Soil and vegetation water content identify the main terrestrial ecosystem changes

ABSTRACT Environmental change is a consequence of many interrelated factors. How vegetation responds to natural and human activity still needs to be well established, quantified and understood. Recent satellite missions providing hydrologic and ecological indicators enable better monitoring of Earth system changes, yet there is no automatic way to address this issue directly from observations. Here, we develop an observation-based methodology to capture evidence of changes in global terrestrial ecosystems and attribute these changes to natural or anthropogenic activity. We use the longest time record of global microwave L-band soil moisture and vegetation optical depth as satellite data and build spatially explicit maps of change in soil and vegetation water content and biomass reflecting large ecosystem changes during the last decade, 2010–20. Regions of prominent trends (from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$-8\%$\end{document} to 9% per year) are observed, especially in humid and semi-arid climates. We further combine such trends with land cover change maps, vegetation greenness and precipitation variability to assess their relationship with major documented ecosystem changes. Several regions emerge from our results. They cluster changes according to human activity drivers, including deforestation (Amazon, Central Africa) and wildfires (East Australia), artificial reforestation (South-East China), abandonment of farm fields (Central Russia) and climate shifts related to changes in precipitation variability (East Africa, North America and Central Argentina). Using the high sensitivity of soil and vegetation water content to ecosystem changes, microwave satellite observations enable us to quantify and attribute global vegetation responses to climate or anthropogenic activities as a direct measure of environmental changes and the mechanisms driving them.

: Detected sensitive regions to SM (left) and VOD (right) changes. The raise (decay) patterns are shown in green (brown). The northeast China coast cluster was removed from the study due to potential interference in microwave data. : Estimated joint and marginal distributions of (a) trends in rainfall and mean annual rainfall, (b) trends in NDVI and mean annual NDVI, (c) trends of SM and rainfall, (d) trends of VOD and NDVI, (e) trends in VOD and mean annual SM, and (f) trends in SM and mean annual VOD, for the identified clusters. The distinct sensitivity to rainfall and trend in VOD of clusters A and D, related to natural reforestation and rain shift or irrigation change, respectively, is evidenced in (a), (c) and (e). Areas with highest mean annual values of VOD related to high biomass are the most sensitive to changes in SM (f). The sensitivity in NDVI and VOD are highly correlated (d), being the exception Cluster D related to rain shift or irrigation change regions with little change in NDVI for the study period (b). Table 1 shows the separability score S between the different clusters (A-E) depending on the different combinations of variable trends (∆SM, ∆VOD, and ∆NDVI). The score S is a standard measure of cluster separability that accounts for intercluster Mahalanobis distances; the lower the value, the most separable the cluster is. We observe that the optimal pair combination is obtained with (∆SM,∆VOD). This combination of variables maximizes cluster separability in all cases and yields the best overall results (lowest OS).  S.5: SMOS-IC L-VOD and soil moisture products The simultaneous retrieval of SM and VOD is based on theoretical analyses showing the possibility of decoupling the effects of SM and VOD and performing 2-Parameters (SM and VOD) retrievals from multi-angular and dual-polarized SMOS L-band observations [1]. These theoretical analyses have been validated from evaluations of both the SM and VOD products. A review of these analyses has been made in the SMOS-IC reference paper [2], and we provide here the key elements of the review. SMOS-IC SM has been evaluated against SM model output simulations at a global scale and in situ observations from the ISMN networks (International Soil Moisture Network, Dorigo21).
Results have shown the high quality of the SMOS-IC SM, which ranked first of all SMOS products in many intercomparison analyses [3,4]. Similarly, the SMOS-IC L-VOD retrievals have been evaluated in numerous studies. As a direct evaluation is not possible (there is no large-scale product of vegetation water content VWC (kg/m2)), we used several proxies Li20. Assuming the average moisture content of vegetation is relatively constant at the interannual scale, L-VOD is closely related to vegetation biomass [5]. For instance, L-VOD is strongly spatially correlated (R > 0.8) to biomass and vegetation height (a proxy of biomass) at continental and global scales using reference biomass maps [2,6]. Interannual variations (IAV) in L-VOD have also been shown to be strongly correlated to the IAV of key factors controlling the IAV of biomass in several forest ecosystems, such as the IAV in forest fraction in the Brazilian Amazon (where forest fraction can be used to monitor deforestation intensity, Qin et al., 2021, Li et al., 2022), and the IAV in burnt areas in the Siberian forests [7] and Australia [8].
Regarding inputs, the SMOS-IC SM and VOD retrievals use very few auxiliary inputs. Contrary to some other retrieval algorithms, SMOS-IC is well known as it does not use any vegetation (as NDVI or LAI) or hydrological variables as inputs, making the independent application studies much more robust. SMOS-IC only uses soil and vegetation temperature parameters estimated from ERA5 model simulations. A detailed analysis of the uncertainties associated with L-VOD retrievals has been made by Fan19. Based on a bootstrap cross-validation method, the analysis considered internal errors (associated with the algorithm process and noise on the SMOS TB observations) and external errors (associated with the reference aboveground biomass (AGB) maps used to calibrate the L-VOD / AGB relationships). Fan et al. showed that external errors are the dominant term in the uncertainties associated with L-VOD. There is an order of magnitude between uncertainties arising from internal and external errors. Considering combined internal and external errors, the relative uncertainties associated with the AGC stocks and changes in the AGC stocks over the tropics are 20-30%. Similar orders of magnitude were found at continental scales.

