A comparison of random forests and linear stepwise regressions to model and map soil carbon in South-Central British Columbia grasslands using normalized difference vegetation index based models
Bibliographic record
Abstract
Industrialization, production and consumption of fossil fuels, and land use changes have resulted in increased concentrations of carbon dioxide (CO2) and other greenhouse gases in the atmosphere causing changes in ecosystem structure and properties. Soil carbon (SC) sequestration, the process of storing CO2 in the soil through crop residues and other organic solids, has been an area under much investigation as it relates to reducing atmospheric carbon (C) and mitigating climate change. Since grasslands predominately sequester C below ground through root growth and consequent soil-building processes, they have a high potential for long term C storage and therefore are of major importance for maintaining Earth’s carbon cycle. Despite advances in SC determination in recent years, it remains a challenge to model and map SC across large regions. There are several factors, both anthropogenic and environmental, that influence C sequestration. Given this complex system, I have used Geographic Information Systems (GIS) data in conjunction with accurate field measurements to examine the mechanisms that affect SC storage in order to produce predictive SC maps for the southern interior grasslands of British Columbia (BC). Soil carbon prediction was based on the Normalized Difference Vegetation Index (NDVI), which has demonstrated high correlation with SC distribution in past studies. The relationship of SC and NDVI was evaluated on two scales using: i) the MOD 13Q1 (250 m/16 day resolution) NDVI data product from the Moderate Resolution Imaging Spectro-radiometer (MODIS) aboard the United States Terra satellite (NDVIMODIS), and ii) a handheld Multispectral Radiometer (MSR16R, Cropscan Inc., 1 m resolution) device (NDVIMSR). Other factors included in the model are: i) grazing, ii) climate data, iii) vegetation community zones, iv) soil classification and drainage, and v) topography. A traditional linear stepwise regression (SR) modelling approach was compared with random forest (RF) modelling, a recursive partitioning technique that employs randomized bagging and bootstrapping of samples. There was a strong relationship between NDVI derived from the MSR with SC in fenced systems (R2=0.41), SOC in fenced systems (R2=0.47), and SOC in grazed systems (R2=0.34). When NDVI data derived from the MSR was used as model input, the percentage of explained variance was greater than for models which used NDVI derived from MODIS data (R2 = 0.68 for SC in 2014 for fenced systems, modelled with SR based on NDVI data derived from MODIS ; R2=0.77 for SC in 2014 for fenced systems, modelled with SR based on NDVI data derived from MSR). These results show the potential of increased model accuracy with higher resolution GIS data and the effectiveness of NDVI based models to predict SC and SOC. Significantly higher SC and SOC was recorded in 2014 as compared to 2013 (p=0.001 for SC and p=0.031 for SOC), demonstrating the potential for C sequestration in BC grasslands as a climate change mitigation tactic. Based on comparisons of R2 and AIC values, SR produces models that explain more variance and are of better quality (R2=0.49-0.77 and AIC = 0.30-0.13 for SR models in 2014; R2=0.36-057 and AIC = 0.36-0.18). This project creates the groundwork for effective monitoring techniques of SC and SOC levels using GIS data in order to develop a carbon offset program for the ranching industry and can be used to help direct land management efforts to increase C sequestration in BC.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".