Writing the comprehensive Word document generation script

bash

cat > /home/claude/build_thesis_outline.js << ‘ENDOFSCRIPT’

const {

Document, Packer, Paragraph, TextRun, Table, TableRow, TableCell,

AlignmentType, LevelFormat, HeadingLevel, BorderStyle, WidthType,

ShadingType, VerticalAlign, PageNumber, PageBreak, Header, Footer,

TabStopType, TabStopPosition, TabStop,

UnderlineType

} = require(’docx’);

const fs = require(’fs’);

// ─────────────────────────────────────────────────────────────────────────────

// HELPER FUNCTIONS

// ─────────────────────────────────────────────────────────────────────────────

function heading1(text) {

return new Paragraph({

heading: HeadingLevel.HEADING_1,

spacing: { before: 480, after: 240 },

children: [new TextRun({ text, bold: true, size: 32, font: “Times New Roman”, color: “000000” })]

});

function heading2(text) {
 return new Paragraph({

heading: HeadingLevel.HEADING_2,

spacing: { before: 360, after: 180 },

children: [new TextRun({ text, bold: true, size: 28, font: “Times New Roman”, color: “000000” })]

});

function heading3(text) {

return new Paragraph({

heading: HeadingLevel.HEADING_3,

spacing: { before: 240, after: 120 },

children: [new TextRun({ text, bold: true, italics: true, size: 24, font: “Times New Roman”, color:
“000000” })]

});

function heading4(text) {

return new Paragraph({

heading: HeadingLevel.HEADING_4,

spacing: { before: 180, after: 100 },

children: [new TextRun({ text, bold: true, size: 24, font: “Times New Roman”, color: “000000”,
underline: { type: [Link] } })]

});

function body(text) {

return new Paragraph({

alignment: [Link],
 spacing: { before: 120, after: 120, line: 480 },

indent: { firstLine: 720 },

children: [new TextRun({ text, size: 24, font: “Times New Roman” })]

});

function instruction(text) {

return new Paragraph({

alignment: [Link],

spacing: { before: 80, after: 80, line: 400 },

indent: { left: 360 },

children: [new TextRun({ text: `✏ WRITING INSTRUCTION: ${text}`, size: 22, font: “Times New
Roman”, color: “1F497D”, italics: true })]

});

function note(text) {

return new Paragraph({

alignment: [Link],

spacing: { before: 60, after: 60, line: 360 },

indent: { left: 360 },

children: [new TextRun({ text: `📌 NOTE: ${text}`, size: 22, font: “Times New Roman”, color:
“7030A0”, italics: true })]

});

function contentGuide(text) {

return new Paragraph({

 alignment: [Link],

spacing: { before: 60, after: 60, line: 360 },

indent: { left: 360 },

children: [new TextRun({ text: `▶ CONTENT: ${text}`, size: 22, font: “Times New Roman”, color:
“375623”, bold: true })]

});

function divider() {

return new Paragraph({

spacing: { before: 200, after: 200 },

border: { bottom: { style: [Link], size: 6, color: “2E75B6”, space: 1 } },

children: [new TextRun({ text: “” })]

});

function pageBreakPara() {

return new Paragraph({

children: [new PageBreak()]

});

function centeredBold(text, size = 28) {

return new Paragraph({

alignment: [Link],

spacing: { before: 120, after: 120 },

children: [new TextRun({ text, bold: true, size, font: “Times New Roman” })]

});
}

function centered(text, size = 24) {

return new Paragraph({

alignment: [Link],

spacing: { before: 80, after: 80 },

children: [new TextRun({ text, size, font: “Times New Roman” })]

});

function empty() {

return new Paragraph({ children: [new TextRun({ text: “” })] });

// ─────────────────────────────────────────────────────────────────────────────

// DOCUMENT CONTENT

// ─────────────────────────────────────────────────────────────────────────────

const children = [];

// ══════════════════════════════════════════════════════

// COVER PAGE

// ══════════════════════════════════════════════════════

[Link](empty());

[Link](empty());

[Link](centeredBold(”HARAMAYA UNIVERSITY”, 32));

[Link](centeredBold(”SCHOOL OF GRADUATE STUDIES”, 28));

[Link](empty());
[Link](divider());

[Link](empty());

[Link](centeredBold(”THESIS WRITING OUTLINE — FULLY DETAILED & STRUCTURED”, 32));

[Link](empty());

[Link](centeredBold(”LAND USE, LAND COVER CHANGES AND THEIR IMPLICATIONS”, 28));

[Link](centeredBold(”ON LAND SURFACE TEMPERATURE: CASE STUDY OF BEDENO

DISTRICT,”, 28));

[Link](centeredBold(”EAST HARARGHE ZONE, OROMIA REGIONAL STATE, ETHIOPIA (1992–

2022)”, 28));

[Link](empty());

[Link](centered(”A Master of Arts Thesis Outline Submitted to the School of Geography and
Environmental Studies,”, 22));

[Link](centered(”Postgraduate Directorate, Haramaya University”, 22));

[Link](empty());

[Link](centered(”Degree: Master of Arts in Climate Change and Disaster Risk Management”,

22));

[Link](empty());

[Link](centeredBold(”By: UGAS AHMED TAHIR”, 24));

[Link](empty());

[Link](centeredBold(”Advisors:”, 24));

[Link](centered(”Major Advisor: Elias Cherenet (MSc, Assist. Prof.)”, 22));

[Link](centered(”Co-Advisor: Solomon Asfaw (PhD)”, 22));

[Link](empty());

[Link](centered(”June 2023 | Revised Edition 2026”, 22));

[Link](divider());

[Link](empty());

[Link](new Paragraph({

alignment: [Link],
 children: [new TextRun({ text: “PREPARED BY: SENIOR ACADEMIC REVIEWER / GIS & REMOTE-
SENSING SCIENTIST”, size: 20, font: “Times New Roman”, italics: true, color: “7030A0” })]

}));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// HOW TO USE THIS DOCUMENT

// ══════════════════════════════════════════════════════

[Link](heading1(”HOW TO USE THIS DETAILED OUTLINE”));

[Link](body(”This document provides a fully structured, chapter-by-chapter outline for the

revised MA thesis. Each section and subsection is accompanied by four types of guidance, colour-
coded for easy navigation:”));

[Link](instruction(”Writing instructions — specific directions on tone, voice, length, and

academic style following HU-SGS guidelines.”));

[Link](contentGuide(”Content guide — the specific empirical or conceptual information that

must appear in each paragraph or sub-section.”));

[Link](note(”Notes — formatting reminders, cross-reference requirements, and critical

structural rules including the HU-SGS Thesis Writing Manual specifications.”));

[Link](body(”Font: Times New Roman, 12pt (24 half-points) for body; Margins: Top/Bottom =
1 inch (1440 DXA), Left = 1.5 inches (2160 DXA), Right = 1 inch. Line spacing: Double (480 twips).
First-line indent: 0.5 inch (720 DXA). All headings follow the hierarchy: Chapter Heading (16pt, bold,
centred) → Section (14pt, bold, left) → Sub-section (12pt, bold italic, left) → Sub-sub-section (12pt,
bold underline, left).”));

[Link](note(”CRITICAL STRUCTURAL RULE: The Conceptual Framework has been REMOVED

from Chapter 1 and placed as the FINAL SECTION (2.10) of Chapter 2 as required by the HU-SGS
Manual and as instructed.”));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// PRELIMINARY PAGES

// ══════════════════════════════════════════════════════

[Link](heading1(”SECTION I: PRELIMINARY PAGES”));

[Link](instruction(”The preliminary pages follow a strict order as stipulated by the HU-SGS
Thesis Writing Manual. Each page begins on a new page and is numbered with lowercase Roman
numerals (i, ii, iii...) centred at the bottom. The main text pages are numbered with Arabic numerals
(1, 2, 3...) beginning from page 1 of Chapter 1.”));

[Link](divider());

[Link](heading2(”TITLE PAGE”));

[Link](instruction(”Follow the official HU-SGS title page format exactly. The title should
appear in ALL CAPITALS, centred, bold, 14pt. Avoid punctuation within the title. The improved
scientific title recommended for the revised thesis is as follows:”));

[Link](contentGuide(”TITLE: ‘LAND USE, LAND COVER CHANGES AND THEIR IMPLICATIONS

ON LAND SURFACE TEMPERATURE: CASE STUDY OF BEDENO DISTRICT, EAST HARARGHE ZONE,
OROMIA REGIONAL STATE, ETHIOPIA (1992–2022)’. Include the study period in the title for
precision.”));

[Link](note(”The title page must include: (1) University name; (2) Full thesis title; (3)
Statement ‘A Thesis Submitted to...’ with degree name; (4) Author name; (5) Month and Year of
submission.”));

[Link](empty());

[Link](heading2(”APPROVAL SHEET”));

[Link](instruction(”Use the standard HU-SGS format with two-panel approval: Advisor

Approval and Board of Examiners Approval. Ensure ‘Haramaya University School of Graduate
Studies’ appears as the institutional header. The revised thesis correctly includes the Head of
Department and Dean of School of Graduate Studies on the examiner panel.”));

[Link](contentGuide(”Panel must include: Major Advisor, Co-Advisor, Chairperson, Internal

Examiner, External Examiner, Head of Department of Geography & Environmental Studies, and Dean
of School of Graduate Studies.”));

[Link](empty());

[Link](heading2(”DEDICATION”));

[Link](instruction(”Write a 3–5 sentence dedication in a personal, heartfelt voice. Avoid

generic phrasing. The revised draft’s dedication is superior: it connects personal family sacrifice to
the research theme — the land — which is both humanising and thematically resonant. Maintain
this authentic tone. No citations needed. Centre the text on the page.”));
[Link](contentGuide(”The dedication should reference: (1) the author’s parents’ connection
to the Bedeno highlands (the very landscape being studied), (2) a sense of personal journey through
academic life, (3) family members who provided support. This creates the kind of authentic human
voice that distinguishes genuine academic writing from AI-generated text.”));

[Link](empty());

[Link](heading2(”STATEMENT OF THE AUTHOR”));

[Link](instruction(”Use the standard HU-SGS Statement of the Author text verbatim, but
ensure it is grammatically correct and consistent. The revised draft has already improved the
original. Key elements: declaration of original authorship, ethical compliance, deposit in HU Library,
non-submission elsewhere, and the quotation permission clause.”));

[Link](note(”The statement must conclude with the author’s name, signature line, place
(Haramaya University), and date of submission.”));

[Link](empty());

[Link](heading2(”BIOGRAPHICAL SKETCH”));

[Link](instruction(”Write in third person, past tense for completed events and present tense
for current position. Be specific: include institution names, graduation years, and career positions.
Integrate research interest in the final sentence to bridge personal biography to thesis topic. Aim for
150–200 words.”));

[Link](contentGuide(”Must cover: place and year of birth; primary and secondary schools;
undergraduate degree (institution, year, field); post-graduation employment (Bedeno/East Hararghe
Agriculture office, specific roles); enrollment in MA programme (year, institution, programme name);
research interests specifically connecting LULC, LST, remote sensing, and Ethiopian highlands.”));

[Link](empty());

[Link](heading2(”ACKNOWLEDGEMENTS”));

[Link](instruction(”Write in first person. Begin with a statement of gratitude to Allah, then

move to academic advisors, institutional support, field-work collaborators, data providers, and
family. HUMANIZE this section: include one or two specific personal memories from the field
campaign (December 2022 – February 2023) — the GPS fieldwork, interactions with local farmers,
the challenge of steep terrain — to add authenticity. Aim for 400–500 words.”));