S.6: Qualitative comparison with previous literature
Many studies have been published trying to characterise the changes in terrestrial ecosystems using trend analysis over various drivers. A direct intercomparison to these studies is challenging, and this is not even possible for many reasons. Here we analyse the works closely related to ours [9,10,11,12,13,14]. Still, crucial differences were observed that preclude a direct comparison because • different considered variables, as none of these works considers SM, and only [12] and [13] consider VOD, but such VOD was retrieved at high frequencies (mostly X-band), which is very sensitive to strong saturation effects in dense vegetation [15] • different study periods, as only [13,14] have an (unfortunately short) coincident period of a few years only; • different and single variables, as changes have been typically identified using a single variable (mainly LAI or NDVI), and the results are assessed by comparison with others-• adopted methodology to estimate trends is typically different. For example, except for [11], the rest of the methods consider non-monotonic trends and do not care for the spatial homogeneity of clusters. We use the method proposed in [11], so a key multiple hypothesis testing correction is applied, which reduces the amount of false positive trends (estimated by as much as 25%, cf. [11]). These properties of the applied approach impact the relevance and statistical significance of the identified groups and consequently make a direct comparison more convoluted.
• spatial resolution used is generally higher [9,11,14] than the 0.25º used in our work, and in the most similar case, i.e. [12], the study period is completely different. This may have an impact when analysing defragmented cropland areas like in Europe.
• issues related to data preprocessing, such as the RFI contamination in mainland China that may affect some L-band estimates and that had to be screened out in our study.
Still, an incomplete (yet reasonable) comparison is possible only with the works [9,11,13], essentially because (a) some similar period but at a higher spatial resolution of LAI trends is studied in [9], which allows us to clarify some divergent patterns, (b) we use the same methodology for computing trends as in [11] over a similar period, but on SM and VOD instead of LAI, and (c) only [13] considers VOD (yet retrieved from high-frequency observations and strongly affected by saturation issues in dense vegetation) at similar spatial and temporal resolutions over a similar period to estimate trends in global terrestrial ecosystems, but using standard linear regression without a non-monotonic or statistical significance test. All these points have been analyzed to compare the methodologie and findings, as well as similarities and contrasting patterns found in the literature, and are reported and summarized in Table 2. We also observe significant changes in Indian crops and capture patterns of natural reforestation in western China. Other croplands, like the dry (southern) US corn belt, are also identified, probably due to irrigation shifts.
We identified the patterns in northeast China (see S2). Still, We decided to remove this region from the study due to potential RFI not being captured by the RFI flags provided and impacting SMOS brightness temperatures and hence the SM and VOD products.
We capture patterns of natural reforestation in southeast China. As in Chen's Fig.2, our work also highlights the southeast China region as being the most active in reforestation.
In Sahel, we identify deforestation, but Chen et al. clearly state a greening cluster compared to previous studies using AVHRR (2000-2016, similar period); the sub-Saharan cluster is missing as in our work.
We did not find statistically significant clustered trends due to croplands in Europe, probably due to the highly heterogeneous (defragmented) croplands, which cannot be captured at the SMOS resolution, or the lack of saturation of L-VOD, unlike in vegetation indices.