Quantifying Soil Carbon Sequestration Potential Through Carbon Farming Practices with RothC Model Adapted to Lithuania

Abstract

Climate change poses one of the greatest challenges of our time, with greenhouse gas (GHG) emissions significantly contributing to global warming. The agriculture, forestry, and land-use (AFOLU) sectors not only emit GHGs but also offer the potential for carbon sequestration, which can mitigate climate change. This study presents a methodological framework for estimating soil organic carbon (SOC) changes based on carbon farming practices in northern Lithuania. Using satellite-derived indicators of cover crops, no-till farming, and residue retention combined with soil and climate data, SOC dynamics were modeled across the Joniškis municipality for the period 2019–2020 using the Rothamsted Carbon Model (RothC) model. The integration of geospatial data and process-based modeling allowed for spatial estimation of SOC change, revealing positive trends ranging from 0.23 to 0.32 t C ha⁻¹ year⁻¹. Higher increases were observed in areas where multiple carbon farming practices overlapped. The proposed workflow demonstrates the potential of combining Earth observation and modeling approaches for regional-scale carbon assessment and provides a basis for future applications in sustainable land management and climate policy support.

1. Introduction

Climate change is the most pressing challenge of our time, significantly driven by GHG emissions [1]. These gases including carbon dioxide (CO₂), methane (CH₄), nitrous oxide (N₂O), and fluorinated gases contribute to global warming by trapping heat [2], leading to increasingly frequent meteorological extremes such as storms, floods, and droughts. Among anthropogenic activities, the agriculture, forestry, and land-use (AFOLU) sector plays a dual role: it is both a source of GHG emissions [3,4,5] and a potential sink through carbon sequestration [1,6].

Carbon sequestration is the process of capturing atmospheric CO₂ and storing it in soil organic matter through biological inputs such as plants and their residues [7,8,9]. When implemented in soils, it is increasingly recognized as an effective strategy to mitigate climate change while simultaneously enhancing soil health and agricultural productivity [1,10,11]. Climate and land management are closely interlinked factors that determine the dynamics of SOC. Climate regulates the turnover of organic matter through temperature and moisture regimes, while land use determines the type, timing, and quantity of carbon inputs into the soil system. Importantly, these two factors do not act independently—their interaction can amplify or offset changes in SOC, depending on local conditions. For example, warming may increase the rate of decomposition, but this may be offset by land management practices that promote biomass input, such as cover crops or agroforestry [12,13,14]. Conversely, intensive land use under adverse climatic conditions can accelerate carbon loss through degradation and reduced vegetation cover [7,15]. Some scientists found that land management has an even greater impact than climate [12]. Nonetheless, understanding the impact of these two factors on the overall process is crucial to developing an effective carbon sequestration strategy in agriculture [7,10,16].

Carbon sequestration can improve soil fertility and structure, increase water retention, and reduce erosion and nutrient runoff. These improvements are primarily driven by enhanced microbial activity, increased soil aggregation, and improved water infiltration capacity, as documented in [16,17,18,19]. These positive changes are mainly due to SOC, which is a key factor in maintaining soil health, and is derived from plant residues, microbial biomass, and organic matter decomposition [20].

Traditional SOC assessment by field sampling is accurate but prohibitively costly and logistically difficult over large areas. In this context, remote sensing (RS) and geographical information systems (GIS) technologies offer promising alternatives for large-scale, consistent monitoring of land cover land use (LCLU), in addition to vegetation dynamics and indicators of sustainable land management practices [21].

Sustainable farming practices that focus on carbon sequestration are commonly referred to as carbon farming. Practices such as cover crops, reduced or no-tillage farming, and residue retention have been shown to enhance SOC levels [13,14,22,23,24,25,26]. However, their effectiveness is still context-dependent, as different research studies show conflicting results over time, depending on the region and the length of implementation. Some field studies show that SOC accumulates without tillage and with cover crops, while others show limited or no long-term effects, most likely due to differences in soil type, climatic conditions, and intensity of farm management. These mixed results highlight the need for region-specific assessments using harmonized data and modeling approaches [13,14,22,27].

Cover crops are planted to maintain soil cover between main crop cycles. They improve soil health by enhancing structure, increasing organic matter, and supporting microbial activity. Typical species include cereal rye (Secale cereale), red clover (Trifolium pratense), and oilseed radish (Raphanus sativus var. Oleiformis), and other legumes or brassicas, which can be annual or perennial [10]. In addition to reducing erosion and nutrient runoff, cover crops help regulate soil moisture and temperature, filter contaminants, and promote biological diversity [28,29]. Their root systems provide a steady source of organic material to soil organisms, enhancing carbon inputs and contributing to long-term SOC accumulation [30].