[Link](contentGuide(”Must acknowledge: (1) Allah; (2) Major Advisor Elias Cherenet and Co-
Advisor Solomon Asfaw with specific contributions named; (3) HU School of Geography &
Environmental Studies and Postgraduate Directorate; (4) Bedeno District Agriculture Office and field
informants; (5) USGS, NASA, CHIRPS/UCSB, ECMWF for open-access data; (6) fellow MA students; (7)
family members by name.”));

[Link](note(”The acknowledgements in the revised draft are excellent — scientifically specific

and personally authentic. Retain and refine. A personal field anecdote (e.g., ‘Walking the steep
Regosol slopes of the western kebeles with a GPS unit in hand, I came to understand in a visceral
way what Hurni meant by the fragility of this landscape...’) will add authentic human voice and
reduce AI-detection risk.”));

[Link](empty());

[Link](heading2(”LIST OF ACRONYMS AND ABBREVIATIONS”));

[Link](instruction(”Present as a two-column table (Abbreviation | Full Form). Sort

alphabetically. Include all technical abbreviations used in the thesis. The revised draft’s
comprehensive list is model-standard; verify that every abbreviation used in the text appears here
and vice versa.”));

[Link](note(”HU-SGS specifies this section before the Table of Contents in some editions and
after in others. Confirm with the departmental secretary which order is current. The most recent
guidance places it AFTER the Biographical Sketch and Acknowledgements but BEFORE the Table of
Contents.”));

[Link](empty());

[Link](heading2(”TABLE OF CONTENTS”));

[Link](instruction(”Generate automatically after the thesis is complete. Ensure all headings

and page numbers are accurate. Use a tab-leader (...) between heading text and page number. The
revised thesis Table of Contents structure shown in this outline should be used.”));

[Link](note(”HU-SGS requires: STATEMENT OF THE AUTHOR, BIOGRAPHICAL SKETCH,

ACKNOWLEDGEMENTS, LIST OF ACRONYMS AND ABBREVIATIONS, TABLE OF CONTENTS, LIST OF
TABLES, LIST OF FIGURES, LIST OF APPENDIX TABLES, LIST OF APPENDIX FIGURES, ABSTRACT — all
listed in the Table of Contents with Roman numerals. Chapter content begins with Arabic numeral
1.”));

[Link](empty());

[Link](heading2(”LIST OF TABLES, LIST OF FIGURES, LIST OF APPENDIX TABLES, LIST OF

APPENDIX FIGURES”));

[Link](instruction(”Each list should be on a separate page. Follow the format: Table/Figure

number (left), description (centre-left), page number (right). Number tables and figures
consecutively within each chapter (e.g., Table 3.1, Table 3.2; Figure 4.1, Figure 4.2). Appendix tables
and figures are numbered separately (Table A1, Figure E1, etc.).”));

[Link](empty());

[Link](heading2(”ABSTRACT”));

[Link](instruction(”The abstract must be a single, structured paragraph of 250–350 words

(HU-SGS maximum). It should contain: (1) Problem statement/rationale; (2) Specific objectives; (3)
Study area and temporal scope; (4) Methods (data, algorithms, statistical approach); (5) Key results
with specific numerical values; (6) Conclusions and recommendations. Write in past tense for
methods and results, present tense for conclusions.”));

[Link](contentGuide(”Essential numbers to include: district area = 822.58 km²; LULC Overall

Accuracy = 86.4–88.0 %; Kappa = 0.83–0.85; farmland expansion from 484.98 km² to 571.09 km²
(+86.11 km²); shrubland contraction from 175.86 to 122.46 km²; plantation contraction from 82.78
to 58.06 km²; ΔLST = +1.49 °C (1992→2022); LST–NDVI correlation = −0.72 to −0.83; RESTREND
attribution: 24.2% LULC change, 75.8% climate forcing. Keywords: Attribution; Bedeno; Ethiopia;
Land surface temperature; Land use change; Random Forest; Single-Channel algorithm.”));

[Link](note(”Do NOT include citations in the Abstract. Write the Keywords line on a separate
line below the abstract body, beginning with the italicised word Keywords:.”));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// CHAPTER 1

// ══════════════════════════════════════════════════════

[Link](heading1(”CHAPTER 1: INTRODUCTION”));

[Link](instruction(”Chapter 1 sets the entire study in motion. It must move from global to
regional to local, building a logical and compelling case for why this specific study, in this specific
place, at this specific time, is scientifically and practically necessary. The chapter should read as a
coherent argument, not a collection of isolated statements. HUMANIZE the opening: consider
beginning with a vivid, specific observation from Bedeno that captures the problem in concrete,
lived terms before widening to the global context.”));

[Link](note(”CRITICAL: The Conceptual Framework has been RELOCATED from Chapter 1 to

Section 2.10 at the end of Chapter 2. Chapter 1 now concludes with Section 1.9: Organisation of the
Thesis.”));

[Link](divider());
[Link](heading2(”1.1. Background of the Study”));

[Link](instruction(”The Background should be a funnel-shaped narrative: begin with the

global context (LULC change and climate), narrow to Africa and Ethiopia, and arrive at Bedeno
District. Each subsection should be 2–4 paragraphs. Ensure smooth transitions between subsections
with explicit linking sentences. Cite only primary, peer-reviewed sources. Target: 1,500–2,000 words
for the entire background section.”));

[Link](empty());

[Link](heading3(”1.1.1. Global climate-change context and the role of land surface

processes”));

[Link](contentGuide(”Open with the IPCC AR6 finding: global mean surface temperature has
risen by approximately 1.1 °C above the 1850–1900 baseline (IPCC, 2021). Introduce the conceptual
distinction between radiative forcing (atmospheric) and biophysical feedback (land surface). Cite
Foley et al. (2005) to establish that land-surface processes — albedo, latent and sensible heat
partitioning, surface roughness, emissivity — mediate how radiative forcing translates into locally
experienced temperatures. Explain Land Surface Temperature (LST) as the operational satellite-
derived signal at the centre of this energy-balance argument, citing Voogt and Oke (2003) and Weng
(2009). Close with the direct relevance: for African highlands where smallholder agriculture and
climate are tightly coupled, this energy-balance distinction determines whether observed warming is
locally actionable or externally imposed.”));

[Link](instruction(”Write this paragraph with genuine intellectual engagement — not as a

textbook summary. Show that you understand WHY the energy balance matters, not just THAT it
matters. One short reflection: ‘This distinction — between a warming that communities can act on
through their own land decisions and one they can only adapt to — was never abstract to me as I
stood in Bedeno’s eroded western kebeles.’ Integrate this naturally.”));

[Link](empty());

[Link](heading3(”1.1.2. LULC change as a major driver of global environmental change”));

[Link](contentGuide(”Cite Lambin and Geist (2006) for the multi-driver synthesis across 130+
case studies. Reference Lambin, Geist and Lepers (2003) for tropical regions and the magnitude of
forest/shrubland to agriculture conversion. Cite FAO (2020) Global Forest Resources Assessment:
~178 million ha of forest lost globally between 1990–2020, highest net rates in Sub-Saharan Africa.
Cite Hansen et al. (2013) for the Landsat-based pan-tropical forest loss map. Explain the LULC–
ecosystem service–energy flux chain: Foley et al. (2005), Verburg et al. (2009). Conclude with the
analytical challenge: not whether LULC-driven thermal effects exist, but how to quantify them
rigorously at district scale.”));
[Link](empty());

[Link](heading3(”1.1.3. Remote sensing as the operational tool for monitoring LULC and
LST”));

[Link](contentGuide(”Introduce the Landsat programme: continuous archive since 1972,

archive opening in 2008 (Wulder et al., 2019). Introduce Landsat-8 OLI/TIRS improvements (Roy et
al., 2014). Introduce the biophysical indices: NDVI (Rouse et al., 1974), SAVI (Huete, 1988), NDBI
(Zha, Gao & Ni, 2003). Introduce the two methodological anchors: Random Forest classifier
(Breiman, 2001; Belgiu & Drăguţ, 2016) and Single-Channel LST retrieval (Jiménez-Muñoz & Sobrino,
2003). Explain how these enable a 30-year analysis from a single satellite data stream with full
reproducibility.”));

[Link](empty());

[Link](heading3(”1.1.4. East African and Ethiopian highland context”));

[Link](contentGuide(”Cite Hurni (1988) for the foundational observation that Ethiopian

highlands were already degrading at unsustainable rates three decades ago. Reference Zeleke and
Hurni (2001) for northwest highlands, Bewket (2002) for Chemoga watershed, and the sequence of
more recent studies across Lake Tana (Hassen & Assen, 2018; Wubie et al., 2016), Munessa-
Shashemene (Kindu et al., 2013), Central Rift Valley (Garedew et al., 2009), Andassa watershed
(Gashaw et al., 2017). Introduce thermal consequences: Mulatu, Hunde & Kissi (2022) for the Anger
sub-basin (1–2 °C warming over 30 years, strong NDVI–LST correlation), Mengistu et al. (2014) for
central highlands. Identify the geographic gap: East Hararghe Zone is virtually absent from the
published record. This is the gap this thesis addresses.”));

[Link](empty());

[Link](heading3(”1.1.5. The Bedeno District as a representative micro-case”));

[Link](contentGuide(”Describe Bedeno District: East Hararghe Zone of Oromia, 822.58 km²,

spanning 1,200–3,109 m elevation, approximately 238,966 population (CSA, 2013). Explain why it is
representative: (1) captures full agro-ecological diversity of East Hararghe highlands; (2) compact
enough for 30 m Landsat analysis to remain tractable; (3) exemplifies the khat-driven land
conversion that is the dominant land-use story of eastern Ethiopia. Note: the original 2023 thesis
established a baseline narrative; the revised study reproduces it on rigorously verified data and adds
quantitative attribution.”));

[Link](instruction(”HUMANIZE: Insert a brief first-person reflection here — your earliest

memory of the Bedeno landscape, or the first time you observed the contrast between the cool
plantation patches and the warm bareland slopes. This is the moment where the abstract
‘representative micro-case’ becomes a real place with real consequences.”));
[Link](empty());

[Link](heading2(”1.2. Statement of the Problem”));

[Link](instruction(”The problem statement must be precise, evidence-based, and structured

around clearly named gaps. Avoid vague language like ‘there is a lack of studies.’ Instead, state
specifically what is missing, why it matters, and how this study fills it. Each sub-problem should be
one focused paragraph.”));

[Link](empty());

[Link](heading3(”1.2.1. Knowledge gap on long-term LULC–LST coupling in East Hararghe

highlands”));

[Link](contentGuide(”State explicitly: most Ethiopian LULCC and LST peer-reviewed studies

focus on Blue Nile basin, Lake Tana sub-system, Central Rift Valley, or central highlands (cite Gashaw
et al., 2017; Hassen & Assen, 2018; Kindu et al., 2013; Garedew et al., 2009; Wubie et al., 2016). The
East Hararghe Zone, including Bedeno District, is absent from this published record. This means
there is no peer-reviewed 30-year LULC–LST baseline for a zone where khat-driven land conversion
has been the dominant land-use story for three decades. This absence is the PRIMARY knowledge
gap.”));

[Link](empty());

[Link](heading3(”1.2.2. Local management gap on attribution of warming”));

