**1. Introduction**

During the last decade, the interest in low-frequency microwave remote sensing and technological advances in instrumentation and space technology have resulted in a series of new mission concepts to measure key components of the water cycle. Soil moisture is one of these key components, controlling the partition of energy at the surface and the interactions between the land surface and the atmosphere at varying temporal and spatial scales [1–4]. Theoretical and experimental

evidence supports the idea that L-band (1 to 2 GHz) microwave radiometry is the optimal technology for measuring global surface soil moisture (SM) on an operational basis. The emission of thermal microwave radiation from soils is strongly dependent on soil moisture content. L-band measurements are insensitive to cloud liquid water and, compared to higher microwave frequencies, they are more sensitive to deeper soil moisture layers (up to 5 cm) and penetrate through denser layers of vegetation canopy [5,6]. The latter is especially important for improvements in climate, numerical weather prediction, water and energy cycle science. As a result, the first two satellite missions specifically designed to measuring SM have an L-band radiometer on-board: ESA launched the Soil Moisture and Ocean Salinity (SMOS) mission in November 2009 [7] and NASA launched the Soil Moisture Active-Passive (SMAP) satellite in early 2015 [8]. SMOS unique payload is an L-band synthetic aperture radiometer with multi-angular and full-polarimetric capabilities that provides synoptic views of the Earth's global SM with a spatial resolution of ∼40 km and a 3-day revisit. SMAP has a real aperture L-band radiometer and an L-band radar to enhance the spatial resolution of the estimates from 36 km (radiometer only) to 9 km (radar-radiometer). However, operations of active-passive products ceased abruptly with the failure of the SMAP radar after about ten weeks of operations. Nonetheless, the radiometer is continuing to make measurements and four annual cycles of measurements are about to be completed.

Also during the last decade, satellite sensors with long technological heritage operating in the low-frequency microwave spectrum, that were initially devoted to atmospheric and/or oceanic sensing, have proved suitable for SM retrieval. As a result, several SM datasets from active and passive microwave sensors at C-band (6 GHz) and X-band (10 GHz) partially covering nearly the last 4 decades have been published and shared openly with the international community. Although these sensors are only sensitive to the top 1 cm of soil and have a larger attenuation in presence of vegetation, they can complement recent L-band missions and allow for a multi-decadal soil moisture observational data record. Spaceborne SM data sets from low-frequency microwaves have been widely validated under different biomes and climate conditions by comparison with ground-based observations (e.g., [9–16] and outputs of land surface models ([17–20]). Relevant for this work, Polcher et al. [20] showed that the rainfall driven structures of SM captured by the ORCHIDEE land surface model and SMOS are compatible, and comprehensive validations of SMOS retrievals have been undertaken showing good agreement with other sensors and consistent results over all surfaces, from very dry (Arizona, Sahel) to wet (tropical rain forests) [15].

Soil moisture was recognized by the Global Climate Observing System (GCOS) to be an Essential Climate Variable (ECV) in 2010. This underscores the potential of SM data sets to support the work of the United Nations Framework Convention on Climate Change (UNFCCC) and the Intergovernmental Panel on Climate Change (IPCC). In this context, the ESA's soil moisture Climate Change Initiative (CCI) is one of the first initiatives to merge the different microwave products available into a single soil moisture climate data record (http://www.esa-soilmoisture-cci.org/). One of the key steps in building a multi-decadal soil moisture data record is that, since different products display different ranges of soil moisture values, data have to be harmonized first using a common climatology. In its current version, the ESA CCI SM product uses the climatology provided by the Global Land Data Assimilation System (GLDAS-Noah) as a reference to scale the individual products. Despite some limitations, the ESA CCI SM product is nowadays the most complete and consistent long-term soil moisture data record available, covering (almost) the 40-year period from 1978 to June 2018 [21]. Recent research has been focused into incorporating SMOS to the climate data record [22]. However, the impact of using a model's climatology in absolute retrievals remains unclear, and the climate community has repeatedly argued for the need for a satellite-based SM record to serve as a reference for verifying land surface model performance and trends.

This study presents an SMOS-based L-band climatology that could potentially serve as a reference for the readily available microwave-based soil moisture data sets (from X-, C- and L-bands) into a long-term climate data record exclusively based on observational data sets. Ideally, an L-band climatology should be built from both SMOS and SMAP data. However, since a combined product is not yet available, and SMAP only covers a short observation period, this paper is focused on an L-band climatology solely based on SMOS observations. We show that this observation-driven climatology provides a close representation of the dominant features of temporal variability in the Earth's SM for the period 2010–2016, and allows identifying areas subjected to seasonal, subseasonal and long-term variability. Previous research showed that, despite being a short data set, SMOS provides coherent and reliable SM variability patterns at both seasonal and interannual scales [23,24]. In this work, the Seasonal Trend decomposition using Loess (STL) procedure [25] is implemented and tailored to the first six annual cycles of SMOS data to decompose the temporal variability of the signal. Our analysis quantifies how much SM variation is due to long-term influences, how much is due to seasonal cycle and how much is dominated by subseasonal short-term influences. This knowledge is critical for understanding how well climate data and land surface models compare with the remotely sensed variable. It also allows distinguishing between linear trends and interannual variability, which could be later related to main phenomena of global weather alterations over land (e.g., El Niño Southern Oscillation (ENSO)).

This paper is organized as follows. The SM dataset and the methodology followed to build the SM climatology and to provide a global assessment of SM variability are described in Section 2. Main results are shown in Section 3. The distribution of SM variance among temporal components is first analyzed at selected target sites representative of terrestrial biomes with distinct vegetation type and seasonality. The SMOS temporal series at these sites are compared to GLDAS-Noah and ERA5 modeled SM and to ground-based estimates available at one of the sites (REMEDHUS network, in Spain) to identify consistencies and potential shortcomings of building a climatology with observational data alone. Subsequently, the main features of SM temporal variability are analyzed at the global scale. Conclusions and perspectives from this work are given in Section 4.

#### **2. Materials and Methods**
