Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent

Understanding the distribution of aboveground biomass (AGB) is vital for evaluating carbon stocks & ecosystem dynamics, especially in regions with diverse landscapes like Indian subcontinent. This study evaluates three machine learning models—Random Forest (RF), Gradient Tree Boost...

Full description

Saved in:
Bibliographic Details
Main Authors: S. Mamgain, B. Ghale, H. C. Karnatak, A. Roy
Format: Article
Language:English
Published: Copernicus Publications 2025-03-01
Series:The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Online Access:https://isprs-archives.copernicus.org/articles/XLVIII-M-5-2024/109/2025/isprs-archives-XLVIII-M-5-2024-109-2025.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850053797624152064
author S. Mamgain
B. Ghale
H. C. Karnatak
A. Roy
author_facet S. Mamgain
B. Ghale
H. C. Karnatak
A. Roy
author_sort S. Mamgain
collection DOAJ
description Understanding the distribution of aboveground biomass (AGB) is vital for evaluating carbon stocks &amp; ecosystem dynamics, especially in regions with diverse landscapes like Indian subcontinent. This study evaluates three machine learning models&mdash;Random Forest (RF), Gradient Tree Boosting (GTB), &amp; Classification and Regression Trees (CART)&mdash;for predicting AGB across the subcontinent. Independent variable in these models is AGB, while dependent variables include a range of vegetation &amp; topographic layers: Normalized Difference Vegetation Index, Enhanced Vegetation Index, Leaf Area Index, Fraction of Photosynthetically Active Radiation, land cover, elevation, aspect, slope, &amp; hillshade. These predictors are essential for capturing ecological &amp; topographical characteristics that influence biomass distribution. The models were evaluated using coefficient of determination (R<sup>2</sup>) &amp; Pearson's correlation coefficient (r) to assess predictive accuracy. RF emerged as most accurate, with an R&sup2; value of 0.834 &amp; r value of 0.913, effectively capturing the spatial variability in AGB across subcontinent&rsquo;s diverse ecosystems, which was then used to predict AGB for 2023. The predictions reveal significant spatial variation in biomass density, reflecting region's diverse ecological zones &amp; land-use patterns. In India, high biomass densities are found in Himalayan foothills, northeastern states, &amp; Western Ghats, while arid regions like Rajasthan &amp; Gujarat have lower values. Pakistan generally exhibits low biomass densities, with higher values near the northern border with India. Nepal &amp; Bhutan show high densities in their forested regions, particularly in the mid-hills, high mountains, &amp; Eastern Himalaya. Bangladesh has moderate to low biomass densities. In Sri Lanka, central highlands &amp; southwestern rainforests have highest biomass densities, while the more arid northern &amp; eastern regions exhibit lower values. This study highlights the importance of using robust machine learning models like RF to accurately capture spatial patterns of biomass distribution, which is crucial for forest management, carbon accounting, &amp; biodiversity conservation in the Indian subcontinent.
format Article
id doaj-art-e2036934ecf24fd08a3910ccb23482f2
institution DOAJ
issn 1682-1750
2194-9034
language English
publishDate 2025-03-01
publisher Copernicus Publications
record_format Article
series The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
spelling doaj-art-e2036934ecf24fd08a3910ccb23482f22025-08-20T02:52:26ZengCopernicus PublicationsThe International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences1682-17502194-90342025-03-01XLVIII-M-5-202410911510.5194/isprs-archives-XLVIII-M-5-2024-109-2025Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian SubcontinentS. Mamgain0B. Ghale1H. C. Karnatak2A. Roy3Indian Institute of Remote Sensing, Indian Space Research Organization (ISRO), Department of Space, 4, Kalidas Road, Dehradun 248001, IndiaIndian Institute of Remote Sensing, Indian Space Research Organization (ISRO), Department of Space, 4, Kalidas Road, Dehradun 248001, IndiaIndian Institute of Remote Sensing, Indian Space Research Organization (ISRO), Department of Space, 4, Kalidas Road, Dehradun 248001, IndiaIndian Institute of Remote Sensing, Indian Space Research Organization (ISRO), Department of Space, 4, Kalidas Road, Dehradun 248001, IndiaUnderstanding the distribution of aboveground biomass (AGB) is vital for evaluating carbon stocks &amp; ecosystem dynamics, especially in regions with diverse landscapes like Indian subcontinent. This study evaluates three machine learning models&mdash;Random Forest (RF), Gradient Tree Boosting (GTB), &amp; Classification and Regression Trees (CART)&mdash;for predicting AGB across the subcontinent. Independent variable in these models is AGB, while dependent variables include a range of vegetation &amp; topographic layers: Normalized Difference Vegetation Index, Enhanced Vegetation Index, Leaf Area Index, Fraction of Photosynthetically Active Radiation, land cover, elevation, aspect, slope, &amp; hillshade. These predictors are essential for capturing ecological &amp; topographical characteristics that influence biomass distribution. The models were evaluated using coefficient of determination (R<sup>2</sup>) &amp; Pearson's correlation coefficient (r) to assess predictive accuracy. RF emerged as most accurate, with an R&sup2; value of 0.834 &amp; r value of 0.913, effectively capturing the spatial variability in AGB across subcontinent&rsquo;s diverse ecosystems, which was then used to predict AGB for 2023. The predictions reveal significant spatial variation in biomass density, reflecting region's diverse ecological zones &amp; land-use patterns. In India, high biomass densities are found in Himalayan foothills, northeastern states, &amp; Western Ghats, while arid regions like Rajasthan &amp; Gujarat have lower values. Pakistan generally exhibits low biomass densities, with higher values near the northern border with India. Nepal &amp; Bhutan show high densities in their forested regions, particularly in the mid-hills, high mountains, &amp; Eastern Himalaya. Bangladesh has moderate to low biomass densities. In Sri Lanka, central highlands &amp; southwestern rainforests have highest biomass densities, while the more arid northern &amp; eastern regions exhibit lower values. This study highlights the importance of using robust machine learning models like RF to accurately capture spatial patterns of biomass distribution, which is crucial for forest management, carbon accounting, &amp; biodiversity conservation in the Indian subcontinent.https://isprs-archives.copernicus.org/articles/XLVIII-M-5-2024/109/2025/isprs-archives-XLVIII-M-5-2024-109-2025.pdf
spellingShingle S. Mamgain
B. Ghale
H. C. Karnatak
A. Roy
Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
title Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
title_full Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
title_fullStr Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
title_full_unstemmed Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
title_short Assessment of Machine Learning Models for Predicting Aboveground Biomass in the Indian Subcontinent
title_sort assessment of machine learning models for predicting aboveground biomass in the indian subcontinent
url https://isprs-archives.copernicus.org/articles/XLVIII-M-5-2024/109/2025/isprs-archives-XLVIII-M-5-2024-109-2025.pdf
work_keys_str_mv AT smamgain assessmentofmachinelearningmodelsforpredictingabovegroundbiomassintheindiansubcontinent
AT bghale assessmentofmachinelearningmodelsforpredictingabovegroundbiomassintheindiansubcontinent
AT hckarnatak assessmentofmachinelearningmodelsforpredictingabovegroundbiomassintheindiansubcontinent
AT aroy assessmentofmachinelearningmodelsforpredictingabovegroundbiomassintheindiansubcontinent