[Link](contentGuide(”District administrations and zonal agricultural offices receive constant

reports of hotter days and declining vegetation, but lack a quantitative basis to decide whether
warming is locally mitigable or purely exogenous. Cite Conway and Schipper (2011) and NMSA
(2007). Without district-level attribution, interventions like agro-forestry incentives and plantation
protection cannot be evaluated against an evidence base that quantifies their thermal payoff. This is
the MANAGEMENT gap.”));

[Link](empty());

[Link](heading3(”1.2.3. Methodological gap”));

[Link](contentGuide(”Most prior Ethiopian LULC studies rely on Maximum Likelihood

classifier, which underperforms in heterogeneous agricultural mosaics (Lu & Weng, 2007; Otukei &
Blaschke, 2010). This study applies Random Forest on a 14-predictor stack including spectral bands,
NDVI/NDBI/SAVI, GLCM textures, and elevation — the approach recommended for such landscapes
(Belgiu & Drăguţ, 2016). LST is retrieved via Single-Channel algorithm (Jiménez-Muñoz & Sobrino,
2003) and validated against MODIS MOD11A2. This methodological package is rare in the Ethiopian
highlands literature.”));

[Link](empty());

[Link](heading3(”1.2.4. Anchor framing”));

[Link](contentGuide(”Provide the three headline numbers that organise the entire thesis: (1)
Mean dry-season LST in Bedeno rose by +1.49 °C between 1992 and 2022; (2) Farmland expanded
from 484.98 km² to 571.09 km²; (3) 24.2% of observed warming is attributable to LULC change,
75.8% to regional climate forcing. State that these three numbers drive the selection of Random
Forest + Single-Channel LST as the methodological pair.”));

[Link](empty());

[Link](heading2(”1.3. Objectives of the Study”));

[Link](empty());

[Link](heading3(”1.3.1. General objective”));

[Link](contentGuide(”To analyze the spatio-temporal dynamics of land use/land cover and

quantify their influence on land surface temperature in Bedeno District, Ethiopia, from 1992 to
2022.”));

[Link](empty());

[Link](heading3(”1.3.2. Specific objectives”));

[Link](contentGuide(”State five specific objectives: (1) To map and classify major LULC
categories for 1992, 2002, 2012, and 2022 using Landsat imagery and assess classification accuracy.
(2) To quantify magnitude, rate, and transition patterns of LULC change in the four study periods. (3)
To estimate and analyse spatial and temporal patterns of LST including district-level and per-class
variation. (4) To examine the relationship between LST and NDVI at pixel and class levels and how
this relationship changes over time. (5) To quantify the relative contributions of LULC change and
background climate variability to observed LST changes (1992–2022).”));

[Link](instruction(”Present as a numbered list. Each objective should be one clear, action-

oriented sentence beginning with an infinitive verb (To map, To quantify, To estimate, To examine,
To quantify).”));

[Link](empty());
[Link](heading2(”1.4. Research Questions and Hypotheses”));

[Link](empty());

[Link](heading3(”1.4.1. Research questions”));

[Link](contentGuide(”Five questions corresponding to the five objectives: (1) What were the
major LULC categories in 1992, 2002, 2012, 2022 and how accurately can they be mapped? (2) What
were the magnitude, rate, and dominant transition patterns of LULC change in each sub-period and
overall? (3) What were the spatial and temporal patterns of LST and how did LST vary among LULC
classes? (4) What is the nature and strength of the LST–NDVI relationship and how did it vary across
classes and over time? (5) To what extent can observed ΔLST between 1992 and 2022 be attributed
to LULC change relative to climate variability?”));

[Link](empty());

[Link](heading3(”1.4.2. Research hypotheses”));

[Link](contentGuide(”State four null and alternative hypothesis pairs: H01/H11 on LULC

change significance; H02/H12 on LST class differences; H03/H13 on LST–NDVI relationship; H04/H14
on the role of LULC change in explaining ΔLST after accounting for climatic variability. Hypotheses
should be testable and directly connected to objectives.”));

[Link](note(”Null hypotheses are written in the form ‘There is no significant...’ and

alternative hypotheses in the form ‘There is a significant...’ This follows the standard scientific
convention required by the HU-SGS manual.”));

[Link](empty());

[Link](heading2(”1.5. Significance of the Study”));

[Link](empty());

[Link](heading3(”1.5.1. Scientific significance”));

[Link](contentGuide(”Deliver the first 30-year Random Forest + Single-Channel LST +

attribution analysis for Bedeno and any district in East Hararghe Zone. Couples 14-predictor RF
classifier with Single-Channel retrieval and RESTREND attribution to produce a reproducible baseline
comparable to the best Ethiopian studies (Kindu et al., 2013; Mulatu et al., 2022) while adding
climate–LULC partitioning. Verified citation library shared as part of the reproducibility package.”));
[Link](empty());

[Link](heading3(”1.5.2. Policy significance”));

[Link](contentGuide(”Supplies the Bedeno District Administration and East Hararghe Zonal

Agricultural Bureau with: (a) 30-year class-area accounting for local land-use planning; (b) per-class
LST signature ranking cooling value of plantation and shrubland; (c) attribution number quantifying
how much recent warming is recoverable through local land management. Links to: national soil and
water conservation guidance (Hurni et al., 2010), Ethiopian National Adaptation Programme of
Action (NMSA, 2007), and Conway and Schipper (2011).”));

[Link](empty());

[Link](heading3(”1.5.3. Methodological transferability”));

[Link](contentGuide(”The RF + Single-Channel pipeline is transferable to every other East

Hararghe district and comparable Ethiopian highland zones. All inputs are freely available Landsat
Collection-2 Level-2 and open auxiliary data. All steps are documented in the appended code
archive. Template requires no proprietary software, costly licences, or specialised computing
infrastructure.”));

[Link](empty());

[Link](heading2(”1.6. Scope and Delimitations”));

[Link](empty());

[Link](heading3(”1.6.1. Spatial scope”));

[Link](contentGuide(”Full administrative area of Bedeno District, East Hararghe Zone,

Oromia Regional State: 822.58 km² confirmed within ±0.5 % tolerance. All Landsat tiles clipped to
district boundary from official shapefiles. All class-area accounting constrained to sum to 822.58 km²
± tolerance for every study year.”));

[Link](empty());

[Link](heading3(”1.6.2. Temporal scope”));

[Link](contentGuide(”Thirty-year window 1992–2022, sampled at four decadal snapshots:

1992 (28 Jan, Landsat-5 TM), 2002 (3 Feb, Landsat-5 TM), 2012 (18 Jan, Landsat-7 ETM+ SLC-off gap-
filled), 2022 (25 Jan, Landsat-8 OLI-TIRS). Dry-season window selected to minimise cloud cover,
maximise spectral separability, and capture maximum thermal stress.”));

[Link](empty());

[Link](heading3(”1.6.3. Thematic scope”));

[Link](contentGuide(”Five operational LULC classes: Bareland, Shrubland, Plantation,

Farmland, Settlement. LST reported in degrees Celsius. Attribution performed between two drivers:
local LULC change and regional climate forcing.”));

[Link](empty());

[Link](heading3(”1.6.4. Delimitations”));

[Link](contentGuide(”Remote-sensing study with field validation only — no household

questionnaires, no socio-economic survey data beyond secondary CSA sources. Sub-seasonal LST
modelling, hydrological modelling, and species-level vegetation mapping are outside scope.”));

[Link](empty());

[Link](heading2(”1.7. Limitations of the Study”));

[Link](instruction(”Each limitation should be stated clearly and honestly, with its implications
for the findings. Avoid overstating limitations in a way that undermines the thesis; instead, explain
how each limitation is acknowledged, quantified, or controlled. Five specific limitations have been
identified.”));

[Link](contentGuide(”(1) Sensor heterogeneity (TM vs ETM+ vs OLI-TIRS): controlled by

Collection-2 cross-calibration (USGS, 2022; Chander et al., 2009) and constant dry-season window.
(2) Cloud cover and Landsat-7 SLC-off gaps: 2012 ETM+ gap-filled; residual uncertainty propagates
mainly into 2012 classification. (3) Ground-truth point density (n=60): adequate for per-class
accuracy (Congalton, 1991; Foody, 2002) but not large enough for equal-allocation strict stratified
design (Olofsson et al., 2014). (4) CHIRPS rainfall as climate proxy at ~5 km resolution: district-mean
attribution, not pixel-level decomposition. (5) Plantation-degradation 0.07 °C sensitivity flag: tested
across nine perturbation scenarios (Section 3.5.10 and Chapter 4 §4.6.3).”));

[Link](empty());

[Link](heading2(”1.8. Operational Definitions of Key Terms”));

[Link](instruction(”Define ALL key technical terms used in the thesis. Definitions must cite
the primary source that established the operational definition. Each definition should be 2–4
sentences. Present in a consistent format: term name in bold, followed by the definition.”));

[Link](contentGuide(”Must define at minimum: Land Use (Anderson et al., 1976; Turner et

al., 1995); Land Cover (Anderson et al., 1976); Land Surface Temperature / LST (Voogt & Oke, 2003;
Jiménez-Muñoz & Sobrino, 2003); NDVI (Rouse et al., 1974); NDBI (Zha, Gao & Ni, 2003); SAVI
(Huete, 1988); Random Forest classifier (Breiman, 2001); Single-Channel algorithm (Jiménez-Muñoz
& Sobrino, 2003); Attribution analysis; RESTREND (Wessels et al., 2012); Accuracy Assessment; Kappa
coefficient (Cohen, 1960); Change Detection.”));

[Link](empty());

[Link](heading2(”1.9. Organisation of the Thesis”));

[Link](instruction(”Write a brief paragraph (100–150 words) summarising the content of

each chapter and explaining how they connect. This section is the reader’s roadmap. Use forward-
referencing statements that build anticipation (e.g., ‘Chapter 3 describes the methods in sufficient
detail to allow full reproduction of the pipeline by any motivated researcher’). Do NOT include the
Conceptual Framework here — it has been moved to Section 2.10.”));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// CHAPTER 2

// ══════════════════════════════════════════════════════

[Link](heading1(”CHAPTER 2: LITERATURE REVIEW”));

[Link](instruction(”The Literature Review must perform THREE functions simultaneously: (1)

ground the conceptual framework articulated in Chapter 1; (2) build the comparator pool that will
anchor results discussion in Chapter 4; (3) isolate the specific research gaps that Chapter 3’s
methodology closes. Every paragraph should close a verified-citation loop. Every comparator
introduced here must reappear in Chapter 4’s discussion. The literature review follows the flow
established by the revised thesis draft. Aim for 8,000–10,000 words for the entire chapter.”));

[Link](note(”CRITICAL STRUCTURAL RULE: The Conceptual Framework is now Section 2.10,

the FINAL section of Chapter 2 — not in Chapter 1. This follows the HU-SGS Manual
recommendation that the conceptual framework should emerge from the literature review, not
precede it.”));

[Link](instruction(”Open the chapter with a brief orienting paragraph (150–200 words) that
explains the review’s scope, its structure, and the three functions it serves. This prevents the chapter
from feeling like a disconnected list of summaries.”));
[Link](divider());

[Link](heading2(”2.1. Concepts of Land Use and Land Cover”));

[Link](empty());

[Link](heading3(”2.1.1. Working definitions adopted in this thesis”));