No-till farming minimizes soil disturbance by avoiding deep ploughing. Seeds are directly inserted into minimally disturbed soil, while previous crop residues remain on the surface as mulch [31]. Long-term studies have shown that deep tillage disrupts SOC in surface layers by increasing soil aeration and microbial decomposition [32,33]. In contrast, reduced tillage has been associated with greater SOC stability and carbon accumulation due to decreased oxidation and organic matter breakdown [7,8,15,34]. Additional benefits include improved water retention, reduced erosion and runoff, enhanced biodiversity, and greater resilience to climate-induced drought [16,35,36].

Residue retention refers to the practice of leaving crop stubble on the field after harvest. This organic layer serves as both a carbon input and a protective mulch, affecting key soil conditions such as moisture, temperature, and microbial activity. Keeping residues slows organic matter decomposition, enhances humus formation, and supports microbial activities that contribute to SOC stabilization [7]. While residue retention was the least commonly observed practice in the study area, its potential for long-term carbon benefits is well supported in the literature.

Different modeling tools have been developed to simulate SOC dynamics under various management scenarios, ranging from highly detailed, process-based models like CENTURY, DNDC, DAISY, CANDY, DayCent, SOCRATES, APSIM, and SOMM [7,37,38,39,40,41,42,43,44,45,46] to those such as RothC [47,48]. Among these, CENTURY and DNDC are highly detailed models that simulate carbon and nitrogen cycles but require extensive input data and calibration, making them less suitable for regional-scale applications in data-scarce environments [49,50]. APSIM, while flexible and widely used in cropping systems, also demands detailed management and crop-specific parameters. In contrast, RothC offers a balance between model complexity and data availability. It requires fewer inputs, operates on a monthly time step, and is well suited for integration with RS data. These features make RothC particularly appropriate for regional-scale assessments, where high-resolution, long-term field data are limited.

Despite the growing number of studies on SOC sequestration, several gaps remain. Many existing assessments are limited to plot-scale experiments or rely on generalized assumptions that do not account for regional variability in soil, climate, and management. Moreover, there is a lack of harmonized approaches that integrate RS with process-based modeling, particularly in Central and Eastern Europe. Contradictory findings regarding the effectiveness of practices such as cover cropping and no-till farming further highlight the need for context-specific, spatially explicit assessments. This study addresses these limitations by combining satellite-derived indicators of land management with RothC modeling to estimate SOC changes for the Joniškis municipality in northern Lithuania. The workflow provides a replicable method for evaluating carbon farming potential in data-limited regions.

It is important to note that this study was conducted without field-based calibration of the RothC model due to the lack of site-specific SOC measurements in the study area. This limitation reflects data availability rather than a constraint of the model itself, and therefore frames the work as a methodological demonstration rather than a fully validated carbon-accounting exercise.

The aim of this study is to develop and test a replicable methodology for assessing SOC sequestration potential at the municipal scale by integrating remote sensing-derived indicators of carbon farming practices with the RothC model. Although RothC is widely used in different climatic zones, the combination of this model with satellite imagery from which carbon farming indicators are derived has not been tested in Lithuania. Moreover, studies that simultaneously consider the dual role of land use and climate in SOC modeling are still scarce, especially in Central and Eastern Europe. This approach directly supports new EU carbon farming initiatives by offering scalable monitoring and evaluation tools. By highlighting the method’s capabilities and limitations in a real-world setting, this study provides a framework that can be replicated in the future over a larger area in data-rich assessments. The research focuses on the Joniškis municipality in northern Lithuania, which is characterized by intensive agricultural activities and fertile soils, using data from 2019–2020. While recognizing the limitations related to the short timeframe, lack of ground-based validation, and reliance on secondary data sources, this study serves as a preliminary methodological assessment, offering insights into the potential and challenges of combining RS, GIS, and SOC modeling approaches in a regional context.

2. Materials and Methods

The study follows a multi-stage methodological approach combining geospatial data analysis, RS interpretation, and SOC modeling. The general workflow consists of six steps: defining the study area and data sources, preparing and harmonizing spatial datasets, detecting carbon farming practices from satellite imagery, integrating environmental and management data into the RothC model, simulating SOC change over a two-year period, and analyzing spatial patterns of SOC sequestration potential.

An overview of the methodological framework is presented in Figure 1.

2.1. Study Area

This study focuses on the Joniškis municipality, which is located in northern Lithuania, between 23°04′ and 23°57′ E longitude and 56°04′ and 56°22′ N latitude, and covers about 1152 km² of predominantly flat Semigallian lowland. Lithuania’s soils are predominantly formed from glacial and post-glacial deposits, with Cambisols, Luvisols, and Gleysols being the most common soil types. In the northern lowlands, fertile loams—chiefly Endogleyic Luvisols and Eutric Cambisols—dominate and are well suited to intensive cropping. Their moderate-to-high clay content and good water-holding capacity moderate organic matter decomposition and microbial activity, thereby influencing SOC dynamics.

