A Review on Soil Moisture Dynamics Monitoring in Semi-Arid Ecosystems: Methods, Techniques, and Tools Applied at Different Scales

Abstract

Soil moisture (SM) plays a crucial role in land–atmosphere interaction systems, directly influencing evapotranspiration, photosynthesis, and the water dynamics of the soil surface. Invariably, SM is negatively impacted by disturbances such as fires, which are becoming more frequent across semi-arid ecosystems. Different ecological restoration activities have been implemented to mitigate the impacts of disturbance that, when left untreated, can worsen the effects of recurrent droughts and accelerate desertification and land degradation processes. To measure and monitor the dynamics of SM, advanced techniques and tools have been developed that integrate remote sensing and in situ measurement. This review encompasses various themes on the application of remote sensing for measuring and monitoring SM dynamics in semi-arid ecosystems at different scales. We focused our analysis on the western United States region and thus have developed a review on the following topics: (a) the different data sources (e.g., satellite, unmanned aerial vehicles), (b) approaches to measure field-based SM, and (c) algorithms and techniques to model SM at different scales. We summarize these topics by emphasizing repeatable approaches for the transparent estimation of this variable, identifying current data gaps, and highlighting future trends to fulfill the expanding demand for SM monitoring strategies.

1. Introduction

Proper monitoring of soil moisture (SM) is important to understand variation in crop growth, performance of vegetation restoration practices, and to understand the impacts of climate change on semi-arid ecosystems [1]. There is a need for detailed scientific knowledge to support agricultural sustainability and ecological restoration in these vulnerable regions [2]. Semi-arid ecosystems, which cover extensive areas globally, are vital for biodiversity, carbon cycling, and ecosystem services [3]. However, they face significant threats from erosion, salinity, and human activities [4]. Climate change has exacerbated these issues by increasing temperature and drought, which reduces vegetation productivity and increases vegetation mortality [5]. This impacts soil moisture and heightens the vulnerability of these ecosystems, underscoring the necessity for effective soil moisture monitoring.

Satellite systems that incorporate remote sensing can provide SM information for large areas at lower costs [6]. Numerous satellite missions have provided publicly available global monitoring of land- and sea-ice, snow cover, and SM. Among these missions the Soil Moisture Active Passive (SMAP) mission from the National Aeronautics and Space Administration (NASA) has made global measurements since 2015 every two to three days [7]. The Soil Moisture and Ocean Salinity (SMOS) mission, launched in 2009, is one of the European Space Agency’s (ESA) missions and is the first to provide global observations of variability in SM and sea surface salinity [8]. The Advanced Microwave Scanning Radiometer (AMSR2), launched in 2012 from the Japan Aerospace Exploration Agency (JAXA), is a passive sensor that measures daily global SM and brightness temperature (BT) through the C-band [9].

Remote sensing with unmanned aerial vehicles (UAVs) offers the option of capturing data with high temporal and spatial resolutions. This approach is increasingly employed to develop precision agriculture and efficient SM monitoring [10]. Different sensors can be used on a UAV, such as thermal infrared sensors [11], red, green, and blue (RGB) imaging [12], light detection and ranging (LiDAR) [13], multispectral (MS) [14], and hyperspectral [15] and microwave sensors (L-band passive microwave) [16]. Using UAVs for measuring and monitoring SM compliments data obtained with ground and satellite sensors and offers wider opportunities to capture higher spatial resolution datasets with a temporal frequency that is limited only by weather conditions.

Numerous portable devices equipped with sensors facilitate in situ SM measurement, providing accuracy, speed, reliability, simplicity, and non-destructive sampling. Commonly utilized methods include tensiometers and granular matrix sensors for direct measurement, as well as soil water content sensors like capacitance-based sensors that use the soil charge storing capacity to calibrate to water content [17], the frequency domain reflectometry (FDR) and time domain reflectometry (TDR). Both FDR and TDR are indirect techniques that estimate SM by measuring the electrical conductivity of pore water [18]. It is clear then that SM monitoring can occur at different dimensions (i.e., spatial, temporal, spectral), and our review here is an attempt to summarize these dimensions in research that has been conducted exclusively on semi-arid lands.

Modeling SM over large- and small-scale areas must consider appropriate statistical techniques to evaluate relationships with predictor covariates. Over the past decade, deep learning (DL) and machine learning (ML) algorithms have significantly advanced the field of SM analytics. These computational techniques enable the extraction of complex patterns and relationships from datasets, revealing correlation and causation relationships between SM and various predictive factors that may not be immediately discernible through traditional statistical methods [19]. These techniques have been used for spatiotemporal SM monitoring, SM downscaling, and upscaling, modeling non-linear relationships between SM and environmental covariates, generation of global and regional spatially continuous high-resolution SM datasets, and testing and validation datasets, among others [20].

Recently, efforts have been made to collect and harmonize in situ SM data generated from different organizations and groups. The International Soil Moisture Network (ISMN) is a collaborative effort that integrates and standardizes data from 80 networks and 3147 stations worldwide [21]. These in situ measurements are essential for validating satellite observations, improving climate models, and supporting hydrological and meteorological research [22]. This extensive database, particularly dense in semi-arid lands, plays a pivotal role in validating and enhancing global satellite observations and is crucial for SM monitoring.

The objectives of this review article are threefold: (a) to compile and summarize the datasets, statistical methods, and available online resources for measuring and monitoring soil moisture (SM) in semi-arid zones; (b) to identify the advantages and limitations of available satellite datasets for measuring and monitoring SM; and (c) to evaluate and systematize available SM platforms, geoportals, measurement networks, and products. Our review highlights the critical role of multidisciplinary approaches in advancing SM research. The primary contributions of this review lie in the identification and systematization of key methods, tools, and models for SM measurement and monitoring at various scales, from large landscapes to smaller plots; furthermore, we highlight the gaps in effective SM monitoring.

2. Materials and Methods

We developed a structured framework to categorize the literature into three main areas: (1) remote sensing of SM (including satellites and UAVs), (2) modeling techniques, auxiliary data as covariates and algorithms for SM modeling, and (3) ground-based SM measurements. This framework includes specific technologies such as UAVs and satellites, machine learning data analysis methods, and various spectral and vegetation indices used in SM modeling.

