Prediction of future Land Use/Land Cover (LULC) changes using random forest model in google earth engine (2035–2045)

Introduction

Land Use/Land Cover (LULC) changes play a major role in global environmental change, affecting local weather, water systems, and ecosystem health (Gaur and Singh, 2023; Zafar et al., 2025) [8,25] . These changes are mainly driven by rapid urban growth and increased agricultural activities, which directly impact surface runoff and water quality in watersheds (Amini et al., 2022; Loukika et al., 2023) [2,12] . In regions like Maharashtra, India, monitoring these changes is essential for sustainable land use planning and protecting important water resources (Patil et al., 2025) [17] . Satellite remote sensing programs such as Landsat and Sentinel provide reliable, long-term data that help track changes in land use over time (Alshehri et al., 2025; Wulder et al., 2016; Shinde and Nandgude, 2024) [1,24,21] . However, traditional methods of analyzing this data can be timeconsuming and require high computational power, making them difficult to use on standard computers (Gorelick et al., 2017; Alshehri et al., 2025) [9,1] . Google Earth Engine (GEE) has transformed this process by offering a cloudbased platform that allows large-scale analysis of satellite data quickly and efficiently (Gorelick et al., 2017; Tassi and Vizzari, 2020) [9,22] . Machine learning techniques like Random Forest (RF) and Support Vector Machines (SVM) are more effective than traditional methods because they can handle complex patterns and noisy data (Breiman, 2001; Kipterer et al., 2025) [6,11] . Among these, RF is widely used in GEE due to its reliability and high accuracy across different landscapes (Alshehri et al., 2025; Bhungeni et al., 2025) [1,4] . For predicting future land use changes, models such as Cellular Automata-Markov (CA-Markov) are useful as they simulate both spatial and temporal changes, helping in planning and decision-making (Osman et al., 2023; Vahid and Aly, 2025) [16, 23] . The Morna watershed in Satara is experiencing increasing human pressure, but there is limited research that combines long-term machine learning-based analysis with future predictions (Ingle, 2022) [10] [5,20] . [25,13] . Topographic information, such as elevation and slope, was obtained from the SRTM Digital Elevation Model (DEM) (Zafar et al., 2025; Vahid and Aly, 2025) [25,23] . Additional supporting data, including distance to roads and drainage networks, were collected from OpenStreetMap (OSM) and Survey of India (SOI) toposheets (Loukika et al., 2023; Sankarrao et al., 2021) [12,19] .

Image Pre-processing

All image pre-processing was carried out within the Google Earth Engine (GEE) platform to maintain consistency over time. Satellite images were filtered to remove those with high cloud cover (less than 10%) and were corrected for atmospheric effects (Nigar et al., 2024; Kipterer et al., 2025) [15,11] . A median reduction method was then used to create clear, cloud-free composite images for each year (Tassi and Vizzari, 2020; Nigar et al., 2024) [22,15] . In addition, spectral indices such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Built-up Index (NDBI) were calculated and added to the dataset to improve the identification of vegetation and urban areas (Kipterer et al., 2025; Loukika et al., 2021) [11,13] .

LULC Classification

LULC classification was carried out using the Random Forest (RF) algorithm (ee.Classifier.smileRandomForest) (Alshehri et al., 2025; Bhungeni et al., 2025) [1,4] . RF is widely used because it can handle large and complex datasets efficiently and reduces the risk of overfitting (Breiman, 2001; Belgiu and Drăguţ, 2016) [6,3] . The study area was classified into five categories: Agriculture, Forest, Waterbodies, Built-up, and Barren land. Training and validation data were prepared using a 70/30 split, with around 350 sample points verified using high-resolution images from Google Earth Pro (Nandgude et al. 2026) [14] .

Accuracy Assessment

The accuracy of both classification and prediction results was assessed using a confusion matrix (Zafar et al., 2025; Foody, 2002) [25,7] . Key evaluation metrics included Overall Accuracy (OA), which shows the proportion of correctly classified pixels, and the Kappa Coefficient (κ), which measures the level of agreement while accounting for random chance (Zafar et al., 2025; Pontius and Millones, 2011) [25,18] . In addition, Producer's Accuracy and User's Accuracy were calculated to identify errors related to missing or misclassified land cover classes (Osman et al., 2023; Foody, 2002) [16,7] .