The Lithuanian climate is classified as humid continental (Köppen Dfb), characterized by cold winters and warm summers. In this classification, “D” indicates a snowy climate with cold winters, “f” denotes no dry season (precipitation is relatively evenly distributed throughout the year), and “b” refers to warm summers where the warmest month remains below 22 °C. The average annual temperature ranges from 6 to 7 °C, and annual precipitation varies between 600 and 900 mm, with the majority falling during the growing season. Seasonal temperature fluctuations and precipitation patterns significantly influence soil moisture regimes and biological activity, which are key drivers of SOC turnover. In the Joniškis area, the relatively mild summer temperatures and adequate rainfall create favorable conditions for biomass production and carbon input into soils.

Figure 2 (left) situates the municipality within Lithuania; Figure 2 (right) shows that agriculture occupies ~70% of the land [51]. Cropping systems are dominated by cereals (wheat, barley, and rye) and oilseed rape. Conventional tillage is still prevalent, but uptake of reduced tillage and cover cropping is increasing under EU agri-environmental schemes. Although no certified carbon-credit projects operate here yet, these emerging practices signal growing interest in climate-smart agriculture, making Joniškis an appropriate case study for evaluating soil carbon sequestration potential.

2.2. Data Sources

A variety of spatial datasets were integrated to estimate SOC changes. These datasets, and their sources, original resolutions, and formats, are summarized in

Table 1. Data sources used in this study.

Data Group	Data Type	Source	Original Spatial Resolution	Units
Soil Data	Total initial 0–30 cm SOC stocks	ESDAC	1 × 1 km	t C ha−1
	Clay content	ESDAC	500 × 500 m	%
Climate Data	Monthly precipitation	CRU	0.5°	mm
	Average monthly air temperature	CRU	0.5°	°C
	Monthly potential evaporation	CRU	0.5°	mm
Satellite Data	Tillage raster	Sentinel-1/Author	10 × 10 m	Binary: 1—applied, 0—not applied
	Crop cover raster	Sentinel-2/Author	10 × 10 m	Binary: 1—applied, 0—not applied
	Residue raster	Sentinel-2/Author	10 × 10 m	Binary: 1—applied, 0—not applied
	Plant cover raster	MODIS	500 × 500 m	Binary: 1—plant cover, 0—bare
Other data	Depth	Constant: 30	-	cm
	Steps	Constant: 24	-	-

Abbreviation: ESDAC—European Soil Data Centre; CRU—Climatic Research Unit, Sentinel-2—High-Resolution Satellite Data, MODIS—Moderate Resolution Imaging Spectroradiometer.

.

Soil variables were obtained from the ESDAC database [53,54], whereas climate variables (precipitation, temperature, and potential evaporation) were taken from the CRU TS v4.08 dataset [51,55]. Monthly plant cover information was derived from the MOD13A1.061 NDVI product and processed in Google Earth Engine. Sentinel-1 synthetic-aperture-radar backscatter and Sentinel-2 multispectral imagery were used to map tillage, cover crop and residue management indicators following Metrikaitytė et al. [56,57] and Environmental Protection Agency guidelines [58].

2.3. Data Processing

Before running the RothC model, all datasets had to be harmonized to ensure compatibility in terms of format, spatial resolution, coordinate reference system, and coverage.

Soil data such as SOC stocks, clay content, and modern data were retrieved from the ESDAC database [53]. These rasters were originally in varying coordinate systems and spatial resolutions (1 km for SOC, 500 m for clay and modern). All layers were reprojected to the WGS84 (EPSG:4326) coordinate system, clipped to the boundaries of the Joniškis municipality, and resampled to a common 500 × 500 m grid using nearest-neighbor interpolation as appropriate.

Climate data such as monthly precipitation, air temperature, and potential evaporation from CRU [55] (0.5° resolution) were extracted for each month in 2019 and 2020 from perennial rasters. These coarse-resolution rasters were clipped to the study area and resampled to match a 500 m grid, acknowledging the limitation in downscaling precision. Each climate layer was stored as monthly raster stack.

Monthly plant cover data were obtained using the MODIS satellite imagery product MOD13A1.061 Terra Vegetation Indices 16-days Global 500 m and processed through Google Earth Engine. The NDVI threshold used to classify vegetated versus bare soil conditions was set at 0.6, based on RothC model assumptions and validated thresholds from previous studies in temperate climates. Pixels with NDVI > 0.6 were classified as vegetated (value = 1), while those with NDVI ≤ 0.6 were considered bare (value = 0). This binary classification was applied to MODIS NDVI composites.