A comprehensive literature review of published, peer-reviewed scientific articles was conducted for the period from 2014 to May 2024 using Harzing’s Publish or Perish software version 8.12. This approach ensured a focus on the most recent methods and advances in SM measurement and monitoring. In addition, Google Scholar was used to search for peer-reviewed articles with keywords like “Soil Moisture”, “Remote Sensing”, “UAV”, and “Semi-arid”. The initial search yielded 896 articles cited more than 34,000 times. After refining the search to eliminate papers not pertinent to the designated ecoregions, algorithms, or remote sensing applications, we identified relevant papers. These selected articles were deemed highly relevant due to their alignment with the core topics and their significant contributions to the state of the art in SM monitoring and modeling. In addition, in some cases, we include gray literature such as technical reports, government documents, and databases from national and global sources to ensure a comprehensive understanding of the topic. We also utilized literature prior to 2014 when contextualizing a relevant subject was necessary.

Each core topic was thoroughly reviewed to assess its current relevance and application, ensuring the review reflects the latest knowledge and practices. The analysis highlighted gaps in current methodologies and technologies and suggested strategies for advancing SM monitoring and measurement in semi-arid regions. Figure 1 illustrates the framework, the topics, and the principal criteria applied in this review.

3. Analysis and Discussion

3.1. Semi-Arid Lands

Semi-arid lands cover approximately 17.7% of the Earth’s terrestrial surface [23,24]. The United Nations Environment Programme (UNEP) characterizes semi-arid regions as areas where the annual precipitation (P) ranges from one-fifth to one-half of the potential evapotranspiration (ET) (0.2 < P/ET≤ 0.5) [25,26].

This paper specifically focuses on these semi-arid lands, following the aridity index (AI) developed by UNEP and is in concordance with the United Nations Convention to Combat Desertification (UNCCD) criteria for defining semi-arid lands [24].

In the central-western region of the Continental United States (CONUS), which is the focus of this article, there are 2.7 million km² of semi-arid lands localized in 17 states and that represent 35% of the total area in the US Central-Western region [27] (Figure 2).

Regarding drought, the US Drought Monitor (USDM) categorizes drought—with a scale from D0 (Abnormally Dry) to D4 (Exceptional Drought)—according to SM, streamflow, and precipitation levels parameters. The USDM reported in April 2023 that 12% of land in the Western states was classified as experiencing extreme or exceptional drought, and an additional 20% was classified as severe. Data reported by the USDM demonstrate that drought incidence in the western United States during the summer of 2023 exceeded all previous droughts in the region since 2000 [28].

The above indicates the complex and increasingly accelerating challenges of drought and land degradation. This also highlights the critical importance of developing strategies to mitigate impacts and improve ecological restoration efforts which focus on enhancing SM conditions in semi-arid lands. Understanding and monitoring SM dynamics is essential for preserving vegetation and soil health in fragile ecosystems, highlighting the need to stay abreast of key advancements, benefits, and gaps in SM monitoring. Additionally, it is crucial to identify and evaluate the most relevant tools, methods, and satellite products, along with field data, for monitoring SM in semi-arid lands, which is the primary focus of this review.

3.1.1. Ecological Restoration in the Western US

Western US semi-arid lands comprise diverse rangeland ecosystems and are often dominated by shrublands at lower elevations and in mountain valleys; these ecosystems face unique threats due to their floristic composition and past disturbances that have degraded vegetation and soils [29] (Figure 3). Of key concern are sagebrush (Artemisia spp. Nutt.)-dominated ecosystems where agricultural expansion, improper livestock grazing management, wildfires, invasive species, and climate change have resulted in considerable reductions since the 1850s [30].

The United States Geological Survey (USGS) in collaboration with the Bureau of Land Management (BLM) launched the Rangeland Condition Monitoring Assessment and Projection (RCMAP) dataset to quantify the percent cover of rangeland components (bare ground, herbaceous, annual herbaceous, litter, shrub, and sagebrush across the western US. Through RCMAP, 2.9 million km² were mapped with the following main results: bare ground 45.5%, shrub 15.2%, sagebrush 4.3%, big sagebrush 2.9%, herbaceous 23.0%, annual herbaceous 4.2%, and litter 15.8% [31]. Rigge et al., (2021) [32] evaluated the trends of rangeland change between 1985 and 2018 and analyzed the relationships with climate drivers, and their results showed the percent net cover of shrubs, sagebrush, and litter significantly (p < 0.01) decreased during this time period.

To rehabilitate ecosystems affected by natural or anthropogenic disturbances, ecological landscape restoration emerges as a widely recommended solution, especially in semi-arid lands [33]. A crucial component of these restoration efforts relates to the development of drought-resistant grasses, legumes, and herbaceous plants capable of competing with invasive weeds while maintaining adequate productivity and quality. [34].

Across the western US, several agencies oversee planning and implementing restoration practices that respond to the specific needs of a particular ecosystem. With regards to rangeland plant material development, the US Department of Agriculture (USDA), Agricultural Research Service (ARS), and Forage and Range Research Laboratory (FRRL) in Logan, Utah has led this process in the region. The FRRL has historically developed plant materials (grass, legume, and forb) that improve the resilience of rangelands and pastures to environmental and anthropogenic stresses [35]. This laboratory currently develops native and non-native plant materials to restore disturbed grasslands [36], recommending plant improvement of the native legume Utah clover (Lotus utahensis Ottley), grasses prairie: Junegrass [Koeleria macrantha (Ledeb.) Schult.], bluebunch wheatgrass (Pseudoroegneria spicata (Pursh) Á. Löve [Poaceae]) and Snake River wheatgrass [(Elymus wawawaiensis) J. Carlson & M. Barkworth]. Likewise, non-native perennial grasses such as crested wheatgrass (Agropyron cristatum, Agropyron spp.), tall wheatgrass (Thinopyrum ponticum [Podp.] Z.-W. Liu & R.-C. Wang), intermediate wheatgrass and prairie brome grass to support sustainable grassland management [35].

