1991-2009

According to the change detection for the land use classification year 1991 and 2009, it can be seen that settlement, forest and grassland area are increasing over time, whereas estate tend to decrease and river remains constant. The trends of land use changes of the area during 1991-2009 are presented in following figure.

Figure 6. Trends of land use change year 1991-2009

2.4.2. Analysis of driving factors of land use changes 2.4.2.1. Logistic regression results

The results of logistic regression between land use and independent variables are presented in this section. Each land use has independent variables or driving factors that influence to its pattern. In logistic regression analysis, five classes of land use are included in the regression calculation.

The selection of the significant and non-significant independent variables is based on a enter procedure (see 2.3.3.3.). The variables, which have coefficient values below 0.01 significant thresholds, are categorized as significant and the

variables above 0.02 will be classified as non-significant and automatically removed in the calculation process. All of significant variables are automatically selected in the results of enter procedure and will be used in the calculation of land use probability. Practically, enter procedure in logistic regression is done by putting all independent variable in the process at the same time in the iteration process. The result can be seen in the Table 10.

Table 10. Results of logistic regression between land use pattern and driving factors

Variable Water Grassland Estate Settlement Forest

Total number of

observed grid cell 88918 88918 88918 88918 88918

POPDEN 2.404 -5.964 -1.374 1.081 -2.913

ELEVS -27.566 .968 1.056 -1.060 -8.369

ROADS -22.757 3.714 -.355 -2.106 .949

PUBFAC ** -6.285 -1.100 -8.848 8.679

EDUFAC ** -.570 -6.851 -5.745 20.718

HEAFAC -11.171 4.251 -2.487 .748 **

COECO 9.398 -.821 1.596 -1.022 **

CONSTANT 1.892*) _{-2.353 1.983 .303 -8.541}

EXP (�) 6.633 0.095 7.264 1.354 5120.462

ROC 0.903 0.701 0.780 0.813 0.994

Enter procedure; Significant at: < 0.01 entry level and > 0.02 removal

*) _{Has significant value above 0.02 (See Appendix 5.); ** Not statistically significant (value above 0.02)}

2.4.2.2. Logistic regression interpretation

The number of driving factors used in this regression analysis is 7 variables. Each driving factor has a different effect on every type of land use. The effect of each driving factor is indicated by the coefficient β in logistic regression result, which presents how much variance from the use of land that can be explained by

the driving factors. A large positive β value indicates a strong positive relationship between the independent and dependent variable (land use change), while a large

negative β value indicates a strong negative correlation with land use change. Besides the effect of driving factors, the results of logistic regression can be used to indicate which driving factor that has the biggest influence to the land use change. How big the influence of each driving factor is indicated in exponential β value that shows the effect of increasing one unit of driving factor to the land use

exponential β for settlement area (Table 10.), it can be interpreted that population density has the biggest influence to the land use change in the area compared to the other driving factors. As an example of interpretation, population density has

coefficient β value 1.081and exponential β value 2.948, which means the increase of 1 unit of population density, will influence 1.081 unit of settlement area to change with the probability value 2.948. In the other words, 1 unit influence of population density has 2.948 probability of settlement area to change. This interpretation also applies to other driving factors and other land use types involved in the logistic regression model.

According to the stepwise procedure in the logistic regression for forest area (Table 10.), some variables have significant values, including population density (POPDEN), elevation (ELEVS) distance to main road (ROADS), distance to public facility (PUBFAC), and distance to education facility (EDUFAC). In the other hand, distance to health facility (HEAFAC) and distance to center of economy (COECO) have non-significant values and removed from the calculation process. In general, the result of logistic regression for forest area is appropriate with the pattern of land use change. In other case for water (Appendix 5.), population density, elevation, distance to main roads, distance to health facility, and distance to central of economy have significant values. In the other hand, distance to public facility and distance to education facility have non-significant value. The interesting in this case is constant for the water case has non-significant value (0.04), that’s mean the equation for water automatically removed in the calculation process, when it is inserted in the model, the probability of river will not change from year to year. It same with the assumption that water area will not change during the model run.