Tillage, cover crops, and residue management practices were identified from Sentinel-1 and Sentinel-2 imagery. These were classified into binary rasters for each year, where 1 means carbon farming practices were applied and 0 means they were not applied, following the remote sensing-based approach described in Metrikiatytė et al. [56,57] and Environmental Protection Agency [58]. The rasters were resampled to the 500 m grid and aligned with the rest of the dataset.

A point grid (centroids of 500 × 500 m cells) was generated to spatially align all data layers. The final harmonized dataset consisted of ~90,000 overlapped point features, each representing a location and time step. All spatial layers were joined to this grid, and the result attributes were exported into .txt format required for RothC model input.

All processing steps were performed using QGIS v3.38 and Python v3.13.1 (via the QGIS Python console and pandas/geopandas libraries), enabling reproducible and automated handling of large datasets.

2.4. Identification of Carbon Farming Practices

The identification of carbon farming practices—cover cropping, tillage, and residue retention—was carried out using Sentinel satellite data, namely, Sentinel-1 (SAR) and Sentinel-2 (MSI) data, following the methodology of Metrikaitytė et al. [56,57] and the national Environmental Protection Agency [58]. This detection relied on time-series imagery and thematic interpretation rules that were tailored to the context of Lithuanian agriculture.

Sentinel-2 multispectral imagery was used to detect the presence of cover crops during the off-season period (post-harvest and pre-sowing). In Lithuania, cover crops must be sown by 15 September and maintained until 15 March of the following year. Normalized Difference Vegetation Index (NDVI)-based vegetation indices were analyzed for October and March, and fields exhibiting vegetative cover in these months were classified as having implemented cover cropping. The classification was carried out using semi-automated scripts developed in Google Earth Engine, supported by local agronomic knowledge.

Reduced or no-tillage practices were identified using Sentinel-1 SAR imagery. Fields were analyzed based on temporal backscatter patterns and surface roughness indicators. Areas with consistently low backscatter values during pre-sowing periods were interpreted as non-ploughed or minimally disturbed. The identification relied on the presence of uniform surface reflectance and temporal signal stability.

Residue management was also interpreted from Sentinel-2 imagery by analyzing spectral reflectance characteristics shortly after harvest. Fields that displayed mixed reflectance signals typical of post-harvest stubble were classified as implementing residue retention. This analysis considered spectral differences between bare soil and non-photosynthetically active organic matter left on the surface.

The outputs of this classification were three binary raster layers representing cover crops, no-till farming, and residue retention, which were subsequently used as management input variables in the RothC model, described in the following section.

2.5. Soil Organic Carbon Modeling with RothC

The RothC model was selected to simulate SOC changes across the Joniškis municipality. RothC is the process-based model developed for non-waterlogged top soils, operating on a monthly timestamp. It decomposes organic matter into five carbon pools: Decomposable Plant Material (DPM), Resistant Plant Material (RPM), Microbial Biomass (BIO), Humified Organic Matter (HUM), and Inert Organic Matter (IOM), with distinct decay rates for each pool [47].

In this study, RothC was adapted for spatial modeling using a harmonized dataset containing monthly climate variables, soil properties, and binary indicators of carbon farming practices (cover cropping, no tillage, and residue maintenance). The simulation was performed on a per-point basis using harmonized monthly inputs for 2019–2020.

Carbon Input Modifications

SOC dynamics in RothC are sensitive to the amount of carbon entering the soil via crop residues, cover crops, and organic inputs. To account for the impact of carbon farming practices, the default annual carbon input (C_inp) was adjusted using binary indicators (1 = practice applied; 0 = not applied) and empirical coefficients from the literature:

$$C_{inp} = C_{base} \times (1 + {\alpha }_1 \times CC + {\alpha }_2 \times R)$$

(1)

where:

• α₁ = 0.8—a coefficient for cover crops;
• α₂ = 0.6—a coefficient for residue retention;
• C_base—baseline carbon input for cropland;
• CC—cover crops indicator (1/0);
• R—residue retention indicator (1/0).