Agencies such as the BLM actively use these improved plant materials in various ecological restoration practices associated with fuel management, soil cultivation, rangeland seeding, invasive plant suppression, prescribed burns, soil stabilization, and wildlife management [37,38]. In addition, the primary types of vegetation targeted for these interventions include pinyon-juniper (Pinus-Juniperus) woodlands, creosote bush scrub, desert scrub, riparian woodlands, and big sagebrush shrubland; these practices aim to restore and maintain the ecological balance and resilience of these crucial ecosystems [39].

Figure 3. (a) Spatial representation of Land cover in western US, (b) principal rangeland in western US, year 2021 (Source: Dewitz, J., 2023) [40].

Figure 3. (a) Spatial representation of Land cover in western US, (b) principal rangeland in western US, year 2021 (Source: Dewitz, J., 2023) [40].

3.1.2. Monitoring and Evaluation of Ecological Restoration

In the western US, disturbances from past land use and invasive species (e.g., cheatgrass; (Bromus tectorum L. [Poaceae]) pose significant challenges to sustainable rangeland management. Many lower-elevation basins in the Intermountain West, previously dominated by big sagebrush (Artemisia tridentata Nutt.) or salt desert shrub ecosystems, are now threatened by annual grass invasions [41]. Recognizing the need for science-based solutions to combat these invasions and their impact on rangeland sustainability, substantial restoration efforts have been made. However, success has been limited, and annual grasses continue to spread beyond the Intermountain West [35].

To monitor the impact of restoration efforts, various performance metrics are employed, including canopy cover (using field-based measurements and also estimates extracted using remotely sensed imagery and machine learning techniques), floristic composition, riches and abundance [42], and soil moisture (SM) [43]. The SM is a crucial metric due to its role in soil-plant-water interactions and land-atmosphere interaction systems [10]. Given the significance of SM in evaluating ecological restoration impacts, this review focuses on analyzing and systematizing literature related to the main data sources available (remote sensing and in situ measurement), mapping techniques, and modeling approaches (algorithms) for effective SM measurement and monitoring in semi-arid ecosystems.

3.2. Measurement and Monitoring of Soil Moisture Dynamics

For the evaluation and monitoring of SM in semi-arid areas, various methodologies have been implemented, including satellite imagery, UAV-mounted cameras, and in situ measurements, or a combination thereof. The selection of an approach is primarily determined by the spatial and temporal scale, spectral resolution of the imagery, type of vegetation, and specific parameters being measured or evaluated. In this review, we analyze and discuss the most relevant SM measurement and monitoring approaches, identifying short- and medium-term research gaps and options in semi-arid areas of the western US.

When selecting satellite imagery or UAV imagery for SM monitoring, we must consider the significant differences in spatial and temporal scales. Satellite platforms such as SMAP offer high temporal resolution (frequency of data collection), capturing data daily, which facilitates continuous monitoring of SM at regional and global scales. However, their spatial resolution (pixel dimension) is relatively low, approximately 9 km, limiting their effectiveness in detailed site-specific studies where small-scale soil and vegetation heterogeneity is critical.

In contrast, UAV imagery provides much finer spatial resolution, often reaching centimeter scales, making it ideal for detailed SM studies in small areas, capturing the impact of microtopography, vegetation, and soil structure. However, UAVs have more restricted temporal coverage due to operational constraints such as site accessibility and weather conditions. The choice between using satellite or UAV imagery for SM monitoring should be guided by the study’s specific spatial and temporal requirements, with satellites being preferred for continuous large-scale monitoring, and UAVs for detailed, high-resolution analyses in specific locations.

3.2.1. Remote Sensing: Satellite Dataset for Monitoring Soil Moisture Dynamics

Satellite imagery serves as a crucial tool for SM measurement, leveraging a diverse array of sensor technologies. These include thermal and optical sensors, alongside active and passive microwave sensors which are widely recognized for their efficacy [44,45]. Thermal infrared (TIR) [46], visible/hyperspectral infrared (VIR/NIR), and radar sensors are among the most utilized technologies for this purpose [47]. Wang et al., (2023) [48] conducted a comprehensive review on the efficacy of empirical, semi-empirical and physical models for SM estimation and found that the relationship between SM and soil reflectance using optical sensors is uncertain due to atmospheric condition limitations. They concluded that currently, there is no single model or data source that provides a universally high-accuracy solution for SM estimation. We identified that radar remote sensing is the most commonly used method to estimate SM in bare soils, utilizing physical, empirical, and semi-empirical approaches [49,50].

Models and Approaches for Monitoring Soil Moisture Dynamics Using Satellite Dataset

(i).

Empirical models: leveraging correlations between remote sensing data—such as surface temperature, radar, vegetation index, and ground-measured data, offer simplicity and ease of data access but suffer from limited interpretability and reliance on specific calibration datasets. In our review, we found that the Water Cloud Model (WCM) and the Dubois Model are commonly used to estimate SM using synthetic aperture radar (SAR) backscatter. [51]. Dubois et al., (1995) [52] evaluated several SAR data sets by comparing derived SM values with in situ measurements collected in a variety of scenes between 1991 and 1994 in space (SIR-C) and Zribi et al., (2019) [53] applied the WCM to the L-band PALSAR/ALOS-2 satellite data for SM estimation. Their findings indicate that increasing the number of polarimetric parameters at the C-band can significantly enhance the robustness of surface SM estimates.

(ii).

Semi-empirical models: integrating both statistical correlations and physical principles, account for factors like vegetation and soil texture, enhancing precision over purely empirical models while still necessitating observed data for calibration [54]. This type of model strikes a balance between accuracy and operational complexity, suitable for applications requiring moderate precision without the comprehensive data demands of physical models [55]. Many semi-empirical models estimate SM by considering the relationship between the backscatter coefficient, surface parameters, and radar parameters [50,56].

(iii).

Physical models: grounded in the fundamental laws of physics like the Integral Equation Model (IEM), provide detailed and interpretable SM estimates, especially for bare soil areas, at the cost of increased complexity and computational demands [57].

