Comparative Evaluation of Machine Learning Models for UAV-Derived Biomass Estimation in Miombo Woodlands DOI Creative Commons
Goodluck S. Melitha, Japhet J. Kashaigili, Wilson Ancelm Mugasha

et al.

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 11, 2024

Abstract Accurately estimating above-ground biomass (AGB) is critical for understanding carbon storage and ecosystem dynamics, which are essential sustainable forest management climate change mitigation. This study evaluated the performance of four machine learning models XGBoost, Random Forest (RF), Gradient Boosting (GBM), Support Vector Machine (SVM) in predicting AGB Miombo Woodlands using UAV-derived spectral height data. A total 52 model configurations were tested, incorporating up to five predictor variables. XGBoost demonstrated superior performance, explaining 99% variance (R² = 0.99), with a low RMSE 9.82 Mg/ha an rRMSE 8.25%. Although it showed slight underestimation bias (-2.48), proved highly reliable handling complex ecosystems like Miombo. also performed well, 91% 0.91), though exhibited higher error rates (RMSE 30.81 Mg/ha). In contrast, GBM SVM weaker R² values 0.23 0.81, respectively. highlights potential UAV data combined advanced models, particularly accurate estimation. Future research should explore integrating technologies LiDAR or satellite imagery further improve prediction accuracy across diverse ecosystems.

Language: Английский

Integration of machine learning and remote sensing for above ground biomass estimation through Landsat-9 and field data in temperate forests of the Himalayan region DOI Creative Commons
Shoaib Ahmad Anees, Kaleem Mehmood, Waseem Razzaq Khan

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102732 - 102732

Published: July 22, 2024

Accurately estimating aboveground biomass (AGB) in forest ecosystems facilitates efficient resource management, carbon accounting, and conservation efforts. This study examines the relationship between predictors from Landsat-9 remote sensing data several topographical features. While provides reliable crucial for long-term monitoring, it is part of a broader suite available technologies. We employ machine learning algorithms such as Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), Random Forest (RF), alongside linear regression techniques like Multiple Linear (MLR). The primary objectives this encompass two key aspects. Firstly, research methodically selects optimal predictor combinations four distinct variable groups: (L1) data, fusion Vegetation-based indices (L2), integration with Shuttle Radar Topography Mission Digital Elevation Model (SRTM DEM) (L3) combination best (L4) derived L1, L2, L3. Secondly, systematically assesses effectiveness different to identify most precise method establishing any potential field-measured AGB variables. Our revealed that (RF) model was utilizing OLI SRTM DEM predictors, achieving remarkable accuracy. conclusion reached by assessing its outstanding performance when compared an independent validation dataset. RF exhibited accuracy, presenting relative mean absolute error (RMAE), root square (RRMSE), R2 values 14.33%, 22.23%, 0.81, respectively. XGBoost subsequent choice RMAE, RRMSE, 15.54%, 23.85%, 0.77, further highlights significance specific spectral bands, notably B4 B5 Landsat 9 capturing spatial distribution patterns. Integration vegetation-based indices, including TNDVI, NDVI, RVI, GNDVI, refines mapping precision. Elevation, slope, Topographic Wetness Index (TWI) are proxies representing biophysical biological mechanisms impacting AGB. Through utilization openly accessible fine-resolution employing algorithm, demonstrated promising outcomes identification predictor-algorithm mapping. comprehensive approach offers valuable avenue informed decision-making assessment, ecological monitoring initiatives.

Language: Английский

Citations

52
Enhancing carbon stock estimation in forests: Integrating multi-data predictors with random forest method DOI Creative Commons
Gabriel E. Suárez-Fernández, J. Martínez-Sánchez, Pedro Arias

et al.

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 102997 - 102997

Published: Jan. 1, 2025

Language: Английский

Citations

3
Correcting forest aboveground biomass biases by incorporating independent canopy height retrieval with conventional machine learning models using GEDI and ICESat-2 data DOI Creative Commons
Biao Zhang, Zhichao Wang, Tiantian Ma

et al.

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 103045 - 103045

Published: Jan. 1, 2025

Language: Английский

Citations

2
Towards carbon neutrality: Enhancing CO2 sequestration by plants to reduce carbon footprint DOI
Dawid Skrzypczak,

Katarzyna Gorazda,

Katarzyna Mikula

et al.