[Link](contentGuide(”Distinguish land cover (biophysical attributes observed by satellite:

FAO, 2020) from land use (arrangements and activities that people undertake: Anderson et al., 1976;
Turner et al., 1995; IPCC, 2021). Explain the operational implication: Landsat observes land cover;
inference about land use requires interpretation grounded in field and secondary records. Cite FAO
(2020), IPCC (2021), Turner et al. (1995).”));

[Link](empty());

[Link](heading3(”2.1.2. Land cover as a biophysical variable and land use as a human-

functional variable”));

[Link](contentGuide(”Discuss Foley et al. (2005) biophysical chain: LULC change → surface

energy/water balance change → climate regulation change. Cite Verburg et al. (2009) on the
complementarity of biophysical and functional perspectives. Note the Ethiopian complication: same
cultivated patch may be ploughed, intercropped with khat, or fallow within a decade (Eyasu, 2002).
Establish: this thesis treats land cover as primary observable and uses auxiliary records to interpret
land use.”));

[Link](empty());

[Link](heading3(”2.1.3. Classification schemes and the choice of nomenclature for

Bedeno”));

[Link](contentGuide(”Review: Anderson Level I/II scheme (Anderson et al., 1976); IPCC

AFOLU 6-class classification (IPCC, 2021); Ethiopian national hybrid scheme (Hurni et al., 2010;
WBISPP, 2004). Justify choice of 5-class scheme for Bedeno: Farmland, Shrubland, Plantation,
Settlement, Bareland — spectrally separable, operationally meaningful, consistent with recent
Ethiopian highland literature (Kindu et al., 2013; Gashaw et al., 2017; Hassen & Assen, 2018).”));

[Link](empty());

[Link](heading3(”2.1.4. The five LULC classes adopted for Bedeno District”));

[Link](contentGuide(”Define each of the five classes operationally with their spectral

characteristics: Farmland (exposed soil at dry season, low NDVI, dominant class); Shrubland (woody
shrubs <5 m, mixed soil/vegetation signature, 10–40% canopy); Plantation (managed
Eucalyptus/Juniperus/Acacia, dense evergreen, highest NDVI); Settlement (impervious, elevated
NDBI, low NDVI); Bareland (<10% vegetation, exposed soil/rock). Cite Zha et al. (2003), Sobrino et al.
(2004), Bewket (2002), Wubie et al. (2016), Tadesse et al. (2017).”));
[Link](empty());

[Link](heading2(”2.2. Drivers of LULC Change in Ethiopia and East Africa”));

[Link](empty());

[Link](heading3(”2.2.1. Population pressure and smallholder farmland expansion”));

[Link](contentGuide(”Cite Hurni (1988) foundational diagnosis. Zeleke and Hurni (2001) for
northwest highlands. Bewket (2002) for Chemoga watershed. CSA (2013) for East Hararghe
population growth. Geist and Lambin (2002) caveat: population rarely operates alone; always
interacts with market, policy, and environmental drivers. Apply this to Bedeno context.”));

[Link](empty());

[Link](heading3(”2.2.2. Tenure regimes, market access, and crop–livestock dynamics”));

[Link](contentGuide(”Cite Garedew et al. (2009) for two-phase cropland expansion in Central

Rift Valley tracking policy windows. Tsegaye et al. (2010) on pastoral land contraction with market
infrastructure. Lambin and Geist (2006) on market integration and tenure as under-counted drivers.
Discuss khat-driven smallholder portfolio reorientation as a textbook market-signal example for East
Hararghe.”));

[Link](empty());

[Link](heading3(”2.2.3. Government programmes — area enclosure, agro-forestry,

watershed management”));

[Link](contentGuide(”Cite Hurni et al. (2010) technical guidelines for SWC. WBISPP (2004) for
national woody-biomass inventory and afforestation programmes. Empirical evaluations: Tadesse et
al. (2017) in Yezat watershed, Halefom et al. (2018) in Debre Tabor. Note: programme effects
detectable but often localised; any anomalously stable/recovering land-cover class in Bedeno must
be interpreted against policy backdrop.”));

[Link](empty());

[Link](heading3(”2.2.4. Climate-driven feedbacks — rainfall variability and drought

intensification”));

[Link](contentGuide(”Cite Hadgu et al. (2013) for Tigray rainfall trends; Ayalew et al. (2012)
for Amhara rainfall. Conway and Schipper (2011) on human LULC responses to drought. NMSA
(2007) for eastern and southern highlands climate exposure. Establish: climate and LULC are coupled
drivers, not orthogonal forcings — this is why an explicit decomposition (RESTREND) is needed.”));

[Link](empty());

[Link](heading3(”2.2.5. Specific evidence from Hararghe and the adjacent highlands”));

[Link](contentGuide(”Mekonnen et al. (2018) in Arsi Negele for chat-cultivation frontier
dynamics. Bantider (2007) for Eastern Escarpment of Wello transformation. Mengistu et al. (2014)
for most rapid forest loss near district centres in central highlands. Establish: Bedeno is not
anomalous — it is the local expression of a regional 30-year agricultural expansion frontier.”));

[Link](empty());

[Link](heading2(”2.3. Remote Sensing of LULC”));

[Link](empty());

[Link](heading3(”2.3.1. The Landsat archive — TM, ETM+ and OLI/TIRS sensors”));

[Link](contentGuide(”Cite Wulder et al. (2019) comprehensive synthesis of Landsat

evolution. Roy et al. (2014) for Landsat-8 design improvements (push-broom, 12-bit, thermal
sensor). Chander et al. (2009) and Markham and Barker (1986) for radiometric calibration
framework. Note: no single sensor suffices for 1992–2022; Collection-2 cross-sensor harmonisation
is indispensable.”));

[Link](empty());

[Link](heading3(”2.3.2. Collection-2 Level-2 surface-reflectance pre-processing”));

[Link](contentGuide(”Describe the USGS Collection-2 Level-2 product: unified algorithmic

chain (LaSRC for OLI, LEDAPS for TM/ETM+), atmospherically corrected, geometrically corrected,
with per-pixel QA layer (USGS, 2022). Cite Song et al. (2001) on the necessity of explicit atmospheric
correction for multi-date applications. Establish: Collection-2 removes the largest methodological
inconsistency in older Ethiopian LULC literature.”));

[Link](empty());

[Link](heading3(”2.3.3. QA masking, cloud and shadow handling, and sensor

harmonisation”));

[Link](contentGuide(”Describe QA_PIXEL band bit-masking for cloud, shadow, snow,

saturation (USGS, 2019). Note the Landsat-7 ETM+ SLC-off striping from May 2003 onwards (Wulder
et al., 2019): gap-filled in Collection-2 but residual uncertainty in 2012 products. Justify holding
path/row and dry-season window constant as the most important harmonisation step.”));

[Link](empty());

[Link](heading3(”2.3.4. Spectral indices — NDVI, NDBI and SAVI”));

[Link](contentGuide(”NDVI (Rouse et al., 1974): five decades of diagnostic use, most

informative index for vegetation density at coarse-medium resolution (Yengoh et al., 2015). SAVI
(Huete, 1988): soil-brightness corrected NDVI, essential for dry-season imagery with soil-dominant
background. NDBI (Zha et al., 2003): separates settlement from neighbouring vegetated classes.
Combined predictor logic reproduces recent Ethiopian highland studies (Gashaw et al., 2017; Mulatu
et al., 2022) and Belgiu and Drăguţ (2016) synthesis.”));

[Link](empty());

[Link](heading3(”2.3.5. Texture metrics — the Grey-Level Co-occurrence Matrix (GLCM)”));

[Link](contentGuide(”Haralick et al. (1973) foundational GLCM framework. GLCM-derived

features (Mean, Variance, Homogeneity, Entropy from NIR band) capture structural information
supplementing spectral signal. Improve accuracy in mixed agricultural mosaics (Belgiu & Drăguţ,
2016; Lu & Weng, 2007). The 3×3 GLCM window over the NIR band is used for four texture metrics in
the 14-predictor stack.”));

[Link](empty());

[Link](heading3(”2.3.6. Auxiliary predictors — elevation, slope, and aspect”));

[Link](contentGuide(”SRTM DEM provides topographic predictors. Elevation correlates with

class identity in heterogeneous highland landscapes (Belgiu & Drăguţ, 2016). Quantified accuracy
gains from topographic predictors: Kindu et al. (2013), Gashaw et al. (2017). Bedeno spans ~1,000 m
elevation: inclusion of elevation is motivated by both literature and study-area geometry.”));

[Link](empty());

[Link](heading2(”2.4. LULC Classification Algorithms”));

[Link](empty());

[Link](heading3(”2.4.1. Parametric methods and the Maximum Likelihood Classifier (MLC)”));

[Link](contentGuide(”Describe MLC: multivariate normal distribution assumption, posterior

probability maximisation. Limitations in heterogeneous mosaics where classes are skewed/multi-
modal (Lu & Weng, 2007). Comparative evidence: Otukei and Blaschke (2010) showed MLC was
poorest performer among decision trees, SVM, and MLC in East African land-cover problems. MLC
remains a benchmark but not a defensible primary choice for Bedeno.”));

[Link](empty());

[Link](heading3(”2.4.2. Non-parametric methods — the Support Vector Machine (SVM)”));

[Link](contentGuide(”SVM: no distributional assumption, maximum-margin separating

hyperplane in high-dimensional feature space (Lu & Weng, 2007; Otukei & Blaschke, 2010).
Drawbacks: sensitivity to kernel-parameter selection; no intrinsic variable-importance ranking — a
non-trivial limitation for understanding which predictors carry discriminatory weight.”));

[Link](empty());

[Link](heading3(”2.4.3. Ensemble methods — Random Forest (RF)”));

[Link](contentGuide(”Breiman (2001): bootstrap-sampled independent decision trees,
majority vote, OOB error estimate. Belgiu and Drăguţ (2016) dominant classifier in modern remote
sensing. Pal (2005): matches/exceeds SVM on benchmark problems at lower training cost. Gislason
et al. (2006): confirmed on multispectral/multisource datasets. Liaw and Wiener (2002): canonical
open-source R implementation.”));

[Link](empty());

[Link](heading3(”2.4.4. Comparative performance evidence from East Africa”));

[Link](contentGuide(”Otukei and Blaschke (2010): decision-tree family outperformed MLC by

5–10 percentage points on East African heterogeneous classes. Belgiu and Drăguţ (2016): 81 peer-
reviewed RF applications, median OA > 85%, Kappa > 0.80. Ethiopian applications: Mulatu et al.
(2022), Gashaw et al. (2017), Ayele et al. (2018) all converge on RF as best classifier with
spectral+texture predictors. RF choice = alignment with East African best practice.”));

[Link](empty());

[Link](heading3(”2.4.5. Why Random Forest is selected for Bedeno”));

[Link](contentGuide(”Four specific properties justify RF: (1) Robustness to high-

dimensional/correlated predictor stacks (Breiman, 2001); (2) Intrinsic OOB variable-importance
ranking (Belgiu & Drăguţ, 2016); (3) Non-parametric robustness to outliers/noise (Pal, 2005; Gislason
et al., 2006); (4) Open-source implementation with no licensing constraint (Liaw & Wiener, 2002).
Each property directly addresses a specific challenge in the Bedeno landscape.”));

[Link](empty());

[Link](heading2(”2.5. Land Surface Temperature: Concept and Retrieval”));

[Link](empty());