High potential has been identified by combining multiple models and data sources to refine SM retrieval accuracy and applicability, leveraging multi-source data to enhance spatial-temporal resolution, extend time series analyses, and improve model interpretability and precision [58].

Products for Monitoring Soil Moisture Dynamics Using Satellite Dataset

• Downscaled products: where the models stand out SMAP/Sentinel (SPL2SMAP) product (1 and 3 km spatial resolution—global) [59], European Space Agency—Climate Change Initiative (ESA-CCI) Surface Soil Moisture (SSM) product (1 km spatial resolution—global) [60], European Remote Sensing Satellite-2 (ERS-2) scatterometer (SCAT) (1 km spatial resolution—Europe) [61], and SMAP-HydroBlocks (30 m spatial resolution—CONUS) [62].
• Microwave remote sensing products: CGLS Sentinel-1 (1 km spatial resolution—Europe) [63], SMAP Enhanced L3 (36 and 9 km spatial resolution—global) [64], SMOS-IC retrieval data set based on the SMOS L-band observations (1 km spatial resolution—global) [65].
• Reanalysis products: European ReAnalysis (ERA5-Land) (0.1° spatial resolution—global) [66], NASA Global Land Data Assimilation System Version 2 (GLDAS-Noah) (0.25° spatial resolution—global), Global Land Surface Satellite Soil Moisture product (GLASS SM) (1 km spatial resolution—global) [67].

Table 1. Comparative analysis of satellite dataset for soil moisture measurement and monitoring.

Dataset	Spatial Resolution	Temporal Availability	Advantages	Limitations
SMAP (Soil Moisture Active Passive)	9 km (SMAP L4, L3_SSM_E 36 km (L3_SSM)	Daily and 3 h	– High accuracy – Effective in vegetative and bare soils (L band)	– Sensitivity to dense vegetation
Sentinel-1	5–20 m	6–12 days	– High spatial resolution Effective under vegetation cover – Use of synthetic aperture radar (C Band)	– Complex data processing required
AMSR-E/AMSR2	5.4 km to 25 km	Daily	– Daily global coverage Less affected by cloud cover (C band)	– Limited spatial resolution for local applications
ERS SAR	Up to 25 km	35 days (modifiable to 3 days in specific modes)	– Vegetation penetration Useful for SM and topography (C band)	– Temporal resolution limitations
PALSAR-2	10–100 m	14 days	– Good vegetation penetration (L band) – High resolution for SM detection	– Data processing complexity – Limited by temporal resolution
Advanced Scatterometer (ASCAT)	12.5 and 25 km grid	2–3 days	– High Wavelength (5.7 cm), and high radiometric accuracy.	– ASCAT retrievals show larger errors over arid regions of the world (Sahara, Western US, deserts of Australia)
SMOS (Soil Moisture and Ocean Salinity)	25 km	2–3 days	– Global coverage (L band) – Can measure SM under moderate vegetation	– Coarser spatial resolution – Sensitivity to high levels of vegetation and rough surfaces

synthesizes satellite options for SM measurement, highlighting how the choice of an instrument and associated model depends on specific project needs, including desired resolution, geographical coverage, and operational limitations.

3.2.2. Ground-Based Soil Moisture Measurements

In situ SM measurements play a pivotal role in the processing, calibration, and validation of geospatially explicit data obtained from remote sensing and models [22]. The major repository of SM data comes from the ISM N, which is a worldwide consortium that integrates and standardizes SM data from various in situ monitoring stations. It offers openly accessible, advanced quality control methods, data essential for the validation of satellite-derived SM estimates, reanalysis of climate models, and support of extensive hydrological and meteorological research [21]. As of April 2024, the ISMN contains data from 80 networks and 3147 stations located all over the globe, with data from 1952 to the present (https://ismn.earth/, accessed on 8 April 2024). Nevertheless, the global in situ SM measurement network is unevenly distributed across the world, with the highest station density concentrated in the arid and semi-arid zones. The ISMN is particularly sparse in the tropics, southern hemisphere, and boreal regions.

In our review, we found many examples of using ground-based SM measurements and produced geospatial products, for example, (Han et al., (2023) [19] developed a new global data set of soil surface moisture (GSSM1 km) at a high spatial resolution (1 km) and temporal (daily), covering the years 2000 to 2020. They used a Random Forest (RF) Model combining meteorological and biophysical co-variable predictors of SM derived from satellite and reanalysis data sets. Predictor variables or co-variables included antecedent precipitation evaporation index (APEI), LST, and various vegetation indices such as NDVI and EVI. Zheng et al., (2023) [60] developed a global surface soil SM model with a daily resolution of 1 km gap-free from 2000 to 2020. They combined the SSM product of the European Space Agency—Climate Change Initiative (ESA-CCI) and ERA5 reanalysis data. They also applied the RF algorithm to disaggregate the coarse-resolution SM into 1 km, with the help of in situ observations from the ISMN and other optical remote sensing datasets. Orth et al., (2021) [20] developed a global, long-term dataset of soil moisture (SoMo.ml), using Machine Learning (ML) models trained on in situ measurements. This dataset extrapolates daily SM dynamics both spatially and temporally, utilizing data from over 1000 global stations. SoMo.ml provides SM data at multiple depths (0–10 cm, 10–30 cm, and 30–50 cm) with a spatial resolution of 0.25° and daily temporal resolution, covering the period from 2000 to 2019.

In the US, the National Integrated Drought Information System (NIDIS) is leading the effort to establish the National Coordinated Soil Moisture Monitoring Network (NCSMMN). NIDIS is working with the US Department of Agriculture (USDA) and other partners to integrate SM data from around the country, and the principal goals is to establish a national “network of networks” that effectively demonstrates data and operational coordination of in situ network to weather monitoring, climate, agricultural and forest/ecological monitoring, remote sensing and model validation and flood forecasting [68].