Regarding to the relationship in settlement case, it can be seen that population density, and distance to health facility have positive effects to the urban built-up area. In other hand, elevation, distance to main roads, distance to public facility, distance to education facility and distance to center of economic have negative effect to the probability of settlement area to change. The relationship of each driving factor with settlement area change in this research is reasonable for the characteristic of highland area like Upstream Cisadane because it is still dominated

from the existing settlement, while distance to main roads, public facility, education facility and center or economy, are far away from settlement area.

Based on the value of the coefficients of these driving factors, the regression equation can be created to indicate the influence of the driving factors and the overall probability for each land use change. By using ROC, the probability of each land use is evaluated and compared to the existing land use. The comparison is based on the number of cells that equal between the maps. The percentage of ROC shows how the regression equation can be used to predict the future land use change based on its probability.

According to the results, the values of ROC for probability of water area, grassland area, estate area, settlement area and forest area are 0.903, 0.701, 0.780 and 0.813 and 0.994 respectively. Based on some literatures, these values are categorized as better (very good). These results are valid, which is confirmed to the results of some similar studies using logistic regression, including Hu and Lo (2007) and Verburg et. al. (2002). The differences are on the direction of the relationship and the most influenced driving factors and land use as explained before. Moreover, this result is reasonable because the influence of driving factors depends on the location and the situation where the model is applied (Xie et al, 2009).

After measuring the goodness of the logistic regression results for every land use type, it can be concluded that the model capable to represent the land use changes within the Upstream Cisadane. However there are some considerations that should be reviewed related to data adequately, data quality, and scale of analysis. The availability of relevant data should placed is the most important consideration because it reflect dynamic of the area being modeled (Verburg et al, 2004). The socio-economic data such as preference of society to choose new settlement location with lower price are not used in this research because of unavailability of data. As these data could improve the understanding of complex land use behavior, their availability should be incorporated in further research.

Scale of analysis is another important issue to be discussed because this research is conducted by using images that have middle scale resolution (50m x 50m). Although this resolution is categorized as large scale in the CLUE-S model

(Verburg et al, 2002), it has possibility to be improved with higher spatial resolution data to acquire more detailed information about Upstream Cisadane behaviors. The more information could be derived from the data, the more understanding about land use dynamic in the location (Cheng, 2003). However, it should be noted that data with large resolution need more capacity for iteration process and when it is combined with the land use model, the capacity data could excess the processing capability of the model.

Furthermore, this research indicates that logistic regression analysis is very sensitive in multistage analysis such as data transformation and spatial sampling. According to Cheng (2003), the logarithmic data transformation and the combination of sampling type could significantly influence the parameter estimation and model accuracy. Related to sampling method, this research has employed appropriate sampling method. The method has been discussed in the conceptual model section, where all of independent variables are normalized to the range 0-1 to create similar range of data and avoid different data dimension. This method is valuable to overcome the sensitivity of logistic regression calculation to the various variables and the result is more reliable in examining the influence of driving factors of land use change.

According to the findings and their interpretation above, the conclusion can be drawn about the usability of statistical analysis by using logistic regression to examine the relationship between driving factors and land use change in study area. The results give evidence about the influence of each driving factor on each land use type, the magnitude of its effect and the most influencing driving factors for each land use type. Regardless the limitation in data, the method used and the results are satisfactory because they capable to reveal the relationships that were not known before and provide new insight about land use behavior in the area, such as the growth of residential area in the adjacent of secondary road and at a specific distance from primary road. Moreover, the combination between logistic regression and CLUE-S framework becomes the solution for the limitation of logistic regression model where it can only be used to predict the location of land use change probability but it cannot be used to predict when the change will occur.

Based on these reasons, logistic regression model in this research is combined with CLUE-S framework to predict future land use change in the area.

2.4.2.3. Land use probability maps based on logistic regression

The final results of the logistic regression model are used to create probability map of each land use, including urban built-up area, agricultural area, mix vegetation, ponds area and river. The value of probability of each land use type is in the range between 0 and 1. The probability maps can be seen in Appendix 7, which show that the darker of the color, have the higher the probability of land use to change. The range of probability water area is 0.870 – 0.984, grassland area is 0

– 0.273, estate is 0 – 0.929, settlement is constant is 0.5, and forest area is 0.002 – 0.966. The interesting from probability of settlement where the constant is 0.5,

that’s mean every pixel in a whole study area has the same probability to change to

settlement (see Appendix 7).