In this study, SOC inputs were differentiated according to the use of three carbon farming practices: cover crops, residue retention, and no-till farming. These were determined using satellite imagery (Section 2.4) and their impact on SOC was incorporated into RothC, replacing the annual carbon input values. The sink rates associated with these practices reported in the literature vary considerably depending on soil type, climate, management intensity, and time period. SOC gains due to carbon farming in arable and grassland systems have been reported to vary typically between 0.2 and 0.8 t C ha⁻¹ per year [38]. When considered in a separate study, cover crops have been found to sequester between 0.5 and 1.5 t C ha⁻¹ per year [13,22,23], retain residue of about 0.6 t C ha⁻¹ per year [13,24], and reduce tillage by about 0.5 t C ha⁻¹ per year [25,59]. These values were selected based on their consistency across multiple meta-analyses and long-term field trials in temperate European climates, which are broadly comparable to Lithuanian conditions. The coefficients (e.g., α₁ = 0.8 for cover crops, α₂ = 0.6 for residues) were derived as conservative mid-range estimates from these studies to reflect realistic, yet not overly optimistic, sequestration potentials. These values provided the scientific basis for estimating the relative variation in carbon inputs under different management combinations in this study. In order to maintain transparency and repeatability, no arbitrary empirical coefficients were used; instead, the observed presence or absence of each practice was used to structure RothC-specific management scenarios [13,22,23,24,25,26,38]. Tillage practices accelerate organic matter decomposition by disturbing soil structure and increasing microbial oxidation. This is represented in RothC via the DPM/RPM ratio. The RothC model was modified to reflect the impact of tillage on SOC decomposition by adjusting the DPM/RPM ratio. The default DPM/RPM ratio was increased by 0.5 in areas where tillage was detected (T = 1), following the equation:

$$({\displaystyle \frac{DPM}{RPM}})=({\displaystyle \frac{DPM}{RP{M}_{base}}})+\Delta tillage\times T$$

(2)

where:

• DPM/RPM_base—the default ratio;
• $\Delta tillage$ = 0.5—the increase in the DPM/RPM ratio due to tillage;
• T—tillage indicator (1/0).

This adjustment reflects the accelerated decomposition of organic matter due to increased aeration and microbial activity under conventional tillage. The baseline DPM/RPM ratio was set according to RothC defaults for arable soils.

SOC decomposition in RothC is controlled by three environmental rate-modifying factors (RMFs), calculated monthly for each point.

• Temperature (Temp) modifier:

$$f_{Temp}={\displaystyle \frac{47.91}{\exp (\frac{106.06}{temp+18.27})+1}}$$

(3)

2.

Moisture (Moist) modifier:

$$f_{Moist}=\min ({\displaystyle \frac{percip}{evap}},{\displaystyle \frac{percip+5}{percip+evap+10}}\cdot (1-0.15\cdot clay))$$

(4)

3.

Plant cover (PC) modifier:

$$f_{PC}=\left\{\begin{array}{c}0.6 if PC = 1\\ 1.0 if PC = 0\end{array}\right.$$

(5)

These modifiers were combined to adjust the decay rates of each SOC pool. The decomposition of each pool is modeled by exponential decay based on rate constants. The formula for decomposition for any pool (e.g., DPM) is:

$$c_t=c_0\cdot e^{-k\cdot f}$$

(6)

where:

• $c_t$—remaining carbon at time;
• C₀—initial pool carbon;
• k—pool specific decay constant (i.e., for DPM k = 10, for HUM k = 0.02, for RPM k = 0.3, for BIO k = 0.66)
• f—combined RMF.

In this study, C₀ is derived from RothC equilibrium runs using C_base inputs (i.e., without additional management practices). Therefore, C₀ reflects the SOC stock under conventional agricultural conditions and is used as the starting point for simulating SOC changes under carbon farming scenarios.

Total SOC at each timestep is the sum of all active pools:

$$SOC = DPM + RPM + BIO + HUM + IOM$$

(7)

Annual SOC change was calculated as:

$$\Delta SOC=SO{C}_{t+1}-SO{C}_t$$

(8)

The resulting SOC change values were exported as geospatial rasters for further analysis and interpretation.

Although the RothC model is often calibrated using site-specific SOC observations, such calibration was not feasible in this study due to the absence of field-measured SOC values for the selected area. Nevertheless, the goal of this work was not to validate RothC outputs at the field scale, but rather to demonstrate a scalable and adaptable framework for integrating remote sensing-derived land management data into a spatial modeling approach. The absence of calibration in this study reflects the lack of field-based SOC measurements in the region, rather than a limitation of the RothC model itself. This approach provides a methodological foundation for future, data-rich studies that could incorporate calibration and validation steps.

Table 2. Example of input data required for modeling in RothC model.