Most long-term SM monitoring networks are operated by federal and state agencies, and this network is expanding. As of the writing of this paper in June 2024 there are more than 2100 stations that are currently in operation and are made up of 29 networks from different agencies (http://nationalsoilmoisture.com, accessed on 17 April 2024). This expansion is part of the National Soil Moisture Network (NSMN). Particularly in Oklahoma, the SM monitoring network has expanded and stations are part of an environmental monitoring network called Mesonet. Of all the networks that make up the National Coordinated Soil Moisture Monitoring Network (NCSMMND), the main ones are the Soil Climate Analysis Network (SCAN), the Snow Telemetry Network (SNOTEL), and the US Climate Reference Network (USCRN).

In situ networks for SM monitoring, while extensive, do not cover all regions of the United States, much less the entire world. To address these coverage gaps, researchers employ computer models that integrate extensive data sets, including readings of precipitation, temperature, humidity, vegetation, and topography. These models simulate SM dynamics in areas lacking direct observation data, thus providing a comprehensive global perspective on SM trends. One notable example of these applications is the research carried out by Vergopolan, et al., (2021) [62], where they developed SMAP-HydroBlocks. This is a satellite-based surface SM dataset with an unprecedented spatial resolution of 30 m for the contiguous United States. This data set was created by combining HydroBlocks model [69] with a Tau-Omega Radiative Transfer Model, using microwave data from the SMAP satellite and in situ observations. For the analysis and validation of the SMAP-HydroBlocks data set, they used more than a thousand observation sites from the Snow Telemetry (SNOTEL) Network. Yuan et al., (2021) [70] evaluated long-term SM trends from 18 general circulation models that are part of the Coupled Model Intercomparison Project Phase Six (CMIP6) and in situ observations over continental United States (CONUS), including 303 in situ SM stations from five networks. Ford et al., (2020) [71] evaluated the fidelity of in situ SM observations from more than 1233 stations in the contiguous United States to derive gridded in situ SM products. They found that 90% of the evaluated stations exhibit high spatial consistency with SM data sets from satellite remote sensing and land surface models. Figure 4 shows the geographic localization of ISMN network, and

Table 2. Comparative Analysis of In Situ Soil Moisture Measurement Networks.

Network Name	Application	Advantages	Limitations	Application in Semi-Arid Lands
ISMN	Global monitoring, validation of satellite-derived SM estimates, climate model reanalysis.	Extensive global coverage, standardized data, advanced quality control methods, and long-term data availability.	Uneven global distribution sparse coverage in the tropics, southern hemisphere, and boreal regions.	Integrate with satellite data for comprehensive analysis. Expand network coverage in underrepresented regions.
NCSMMN	US-wide monitoring, weather forecasting, climate, agricultural, and ecological monitoring.	Large number of stations, integration with various data networks, and real-time data availability.	Limited to the US, varying sensor depths and data quality among different networks.	Standardize sensor depths and data quality. Use for national-scale SM monitoring and model validation.
SCAN	Agricultural and ecological monitoring in the US.	Long-term data, multiple sensor depths, and integration with other networks.	Limited to agricultural areas, data quality varies with station maintenance.	Regularly maintain and calibrate sensors. Use for detailed agricultural SM assessments.
SNOTEL	Monitoring snowpack and soil moisture in mountainous regions of the US.	High-quality data, critical for water resource management, and integration with SM networks.	Limited to mountainous regions, primarily focused on snowpack data.	Combine with other networks for comprehensive hydrological studies.
USCRN	Long-term climate monitoring, providing high-quality climate data, including SM.	High data accuracy, multiple sensor depths, and broad spatial coverage in the US.	High maintenance costs are limited to the US.	Use for long-term climate and SM trend analysis. Integrate with global networks for comprehensive climate studies.
Mesonet Networks (e.g., Oklahoma Mesonet)	Regional environmental monitoring, including SM, across various US states.	High spatial and temporal resolution, real-time data, integration with weather and climate models.	Regional focus, varying sensor depths, and data quality.	Standardize data collection protocols across states. Use for high-resolution regional SM monitoring and agricultural planning.

provides an overview of SM monitoring networks in the United States.

In situ measurements are essential for the calibration and validation of remote sensing data and models, providing precise information for SM assessment, which is crucial for ecological restoration and sustainable land management in semi-arid lands. Table 2 organizes and highlights the importance of in situ measurements in monitoring SM dynamics in semi-arid lands, including a detailed comparative analysis of the main in situ SM monitoring networks.

3.2.3. Integration of Satellite and Ground-Based Observational Data for SM Monitoring Dynamics

SM monitoring products and models derived from reanalysis and data assimilation processes are advancing rapidly with the introduction of innovative in situ measurement, satellite, and other remote sensing technologies alongside improved modeling capabilities [68]. Given that many SM datasets are derived from models, the application of satellite remote sensing has grown to supplement the gaps in observational data that models alone cannot fill [71].

The availability of remote sensing products and SM monitoring networks is extensive, particularly in the semi-arid regions of the Western US, where a high-density SM measurement station network is found. Leveraging satellite data and SM monitoring network data, a range of global reanalysis products, especially in the USA, have been developed. These reanalysis products integrate satellite and ground-based observational data products and aim to create gridded data for entire territories, employing semi-empirical and physical models to enhance measurement precision by applying the modeling and data assimilation framework (Table 2).

Among the main products that integrate terrestrial and satellite observation data for SM monitoring we find the Climate Prediction Center’s (CPC) (https://www.cpc.ncep.noaa.gov/, accessed on 24 April 2024) [72]. The CPC encompasses global monthly SM, evaporation, and runoff data starting from January 1948, offering both daily and monthly updates from 2008 to the present. This product leverages the Leaky Bucket hydrological model, utilizing observed precipitation and temperatures to estimate SM [73]. Also the “Groundwater and Soil Moisture Conditions from GRACE Data Assimilation” delivers weekly drought indicators for groundwater and SM, based on Gravity Recovery and Climate Experiment (GRACE) satellite data and other observations, integrated through a numerical model [74]. The NASA SPoRT-LIS Soil Moisture Products provide high-resolution (~3 km) real-time SM data to support regional and local modeling, employing the Noah land surface model, updated every 6 h since 2010 [75]; and the Soil Climate Analysis Network (SCAN), initiated in 1991, focuses on US agricultural areas, with over 210 stations collecting comprehensive SM and atmospheric data [76].