Model Prediction

The future LULC prediction was carried out using a datadriven machine learning approach, where past land use patterns were used to simulate future scenarios. Unlike traditional methods, this approach does not rely on predefined transition rules but learns directly from historical data (Zafar et al., 2025) [25] . The prediction for 2035 represents a short-term projection, primarily influenced by recent land use trends (2015-2025). In contrast, the prediction for 2045 represents a long-term projection, influenced by cumulative changes propagated through the recursive modeling process. The recursive nature of prediction introduces slight uncertainty but is essential for simulating realistic future dynamics. It allows the model to account for feedback effects, such as, expansion of built-up areas, persistence of barren land and stability of forest regions etc. The results indicate that land use changes are not random but follow systematic transition patterns, which can be effectively captured using machine learning techniques. www.extensionjournal.com

Results and Discussion

Accuracy assessment

The accuracy assessment of the Random Forest (RF) model for the Morna watershed for the year 2025 shows an Overall Accuracy (OA) of 71.6% and a Kappa coefficient of 0.61 (Nandgude et al. 2026) [14] . These results indicate that the model performs at a moderate to good level in classifying land use and land cover (LULC) classes. An overall accuracy of 71.6% means that the model correctly classified approximately seven out of every ten pixels, suggesting a reasonably reliable representation of the actual ground conditions. Meanwhile, the Kappa coefficient of 0.61 reflects a substantial level of agreement between the predicted and reference data beyond chance, reinforcing the consistency of the model's performance. Together, these metrics suggest that the RF model is effective for LULC mapping in the Morna watershed, though there is still room for improvement.

LULC change prediction of 2035

The projected LULC changes for the Morna watershed between 2025 and 2035 reveal a landscape that is partially stable but undergoing noticeable anthropogenic transformations, which is shown in fig. 2. Forest land exhibits the highest stability (31.69%), indicating strong ecological resilience and confirming its dominance in the region. Barren land also shows significant persistence (20.85%), which reflects ongoing land degradation or limited land productivity. Agricultural land demonstrates moderate stability (11.13%), suggesting that while farming remains important, it is vulnerable to change.

A key finding during this period is the continuous expansion of built-up areas. The transition from agriculture to built-up land (3.08%) and from barren land to built-up land (1.16%), along with the persistence of existing built-up areas (6.12%), clearly indicates steady urban growth. This trend highlights increasing developmental pressure, particularly on agricultural land, which is gradually being converted for urban uses.

Forest degradation is present but relatively controlled, as seen in the transitions from forest to agriculture (1.35%) and forest to barren land (2.10%). This suggests localized disturbances rather than widespread deforestation. Meanwhile, water bodies remain largely stable (1.75%), indicating minimal changes in the hydrological system during this period. A major concern during this period is the continued persistence of barren land (22.24%). This highlights significant environmental challenges such as soil degradation, reduced land productivity, and possibly inadequate land management practices. The increasing stability of barren land suggests that degraded areas are not being effectively restored. Urban expansion is the most prominent and continuous change, with built-up stability rising from 6.12% to 6.53%, along with ongoing conversion from agriculture (3.08% to 2.65%). This reflects sustained development pressure driven by population growth and infrastructure expansion.

Agricultural land shows moderate stability (11.13% to 12.80%) but faces continuous pressure due to conversion into built-up areas, raising concerns about its long-term sustainability.

Barren land remains consistently high, increasing from 20.85% to 22.24%, highlighting persistent land degradation, possibly due to poor land management and environmental stress.

Water bodies show minimal variation, with stability around 1.75%, indicating that the hydrological system of the watershed remains largely stable over time.

Conclusion

The study successfully demonstrates the effectiveness of machine learning models in predicting future LULC changes in the Morna watershed. The projected scenarios for 2035 and 2045 clearly indicate continuous urban expansion, with built-up land increasing from 6.12% to 6.53%, along with notable conversion from agriculture (3.08% to 2.65%), reflecting growing developmental pressure.

Agricultural land shows moderate stability (11.13% to 12.80%) but is gradually declining due to urban encroachment, raising concerns about its long-term sustainability. In contrast, forest areas remain relatively stable and dominant, increasing from 31.69% to 34.10%, which highlights strong ecological resilience within the watershed. However, the persistent and increasing extent of barren land (20.85% to 22.24%) indicates ongoing land degradation and emphasizes the urgent need for sustainable land management and restoration strategies. Water bodies remain largely stable (~1.75%), suggesting minimal impact on the hydrological system.