Latitude	Longitude	Year	Month	Temp	Precip	Evap	Cbase	PC	DPM/RPMbase	Clay	Depth
56.260449	23.078712	2019	1	-3.7	55.8	0.3	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2019	2	0.6	32.6	0.4	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2019	3	2.6	55.0	0.9	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2019	4	8.2	1.9	2.8	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2019	5	12.3	56.1	2.9	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	6	18.7	32.9	4.7	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	7	16.9	95.5	3.6	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	8	17.9	63.8	3.0	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	9	13.1	69.5	1.8	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	10	8.7	68.7	0.7	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	11	4.3	53.0	0.3	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2019	12	2.2	51.3	0.3	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2020	1	2.3	50.6	0.3	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2020	2	2.0	56.7	0.5	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2020	3	3.3	42.2	1.2	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2020	4	6.6	14.4	2.3	0.5	0.0	1.44	10.08	30
56.260449	23.078712	2020	5	10.1	49.5	3.0	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2020	6	18.1	96.4	4.1	0.5	1.0	1.44	10.08	30
56.260449	23.078712	2020	7	17.1	89.7	3.6	0.5	1.0	1.44	10.08	30

Abbreviation: Temp—air temperature, Precip—precipitation, Evap—evaporation, PC—plant cover.

provides an example from the input dataset used for RothC modeling, showing monthly values for selected parameters.

3. Results

3.1. Results of Identified Carbon Farming Practices

Carbon farming practices were successfully identified using Sentinel-1 and Sentinel-2 satellite data. Figure 3 presents the spatial distribution of detected practices for 2019 and 2020. The presence of cover crops was determined by analyzing satellite images from October to March, which corresponds to the period laid down by Lithuanian legislation (15 September to 15 March). The presence of no-till farming was determined by using the backscatter of radar satellite images during the pre-sowing period from April to May. Residue retention classification was carried out immediately after harvest, usually in August–September, based on Sentinel-2 multispectral satellite imagery.

Residue retention was the least commonly observed practice in both years, which is probably due to the region’s intensive farming system and the farmers’ own habits, which are difficult to change quickly. Local farmers often prefer to sow winter crops shortly after harvest rather than leave residues in the field over the winter. Cover crops appeared more widespread in 2020 compared to 2019, especially in the eastern and southeastern parts of the municipality. The extent of no-till farming increased by approximately 4% from 2019 to 2020, indicating a gradual shift toward reduced tillage systems. The overlapping of practices was more common in fields with cover crops and no-till farming, which suggests higher cumulative sequestration potential in these areas. These observations were based on binary raster classification derived from spectral and radar features, as described in the methodology.

3.2. Modeled Soil Organic Carbon Change

The RothC model simulations estimated the annual change in SOC between 2019 and 2020 under the identified land management conditions. The results are presented in Figure 4.

SOC sequestration rates varied spatially, ranging from approximately 0.23 to 0.32 t C ha⁻¹ year⁻¹ across the municipality. The lowest SOC changes were found in areas with minimal or no carbon farming practices, while the highest increases were concentrated in parcels where multiple practices overlapped. The higher SOC changes observed in these zones are mainly attributed to the combined application of carbon farming practices rather than inherent soil differences, as this region is largely homogeneous in terms of soil types and predominantly dedicated to cereal crop production.

These figures reflect model outputs generated with literature-derived coefficients and without site-specific calibration. Therefore, the reported SOC changes should be interpreted as indicative rather than precise. Key uncertainties include the spatial resolution of satellite and climate datasets, the binary simplification of management practices, and the absence of ground-based SOC measurements. While not representing audited carbon balances, these results provide a useful spatial baseline that can be refined as higher-resolution or field data become available.

The observed spatial variability in SOC gains highlights the potential of combining multiple practices. Areas where cover crops, residue retention, and reduced tillage were implemented simultaneously showed the highest SOC increases. These patterns agree with long-term experiments, which report a synergistic effect of integrated management strategies on SOC [7,25].

The comparative analysis of SOC changes across different management categories (cover crops, no-till, and residue retention) revealed that each practice contributed positively to SOC accumulation, with the highest gains observed in areas where multiple practices overlapped. This reinforces evidence from temperate long-term trials that integrated strategies outperform single measures.

To estimate the magnitude of the simulated SOC gain (0.23–0.32 t C ha⁻¹ year⁻¹), we visually compared our resulting raster with the global ISRIC “Soils Revealed” SOC change layer for 2018–2023 [60]. Dividing that five-year cumulative map by five yielded an approximate annual rate. Hotspots (>0.25 t C ha⁻¹ yr⁻¹) in the ISRIC layer coincided with our highest-gain parcels (Figure 4; see also Figure 5), suggesting that RothC did not systematically overestimate sequestration despite coarser inputs and differing time windows.

It should be noted that the effects of individual carbon farming practices were not modeled separately due to their frequent spatial overlap and the binary nature of the input data. As a result, the estimated SOC changes reflect the cumulative impact of combined practices rather than isolated effects. Future studies could explore factorial modeling or field-based validation to better distinguish the contribution of each practice.

