The Value of L-Band Soil Moisture and Vegetation Optical Depth Estimates in the Prediction of Vegetation Phenology

Bonan Li,Stephen P Good,Dawn R Urycki

doi:10.3390/rs13071343

Abstract

Vegetation phenology is a key ecosystem characteristic that is sensitive to environmental conditions. Here, we examined the utility of soil moisture (SM) and vegetation optical depth (VOD) observations from NASA’s L-band Soil Moisture Active Passive (SMAP) mission for the prediction of leaf area index (LAI), a common metric of canopy phenology. We leveraged mutual information theory to determine whether SM and VOD contain information about the temporal dynamics of LAI that is not contained in traditional LAI predictors (i.e., precipitation, temperature, and radiation) and known LAI climatology. We found that adding SMAP SM and VOD to multivariate non-linear empirical models to predict daily LAI anomalies improved model fit and reduced error by 5.2% compared with models including only traditional LAI predictors and LAI climatology (average R2 = 0.22 vs. 0.15 and unbiased root mean square error [ubRMSE] = 0.130 vs. 0.137 for cross-validated models with and without SM and VOD, respectively). SMAP SM and VOD made the more improvement in model fit in grasslands (R2 = 0.24 vs. 0.16 and ubRMSE = 0.118 vs. 0.126 [5.7% reduction] for models with and without SM and VOD, respectively); model predictions were least improved in shrublands. Analysis of feature importance indicates that LAI climatology and temperature were overall the two most informative variables for LAI anomaly prediction. SM was more important in drier regions, whereas VOD was consistently the second least important factor. Variations in total LAI were mostly explained by local daily LAI climatology. On average, the R2s and ubRMSE of total LAI predictions by the traditional drivers and its climatology are 0.81 and 0.137, respectively. Adding SMAP SM and VOD to these existing predictors improved the R2s to 0.83 (0.02 improvement in R2s) and reduced the ubRMSE to 0.13 (5.2% reduction). Though these improvements were modest on average, in locations where LAI climatology is not reflective of LAI dynamics and anomalies are larger, we find SM and VOD to be considerably more useful for LAI prediction. Overall, we find that L-band SM and VOD observations can be useful for prediction of LAI, though the informational contribution varies with land cover and environmental conditions.

Highlights

Vegetation phenology is the study of plant life cycles such as budburst, flowering and leaf senescence, and the impact of seasonal and inter-annual climate variability on the timing and magnitude of these events [1]
We find that L-band soil moisture (SM) and vegetation optical depth (VOD) observations can be useful for prediction of leaf area index (LAI), though the informational contribution varies with land cover and environmental conditions
This study evaluated and quantified how informative L-band VOD and SM products are for the prediction of LAI using a machine learning approach

Summary

Introduction

Vegetation phenology is the study of plant life cycles such as budburst, flowering and leaf senescence, and the impact of seasonal and inter-annual climate variability on the timing and magnitude of these events [1]. Recent climate warming has impacted vegetation phenology, which may have profound implications for agriculture and forest productivity [2]. Tracking land surface vegetation phenology is critical to enhance our understanding of the water–food–energy nexus and carbon exchanges between terrestrial ecosystems and the atmosphere [3]. Satellite derived vegetation indices are frequently used to track and map the spatial and temporal dynamics of vegetation phenology at large scales and can overcome the limitations of traditional phenology studies that heavily rely on in situ observations [4]. Various vegetation indices have been applied to assess and track vegetation phenology.

Objectives

Methods

Results

Discussion

Conclusion