2.4.2.4. Model Validation

In order to achieve a valid land use simulation, the validity of this model is examined by using the observed land use map year 2009, which has been created from image interpretation. Land use map year 2009 resulted from the simulation is compared to the observed land use map year 2009 to measure the number of equal grid cells in both of maps and the similarity of land use pattern. Similar with the method done by Zhu et. al. (2009), this approach results a overall accuracy which indicate the degree of similarity between those two maps and Kappa accuracy that depicts the degree of pattern similarity. According to Landis and Koch (1977), Kappa accuracy is useful to calculate the agreement of two maps. It is stated that Kappa values > 80% is categorized as fit, 60% – 80% is high agreement, 40% – 60% is moderate and <40% is poor. Therefore Kappa accuracy is used in this research to measure pattern similarity.

After comparing both maps, the results show that overall accuracy and Kappa accuracy for land use simulation year 2009 is 90.83 % and 86.00 % or categorized

pattern in study area and it can be used to predict future land use pattern. Furthermore, related to the uncertainty of land use pattern in the future, 90.83 % overall accuracy and 86.00 % Kappa accuracy values show that all of driving factors capable to reduce the uncertainty because they capable to describe the land use behavior that will shape the future land use condition. The results are confirmed to the statement of Pontius and Neeti (2009) where high agreement resulted from validation process indicates that the processes of land use change during the calculation are stable trough the interval of validation and suitable to be used in simulation process. The result of Kappa measurement can be seen in the table below.

Table 11. Results of KAPPA Analysis for Comparison between Land use map between interpretation and simulation for year 2009.

Land Use User Total _Accuracy User % W G E S F

Pr

o

d

u

c

e

r W 374 0 0 0 0 374 1.000 100.000

G 0 6971 832 76 1375 9254 0.753 75.330 E 0 830 42608 1419 315 45172 0.943 94.324 S 0 172 1407 12222 17 13818 0.884 88.450 F 0 1280 325 102 18593 20300 0.916 91.591 Total 374 9253 45172 13819 20300 88918

Producer

Accuracy 1 0.75 0.94 0.88 0.92 % 100 75.34 94.32 88.44 91.60 Overall Accuracy 90.83 %

Kappa Accuracy 86.00 %

2.4.3. Scenario-based land use modeling by using CLUE-S model

Scenario-based land use modeling is used in this research to understand the phenomena of Upstream Cisadane Watershed dynamic, especially related to land use change. In order to understand the dynamic of land use in recent and future time, the model will be used as a tool to link between present and future of land use condition and the scenarios will be built to design different alternative conditions of land use. In order to design the plausible scenarios of highland use in the future, the most important things are the understanding of relationship between land use and the driving factors and the understanding of pressures that caused by those driving factors.

In CLUE-S model, the demand of land use is specified on yearly basis before the iteration and used as direct input for the allocation module. The allocation of land use in this framework is based on combination between empirical analysis, spatial analysis and dynamic modeling in raster base iterative procedure. This procedure allows continuous interaction between macro scale demand and local land use suitability as derived from logistic regression. Macro scale demand or regional demands influence the actual allocation change together with the local highest probability derived from logistic regression.

Based on the iteration variable, preliminary allocation is created by considering conversion matrix (see Table 6.) and the total allocated area is calculated by using land use requirements from the regional demand. The iteration process is repeated until all of demands are balance and correctly allocated.

In this research, four combinations of land use change are designed. These scenarios have been explained in the section 2.3.3.4, including baseline scenario and increasing population growth scenario. These scenarios are demonstrated to verify the usability of scenario based modeling to support land use analysis in Upstream Cisadane. Each scenario and the results will be explained in the following section.

2.4.2.1. Baseline scenario

In this scenario, it is assumed that land use change in future is the continuity of land use change in the past time. Land use change in the past time has been calculated in the land use pattern section. In this model, the initial land use is land use map year 1991 created from image interpretation. The demand of land use in the future (until the year 2030) is the continuity of the demand of land use change 2001-2009. Moreover, with no spatial policies applied, it’s mean that all of areas inside the study area are possible to change to another land use. Simulated maps for selected years 2000, 2010, 2020, and 2030 can be seen in the following figure.