These products are typically utilized for the monitoring of regenerative agriculture, particularly in grasslands/farmlands as well as for assessing soil conditions, crop growth, drought monitoring over large spatial areas, and various metrics and indicators of land degradation in semi-arid lands. We identified that this product presents valuable opportunities to develop methods for transparent, consistent, and replicable monitoring and reporting systems for SM at a landscape scale.

Table 3. Soil moisture products and models derived from reanalysis and data assimilation processes.

Product	Temporal Resolution	Time	Data	Description	Application in Semi-Arid Lands
National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center Soil Moisture Products	Daily and monthly	2008–Present	Land-based Station	The model takes as forcing observed precipitation and temperature and calculates SM, evaporation, runoff, and snowpack www.cpc.ncep.noaa.gov, accessed on 24 April 2024	Use for large-scale climate studies and trend analysis. Integrate with satellite data for comprehensive assessments.
Groundwater and Soil Moisture Conditions from GRACE Data Assimilation	7 days	2003–Present	Satellite	Generate groundwater and SM drought indicators each week. https://nasagrace.unl.edu/, accessed on 15 April 2024	Combine with ground-based measurements for enhanced accuracy.
NASA SPoRT-LiS Soil Moisture Products	Daily	2010–Present	Model Radar Satellite	Provides high-resolution (about 3-km) gridded SM products in real-time to support regional and local modeling and improve situational awareness. https://weather.ndc.nasa.gov/sport/, accessed on 8 April 2024	Apply for regional SM monitoring, integrate with local in situ measurements for model validation.
Soil Moisture Monitoring Network.	Near real-time	2018–present	Model	Provides high-resolution gridded SM products (4 km grid) derived from in situ SM measurements http://nationalsoilmoisture.com/, accessed on 7 May 2024	Utilize for high-resolution local and regional studies. Ensure regular calibration and maintenance of in situ sensors.
Soil Climate Analysis Network (SCAN)	Near real-time	1991–Present	Land-based Station	SM measurements are observed at the depths of 5, 10, 20, 50, and 100 cm. www.nrcs.usda.gov, accessed on 7 May 2024	Focus on agricultural SM monitoring, integrate with other networks for broader applications.
North-American Land Data Assimilation System (NLDAS)	Near real-time	1979–Present	Model	Reduce the errors in the stores of SM and energy which are often present in numerical weather prediction models. www.cpc.ncep.noaa.gov, accessed on 15 May 2024	Employ for improving weather prediction models, integrate with satellite data for comprehensive climate studies.
Crop Condition and Soil Moisture Analytics Tool (Crop-CASMA)	daily with a 3-day delay.	2015–present	Satellite	SM and vegetation index data derived from NASA SMAP and MODIS missions to assess conterminous US crop vegetation and SM conditions. https://nassgeo.csiss.gmu.edu/CropCASMA/, accessed on 17 May 2024	Use for detailed agricultural SM monitoring, integrate with other agricultural data for comprehensive analysis.

systematizes and shows the main and most used products that integrate satellite and ground observation data for SM monitoring.

3.2.4. Remote Sensing: Unmanned Aerial Vehicle (UAV) for Monitoring Soil Moisture Dynamics

The principal advantages of using drone-based SM monitoring include precision, efficiency, timeliness, scalability, and cost-effectiveness. UAVs can be equipped with diverse sensors, including optical, infrared, and laser imaging detection and ranging (LIDAR), serving as a crucial platform for remote sensing observations. Through UAV-based remote sensing, high-resolution imagery can be acquired, achieving pixel sizes down to the centimeter or millimeter scale, enabling detailed and precise SM monitoring and analysis [15].

In our review, we identified the most prevalent sensors equipped on UAVs for SM monitoring:

• RGB cameras: which capture images in the visible spectrum primarily for vegetation mapping, offering high-quality RGB band images at a relatively low cost [77];
• Multispectral and hyperspectral cameras: capturing images across multiple spectral bands to compute various spectral indices (Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI), Normalized Difference Vegetation Index (NDVI), and others.) for assessing vegetation conditions, soil physical parameters, and SM content; this image generally require radiometric and atmospheric corrections [78,79];
• Thermal cameras: commonly used for mapping temperature, evapotranspiration, and SM by capturing infrared radiation emitted by the targets, allowing for temperature differences and thermal anomalies identification. Thermal cameras have been introduced for SM mapping by utilizing the relationship between water content and land surface temperature (LST) [80], with their most common application in agriculture for monitoring crop water stress and irrigation management [81];
• Shortwave near infrared (SWIR) camera: helps generate various SWIR reflectance indices that correlate more closely with the total water content of the vegetation pores and infer SM content [82];
• LIDAR: significantly applied for 3D reconstruction, flood monitoring, snow depth, and erosion assessments, though they are costlier and require ground filtering corrections [83].

Given the nature of SM measurement via UAV remote sensors, it is essential to develop models linking SM content to derivable remote sensing indicators like surface temperature and vegetation indices [84]. For instance, Zhang et al., (2020) [85] mounted multi-type cameras at plot scale (visible RGB camera, near-infrared camera, and thermal infrared camera) and calculates the GNDVI (green normalized difference vegetation index) in the Tibetan semi-arid lands to analyze the role of the size and type of patchiness in grassland to monitoring the SM. Similarly, Lu et al., (2020) [86] in China’s semi-arid lands found a significant correlation with the brightness of UAV visible images and SM, suggesting the SM can be estimated using image brightness combined with vegetation coverage. Paridad et al., (2022) [77] also demonstrated significant correlations between soil texture properties and surface temperature for SM estimation in Iran’s arid lands, using RGB and thermal imaging sensors from UAV platforms. Figure 5 shows the principal sensors carried by UAVs used for monitoring SM at project scales.

