In The Science of the total environment
Phytoliths are known to play a significant role in the global carbon cycle by sequestering atmospheric carbon dioxide as phytolith-occluded carbon (PhytOC) for a long time. Given the resistant nature of phytolith to decomposition, PhytOC can represent up to 82 % of total carbon in some soil and sediments even after 2000 years of litter decomposition. Hence, forests with high PhytOC sequestration rates could play a critical role in increasing terrestrial carbon storage. In this study, we quantified the variation in PhytOC concentrations in bamboo leaves, branches and culms with forest types in the Eastern Indian Himalayas as bamboos are efficient accumulator of phytolith and PhytOC due to their fast growth and high biomass accumulation rates. Using nine different machine learning techniques, we also investigated the determinants of PhytOC production in bamboo stands in the study area of India. The results revealed that the PhytOC concentrations in bamboo stands were in the order of leaf (3.0 g kg-1) > culm (1.0 g kg-1) > branch (0.2 g kg-1) across forest types. The highest PhytOC stock (53.8 kg ha-1) was found in bamboo stands in the subtropical pine forests (1900-3500 m elevation), while the lowest (28.0 kg ha-1) was in the tropical evergreen forests (<900 m elevation). Machine learning techniques established a positive correlation of PhytOC content in leaf and total PhytOC content with soil available phosphorus, elevation, total nitrogen, exchangeable potassium, atmospheric humidity, SOC content, CEC and pH. Numerical evaluation criteria and graphic methods identified artificial neural network (ANN) and support vector regression as the superior techniques with a root mean square value of 0.52 kg ha-1 and 0.59 kg ha-1 respectively. The results of these two models were found to be better among all the nine machine learning algorithms used. The high PhytOC storage in the bamboo stands in the Indian Himalayan region suggests that forest management could secure a stable carbon sink on a millennial scale.
Debnath Nirmal, Nath Amitabha, Sileshi Gudeta W, Nath Arun Jyoti, Nandy Subrata, Das Ashesh Kumar
Bamboo stands, Carbon sequestration, Forest ecosystems, Machine learning, Modelling