2.4.2.2. Scenario 2

In this scenario, the demand of the land use is still the same with baseline. The difference, in this scenario spatial policies is apply where forest area based on forest designation map from Ministry of Forestry is not allowed to change to other land use. Simulated maps based on this scenario in selected years 2000, 2010, 2020, and 2030 can be seen in the following figure.

Figure 9. Simulated land use maps with Scenario 2.

2.4.2.3. Scenario 3

In this scenario, the demand of the land use is doubled compared with

baseline. It’s mean that the demand for each land use also doubled from baseline. Spatial policies are not applied in this scenario. Simulated maps based on this

scenario in selected years 2000, 2010, 2020, and 2030 can be seen in the following figure.

Figure 10. Simulated land use maps with Scenario 3

2.4.2.4. Scenario 4

In this scenario, the demand of the land use is doubled compared with baseline, same with Scenario 3. The difference, spatial policies are applied in this scenario. Simulated maps based on this scenario in selected years 2000, 2010, 2020, and 2030 can be seen in the following figure.

Figure 11. Simulated land use maps with Scenario 4

2.4.2.5. Scenario-base land use modeling interpretation

Based on the results of land use scenario modeling using CLUE-S framework, it can be seen that the combination between empirical analysis, rule-based modeling and cellular automata is valuable in depicting future land use distribution and can be use to answer how does the future land use ‘might be’ occurs in the study area. It means that it can provide possibilities of future land use configuration before it happens. However, due to the uncertainty in the future and the complexity of land use and the drivers, this scenario-based modeling is not only addressed to make predictions but more in improving understanding about land use behavior on upstream watershed zones based on the influencing factors and what might happen in the future based on these known underlying factors. In the planning process, the

understanding about land use behavior as well as future land use distribution is important for planners and managers to justify a spatial planning decision and avoid inappropriate decision (Pontius and Neeti, 2009). Therefore, in order to provide an appropriate explanation, the critical discussion in this research will be conducted by considering the findings and its appropriateness with some literatures.

According to the change detection for the land use classification year 1991, simulated year 2000, year 2010, year 2020 and 2030, it can be seen that settlement, forest and grassland area are increasing over time, whereas estate tend to decrease and water remains constant. The trends of land use changes of the area during 1991

– 2030 are presented in following figure.

Figure 12. Trends of land use change based on Scenario 1.

Figure 14. Trends of land use change based on Scenario 3.

Figure 15. Trends of land use change based on Scenario 4.

In case estate, the trend show that estate always decreased from time to time. Some probability could be happen, such as plants that have been grown, with a high density is often identified as a forest. In other probability, in the areas that have been harvested, the land cover looks like a bare land or grassland. It is most likely to occur, due to the habit of people in Upstream Cisadane plant fast-growing plantation and harvest it as soon as possible. However, although the area of estate decrease from time to time, estates still dominates the land use in Upstream Cisadane.

Settlement has increased significantly with the annual average change 0.73 % per annum. This is because people need space to build residential area and other building facility in line with the increasing of population. Grassland and forest are

Logistic Regression Result

Logistic Regression for Water

Stepwise Variable in The Equation

B S.E. Wald df. Sig Exp(B)

Step 1a _{POPDEN 2.404 .365 43.314 1 .000 11.073}

ELEVS -27.566 2.718 102.842 1 .000 .000 ROADS -22.757 1.653 189.633 1 .000 .000 PUBFAC 4.075 1.810 5.072 1 .024 58.879 EDUFAC -5.056 1.671 9.157 1 .002 .006 HEAFAC -11.171 .818 186.593 1 .000 .000 COECO 9.398 .642 214.340 1 .000 12070.235 CONSTANT 1.892 .648 8.516 1 .004 6.633

ROC 0.903

Logistic Regression for Grassland

Stepwise Variable in The Equation

B S.E. Wald df. Sig Exp(B)

Step 1a _{POPDEN -5.964 .144 1711.990 1 .000 .003}

ELEVS .968 .097 100.657 1 .000 2.634 ROADS 3.714 .101 1349.627 1 .000 41.019 PUBFAC -6.285 .249 638.102 1 .000 .002

EDUFAC -.570 .236 5.809 1 .016 .566

HEAFAC 4.251 .121 1227.966 1 .000 70.184

COECO -.821 .142 33.490 1 .000 .440

CONSTANT -2.353 .046 2670.907 1 .000 .095

ROC 0.701

Logistic Regression for Estate

Stepwise Variable in The Equation

B S.E. Wald df. Sig Exp(B)