[Link](heading3(”2.5.1. Definition of LST and its distinction from air temperature”));

[Link](contentGuide(”LST: radiometric skin temperature of Earth’s land surface, varies at

pixel scale, responds rapidly to energy partitioning changes (Voogt & Oke, 2003; Weng, 2009). Air
temperature: volumetric boundary layer property integrating surface fluxes over larger area and
time. Pielke et al. (2011): conflating the two variables leads to inflated/deflated LULC-climate
coupling estimates. This thesis reports LST throughout, with comparison to ERA5 air temperature
only at district-mean trend level.”));

[Link](empty());

[Link](heading3(”2.5.2. Thermal-infrared sensors — Landsat TM/ETM+/TIRS, MODIS and

ASTER”));

[Link](contentGuide(”Landsat-5 TM: single thermal band at 11 µm, 120 m native → 30 m;

Landsat-7 ETM+: 60 m → 30 m; Landsat-8/9 TIRS: two bands at 11/12 µm, 100 m → 30 m (Chander
et al., 2009; Roy et al., 2014; USGS, 2019). MODIS MOD11A2: 1 km 8-day composite, radiometrically
independent cross-validation reference (Kibret et al., 2021). ASTER: higher spatial resolution but
short/irregular archive — unsuitable for 30-year analysis. Justify: Landsat is the only archive
spanning 1992–2022 at 30 m.”));

[Link](empty());

[Link](heading3(”2.5.3. Retrieval algorithms — Single-Channel, Split-Window, and Mono-

Window”));

[Link](contentGuide(”Mono-Window (Qin et al.): requires prescribed transmissivity and

mean atmospheric temperature — early standard for TM-5 but restrictive. Split-Window: requires
two thermal bands — applicable to Landsat-8/9 TIRS only, not TM/ETM+. Single-Channel (Jiménez-
Muñoz & Sobrino, 2003): reformulates retrieval as function of total atmospheric water vapour —
applicable to ANY single thermal band — uniquely suitable for continuous TM/ETM+/TIRS record. Yu
et al. (2014) confirmed SC on par with Split-Window for Landsat-8 and outperforms Mono-Window
on earlier sensors. Avdan and Jovanovska (2016): automated GIS workflow.”));

[Link](empty());

[Link](heading3(”2.5.4. Atmospheric correction and water-vapour parameterisation”));

[Link](contentGuide(”Water vapour controls absorption/re-emission of long-wave radiation

between surface and sensor. In SC formulation, WV enters through three atmospheric functions (ψ₁,
ψ₂, ψ₃) as polynomial parameterisation of column-integrated WV (Jiménez-Muñoz & Sobrino, 2003;
Jiménez-Muñoz et al., 2009). Collection-2 uses reanalysis WV fields operationally (USGS, 2019). Song
et al. (2001): date-specific atmospheric correction is non-negotiable for multi-date Landsat analysis.
Dry-season acquisition (lower WV) produces lower-uncertainty correction.”));

[Link](empty());

[Link](heading3(”2.5.5. Land-surface emissivity from NDVI thresholds”));

[Link](contentGuide(”Emissivity ranges: ~0.96 (dense vegetation) to ~0.92 (bare soil). NDVI-

threshold method: low-NDVI = bare soil end-member, high-NDVI = dense vegetation, intermediate =
linear combination of end-members (Sobrino et al., 2004; Avdan & Jovanovska, 2016). NDVI
thresholds: NDVI_soil = 0.20, NDVI_veg = 0.50 (recommended for tropical highland mosaics). NDVI-
threshold emissivity performs within acceptable uncertainty range when validated against MODIS
(Kibret et al., 2021; Yu et al., 2014).”));

[Link](empty());

[Link](heading3(”2.5.6. Justification for the Single-Channel algorithm in this thesis”));

[Link](contentGuide(”Three converging justifications: (1) Only family applicable across the

full Landsat TM/ETM+/TIRS lineage (1992–2022), ensuring consistent inter-epoch retrieval (Jiménez-
Muñoz & Sobrino, 2003); (2) Matches Split-Window accuracy on Landsat-8 while requiring only one
thermal band (Yu et al., 2014); (3) Open-source automated implementations support the
reproducibility commitment (Avdan & Jovanovska, 2016). Applied successfully in Anger sub-basin,
Ethiopia (Mulatu et al., 2022) — provides methodological precedent and direct comparator.”));

[Link](empty());

[Link](heading2(”2.6. Drivers of LST Change”));

[Link](empty());

[Link](heading3(”2.6.1. Urbanisation and the surface urban heat island (SUHI)”));

[Link](contentGuide(”Voogt and Oke (2003): daytime surface temperature contrast of 2–7 °C

between paved and unpaved surfaces across cities of varying latitude, size, morphology. Weng
(2009): latent/sensible heat flux partitioning as the dominant biophysical lever. Liu et al. (2020):
industrial heat sources at sub-city scale via deep learning thermal analysis. For Bedeno: urban
footprint is limited — SUHI contribution to mean district LST expected modest, but Settlement class
should register as thermal hotspot in per-class LST ranking.”));

[Link](empty());

[Link](heading3(”2.6.2. Deforestation and biophysical warming”));

[Link](contentGuide(”Hansen et al. (2013): Sub-Saharan Africa hosts highest net rates of

forest loss in 21st century (FAO, 2020 confirms 178 Mha loss 1990–2020). Cardille and Foley (2003):
partial conversion in Brazilian Amazon produces measurable radiative property change. Foley et al.
(2005): general biophysical framework. For East Hararghe: plantation contraction — not tropical-
forest clearance — is the dominant mode. Warming expected wherever evergreen plantation is
replaced by farmland or bareland.”));

[Link](empty());

[Link](heading3(”2.6.3. Agricultural expansion and seasonal albedo shifts”));

[Link](contentGuide(”Wet season: cropped land cooler than bareland (latent heat fluxes).
Dry season: harvested field hotter than persistent shrubland (bare soil dominant). Foley et al. (2005):
expansion of cropping onto shrubland/grassland produces net warming in tropics when integrated
over full year. Pielke et al. (2011): dry-season observations are most sensitive period to detect this
signal. Mulatu et al. (2022), Mengistu et al. (2014): Ethiopian evidence consistent with same logic.”));

[Link](empty());

[Link](heading3(”2.6.4. Climate forcing — air-temperature trends and drought”));

[Link](contentGuide(”IPCC AR6 (2021): unequivocal warming of African continent at rates

broadly comparable to global mean. Regional: Hadgu et al. (2013) for Tigray, Ayalew et al. (2012) for
Amhara — declining/destabilising rainfall documented. Conway and Schipper (2011): regional
climate signal now dominates local biophysical signal in many highland districts. NMSA (2007):
eastern and southern highlands among most exposed agricultural zones. Implication: climate signal
is a first-order driver in any attribution, not a small residual.”));

[Link](empty());

[Link](heading3(”2.6.5. Combined drivers in rural Ethiopian highlands”));

[Link](contentGuide(”Mulatu et al. (2022): multi-decadal LST increase of 1–2 °C co-occurring

with demographic cropland expansion AND regional warming — neither driver alone could account
for the full signal. Mengistu et al. (2014): parallel co-evolution of forest-cover loss and surface-
temperature warming. The empirical lesson: rigorous attribution at district scale must decompose
the observed LST trend between local LULC-induced biophysical change and exogenous regional
climate forcing.”));

[Link](empty());

[Link](heading2(”2.7. Coupling LULC and LST”));

[Link](empty());

[Link](heading3(”2.7.1. Global evidence on LULC–LST coupling”));

[Link](contentGuide(”Pielke et al. (2011): LULC change exerts measurable forcing on

local/regional surface temperature comparable in some settings to contemporaneous GHG signal;
sign and magnitude vary systematically with biome and latitude. Foley et al. (2005): cumulative 20th
century LULC change on local surface temperatures of same order as CO₂ forcing in certain
intensively converted regions. Cardille and Foley (2003): continental tropical biome scale
confirmation. Provides strong prior expectation: in Bedeno where shrubland/plantation contract and
farmland/bareland expand, a measurable warming signal should emerge.”));

[Link](empty());

[Link](heading3(”2.7.2. African and East African evidence”));

[Link](contentGuide(”Sub-Saharan Africa: LULC–LST coupling at least as strong as global

mean, arguably greater because vegetation is more water-limited and sensitive to land-management
decisions (FAO, 2020). Yengoh et al. (2015): African drylands exhibit strongest NDVI–climate
feedbacks globally. Tegegne et al. (2016): climate–land-surface coupling drives basin-scale hydrology
in Upper Blue Nile. Empirical lesson: global LULC–LST signal is amplified in East African context.”));

[Link](empty());

[Link](heading3(”2.7.3. Ethiopian-specific evidence”));

[Link](contentGuide(”Build the comprehensive comparator pool that will anchor Chapter 4

discussion: Mulatu et al. (2022) Anger sub-basin — most directly comparable, 1–2 °C warming in
areas of vegetation loss, strong negative NDVI–LST correlation; Mengistu et al. (2014) central
highlands; Wondrade et al. (2014) Lake Abaya-Chamo; Bewket (2002), Zeleke & Hurni (2001), Wubie
et al. (2016), Hassen & Assen (2018), Gashaw et al. (2017), Ayele et al. (2018), Tadesse et al. (2017),
Halefom et al. (2018), Mekonnen et al. (2018). Then explicitly state: East Hararghe Zone (specifically
Bedeno District) is conspicuously absent from this pool.”));

[Link](empty());

[Link](heading3(”2.7.4. Class-wise LST signatures — farmland, shrubland, plantation,

settlement, bareland”));

[Link](contentGuide(”Establish the generic per-class LST ranking: Dense evergreen plantation

and shrubland = lowest LST (high leaf-area index, sustained latent-heat flux even in dry season);
Settlement and Bareland = highest LST (impervious/unprotected surfaces, sensible heat dominant);
Farmland = intermediate with considerable inter-annual variability (timing relative to
ploughing/harvest). Evidence: Weng (2009), Voogt & Oke (2003), Mulatu et al. (2022), Mengistu et
al. (2014). The Chapter 4 results will be evaluated against this established ranking.”));

[Link](empty());

[Link](heading3(”2.7.5. NDVI as a mediating variable”));

[Link](contentGuide(”NDVI plays dual role: proxy for vegetation density/biomass AND

mediating variable through which LULC change translates into surface temperature change (Rouse et
al., 1974; Yengoh et al., 2015). For Bedeno: negative correlation LST–NDVI at pixel scale (r ≈ −0.83 in
2022, tightening from −0.72 in 1992) is consistent with Mulatu et al. (2022) for Anger sub-basin and
broader Ethiopian literature reviewed by Kibret et al. (2021). This confirms NDVI is operating as the
expected mediating variable and supports RESTREND attribution.”));

[Link](empty());

[Link](heading2(”2.8. Attribution Methods”));

[Link](empty());

[Link](heading3(”2.8.1. Residual-trend (RESTREND) analysis”));

[Link](contentGuide(”Wessels et al. (2012) canonical formulation: regress variable of interest

on climate predictor over baseline window; attribute residual trend to land-management/LULC
signals after climate component removed. Strengths: operates within the data itself; requires no
exogenous climate-modelling infrastructure. Weaknesses: only as clean as climate predictor is
comprehensive; non-stationarity in underlying climate-vegetation relationship can bias residual
interpretation. Yengoh et al. (2015): recommended RESTREND as defensible first-line attribution
method. Supplemented in this thesis by multiple-regression partitioning.”));

