Comparison of parametric regression model and a machine learning algorithm for the prediction of the above ground biomass in a regional tropical forest landscape, Western Ghats - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Comparison of parametric regression model and a machine learning algorithm for the prediction of the above ground biomass in a regional tropical forest landscape, Western Ghats

Résumé

Forest ecosystems are critical to the global water cycle, biodiversity, and the global carbon balance. Carbon accounts for roughly 50% of the dry weight of trees, often known as biomass. Monitoring forests, specifically monitoring forest biomass, is crucial for tracking changes in carbon stocks and the global carbon cycle. Despite multiple global initiatives (Reducing Emissions from Deforestation and Forest Degradation (REDD+), accurate assessment of forest biomass at the local and regional levels remains difficult. In recent years, there has been a lot of interest in combining machine learning (ML) techniques with satellite observations for ecological monitoring and biomass estimation. This is especially beneficial in terrestrial environments where vegetation growth varies spatially and temporally due different factors. In such circumstances, combining ML approaches with satellite data aids in capturing the complexity of forest ecosystems at various spatiotemporal scales. Therefore, the goal of this study is to estimate aboveground biomass in the landscape through parametric regression model (Multiple linear regression; MLR) and a machine learning model (Random Forest; RF). Field-based biomass estimation was generated from one hundred and forty, 0.1 ha plots laid in Biligiri Ranganthaswamy Temple Tiger Reserve, Western Ghats and spatial datasets were built by merging multi-spectral bands of Sentinel-2 satellite imagery and various vegetative indices and topography variables. The different variables include Soil Adjusted Vegetation Index, Land Surface Water Index, Normalized Difference Water Index, Enhanced Vegetation Index, Normalized Difference Vegetation Index, elevation, Height (Global Ecosystem Dynamics Investigation (GEDI)) and sentinel-2 multispectral bands (B2-B8, B8A, B11 & B12). Multicollinearity among independent variables were checked using Boruta algorithm and Variance Inflation Factor (VIF) and variables with VIF >5 were removed from the analysis to reduce the multicollinearity. Insitu biomass estimation was done using the allometric equation provided by Chave et al. (2014). The satellite image was extracted from United States Geological Survey (USGS) Earth Explorer which was atmospherically corrected followed by the resampling of satellite bands to 10 m spatial resolution. The vegetation indices were derived using Google EarthEngine. In the study, Random Forest model was developed and compared with the MLR. Model evaluation and accuracy assessment was done using 10-fold cross-validation techniquewith ten repetitions utilizing the caret package in open-source R statistical software. The data was split into two sections at random, with 80% of the training and 20% for the model’s validation and evaluation (testing). Statistical metrics like Correlation Coefficient (r), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) was used for estimating the performance of each model. The results revealed that the RF performed well for estimating biomass in the study area with an accuracy of 92.6%, than the MLR (accuracy of 75.3%). The r, RMSE, MAE values of RF model were 0.96,5.15 Mg/0.1 ha and 3.57 Mg/0.1 ha respectively whereas it was 0.86, 9.42 Mg/0.1 ha, 6.81 Mg/0.1 ha for MLR. Therefore, these findings demonstrate the use of Machine learning in situations where prediction accuracy is crucial for proper conservation and management of the study area. The spatial prediction biomass map generated for the study can be used as a baseline map for further investigation for tracking changes in carbon stocks at the regional scale.
Fichier non déposé

Dates et versions

hal-03992247 , version 1 (16-02-2023)

Identifiants

  • HAL Id : hal-03992247 , version 1

Citer

Kurian Ayushi, Narayanan Ayyappan, Kanda Naveen Babu. Comparison of parametric regression model and a machine learning algorithm for the prediction of the above ground biomass in a regional tropical forest landscape, Western Ghats. International Conference on Applied Geoinformatics for Society and Environment – 2022, Digital University of Kera Thiruvananthapuram, India, Nov 2022, Thiruvananthapuram, India. ⟨hal-03992247⟩
36 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More