3.3. Modeling

3.3.1. Algorithms

Artificial Intelligence (AI) models are increasingly applied in soil science to predict local-scale near-surface SM data. These models use coarse-scale predictors such as topographic features, evapotranspiration, land surface temperature, and vegetation indices from remote sensing products to enhance the resolution and accuracy of SM estimates [87]. The use of ML techniques can significantly improve SM estimation through various applications such as those described below:

• Pedotransfer Functions: ML is particularly effective in predicting soil hydraulic properties [88,89].
• Predictive Models for SM Estimation: These models are crucial for meteorological studies, runoff and soil erosion estimates, and predicting droughts and floods [6,90,91].
• Remote Sensing for SM Retrieval: Utilizes optical, infrared, and primarily microwave wavelengths to estimate SM content without physical soil contact [20,21].
• Downscaling Satellite-Derived SM Products: Improves the resolution of satellite-based SM data, enhancing usability for detailed local applications [60,62].

The use of ML algorithms to downscale passive microwave surface soil moisture (SSM) products, like SMAP and SMOS, is increasing due to their typically low resolution, which limits their application and validation at regional and local scales. Vergopolan et al., (2021) [62] employed RF to enhance these SM products to 30 m of spatial resolution and Han et al., (2023) [19] provided SSM (0–5 cm) at 1 km spatial and daily temporal resolution using the RF algorithm and demonstrating the potential of ML to improve the usability of coarse-resolution data for more precise regional studies.

In modeling the complex, non-linear relationships impacting SM, we found the most prominent algorithms include RF, support vector machine (SVM), and artificial neural network (ANN). These methods are favored for their robustness in handling variable interactions and their effectiveness in deriving accurate SM predictions, e.g., in semi-arid areas, the multiscale extrapolative learning algorithm (MELA) was applied using hydroclimatic data obtained from remote sensors to extrapolate local monthly SM measurements to multiple depths from 2015–2021 [91]. Adab et al., (2020) [90] used optical and thermal sensors from Landsat 8 in Iran’s semi-arid restoration areas, and they identified that the RF method outperformed others, achieving the highest Nash–Sutcliffe efficiency value [92]. Han et al., (2023) [6] evaluated eight ML algorithms for estimating SSM, and found that the ensemble model with K-neighbors regressor (KNR), RF, and extreme gradient boosting (XB) had the best performance.

Paradian et al., (2020) [93] observed an overall increase in the use of ML models in soil science yet noted a relative decline in the usage of certain models like SVM, multivariate adaptive regression splines (MARS), and classification and regression trees (CART). This shift reflects a trend towards adopting models such as RF, which are perceived as offering enhanced capabilities in handling complex soil science data. The RF algorithm is used in ML due to its capability to handle low interpretability scenarios [94]. However, its popularity also stems from its ability to facilitate easier interpretation through the identification of important variables, enhancing its utility in various applications [95].

The application of interpolation methods for SM measurement and monitoring is infrequent due to the scarcity of in situ observations and the significant spatial variability of hydrometeorological conditions and soil characteristics. Yuan et al., (2017) [96] employed the reduced optimal interpolation (ROI) method using data from 65 stations in the Oklahoma Mesonet Network and compared it with other methods such as Co-kriging and Inverse Distance Weighting (IDW). They found that ROI was significantly more accurate than IDW and outperformed Co-kriging in their results.

In a comprehensive review of ML for SM assessment, Rani et al., (2022) [97] concluded that machine learning algorithms like ANN and SVM disfavor the understanding of the relationship between dependent target variables and independent input variables due to their complex model structures; in contrast, algorithms such as CART and RF are favored for their transparency and ease of interpretation, facilitating a clearer comprehension of how input variables influence model outcomes.

The performance of algorithms for SM measurement significantly depends on the quality and volume of available data. High-resolution spatial and temporal data, which are accurate and consistent, including multispectral and hyperspectral information along with in situ measurements, are crucial. While no specific data quantity threshold for SM modeling is established, extensive datasets with broad temporal and spatial observations are preferred. Long-term satellite data with high temporal frequencies and repeated UAV/in situ measurements across different seasons are recommended. Employing cross-validation techniques that integrate satellite, UAV, and in situ data is essential to reduce uncertainty and increase the reliability of SM models.

3.3.2. Numerical Modeling

Numerical modeling is crucial for estimating SM, providing robust frameworks to simulate SM dynamics under varying climatic and soil conditions [98]. Prominent models include the Soil and Water Assessment Tool (SWAT), the Community Land Model (CLM), the Variable Infiltration Capacity (VIC) model, and Hydrus-1D model. The SWAT model predicts the impact of land management on water, sediment, and chemical yields in large watersheds [99]. CLM offers accurate SM simulations by integrating biogeophysical and biogeochemical processes [100]. VIC is suitable for large-scale SM estimation, balancing water and energy while emphasizing spatial heterogeneity [101]. Hydrus-1D effectively profiles soil moisture by simulating water, heat, and solute movement in variably saturated media [102].

These models utilize key equations such as Richard’s Equation, Van Genuchten-Mualem’s Equation, Darcy’s Law, and Brooks-Corey’s Equation. Advantages include the ability to integrate diverse data sources, flexibility in simulating different scenarios, and robustness in handling complex soil, vegetation, and climate interactions. However, these models often require extensive calibration, are computationally intensive, and may lack flexibility in adapting to new data without significant reconfiguration.

Figure 6 shows a framework for SM measurement and monitoring, including the integration of instruments, environmental data, numerical model, and machine learning model.

3.3.3. Soil Moisture Modeling: Relationship to Soil, Vegetation, and Climate

The assessment of SM is intrinsically linked to the interaction between soil properties, vegetation, and climate. Remote sensing technologies, such as satellite imagery and UAV data collection, predominantly focus on surface soil moisture (SSM), capturing data from the uppermost soil layer. These measurements are significantly influenced by factors like soil texture, surface roughness, and vegetation cover, all of which are crucial for determining the accuracy and reliability of the moisture data obtained.

Evapotranspiration (ET), a variable that integrates the effects of soil, vegetation, and climate, adds further complexity to this relationship. High-resolution UAV data can capture detailed variations in ET at the plot or field level (small areas), reflecting microtopographic features, vegetation density, and soil structure. This data is vital for understanding the dynamic processes of SM in heterogeneous landscapes, where small-scale variations in vegetation and soil properties are pivotal to the overall moisture balance. Conversely, satellite-based measurements, such as MODIS satellite products provide ET estimates with a spatial resolution of 500 m to 1 km and daily temporal resolution; while effective for regional and global assessments, this low resolution may overlook the fine-scale interactions between soil, vegetation, and climate that are essential for site-specific SM management and ecological studies.