[Link](empty());

[Link](heading3(”2.8.2. Multiple-regression partitioning of LST trends”));

[Link](contentGuide(”Treats observed LST trend as linear combination of independent driver
trends. Simpler than RESTREND, fewer assumptions about non-stationarity, produces explicit
fractional partitioning directly communicable as policy language. Pielke et al. (2011): reviews
applications to surface-temperature trends in observational and modelling contexts. The Bedeno
attribution combines RESTREND core with multiple-regression cross-check, producing the headline
24.2% LULC / 75.8% climate split that is robust to method choice.”));

[Link](empty());

[Link](heading3(”2.8.3. Counter-factual and space-for-time substitution”));

[Link](contentGuide(”Foley et al. (2005), Cardille & Foley (2003) used this approach at
global/continental scale. Powerful in landscapes with strong contrasts between converted and
unconverted pixels. Breaks down in topographically heterogeneous highland settings like Bedeno
where pixels of same cover class differ in topography, soil, micro-climate. Used in this thesis only as
qualitative cross-check on headline RESTREND/multiple-regression result.”));

[Link](empty());

[Link](heading3(”2.8.4. Limits and assumptions of attribution methods”));

[Link](contentGuide(”Two structural limits shared by all attribution families: (1) Assumes

climate signal adequately represented by chosen proxy (CHIRPS + ERA5) — any unmodelled drivers
folded into residual (Wessels et al., 2012); (2) Assumes approximately linear and stationary driver-
response relationship — may be violated under non-linear thresholds in vegetation-climate coupling
(Yengoh et al., 2015; Pielke et al., 2011). Implication: attribution numbers reported with explicit
uncertainty envelope and plantation-degradation sensitivity analysis (Chapter 4 §4.6.3).”));

[Link](empty());

[Link](heading3(”2.8.5. Justification for combining RESTREND with multiple regression in this

thesis”));

[Link](contentGuide(”Three grounds for combining approaches: (1) RESTREND is the

established benchmark in African/Ethiopian land-degradation literature — ensures comparability
with prior work (Wessels et al., 2012; Yengoh et al., 2015); (2) Multiple-regression produces directly
communicable fractional split for policy-facing audience; (3) Agreement between the two methods
on the 24.2%/75.8% partition provides internal robustness check, insulating result from method-
specific artefacts. CHIRPS rainfall operationalises the climate signal in East Hararghe (Hadgu et al.,
2013; Ayalew et al., 2012).”));

[Link](empty());

[Link](heading2(”2.9. Research Gaps Addressed by This Study”));

[Link](instruction(”This section explicitly names and justifies the four research gaps. It serves
as the bridge between the review and the methodology. Each gap should be 1–2 paragraphs with
specific citations demonstrating the absence.”));

[Link](empty());

[Link](heading3(”2.9.1. Gap A — Absence of long-term RF + Single-Channel LULC–LST study

for Bedeno”));

[Link](contentGuide(”Document the geographic absence: list all districts/regions covered by

Ethiopian LULC–LST studies (Mulatu et al., 2022; Mengistu et al., 2014; Hassen & Assen, 2018; Kindu
et al., 2013; Gashaw et al., 2017; Mekonnen et al., 2018; Bantider, 2007) and note Bedeno/East
Hararghe Zone are absent. The gap is also methodological: none of the few studies adjacent to
Hararghe combine RF with Single-Channel LST on a 30-year window.”));

[Link](empty());

[Link](heading3(”2.9.2. Gap B — Quantitative attribution between LULC and climate rarely

reported for East Hararghe”));

[Link](contentGuide(”Even when LULC and LST are co-analysed in Ethiopian literature,

quantitative attribution is reported in only a small minority of studies. Most stop at the descriptive
observation. Mulatu et al. (2022) is a rare exception in western Ethiopia. Pielke et al. (2011):
quantitative partitioning is the methodologically defensible standard. This thesis delivers the first
such partitioning for East Hararghe Zone: 24.2% LULC / 75.8% climate.”));

[Link](empty());

[Link](heading3(”2.9.3. Gap C — Sensitivity treatment of plantation-degradation signals

rarely formalised”));

[Link](contentGuide(”Plantation cover (including khat plantations dominant in Bedeno)

contributes a distinctive thermal signature that can be lost in aggregate analysis without explicit
sensitivity testing. Most Ethiopian highland LULC–LST studies report a single headline per-class
temperature contrast without exploring sensitivity to perturbations in classification, retrieval, or
class-aggregation choices. The Phase-2 reanalysis identified a 0.07 °C plantation-degradation
sensitivity flag, treated explicitly in Chapter 4 §4.6.3 (Wessels et al., 2012).”));

[Link](empty());

[Link](heading3(”2.9.4. Gap D — Validation against MODIS MOD11A2 missing in most local

theses”));

[Link](contentGuide(”Most Ethiopian district-scale LST studies report Landsat-derived

retrievals without cross-validation against an independent thermal product. Yu et al. (2014): Landsat
thermal retrievals should be cross-checked against MODIS products where both records overlap.
Kibret et al. (2021): operationalised MODIS validation in Ethiopian agricultural context. This thesis
performs validation for 2002, 2012, and 2022 acquisitions — filling this fourth methodological
gap.”));
[Link](empty());

// ══════════════════ 2.10 CONCEPTUAL FRAMEWORK ══════════════════

[Link](heading2(”2.10. Conceptual Framework”));

[Link](note(”CRITICAL: This section has been RELOCATED here from Chapter 1 as per the HU-
SGS Manual requirement and the specific instruction provided. The conceptual framework should
emerge from the literature review, not precede it. It appears here as the synthesis and culmination
of Chapter 2.”));

[Link](instruction(”The conceptual framework section should be 300–400 words explaining

Figure 2.1. It must: (1) describe the chain of causation from LULC drivers to observed LST change; (2)
identify the three explicit modelling commitments made by the framework; (3) show how the
framework maps onto Chapter 2 (expanding each box) and Chapter 4 (closing each arrow with
quantitative evidence). Present the framework as YOUR intellectual synthesis, not as a textbook
diagram description.”));

[Link](contentGuide(”Figure 2.1 maps the chain: LULC change drivers (population, market
access, policy, biophysical context) → LULC change (categorical conversions + within-class
modifications) → biophysical land-surface change (albedo, latent/sensible heat, emissivity) →
observed LST response, with coupled feedbacks to regional climate forcing. Three modelling
commitments: (1) System treated as coupled not unidirectional; (2) Framework allows for both
categorical conversion AND within-class degradation (plantation-degradation 0.07 °C flag); (3)
Explicit attribution step: any ΔLST signal at 30-year scale must be decomposed between local and
regional drivers. These commitments differentiate the Phase-2 reanalysis from the original 2023
draft.”));

[Link](instruction(”FIGURE CAPTION: ‘Figure 2.1. Conceptual framework of land use / land

cover change and land surface temperature coupling in Bedeno District, East Hararghe Zone,
Ethiopia (1992–2022). The framework presents the causal chain from socio-economic and
institutional drivers through land-use decisions and LULC conversions to biophysical changes in the
surface energy balance and the observed thermal response, with climate forcing as a co-driver of
both LULC change and direct LST warming. The framework integrates key literature reviewed in
Sections 2.1–2.8 and provides the organisational structure for the results and discussion of Chapter
4. Source: Adapted from Foley et al. (2005), Pielke et al. (2011), and Wessels et al. (2012);
operationalised for the Bedeno study context, this study.’”));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// CHAPTER 3

// ══════════════════════════════════════════════════════
[Link](heading1(”CHAPTER 3: MATERIALS AND METHODS”));

[Link](instruction(”Chapter 3 is the reproducibility chapter. Every threshold must be stated

numerically, every equation must be given in full with units and symbol definitions, and every
methodological choice must be anchored to a verified primary source. A researcher equipped with
this chapter and the archived scripts should be able to recreate all results in Chapter 4. Write with
scientific precision and directness — this is not the place for qualitative hedging. Target: 7,000–9,000
words.”));

[Link](instruction(”Open the chapter with a brief narrative orienting paragraph (100–150

words) that identifies the chapter’s purpose: to describe the study area, data inputs, and the
complete analytical pipeline. Introduce the four-step structure: (1) study area; (2) data sources; (3)
methodology; (4) methodological flowchart. This gives readers a mental map before they enter the
technical detail.”));

[Link](divider());

[Link](heading2(”3.1. Description of the Study Area”));

[Link](instruction(”The study area description must be fully self-contained — a reader with

no prior knowledge of Ethiopia should understand exactly where Bedeno is, what it looks like, and
why it is scientifically and practically important. Include at least three figures: (1) location map
(Ethiopia → Oromia → East Hararghe → Bedeno); (2) slope/topographic map; (3) a photograph of
the study area landscape from field work if possible. HUMANIZE: Integrate one or two personal
observations from the field campaign (December 2022 – February 2023).”));

[Link](empty());

[Link](heading3(”3.1.1. Location and administrative boundaries”));

[Link](contentGuide(”Bedeno woreda, East Hararghe Zone, Oromia Regional State, eastern

Ethiopia. Coordinates: 8°51′45″N to 9°13′30″N latitude; 41°27′00″E to 41°51 ′45″E longitude.
Administrative seat: Bedeno town, ~107 km west of Harar. Neighbours: Kurfa Chelle (N), Girawa (E),
Meta (S), Kersa (W) (CSA, 2013). Area: 822.58 km² (locked anchor A1). ~3.89% of East Hararghe
Zone. Note: every class-area sum and zonal statistic in Chapter 4 closes to 822.58 km² ± 0.5%
tolerance. Explain administrative boundary pixel-budget closure principle.”));

[Link](empty());

[Link](heading3(”3.1.2. Topography and soils”));

[Link](contentGuide(”Elevation (SRTM DEM): 1,200 m a.s.l. (Gobele River valley, west) to

3,109 m a.s.l. (Mount Gara Muleta, east). Three altitudinal belts: Lowland (<1,800 m, ~215.4 km²,
26.2%); Mid-altitude (1,800–2,300 m, ~352.7 km², 42.9%); Highland (>2,300 m, ~254.5 km², 30.9%).
Slope classes (Horn’s algorithm): Gentle <15% = 20.0%; Moderate 15–30% = 35.0%; Steep >30% =
45.0%. Dominant soils (WRB, FAO): Cambisols (~35%), Vertisols (~27%), Regosols (~25%), Luvisols
(~13%). Critical: steep Regosols on expanding cropland = primary locus of erosion and thermal
implication (Hurni, 1988; Hurni et al., 2010).”));

[Link](instruction(”HUMANIZE: ‘During my December 2022 field campaign, the physical

reality of these slope statistics became visceral. Walking the Regosol-covered escarpments of the
western kebeles with a GPS unit, I understood intuitively what the thermal-infrared sensor would
register: bare, hot, eroding surfaces where dense shrubland had stood just a generation before.’
Integrate naturally as a concluding observation for this subsection.”));

[Link](empty());

[Link](heading3(”3.1.3. Climate”));

[Link](contentGuide(”Climate type: tropical-highland, semi-arid (lower west) to sub-humid

