Inferring the relationship between soil temperature and the normalized difference vegetation index with machine learning

TL;DR

This study investigates how soil temperature and other meteorological drivers shape NDVI-derived vegetation phenology in subarctic grasslands. By fitting a double logistic curve to NDVI data, the authors extract intra-annual phenology metrics ($SOS$, $POS$, $PEAK$) and compare linear regression with ML (MLP) approaches, interpreting results with SHAP values. They find a robust, negative relation between $T_{soil}$ and both $SOS$ and $POS$, indicating earlier greening with warming, while $PEAK$ increases only slightly; meteorological variables largely govern inter-annual variation and model predictions. The work demonstrates the utility of ML and SHAP for elucidating drivers of vegetation phenology and provides a foundation for future, more generalized ML-based phenology studies in the face of climate change.

Abstract

Changes in climate can greatly affect the phenology of plants, which can have important feedback effects, such as altering the carbon cycle. These phenological feedback effects are often induced by a shift in the start or end dates of the growing season of plants. The normalized difference vegetation index (NDVI) serves as a straightforward indicator for assessing the presence of green vegetation and can also provide an estimation of the plants' growing season. In this study, we investigated the effect of soil temperature on the timing of the start of the season (SOS), timing of the peak of the season (POS), and the maximum annual NDVI value (PEAK) in subarctic grassland ecosystems between 2014 and 2019. We also explored the impact of other meteorological variables, including air temperature, precipitation, and irradiance, on the inter-annual variation in vegetation phenology. Using machine learning (ML) techniques and SHapley Additive exPlanations (SHAP) values, we analyzed the relative importance and contribution of each variable to the phenological predictions. Our results reveal a significant relationship between soil temperature and SOS and POS, indicating that higher soil temperatures lead to an earlier start and peak of the growing season. However, the Peak NDVI values showed just a slight increase with higher soil temperatures. The analysis of other meteorological variables demonstrated their impacts on the inter-annual variation of the vegetation phenology. Ultimately, this study contributes to our knowledge of the relationships between soil temperature, meteorological variables, and vegetation phenology, providing valuable insights for predicting vegetation phenology characteristics and managing subarctic grasslands in the face of climate change. Additionally, this work provides a solid foundation for future ML-based vegetation phenology studies.

Figures (7)

• Figure 1: Overview of all available variables for plot GN1A (unwarmed control plot). Whereas the NDVI and soil temperature (upper two figures) are unique for all 50 plots, the meteorological variables (bottom three figures) are the same for every plot.
• Figure 2: The SOS is estimated based on the second derivative of the fitted NDVI curve. The SOS is defined as the week when the NDVI curvature increases the most, and is indicated with a red line.
• Figure 3: Linear model that predicts the start of the season (a), the peak date of the season (b) and the peak value of NDVI (c), based on the average annual soil temperature. The color indicates the soil warming category where the blue points are A plots, the red points are B plots, the yellow points are C plots, the green points are D plots, and the orange points are E plots. All models had a significant relationship between the average soil temperature and the studied NDVI curve parameter. (See \ref{['tab:fa:linregequations']})
• Figure 4: SHAP values of multi-layer perceptron that predicts the start of the greening season based on the average soil temperature, air temperature, precipitation, and radiation. The color indicates the soil warming category where the blue bars are A plots, the red bars are B plots, the yellow bars are C plots, the green bars are D plots, and the orange bars are E plots.
• Figure 5: SHAP values of multi-layer perceptron that predicts the peak of the greening season (POS) based on the average soil temperature, air temperature, precipitation, and radiation. The color indicates the soil warming category where the blue bars are A plots, the red bars are B plots, the yellow bars are C plots, the green bars are D plots, and the orange bars are E plots.