Table 4. Soil moisture modeling techniques and their relation to soil and vegetation characteristics.

Measurement Method	Influence of Soil Surface Characteristics	Relevant Soil Properties
Satellite Remote Sensing	Captures surface SM, influenced by soil texture, roughness, and vegetation	Soil texture, surface roughness, vegetation cover [64].
UAV-Based Data Collection	High-resolution imagery sensitive to surface conditions, vegetation, and topography	Surface roughness, vegetation density, soil texture [15].
In situ Measurements	Direct measurements that can be calibrated against soil structure and SM content	Soil porosity, texture, organic matter content [21].
Empirical Models	Relies on statistical relationships, influenced by surface reflectance and vegetation	Reflectance properties, vegetation indices (NDVI, etc.) [51,103].
Semi-Empirical Models	Combines physical principles and statistical relationships, considers surface conditions	Soil texture, surface roughness, vegetation cover [50,56].
Physical Models	Based on fundamental physics, requires detailed surface properties for accuracy	Soil dielectric properties, surface roughness, moisture [57].
Microwave Remote Sensing	Sensitive to surface SM content and roughness, can penetrate vegetation to some extent	SM content, vegetation cover, roughness [83].

provides a descriptive summary of how soil moisture modeling techniques relate to soil and vegetation characteristics.

4. Conclusions and Recommendations

4.1. Conclusions

• In our review, we identified a significant gap in the application of SM as a performance indicator to assess the impact of ecological restoration in areas affected by disturbances in semi-arid zones. Although we identified great technological development to measure SM, this gap highlights the need to focus research and develop methodologies that incorporate SM measurements to evaluate and improve restoration results in semi-arid ecosystems; e.g., Proctor et al., (2022) [104] highlighted the importance of SM instead of precipitation for crops yields and other purposes.
• The integration of DL and ML algorithms has significantly enhanced SM analytics, facilitating detailed spatiotemporal monitoring; additionally, the ISMN plays a pivotal role in harmonizing in situ data, essential for validating satellite observations.
• ML techniques, such as RF, SVM, and ANN, are increasingly applied due to their robustness in handling complex, non-linear relationships between SM and various environmental covariates; the integration of these algorithms enhances the resolution and accuracy of SM estimates.
• Our review underscores the substantial potential of ML algorithms in the geospatial monitoring of SM across regional and local scales, integrating in situ and satellite remote sensing data, notably through the widespread use of the RF algorithm in conjunction with the SMAP satellite product.
• There is also a promising trend towards developing algorithms that reduce reliance on auxiliary data, or utilize readily available remote sensing observations to improve SM monitoring capabilities.
• Ground-based soil moisture (SM) measurements are indispensable for the calibration and validation of remote sensing data and models, providing precise and localized information critical for ecological restoration and sustainable land management in semi-arid ecosystems. Our analysis underscores the extensive yet varied global distribution of in situ SM measurement networks, highlighting their pivotal role in enhancing the reliability of satellite-derived SM estimates and supporting climate model reanalysis. Notable examples, such as the ISMN and NCSMMN, demonstrate the integration of large-scale networks that offer real-time, high-quality data essential for comprehensive SM monitoring.
• The use of empirical, semi-empirical, and physical models significantly enhances the accuracy and utility of soil moisture (SM) estimates. Furthermore, the integration of satellite products such as SMAP/Sentinel, CGLS Sentinel-1, and ERA5-Land contributes to robust, detailed, and precise SM monitoring. This comprehensive approach, combining diverse data sources and modeling techniques, establishes an effective framework for addressing the complex dynamics of SM.
• UAVs offer promising potential for detailed SM monitoring due to their high-resolution capabilities. However, their high costs, expertise requirements, and dependence on weather conditions, along with limited resources, necessitate the development of methods to integrate UAV data with satellite observations. This approach can provide reliable estimates over vast tracts of land typical of the western US, optimizing both feasibility and cost-effectiveness in large-scale SM monitoring. Additionally, a gap was identified to demonstrate how RGB/multispectral imaging can be used to “substitute” the use of expensive shortwave infrared (SWIR) sensors. Demonstrating how RGB or multispectral imaging can effectively replace SM estimation with SWIR in UAVs presents an important area for future research.
• Despite advancements, challenges remain, particularly in the application of interpolation methods and the need for high-resolution, precise data for disturbed vegetation in semi-arid regions.

4.2. Recommendations for Future Research

• Future efforts should focus on standardized protocols and guidelines for the integration of SM measurements into ecological restoration projects with a particular emphasis on semi-arid landscapes. These protocols should include the development of comprehensive methodologies that not only incorporate advanced remote sensing technologies but also ensure the consistency and comparability of data across different scales and regions. Establishing such protocols will enhance the reliability of SM as a performance indicator, enabling more accurate assessment and improvement of restoration outcomes in semi-arid ecosystems.
• Upscaling applications (from the field measurements to the UAV to the satellite) should be strengthened so that SM monitoring can be implemented at different scales. While UAV estimations of SM provide high-resolution data, their limited scope is insufficient for covering the vast landscapes characteristic of the western USA. Future research should focus on integrating UAV-derived SM data with satellite observations to enhance spatial coverage and data reliability. Additionally, the potential of using cost-effective RGB or multispectral imagery as substitutes for expensive SWIR sensors in UAVs should be investigated to ensure economic feasibility without compromising data quality.
• There are various national and regional land use and cover mapping efforts for the Western US that provide estimates of canopy cover useful for assessing ecological restoration performance. However, it is crucial to identify and demonstrate which of these products offer the highest accuracy.
• Assessing the spatiotemporal dynamics of soil moisture (SM) as a performance indicator of ecological restoration is a significant future research challenge and opportunity. For instance, it is relevant to identify and measure which specific types of restoration practices effectively improve SM in particular landscapes.