Pemodelan Spasial Suhu Permukaan Lahan Menggunakan Algoritma XGBoost dan Cellular Automata-Markov Chain di Kawasan Perkotaan Yogyakarta dan Sekitarnya Tahun 1999-2029

PRAMITHA DEWI, Karen Slamet Hardjo, S.Si., M.Sc.

2025 | Tugas Akhir | D4 SISTEM INFORMASI GEOGRAFIS

Memanasnya temperatur permukaan semakin menunjukkan intensitas yang mengkhawatirkan. Anomali tersebut menyebabkan gletser dan salju di puncak tertinggi Indonesia menyusut drastis hingga Badan Meteorologi Klimatologi dan Geofisika (BMKG) mengangkat headline bertajuk Jelang Kepunahan Salju Abadi di Pegunungan Jayawijaya. Hal ini memantik pertanyaan; jika peningkatan temperatur global saat ini mampu menyasar dataran tinggi, lantas bagaimana nasib dataran rendah yang secara alami lebih panas? Penelitian ini bertujuan untuk mengetahui distribusi spasio-temporal suhu permukaan lahan (LST) di dataran rendah Kawasan Perkotaan Yogyakarta dan sekitarnya pada tahun 1999–2029.

Data historis LST tahun 1999–2024 pada musim kemarau (April–Oktober) diekstraksi dari dataset citra Landsat Surface Reflectance di Google Earth Engine menggunakan metode Single-Channel. Sementara data prediksi LST tahun 2029 diperoleh melalui pemodelan machine learning berbasis XGBoost di Google Colab. Fitur yang digunakan, yakni data elevasi dan slope serta data penutup lahan dan indeks (NDVI, NDBI, dan NDMI) pada periode yang sama. Data penutup lahan tahun 2029 diperoleh dari pemodelan Cellular Automata-Markov Chain, sedangkan indeks tahun 2029 dari pemodelan XGBoost. Hasil seluruh pemodelan divisualisasikan dalam web app Streamlit berbasis Python.

Model prediksi LST berbasis XGBoost menghasilkan akurasi yang sangat baik dengan nilai R2 > 0,96 dan RMSE 0,7994 serta MAE 0,6162 (toleransi error prediksi LST ± 2°C). Nilai LST diklasifikasikan ke dalam empat kelas, yaitu sangat rendah, rendah, sedang, dan tinggi. Pada tahun 1999–2009, kelas LST tinggi terpusat di Kota Yogyakarta dan Kapanewon Depok yang meluas ke kapanewon di sekitarnya mulai tahun 2014–2029. Kelas LST sangat rendah secara konsisten berada di Kapanewon Pakem, Turi, dan Cangkringan. Nilai rata-rata LST terendah di lokasi penelitian selama periode 1999–2029 adalah 34,53°C pada tahun 1999, sedangkan nilai LST tertinggi tercatat sebesar 38,35°C pada prediksi tahun 2029.

Global surface temperatures have been rising at a significant rate. This anomaly has caused glaciers and snow on Indonesia’s highest peak to shrink drastically, to the extent that the Badan Meteorologi Klimatologi dan Geofisika (BMKG) issued a headline warning of Imminent Extinction of Eternal Snow in the Jayawijaya Mountains. This phenomenon raises a critical question; if global temperature increases can now affect high-altitude regions, what will become of lowland areas that area naturally warmer? This study aims to examine the spatio-temporal distribution of Land Surface Temperature (LST) in the lowland areas of Yogyakarta Urban Area and its surroundings from 1999 to 2029.

Historical LST data from 1999 to 2024 during the dry season (April–October) were extracted using the Single-Channel method from Landsat Surface Reflectance imagery datasets in Google Earth Engine. Meanwhile, LST predictions for 2029 were obtained through XGBoost machine learning modeling in Google Colab. The features utilized included elevation and slope data; land cover; NDVI, NDBI, and NDMI for the corresponding periods. Land cover for 2029 was derived from Cellular Automata-Markov Chain modelling, while the 2029 indices were generated through XGBoost modeling. All modeling results were visualized in a web application built on Python-based Streamlit.

The XGBoost-based LST prediction model demonstrated excellent accuracy with an R2 exceeding 0,96, along with an RMSE of 0,7994 and MAE of 0,6162 (prediction error tolerance of LST ± 2°C). LST values were classified into four categories: very low, low, moderate, and high. From 1999 to 2009, high LST classes were concentrated in Yogyakarta City and Depok sub-district, expanding to surrounding sub-districts from 2014 to 2029. Very low LST classes were consistently located in Pakem, Turi, and Cangkringan sub-district. The lowest mean LST in the study area during the 1999–2029 period was 34,53°C in 1999, while the highest mean LST was recorded at 38,35°C in the 2029 prediction.

Kata Kunci : Suhu Permukaan Lahan, Google Earth Engine, XGBoost, Prediksi, Analisis Spasio-temporal