References

1. Alshehri B, Zhang Z, Liu X. A review of Google Earth Engine for land use and land cover change analysis: trends, applications, and challenges. ISPRS Int J Geo Inf. 2025;14:416. https://doi.org/10.3390/ijgi14110416
2. Amini S, Saber M, Rabiei-Dastjerdi H, Homayouni S. Urban land use and land cover change analysis using random forest classification of Landsat time series. Remote Sensing. 2022;14(11):2654. https://doi.org/10.3390/rs14112654
3. Belgiu M, Drăguţ L. Random Forest in remote sensing: a review of applications and future directions. ISPRS J Photogramm Remote Sens. 2016;114:24-31. https://doi.org/10.1016/j.isprsjprs.2016.01.011
4. Bhungeni O, Gebreslasie M, Ramjatan A. LULC change detection and future LULC modelling using RF and MLPNN-Markov algorithms in the uMngeni catchment, KwaZulu-Natal, South Africa. Front Environ Sci. 2025;13:1543524. https://doi.org/10.3389/fenvs.2025.1543524
5. Bowlekar A, Sawant H, Nandgude S, Mahale D, Scholar B. Morphometric analysis of Morna River catchment using geographic information system. Agric Eng Today. 2019;43(1):54-61.
6. Foody GM. Status of land cover classification accuracy assessment. Remote Sens Environ. 2002;80(1):185-201.
7. Gaur S, Singh R. A comprehensive review on land use/land cover (LULC) change modeling for urban development: current status and future prospects. Sustainability. 2023;15(2):903. https://doi.org/10.3390/su15020903
8. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. Google Earth Engine: planetaryscale geospatial analysis for everyone. Remote Sens Environ. 2017;202:18-27. https://doi.org/10.1016/j.rse.2017.06.031
9. Ingle NB. Morphometric analysis of Morna River sub basin using geospatial approach. Int J Multidiscip Res Dev. 2022;9(1):71-78.
10. Kipterer JK, Boitt MK, Mundia CN. Comparative analysis of machine learning algorithms for optimal land use and land cover classification: guiding method selection for resource-limited settings in Tiaty, Baringo County, Kenya. J Geosci Environ Prot. 2025;13:393414. https://doi.org/10.4236/gep.2025.134021
11. Loukika KN, Keesara VR, Buri ES, Sridhar V. Future prediction of scenario based land use land cover (LU&LC) using DynaCLUE model for a river basin. Ecol Inform. 2023;77:102223. https://doi.org/10.1016/j.ecoinf.2023.102223
12. Loukika KN, Keesara VR, Sridhar V. Analysis of land use and land cover using machine learning algorithms on Google Earth Engine for Munneru River Basin, India. Sustainability. 2021;13(24):13758. https://doi.org/10.3390/su132413758
13. Nandgude SB, Desai AV, Ganachari HS, Shinde SS, Wankhede PS. Land use land cover change detection using machine learning in Google Earth Engine of Morna Watershed. Int J Res Agron. 2026;SP-9(4):163168. https://doi.org/10.33545/2618060X.2026.v9.i4Sb.5386
14. Nigar A, Li Y, Jat Baloch MY, Alrefaei AF, Almutairi MH. Comparison of machine and deep learning algorithms using Google Earth Engine and Python for land classifications. Front Environ Sci. 2024;12:1378443. https://doi.org/10.3389/fenvs.2024.1378443
15. Osman MA, Abdel-Rahman EM, Onono JO, Olaka LA, Elhag MM, Adan M, et al. Mapping, intensities and future prediction of land use/land cover dynamics using Google Earth Engine and CA-artificial neural network model. PLoS One. 2023;18(7):e0288694. https://doi.org/10.1371/journal.pone.0288694
16. Patil SS, Wandre SS, Patil MA, Gadakh S, Bansude SN, Gavit BK, Dhekale JS. Land use and land cover change detection using geospatial techniques in the Warana River Basin (WRB), Western Maharashtra, India. J Geogr Environ Earth Sci Int. 2025;29(11):4963. https://doi.org/10.9734/jgeesi/2025/v29i11967 www.extensionjournal.com
17. Pontius RG, Millones M. Death to Kappa: birth of quantity disagreement and allocation disagreement for accuracy assessment. Int J Remote Sens. 2011;32(15):4407-4429. https://doi.org/10.1080/01431161.2011.552923
18. Sankarrao L, Ghose DK, Rathinsamy M. Predicting land-use change: intercomparison of different hybrid machine learning models. Environ Model Softw. 2021;145:105207. https://doi.org/10.1016/j.envsoft.2021.105207
19. Shinde S, Nandgude S. Soil organic carbon stocks assessment under different land use land cover in Morna Watershed, India using GIS technique. Mysore J Agric Sci. 2024;58(3).
20. Shinde SS, More DD, Shendkar DA, Dahiphale MS, Thange AB, Nandgude SB. Temporal and spatial dynamics of land use and land cover in Shirur Kasar Tehsil, Maharashtra using geospatial technology. Int J Agric Invent. 2024;9(1):193-204. https://doi.org/10.46492/IJAI/2024.9.1.25
21. Tassi A, Vizzari M. Object-oriented LULC classification in Google Earth Engine combining SNIC, GLCM, and machine learning algorithms. Remote Sensing. 2020;12(22):3776. https://doi.org/10.3390/rs12223776
22. Vahid R, Aly MH. An integrated CA-Markov modeling framework for forecasting land use and land cover dynamics in Arkansas, USA. Geomatics. 2025;5(4):62. https://doi.org/10.3390/geomatics5040062
23. Wulder MA, White JC, Loveland TR, Woodcock CE, Belward AS, Cohen WB, et al. The global Landsat archive: status, consolidation, and direction. Remote Sens Environ. 2016;185:271-283. https://doi.org/10.1016/j.rse.2015.11.032
24. Zafar SM, Khan JA, Mobeen A, Jaafar WSWM, Kemarau RA. Estimating the plausible projections of land use/land cover dynamics in Jhelum and Chenab River basins using satellite imageries and machine learning models in Google Earth Engine. Geocarto Int. 2025;40(1):2491640. https://doi.org/10.1080/10106049.2025.2491640