Step 1a _{POPDEN -1.374 .075 332.026 1 .000 .253}

ELEVS 1.056 .073 211.273 1 .000 2.874

ROADS -.355 .077 21.314 1 .000 .701

PUBFAC -1.100 .184 35.808 1 .000 .333 EDUFAC -6.851 .174 1555.981 1 .000 .001 HEAFAC -2.487 .078 1019.358 1 .000 .083 COECO 1.596 .089 319.063 1 .000 4.934 CONSTANT 1.983 .032 3843.692 1 .000 7.264

ROC 0.780

Logistic Regression for Settlement

Stepwise Variable in The Equation

B S.E. Wald df. Sig Exp(B)

Step 1a _{POPDEN 1.081 .089 149.088 1 .000 2.948}

ELEVS -1.060 .119 79.428 1 .000 .346 ROADS -2.106 .132 256.184 1 .000 .122 PUBFAC -8.848 .303 851.477 1 .000 .000 EDUFAC -5.745 .268 460.095 1 .000 .003 HEAFAC .748 .106 50.271 1 .000 2.113 COECO -1.022 .112 83.173 1 .000 .360 CONSTANT .303 .049 38.357 1 .000 1.354

ROC 0.813

Logistic Regression for Forest

Stepwise Variable in The _Equation B S.E. Wald df. Sig Exp(B)

Step 1a _{POPDEN -2.913 .483 36.407 1 .000 .054}

ELEVS -8.369 .290 830.462 1 .000 .000

ROADS .949 .253 14.024 1 .000 2.584

PUBFAC 8.679 .520 278.896 1 .000 5876.518 EDUFAC 20.718 .625 1097.218 1 .000 9.945E8

HEAFAC .256 .221 1.348 1 .246 1.292

COECO .927 .407 5.188 1 .023 2.528

ROC Curve for each Land Use Type

(a) ROC = 0.903 for Water (b) ROC = 0.701 for Grassland (c) ROC = 0.780 for Estate

Land use probability maps resulted from logistic regression

Definition of Hydrologic Soil Groups

HSG

Soil Group Characteristic

A

Soils having high infiltration rates, even when thoroughly wetted and

consisting chiefly of deep, well to excessively drained sands or

gravels. These soils have a high rate of water transmission.

B

Soils having moderate infiltration rates when thoroughly wetted and

consisting chiefly of moderately deep to deep, moderately fine to

moderately coarse textures. These soils have a moderate rate of water

transmission.

C

Soils having slow infiltration rates when thoroughly wetted and

consisting chiefly of soils with a layer that impedes downward

movement of water, or soils with moderately fine to fine texture.

These soils have a slow rate of water transmission.

D

Soils having very slow infiltration rates when thoroughly wetted and

consisting chiefly of clay soils with a high swelling potential, soils

with a permanent high water table, soils with a clay pan or clay layer

at or near the surface, and shallow soils over nearly impervious

material. These soils have a very slow rate of water transmission.

Source: US SCS 1972 in Asdak, 2007.

Appendix 9.

Land Use Categories and Associated Curve Numbers

Description Average %

Impervious

Curve Number (CN) by

Hydrologic Soil Group Typical Land Uses

A B C D

Water 100 100 100 100 100 Water Bodies, Lakes, Ponds, Wetlands Grassland 5 35 60 73 79 Parks, Golf Courses, Greenways,

Grazed, Pasture, Hay Fields, Tall Grass, Un grazed Pasture

Estate 5 67 77 83 87 Cultivated Land, Row, crops, Broadcast, Legumes

Settlement 45 71 81 87 90 Multi-family, Apartments, Condos, Trailer Parks, Single-Family, Lot Size ¼ to 1 acre Single-Family, Lot Size 1 acre and Greater, Strip Commercial, Shopping, Convenience Stores, Light Industrial, Schools, Prisons, Treatment Plants, Gravel Parking, Quarries, Land Under Development.

Forest 5 37 60 73 80 Forest Litter and Brush adequately cover soil, Light Woods, Woods- Grass combination, Tree, Farms

Soil Distribution Map, Hydrology Soil Group (HSG) Map, and Thiessan Polygon

for Rain Fall Data.