4. Discussion

The modeling results obtained in this study suggest that the application of carbon farming practices can lead to measurable increases in SOC, with estimated annual sequestration rates ranging from 0.23 to 0.32 t C ha⁻¹ across the Joniškis municipality. These values align with reported literature ranges for similar agricultural practices in European contexts, although variation is expected due to differences in soil types, climate conditions, and implementation intensity.

Beyond productivity, the observed SOC increases also carry important ecological implications. Higher SOC levels are associated with improved soil structure, enhanced microbial activity, and more efficient nutrient cycling, all of which contribute to long-term soil fertility. From an ecosystem services perspective, SOC accumulation supports provisioning services (e.g., food production), regulating services (e.g., water retention, erosion control, and climate regulation), and supporting services (e.g., biodiversity habitat and nutrient cycling). In agricultural regions like Joniškis, even modest SOC gains can improve agroecosystem resilience to climate extremes, such as droughts or heavy rainfall, thereby strengthening the ecological stability of the landscape. Carbon farming not only mitigates climate change, but also enhances food security and long-term productivity.

Among the practices assessed, cover cropping and no-till farming showed the strongest association with higher SOC gains, both in terms of spatial extent and modeled sequestration rates. Cover crops contribute to SOC by increasing biomass input during the off-season, while no-till farming reduces decomposition by minimizing soil disturbance. Residue retention, although less widespread, also plays a key role by maintaining surface organic matter and supporting microbial activity. The highest SOC increases were observed in areas where multiple practices overlapped, suggesting that integrated management strategies are more progressive and effective than single practices alone. The synergistic effect of combining cover crops, residue retention, and reduced tillage likely results from complementary mechanisms: cover crops increase biomass input and root exudates, residue retention maintains surface organic matter and moderates soil microclimate, while reduced tillage minimizes disturbance and preserves soil structure. Together, these practices enhance microbial activity, reduce decomposition rates, and promote stable carbon pool formation [13,16,24].

While this study focused on croplands, it is important to acknowledge that other land uses such as grasslands and forests also sequester carbon and contribute to the overall carbon balance. These land uses were not assessed due to the lack of harmonized RS indicators and the study’s focus on agricultural management. Future research could expand the scope to include comparative assessments across land-use types to better understand their relative contributions to SOC dynamics and climate mitigation.

The credibility of the simulated 0.23–0.32 t C ha⁻¹ yr⁻¹ gain is constrained by several data issues. Model inputs were resampled from coarse-resolution grids (e.g., 0.5° CRU climate), so local temperature- and moisture-driven decomposition rates may be under- or overestimated. Management layers were encoded as binary masks that ignore variability in timing, intensity and farmer compliance, while remote sensing misclassification (clouds, spectral confusion) can further distort practice extent. Finally, the two-year window and lack of site-specific SOC calibration mean that absolute values are indicative rather than audited. Finer products such as ERA5-Land (≈9 km, hourly) and longer monitoring series would reduce these uncertainties; ground plots are essential for calibrating rate-modifying factors and validating outputs [60].

Despite the above limitations, our spatial pattern and magnitude agree with independent evidence. Figure 5 shows that ISRIC’s “Soils Revealed” 2018–2023 layer—annualized for comparison—locates hotspots (>0.25 t C ha⁻¹ yr⁻¹) in the same central-western zones as our RothC map. Published field and modeling studies from temperate Europe report similar means (≈0.25–0.35 t C ha⁻¹ yr⁻¹), and long-term ryegrass and cover crop trials in Sweden and Germany fall squarely within our range. This convergence suggests that, while coarse, the workflow does not systematically overestimate sequestration. Reported sequestration rates from temperate croplands converge on values similar to those obtained here. Process-based simulations for 1267 German sites predicted 0.28–0.33 t C ha⁻¹ year⁻¹ under extensive winter cover crops [61]. A European meta-analysis of 71 field trials found a mean mineral-associated SOC increase of 23% (95% CI 14–34%), equivalent to ≈0.25–0.35 t C ha⁻¹ year⁻¹ for typical arable soils [62]. Long-term undersown ryegrass experiments in southern Sweden reported 0.32 ± 0.28 t C ha⁻¹ year⁻¹ over 16–24 years [63], while a global meta-analysis covering 61 maize-based studies showed a median SOC gain of 7.3%, translating to ≈ 0.30 t C ha⁻¹ year⁻¹ when expressed absolutely [64]. Together with the consistency check against the Soils Revealed map, these independent lines of evidence reinforce the credibility of the simulated sequestration rates and their spatial distribution, albeit within the uncertainty introduced by divergent climates, management histories, and model structures.

