Geospatial and field-based digital mapping of soil organic carbon in Taiwan's agricultural and forest landscapes
Journal
Ecological Informatics
Journal Volume
93
Start Page
103557
ISSN
1574-9541
Date Issued
2026-02
Author(s)
農業部農業試驗所
Miguel Conrado Valdez
Chi-Farn Chen
Santos Daniel Chicas
Chun-Chien Yen
Youg-Sin Cheng
Nobuya Mizoue
Jui-Han Yang
Abstract
Soil carbon stocks are the second largest natural reservoir of carbon globally, after oceans. Soil carbon sequestration is a key process that supports both climate change mitigation and adaptation strategies. Thus, accurate mapping of soil organic carbon (SOC) stocks is essential for estimating their regional distribution and assessing soil functions. Regular ground sampling for estimating SOC is often expensive, prompting exploration of alternative, cost-effective techniques such as remote sensing. Many studies have applied remote sensing and machine learning to identify factors influencing SOC and predict its levels, but they often rely on limited data, such as single-year soil samples and small sample sizes, restricting their ability to capture SOC's variability and dynamics. This study aims to combine high sampling density (1.15 samples/km(-2) in agricultural areas and 0.38/km(-2) for forest areas) and long-term SOC data (2010-2021) with remote sensing, topographic, and climatic data using random forest (RF) modeling to estimate SOC stocks in surface soils (0-30 cm) and identify key influencing factors across Taiwan's agricultural and forest lands. Initially, 22 variables were considered, and the least important ones were excluded. Two models were developed: one for agricultural land and one for forest land, with 14 variables selected, in both models. The agricultural model emphasized elevation, climate, and green soil-adjusted vegetation index (GSAVI), while the forest model focused on vegetation indices and elevation. The models successfully predicted SOC stocks (in kg m(-2)), showing a strong correlation (R-2 > 0.75). Results revealed higher SOC in mountainous regions and lower SOC in suburban and low-lying agricultural areas, with SOC stocks ranging from 0.64 to 8.83 kg m(-2) in agriculture land and 2.41 to 21.34 kg m(-2) in forest land. This research significantly contributes to SOC monitoring and informs carbon sequestration policy development in Taiwan. The findings from this study offer important insights for the long-term tracking of soil organic carbon stocks in Taiwan and provide a useful foundation for shaping future carbon sequestration policies.
Subjects
Soil organic carbon stocks
Remote sensing
Digital soil mapping
Taiwan
Agriculture
Forest
Random forest
SDGs
Publisher
Elsevier
Type
journal article