The Science of The Total Environment, Journal Year: 2025, Volume and Issue: 966, P. 178763 - 178763

Published: Feb. 1, 2025

Language: Английский

Citations

2
From simple linear regression to machine learning methods: Canopy cover modelling of a young forest using planet data DOI Creative Commons
Arun Gyawali, Hari Adhikari, Mika Aalto

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102706 - 102706

Published: July 4, 2024

Accurate canopy cover estimation is essential for mature and early-stage young forests, as it guides forest management silvicultural activities necessary their growth regeneration. However, obtaining precise measurements of in the field time-consuming challenging, especially at regional landscape levels. Remote sensing techniques offer a promising alternative to traditional field-based estimating cover. In this study, our objective estimate using vegetation indices derived from multispectral bands PlanetScope (Planet Lab, Inc., San Francisco, CA, USA). To best knowledge, first study utilise imagery data boreal forests. Based on analysis four (green, blue, red, near-infrared) imagery, 43 indices, including spectral 13 salinity were computed select predictors modelling. Six regression models employed model cover: linear, elastic net, support vector machine, random forest, extreme gradient boosting, light boosting machine. All demonstrated good performance both training dataset (R2 = 0.58–0.69) testing 0.59–0.64, RMSE 0.16–0.18, rRMSE 22%–23%, MAE 0.12–0.14). fit statistics datasets paired t-test, identified machine most suitable predicting For R2 value was 0.69 (training), data, 0.64, 0.16, 22%, 0.12. Therefore, we recommend that future researchers Planet higher spatial resolution. exploring additional learning algorithms explicitly methods when computing satellite remote strongly advised.

Language: Английский

Citations

8
Explainable machine learning-based fractional vegetation cover inversion and performance optimization – A case study of an alpine grassland on the Qinghai-Tibet Plateau DOI Creative Commons
Xinhong Li, Jianjun Chen, Zizhen Chen

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102768 - 102768

Published: Aug. 10, 2024

Fractional Vegetation Cover (FVC) serves as a crucial indicator in ecological sustainability and climate change monitoring. While machine learning is the primary method for FVC inversion, there are still certain shortcomings feature selection, hyperparameter tuning, underlying surface heterogeneity, explainability. Addressing these challenges, this study leveraged extensive field data from Qinghai-Tibet Plateau. Initially, selection algorithm combining genetic algorithms XGBoost was proposed. This integrated with Optuna tuning method, forming GA-OP combination to optimize learning. Furthermore, comparative analyses of various models inversion alpine grassland were conducted, followed by an investigation into impact heterogeneity on performance using NDVI Coefficient Variation (NDVI-CV). Lastly, SHAP (Shapley Additive exPlanations) employed both global local interpretations optimal model. The results indicated that: (1) exhibited favorable terms computational cost accuracy, demonstrating significant potential tuning. (2) Stacking model achieved among seven (R2 = 0.867, RMSE 0.12, RPD 2.552, BIAS −0.0005, VAR 0.014), ranking follows: > CatBoost LightGBM RFR KNN SVR. (3) NDVI-CV enhanced result reliability excluding highly heterogeneous regions that tended be either overestimated or underestimated. (4) revealed decision-making processes perspectives. allowed deeper exploration causality between features targets. developed high-precision scheme, successfully achieving accurate proposed approach provides valuable references other parameter inversions.

Language: Английский

Citations

8
Analysis of vegetation dynamics from 2001 to 2020 in China's Ganzhou rare earth mining area using time series remote sensing and SHAP-enhanced machine learning DOI Creative Commons
Ming Lei, Yuandong Wang, Guangxu Liu

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 84, P. 102887 - 102887

Published: Nov. 9, 2024

Language: Английский

Citations

8
Comparative evaluation of machine learning models for UAV-derived biomass estimation in Miombo Woodlands DOI
Goodluck S. Melitha, Japhet J. Kashaigili, Wilson Ancelm Mugasha

et al.

Earth Science Informatics, Journal Year: 2025, Volume and Issue: 18(3)

Published: Feb. 19, 2025

Language: Английский

Citations

1
Inferring the relationship between soil temperature and the normalized difference vegetation index with machine learning DOI Creative Commons
Steven Mortier, Amir Hamedpour,

Bart Bussmann

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: 82, P. 102730 - 102730

Published: July 20, 2024

Language: Английский

Citations

6
Machine learning methods for basal area prediction of Fagus orientalis Lipsky stands based on national forest inventory DOI
Seyedeh Fatemeh Hosseini, Hamid Jalilvand, Asghar Fallah

et al.

Trees, Journal Year: 2025, Volume and Issue: 39(2)

Published: March 19, 2025

Language: Английский

Citations

0