Empirical information on SOC dynamics in Lithuania is still limited: recent work has produced high-resolution but time-invariant SOC maps of the moraine plains [65] and one-off estimates of national stocks prepared for GHG accounting reports [66], and a comprehensive evaluation of agricultural sequestration potential has synthesized long-term field experiments without spatial modeling [67]. In the absence of a long-term monitoring network, none of these sources can reveal how management changes SOC trajectories. By applying the RothC model at the municipality scale, our study provides potentially the first dynamically modeled estimates of SOC changes on Lithuanian croplands, filling an important evidence gap for both science and policy.

This study demonstrates a practical workflow for integrating RS-derived land management indicators into a well-established SOC model, RothC. By harmonizing data from Sentinel satellites, and MODIS, ESDAC, and CRU sources, the model was adapted for spatial application at a regional scale. The methodological approach is transferable and can be extended to other regions using publicly available datasets. This positions the model as a valuable foundation for building a national soil carbon monitoring system in Lithuania.

Second, model-inherent limitations stem from the simplified structure of the RothC model itself. For example, RothC does not simulate nitrogen dynamics, crop growth, or microbial interactions, and assumes uniform soil depth and decomposition rates. These assumptions may lead to under- or overestimation of SOC changes in areas with heterogeneous soil profiles or variable decomposition environments, and should be considered when interpreting the spatial patterns of modeled sequestration. The structural simplifications limit the model’s ability to capture complex feedbacks in agroecosystems. While RothC remains a practical tool for regional-scale assessments, its outputs should be interpreted with these assumptions in mind. However, its simplicity and compatibility with RS inputs make it a valuable tool for first-order assessments and spatial prioritization in carbon farming strategies.

The modeling approach did not include calibration due to the lack of field-measured SOC data for the study area. As a result, the empirical coefficients used to adjust carbon inputs and decomposition rates were drawn from the literature and represent approximate average values rather than locally validated figures. In addition, management practices were simplified as binary inputs (applied or not applied), which may not capture their true variation in intensity or duration. To overcome these limitations, future research should prioritize the collection of field-based SOC measurements across representative sites in the study area. This would enable calibration and validation of RothC outputs, improving model accuracy and reliability. Additionally, integrating more detailed management data—such as the duration, intensity, and crop-specific effects of carbon farming practices—would allow for more nuanced modeling scenarios. Advances in remote sensing, including higher-resolution imagery and machine learning–based classification, could also reduce misclassification errors and better capture spatial variability. Finally, collaboration with local farmers and extension services could support participatory data collection and ensure that modeling assumptions reflect real-world practices.

The findings of this study highlight the feasibility of integrating Earth observation and process-based modeling to estimate management-sensitive SOC change at the regional scale. While the modeled sequestration values (0.23–0.32 t C ha⁻¹ yr⁻¹) are preliminary due to lack of calibration and practice simplifications, they reflect spatial patterns consistent with both global datasets and reported field ranges in Europe. This strengthens confidence in the underlying approach, especially for regions where field-based SOC monitoring is unavailable.

5. Conclusions and Recommendations

This study demonstrates a practical workflow for integrating RS-derived land management indicators into the RothC model and adapting it for spatial application at the regional scale. By harmonizing data from Sentinel satellites, and MODIS, ESDAC, and CRU sources, the approach is transparent, replicable, and readily extendable to other regions using publicly available datasets. This positions the model as a valuable foundation for building a national soil carbon monitoring system in Lithuania.

Gridded SOC-change maps can support national GHG-mitigation strategies by identifying priority areas with high sequestration potential, enabling more targeted and efficient allocation of resources. They also inform land managers about where to promote specific carbon farming practices, thereby enhancing agri-environment measures and accelerating the transition to climate-smart agriculture. At a broader scale, the workflow lays the technical groundwork for integrating SOC monitoring into emerging carbon-credit verification systems and aligns with the EU’s carbon farming initiatives by offering a transparent, scalable method based on open data.

While the present analysis focused on croplands, grasslands and forests also sequester carbon and contribute to the regional carbon balance. Future assessments should therefore incorporate these land-use categories to clarify their relative roles in SOC dynamics and climate-change mitigation.

To reduce uncertainty and move from indicative mapping to verifiable SOC accounting, we recommend the following:

• Establishing long-term SOC monitoring plots for calibration and validation;
• Improving climate and soil-input resolution (e.g., ERA5-Land) and extending simulations to multi-year periods;
• Encoding management intensity on a continuous rather than binary scale;
• Integrating farmer-reported data and participatory mapping for ground-truthing;
• Applying machine-learning classifiers to refine remote sensing detection of tillage and residue cover.

Implementing these steps will tighten uncertainty bounds, support field-level decision-making, and enable credible inclusion of soil carbon gains in carbon-credit and agri-environment schemes.