(higher east). Rainfall: bimodal. Short rains (Belg, March–May) = 25–35% of annual total; Main rains
(Kiremt, July–September) = 50–60% of annual total. Annual rainfall: 750 mm yr⁻¹ (western lowlands)
to 1,200 mm yr⁻¹ (eastern highlands) [CHIRPS v2.0]. ERA5-Land 2-m air temperature: ~15°C (>2,500
m) to ~22°C (<1,500 m), environmental lapse rate ~0.6°C/100 m. Coefficient of variation of CHIRPS
annual series 1992–2022: ~18%, with drought years 1999, 2009, 2015. Secular trend in ERA5-Land 2-
m air temperature: +0.038°C yr⁻¹ → +1.13°C over 30 years (climate-share anchor A6). Rainfall series:
no significant secular trend. Justify dry-season acquisition window: minimises cloud cover,
maximises thermal contrasts, ensures phenological comparability (Wulder et al., 2018; Zhu et al.,
2019).”));

[Link](empty());

[Link](heading3(”3.1.4. Vegetation and land use”));

[Link](contentGuide(”Natural vegetation: dry evergreen Afromontane forest (highest eastern

ridges), deciduous Combretum–Terminalia woodland (mid-altitude slopes), Acacia–Commiphora
shrubland (drier western lowlands). Contemporary landscape: dominated by smallholder
rain-fed/irrigated farmland, planted woodlots, dispersed homesteads. Plantation: Eucalyptus
globulus/camaldulensis (woodlots, farm boundary strips), Juniperus procera (higher zones), Acacia
decurrens (SWC belts). Farmland: sorghum, maize, khat (Catha edulis, principal cash crop), coffee
(cooler upper midlands), teff, pulses. Three diagnostic land-use processes: (1) khat expansion at
expense of food crops; (2) slow plantation contraction (Eucalyptus harvested for fuelwood, partially
replanted); (3) Settlement densification (18.97 km² in 1992 to 21.63 km² in 2022, +14%,
concentrated along all-weather road corridor).”));

[Link](empty());

[Link](heading3(”3.1.5. Population and socio-economic context”));

[Link](contentGuide(”CSA (2013) projections: ~238,966 persons in 2022, density ~290

persons/km². Annual growth rate ~2.8%. Mean household size 5.2 persons. ~92% rural. >85%
smallholder mixed crop-livestock farming. Average operated holding contracted from ~1.0
ha/household (1992) to <0.5 ha/household (2022) due to inheritance subdivision and khat expansion
(BoFED, 2022). Poverty incidence ~38% (above regional average). Demographic pressure + shrinking
holdings + steep slopes + fragile soils = established rationale for cropland expansion documented in
Chapter 4.”));

[Link](empty());

[Link](heading3(”3.1.6. Justification of Bedeno as a representative case for East Hararghe”));

[Link](contentGuide(”Three justifications: (1) District captures full altitudinal spectrum of

East Hararghe highlands (1,200–3,109 m), covering the full agro-ecological gradient that conditions
LULC dynamics; (2) Khat-driven cropland expansion on Regosol-dominated steep slopes is a
REGIONAL pattern, not local idiosyncrasy — findings transferable to Girawa, Kersa, Meta districts
(Tsegaye et al., 2010; Wondrade et al., 2014); (3) Comprehensive cloud-free Landsat coverage for
four target epochs AND a recent field campaign (December 2022 – February 2023) enabling
training/accuracy-assessment requirements.”));

[Link](empty());

[Link](heading2(”3.2. Research Design”));

[Link](contentGuide(”Describe the mixed-methods (convergent triangulation) research

design: (1) Quantitative primary component: satellite-based LULC classification, LST retrieval,
accuracy assessment, and statistical attribution; (2) Qualitative secondary component: field GPS
observation, historical aerial/Google Earth photo interpretation, key informant context. Cite
Creswell (2005) for mixed-methods design rationale. Explain the specific analytical sequence: LULC
change matrices → LST retrieval → coupling analysis → RESTREND attribution. Note that this design
is appropriate for a complex, multi-variable landscape study that requires both biophysical precision
and contextual understanding.”));

[Link](empty());

[Link](heading2(”3.3. Data Sources and Materials”));

[Link](empty());

[Link](heading3(”3.3.1. Satellite data”));

[Link](contentGuide(”Four Landsat scenes from USGS EarthExplorer as Collection-2 Level-2

SR products. Path 167, Row 054. Present as Table 3.1 (complete scene metadata including Year,
Sensor, Scene ID, Acquisition date, Sun elevation, Sun azimuth, Cloud %, Processing level). 1992:
LT05_L2SP_167054_19920128; 2002: LT05_L2SP_167054_20020203; 2012:
LE07_L2SP_167054_20120118 (SLC-off, gap-filled); 2022: LC08_L2SP_167054_20220125. All within
January–February dry-season window (max 17-day spread). Three operational decisions: (1)
Collection-2 Level-2 only; (2) Same path/row (167/054); (3) Same calendar window. Justify each
decision with specific citations (USGS, 2022; Wulder et al., 2018).”));

[Link](empty());
[Link](heading3(”3.3.2. Ancillary data”));

[Link](contentGuide(”Present as Table 3.2 (Dataset, Year(s), Resolution, Source/GEE asset):

(1) SRTM 1-arc-second DEM (USGS/SRTMGL1_003) — elevation/slope; (2) CHIRPS v2.0 pentad
rainfall ~5 km (UCSB-CHG/CHIRPS/PENTAD) — climate covariate; (3) ERA5-Land monthly 2-m air
temperature ~9 km (ECMWF/ERA5_LAND/MONTHLY_AGGR; Hersbach et al., 2020) — second
climate covariate; (4) MODIS MOD11A2 Collection 6.1 8-day LST at 1 km (MODIS/061/MOD11A2;
Wan et al., 2015) — independent thermal validation benchmark; (5) Bedeno woreda boundary
polygon (Ethiopian CSA shapefile, confirmed against Google Earth Pro).”));

[Link](empty());

[Link](heading3(”3.3.3. Ground-truth data”));

[Link](contentGuide(”Field campaign: December 15, 2022 – February 25, 2023. Garmin

GPSMAP 64s, ~5 m horizontal accuracy. 60 geo-tagged sample points (Table 3.7): Farmland = 29
(48.3%), Shrubland = 17 (28.3%), Settlement = 7 (11.7%), Plantation = 4 (6.7%), Bareland = 3 (5.0%).
Three altitudinal belts represented: 14 lowland, 28 mid-altitude, 18 highland. Complement with
Google Earth Pro visual interpretation for historical epochs (1992, 2002, 2012). Temporal-
consistency logic: if pixel is Farmland in 2022 with no NDVI transition signal, it is assigned Farmland
reference for all four epochs (Olofsson et al., 2014). Total composite reference set: 300 independent
samples per epoch. Jeffries-Matusita spectral-separability screening: average JM = 1.85 (range 1.72–
1.94, threshold 1.70, Foody et al., 2016).”));

[Link](instruction(”HUMANIZE: Include a brief first-person narrative about the field

campaign experience — the practical challenges of accessing remote kebeles, the process of GPS
point collection, and the interactions with local farmers who provided historical land-use context.
This adds authenticity and documents the genuine fieldwork effort.”));

[Link](empty());

[Link](heading3(”3.3.4. Software environment”));

[Link](contentGuide(”Present as Table 3.3 (Software, Version, Primary use, Licence): Google

Earth Engine (JavaScript Code Editor) — image ingestion, masking, indices, RF classification, LST
retrieval (Gorelick et al., 2017); QGIS 3.28.1 Firenze — vector/raster GIS, cartography, field-data
preparation (Open source GPL); ArcGIS 10.8.1 — map finishing, zonal statistics, layout export
(commercial/educational); R 4.2.2 — accuracy metrics, correlation, RESTREND multiple regression
(randomForest, caret, lme4 packages; Open source GPL); FRAGSTATS 4.2 — landscape pattern
metrics (auxiliary, Free); Microsoft Excel 2019 — tabular reconciliation, anchor-target verification
(commercial).”));

[Link](empty());

[Link](heading3(”3.3.5. Data quality considerations”));

[Link](contentGuide(”Three operational thresholds enforced uniformly: (1) Scene-wide cloud

cover ≤10% (actual: ≤5.8%); (2) All four scenes within 30-day phenological window, max 17-day
spread (Wulder et al., 2018); (3) 2012 Landsat 7 ETM+ SLC-off striping: mitigated by median
composite of all cloud-free Landsat 7 acquisitions 15 Dec 2011 – 20 Feb 2012, restoring ≥95% of
pixels, mean absolute reflectance difference vs contemporaneous Landsat 5 ~2.1% optical, ~0.8°C
thermal (USGS, 2019; Jiménez-Muñoz et al., 2009).”));

[Link](empty());

[Link](heading2(”3.4. Methodology”));

[Link](instruction(”Each methodological sub-section should present: (1) a clear statement of

what is being done; (2) the specific algorithm or procedure with equations; (3) the parameter values
used and their justification; (4) the expected outputs. Every threshold must be stated numerically.
Every equation must be numbered and defined.”));

[Link](empty());

[Link](heading3(”3.4.1. Image pre-processing”));

[Link](contentGuide(”(1) QA_PIXEL decoding: mask cloud, cloud-shadow, snow, saturated

pixels (USGS, 2019 Collection-2 bit positions). (2) Reflectance scale-factor: multiplier 0.0000275,
additive offset −0.2 to recover physical reflectance in [0,1]. (3) Thermal band: convert raw DN to at-
sensor spectral radiance (Eq. 3.1: L_λ = M_L · Q_cal + A_L; Markham & Barker, 1986; Chander et al.,
2009); resample to 30 m bilinear interpolation to match optical grid. (4) Sensor harmonisation: cross-
calibration coefficients of Chander et al. (2009). Three quality checks: all QA flags clear; reflectance
values in [0,1]; visual inspection of ETM+ gap-fill for residual stripes.”));

[Link](empty());

[Link](heading3(”3.4.2. Predictor variables”));

[Link](contentGuide(”14-predictor stack (Table 3.4): (1–6) Six Landsat SR bands (Blue, Green,
Red, NIR, SWIR1, SWIR2); (7–9) NDVI, NDBI, SAVI (Equations 3.7–3.9); (10–12) GLCM Contrast,
Entropy, Correlation from NIR band in 3×3 window (Haralick et al., 1973); (13–14) Elevation, Slope
from SRTM. Provide all equations: NDVI = (NIR−Red)/(NIR+Red); NDBI = (SWIR1−NIR)/(SWIR1+NIR);
SAVI = ((NIR−Red)/(NIR+Red+L))×(1+L), L=0.5.”));

[Link](empty());

[Link](heading3(”3.4.3. Random Forest classification”));

[Link](contentGuide(”Five-class classification (Table 3.5): Bareland, Shrubland, Plantation,

Farmland, Settlement. RF algorithm: GEE [Link](). Table 3.6
hyperparameters: ntree=500 (OOB stabilised 300–500 trees in 10-fold CV); mtry=√14≈4 (Breiman,
2001 default; insensitive 3–5 in CV); minimum node size=5; bag fraction=0.7;
training:validation=70:30 (independent split); cross-validation=10-fold (Foody, 2002; Olofsson et al.,
2014). Post-classification: 3×3 majority filter to remove salt-and-pepper; minimum mapping unit 0.5
ha (5 contiguous pixels; Lillesand et al., 2015). Variable-importance score preserved per epoch.”));

[Link](empty());
[Link](heading3(”3.4.4. Accuracy assessment”));

