Geographia Technica, Vol 21(2), Special Issue: Artificial Intelligence Applications in Geography, 2026, pp. 19-34

PROJECTING LULC CHANGE (2025–2039) AND MODELING LANDSLIDE SUSCEPTIBILITY USING CELLULAR AUTOMATA–MARKOV CHAIN (CA–MC) IN THE KAMBANG WATERSHED, WEST SUMATRA, INDONESIA

Triyatno TRIYATNO , Lailatur RAHMI , Syafri ANWAR , Beni AULIA , Sumayyah Aimi Mohd NAJIB

ABSTRACT: This study aims to map and quantify Land Use/Land Cover (LULC) change across 2025, 2029, and 2039 to develop a Frequency Ratio (FR) based landslide hazard model, to integrate LULC projections with key environmental parameters, and to delineate priority hotspot areas for mitigation based on scenario-specific susceptibility patterns. This study applies a quantitative, spatial approach to examine how LULC change reshapes landslide risk. Supervised classification was performed to produce the 2025 LULC map, which serves as the base year for projection. Future LULC (2029, 2039) was simulated using Cellular Automata–Markov Chain (CA–MC) formulations. A Random Forest (RF) model was used to predict Landslide Potential (LP) based on environmental variables and distance to mapped landslides. The results show primary forest is projected to decline steadily across the 2025, 2029, and 2039 years. The LULC maps achieve overall accuracies of 0.92 (2025), 0.88 (2029), and 0.86 (2039). The high-hazard class increases from 4,273.29 ha (8.90%) in 2025 to 4,782.06 ha (9.96%) in 2029 (+508.77 ha; +1.06 percentage points), and to 5,041.89 ha (10.50%) in 2039 (+259.83 ha; +0.54 percentage points). Combined, the moderate + high categories grow from 10,271.25 ha (21.39%) in 2025 to 15,590.11 ha (32.47%) in 2029 (+5,318.86 ha; +11.08 percentage points), and to 16,155.00 ha (33.65%) in 2039 (+564.89 ha; +1.18 percentage points). Landslide-potential models yield Area Under the Curve (AUC) of 0.95 (2025), 0.87 (2029), and 0.84 (2039). The results indicate a clear and coherent trajectory linking LULC change to evolving landslide susceptibility in the Kambang watershed. To mitigate future risks, priority actions should include protecting steep-slope forests, deploying deep-rooted revegetation and targeted slope stabilization, upgrading surface–subsurface drainage on hillslopes and along access roads, and incorporating the 2025/2029/2039 susceptibility maps into risk-sensitive spatial planning.

Keywords: LULC; Landslide; Susceptibility; Cellular Automata–Markov Chain (CA–MC); Random Forest; Watershed.

Full article here