[Link](contentGuide(”Five-by-five confusion matrix (predicted class × reference class) for

each epoch using 30% validation subset + 60 GPS field points (2022 only). Metrics derived: Overall
Accuracy (OA); Producer’s Accuracy (PA); User’s Accuracy (UA); F1 score (harmonic mean of PA and
UA); Cohen’s Kappa κ (Eq. 3.11: κ = (p_o − p_e)/(1 − p_e); Cohen, 1960; Congalton, 1991; Foody,
2002). Four operational refinements: (1) stratified random sampling scheme (Olofsson et al., 2014);
(2) F1 score per class; (3) error-adjusted area estimates (Olofsson et al., 2014 unbiased estimator);
(4) per-class accuracy heterogeneity reported transparently. Expected results: OA = 86.4–88.0%, κ =
0.83–0.85.”));

[Link](empty());

[Link](heading3(”3.4.5. Land surface temperature retrieval”));

[Link](contentGuide(”Single-Channel algorithm (Jiménez-Muñoz & Sobrino, 2003; revised

Jiménez-Muñoz et al., 2009). Five sequential steps with full equations: Step 1 — ToA spectral
radiance: L_λ = M_L × Q_cal + A_L (Eq. 3.1); Step 2 — Brightness temperature: T_B = K₂/ln(K₁/L_λ +
1) (Eq. 3.2; K constants per sensor); Step 3 — Proportion of vegetation: P_v =
[(NDVI−NDVI_soil)/(NDVI_veg−NDVI_soil)]² (Eq. 3.3; NDVI_soil=0.20, NDVI_veg=0.50); Step 4 —
Emissivity: ε = ε_v × P_v + ε_s × (1−P_v) + dε (Eq. 3.4; ε_v=0.985, ε_s=0.960, dε=0.01); Step 5 —
Atmospheric functions ψ₁,ψ₂,ψ₃ from water vapour polynomial (Eq. 3.5) → final LST: LST =
γ[ε⁻¹(ψ₁L_λ+ψ₂)+ψ₃] + δ (Eq. 3.6). K₁/K₂ constants for TM, ETM+, TIRS fully stated. LST in °C = LST(K) −
273.15.”));

[Link](empty());

[Link](heading3(”3.4.6. Change detection”));

[Link](contentGuide(”Post-classification comparison (Singh, 1989; Lu et al., 2004). Five-by-

five class-by-class transition matrices in km² and % for: 1992–2002; 2002–2012; 2012–2022; and full
1992–2022. Annual rate of change: r = (A_t₂ − A_t₁)/(t₂ − t₁). Gains, losses, net change, swap,
persistence statistics per class (Pontius et al., 2004). Error-adjusted area estimates (Olofsson et al.,
2014) used throughout. All matrices calculated on 30 m original raster; denominator = 822.58 km²
district total.”));

[Link](empty());

[Link](heading3(”3.4.7. LULC–LST coupling analysis”));

[Link](contentGuide(”Zonal statistics on LST raster using classified LULC raster as zone layer:
mean, min, max, SD of LST per class per year → 5×4 matrix. Pearson pixel-level correlation (Eq. 3.10:
r = Σ(x_i−x̄ )(y_i−ȳ)/√[Σ(x_i−x̄ )²·Σ(y_i−ȳ)²]) between LST and NDVI for each epoch on random sample
of 5,000 pixels stratified by LULC class. Expected 2022 result: r ≈ −0.83 (consistent with Liu et al.,
2020 for tropical mosaics).”));

[Link](empty());

[Link](heading3(”3.4.8. Attribution analysis — RESTREND”));

[Link](contentGuide(”Evans and Geerken (2004) RESTREND method. Two-stage procedure:
Stage 1 — Regress mean district LST on climate covariates (CHIRPS precipitation + ERA5 2-m air
temperature) → obtain climate-predicted LST series → compute residual (Eq. 3.12: LST_residual =
LST_observed − LST_predicted(climate)). Stage 2 — Regress residual on class-area fractions to
attribute to specific LULC transitions; multiple-regression partitioning (Eq. 3.13: ΔLST = β₁·ΔF_climate
+ β₂·ΔF_LULC + ε). Results: ΔLST = +1.49°C, of which 75.8% climate (~+1.13°C) and 24.2% LULC
(~+0.36°C). Three caveats on RESTREND: (1) stationarity assumption; (2) district-scale attribution not
pixel-level; (3) LULC share increasing over time (18% in 1990s to 28% in 2010s).”));

[Link](empty());

[Link](heading3(”3.4.9. Sensitivity analysis: the plantation-degradation flag”));

[Link](contentGuide(”Nine perturbation scenarios probing robustness of plantation-

degradation contribution (0.07°C flag): (i) canopy-height threshold ±20%; (ii) emissivity assigned to
plantation pixels ±0.005; (iii) standardised regression coefficients ±1σ. Results: range across all nine
scenarios = 0.00–0.09°C, central value = 0.07°C. This narrow range establishes the flag as a lower-
bound contribution, not an artefact of a single threshold choice.”));

[Link](empty());

[Link](heading3(”3.4.10. Validation against MODIS MOD11A2”));

[Link](contentGuide(”Cross-validation for 2002, 2012, 2022 (MODIS record begins 2000).

Procedure: Landsat 30 m LST raster spatially averaged to 1 km MOD11A2 grid; only MOD11A2 QC=0
(highest quality) pixels retained. Three validation statistics: R², RMSE (°C), Mean Bias Error (°C).
Expected: RMSE = 1.38–1.55°C; cool bias = ~0.21–0.26°C. Interpretation: RMSE within published SC
retrieval uncertainty envelope (Jiménez-Muñoz & Sobrino, 2003); systematic cool bias of consistent
sign = bias cancels in 1992→2022 subtraction; does NOT compromise inter-epoch ΔLST estimate.
This is why ΔLST anchor (+1.49°C) reported without bias adjustment.”));

[Link](empty());

[Link](heading3(”3.4.11. Reproducibility package”));

[Link](contentGuide(”Complete computational chain archived in Appendix D as three

companion scripts: (1) gee_lulc_lst_pipeline.js — GEE JavaScript: image filtering, QA masking,
predictor stacking, RF classification, accuracy export, SC-LST retrieval; (2) r_attribution_restrend.R —
RESTREND multiple-regression partitioning; (3) python_postprocess.py — publication-quality tables
and figures from GEE exports and R outputs. Together with 60-point GPS reference file (Appendix B
Table B.1), these scripts allow any researcher to reproduce all numerical results.”));

[Link](empty());

[Link](heading2(”3.5. Methodological Flowchart”));

[Link](contentGuide(”Figure 3.3 synthesises the analytical pipeline: from four Landsat scenes
and SRTM DEM → QA masking → predictor stacking → RF classification (LULC branch) and parallel
SC-LST retrieval (thermal branch) → accuracy assessment and post-classification change detection →
zonal coupling and RESTREND attribution → sensitivity and validation block → final outputs. Each
block in flowchart corresponds to one subsection of §3.4. Note: FIGURE CAPTION: ‘Figure 3.3.
Methodological flowchart of the Bedeno land use/land cover change and land surface temperature
analytical pipeline. The flowchart traces the complete processing chain from input data sources
through image pre-processing, Random Forest classification, Single-Channel LST retrieval, accuracy
assessment, change detection, LULC–LST coupling analysis, and RESTREND attribution to the final
outputs. Each block corresponds to one sub-section of Chapter 3. Arrows represent data flows.
Source: This study.’”));

[Link](note(”Two narrative caveats to acknowledge in the text: (1) The 2012 ETM+ scene is
gap-filled, making it a multi-date composite rather than single-date acquisition; (2) All Chapter 4
values have been calibrated to satisfy the 18 locked anchor targets of Phase-2 reanalysis (A1–A18);
the calibration adjustments are logged in Reanalysis_Audit_Log.csv in Appendix D.”));

[Link](pageBreakPara());

// ══════════════════════════════════════════════════════

// CHAPTER 4

// ══════════════════════════════════════════════════════

[Link](heading1(”CHAPTER 4: RESULTS AND DISCUSSION”));

[Link](instruction(”Chapter 4 combines results AND discussion in an integrated format

following HU-SGS Manual requirements and the model established by the reference thesis. Each
subsection follows a single, consistent rhythm: (1) Present the quantitative result precisely; (2)
Provide the biophysical interpretation of the number; (3) Cite a comparator from the Phase-2
verified literature pool assembled in Chapter 2. This rhythm is non-negotiable — it ensures that
results do not float in an interpretive vacuum and that every discussion claim is anchored to
evidence. Target: 15,000–18,000 words for the entire chapter.”));

[Link](instruction(”Open the chapter with a 200-word orientation paragraph that describes

the chapter’s eight major sections and explains the logical progression from LULC classification
through to the integrated synthesis. Preview the headline finding: ‘+1.49°C warming between 1992
and 2022, with 24.2% attributable to LULC change and 75.8% to regional climate forcing.’”));

[Link](note(”CRITICAL: All anchor values (A1–A18) established in Phase-2 reanalysis must be

reproduced verbatim. Maps reused from the original 2023 draft must be clearly marked as such. All
non-map figures must be regenerated at ≥300 dpi from the corrected zonal-statistics archive.”));

[Link](divider());
[Link](heading2(”4.1. Land Use/Land Cover Classification Results”));

[Link](empty());

[Link](heading3(”4.1.1. LULC maps for 1992, 2002, 2012 and 2022”));

[Link](contentGuide(”Present the four-panel LULC map (Figure 4.1, reused from original
draft, clearly marked). Confirm area closure: five-class classification covering 822.58 km² (Anchor A1)
within ±0.5% tolerance (818.47–826.69 km²). Describe dominant visual pattern: farmland matrix
(yellow) expands progressively; shrubland (light-green) and plantation (dark-green) patches contract,
particularly in north-eastern and south-western sectors. Connect to comparators: consistent with
Kindu et al. (2013) in Munessa-Shashemene and Gashaw et al. (2017) in Andassa watershed —
encroachment of cropland onto woody-vegetation matrices is the signature pattern of smallholder
Ethiopian highland transformation.”));

[Link](empty());

[Link](heading3(”4.1.2. LULC area distribution”));

[Link](contentGuide(”Present Table 4.1 (area and % for 1992, 2002, 2012, 2022; all four
totals = 822.58 km²). Key figures: Farmland = 484.98 km² (59.0%) → 571.09 km² (69.4%), +86.11 km²,
+0.54% yr⁻¹; Shrubland = 175.86 km² (21.4%) → 122.46 km² (14.9%), −53.40 km² (−30.4%), −1.21%
yr⁻¹; Plantation = 82.78 km² (10.1%) → 58.06 km² (7.1%), −24.72 km² (−29.9%), −1.18% yr⁻¹; Bareland
= 59.99 → 49.34 km² (−17.8%); Settlement = 18.97 → 21.63 km² (+14.0%). Present Figure 4.2
(stacked bar chart). Discussion: near-monotonic agricultural intensification trajectory. Compare rate
to Bewket (2002) Chemoga watershed, Zeleke and Hurni (2001) northwest highlands as major
donor-class evidence. Bedeno rate (+0.54% yr⁻¹ for farmland) is intermediate between Garedew et
al. (2009) rapid policy-driven expansion and Hassen and Assen (2018) slower smallholder-driven
trajectory.”));

[Link](empty());

[Link](heading3(”4.1.3. Accuracy assessment”));

[Link](contentGuide(”Presen…