Simulation of the Growth and Yield of Maize (Zea mays L.) on a Loosened Plinthosol Amended with Termite Mound Material in the Lubumbashi Region

Abstract

The low fertility of plinthosols is a major constraint on agricultural production, largely due to the presence of plinthite, which restricts the availability of water and nutrients. This study aimed to simulate the growth and yield of grain maize on a loosened plinthosol amended with termite mound (from Macrotermes falciger) material in the Lubumbashi region. A 660-hectare perimeter was established, subdivided into ten maize blocks (B1–B10) and a control block (B0), which received the same management practices as the other blocks except for subsoiling and termite mound amendment. The APSIM model was used for simulations. The leaf area index (LAI) was estimated from Sentinel-2 imagery via Google Earth Engine, using the Simple Ratio (SR) spectral index, and integrated into APSIM alongside agro-environmental variables. Model performance was assessed using cross-validation (2/3 calibration, 1/3 validation) based on the coefficient of determination (R²), Nash–Sutcliffe efficiency (NSE), and root mean square error (RMSE). Results revealed a temporal LAI dynamic consistent with maize phenology. Simulated LAI matched observations closely (R² = 0.85 − 0.93; NSE = 0.50 − 0.77; RMSE = 0.29 − 0.40 m² m⁻²). Maize grain yield was also well predicted (R² = 0.91; NSE > 0.80; RMSE < 0.50 t ha⁻¹). Simulated yields reproduced the observed contrast between treated and control blocks: 10.4 t ha⁻¹ (B4, 2023–2024) versus 4.1 t ha⁻¹ (B0). These findings highlight the usefulness of combining remote sensing and biophysical modeling to optimize soil management and improve crop productivity under limiting conditions.

1. Introduction

In the Democratic Republic of the Congo (DRC), agriculture remains a cornerstone of the national economy, employing over 60% of the population and contributing around 17.4% of GDP [1]. Yet, despite its central role, agricultural production continues to face major challenges in ensuring food security and generating sufficient income for rural households [2,3,4,5,6]. The main staple crops include cassava, plantains, maize, groundnuts, and rice. Most farmers practice subsistence agriculture, while commercial production remains marginal [7,8,9,10,11].

In the south-eastern DRC (Haut-Katanga province), food deficits persist, and the population relies heavily on maize imports from neighboring southern African countries to meet local demand [12,13,14]. Maize is particularly critical, as it constitutes the main dietary energy source and plays a key role in local food systems. Therefore, improving local maize productivity is essential for achieving food self-sufficiency. However, one of the main constraints limiting yield potential is soil degradation, largely driven by the dominance of highly weathered and nutrient-depleted soils [15,16,17,18]. These soils are typically characterized by low chemical fertility, the frequent presence of plinthite at or near the surface [18], and the widespread occurrence of large termite mounds [19,20]. The shallow nature of plinthic soils restricts root development and nutrient availability, thus reducing their agricultural potential [21,22,23,24,25].

To address these constraints, an innovative soil management strategy has been implemented in the Lubumbashi region. It combines subsoiling—to disrupt the hardened plinthite layer—with the use of termite mound materials (from Macrotermes falciger) as a natural amendment. These materials, rich in fine particles and exchangeable bases [26,27,28,29,30,31,32,33,34,35], are excavated and spread across subsoiled plots (Figure 1), improving soil structure, enhancing fertility, and facilitating mechanized maize cultivation. Unlike conventional soil rehabilitation techniques that rely mainly on organic or mineral fertilizers, this approach directly modifies soil physical constraints while simultaneously improving nutrient status, making it particularly suited to plinthic landscapes.

Despite its increasing adoption by local farmers, few scientific studies have evaluated the agronomic and modeling implications of this practice. Most previous works on plinthosols have focused on soil characterization or amendment effects under controlled conditions but lack process-based modeling to assess long-term crop responses under real field variability. This knowledge gap justifies the need for an integrated assessment combining field data with crop simulation.

Crop models offer a powerful means to analyze and predict the effects of management practices on soil–plant–climate interactions, supporting more sustainable and resilient agricultural systems. Among them, APSIM (Agricultural Production Systems sIMulator) is widely recognized for its ability to integrate dynamic processes governing crop growth, soil water and nutrient fluxes, and management practices [36,37,38,39,40,41,42,43,44]. The model has been successfully applied to assess the effects of organic and mineral amendments [45,46,47,48,49,50] and soil tillage techniques, including subsoiling [51,52].

This study therefore aims to evaluate the performance of the APSIM model in simulating maize growth and yield on plinthosols improved through subsoiling and amendment with termite mound materials in the Lubumbashi region. It specifically seeks to answer the following question: How does the combined use of subsoiling and termite mound amendments influence maize grain yield performance, and how accurately can APSIM reproduce these effects under local conditions?

The study is based on the hypothesis that the combination of subsoiling and termite mound amendment enhances maize productivity by improving soil physical conditions and nutrient availability, and that the APSIM model can accurately simulate these effects when properly calibrated for plinthosols.

2. Materials and Methods

2.1. Study Sites

This study was conducted in Lubumbashi, Haut-Katanga province, Democratic Republic of the Congo, specifically at the FarmCo farm. The study site is located in the suburbs of Lubumbashi, approximately 60 km east of the city, along the Kasenga road in the Kifumanshi river valley (Figure 2). The study area covers approximately 660 hectares, subdivided into ten blocks. The pedological cover of the region is dominated by plinthitic Ferralsols, according to the WRB classification [18].

A detailed soil survey conducted before deep tillage revealed surface-exposed plinthite in several blocks (notably B3, B7, B8, B9, and B10), sometimes occurring at depths shallower than 10 cm. In the remaining blocks, plinthite was present in all profiles but generally appeared at depths greater than 10 cm.

The soils of the site are characterized by a shallow effective depth, high compaction of subsoil horizons, and marked textural variability, making their agricultural use difficult without mechanical intervention or suitable amendments.

To overcome these constraints, systematic excavation of lateritic layers was carried out across the entire area to fragment compact horizons. This operation was followed by deep mechanical subsoiling, aimed at improving soil structure, water percolation, and root development. In parallel, materials from inactive giant termite mounds, which are abundant in the area, were excavated, leveled, and spread over the subsoiled plots to expand the cultivated surface, facilitate the use of agricultural machinery, and serve as organo-mineral amendments.

In this region, the average density of inactive termite mounds (from Macrotermes falciger) is approximately seven units per hectare, with an individual volume of about 256 m³ [19]. The total mobilizable volume is therefore estimated at 1800 m³ ha⁻¹.

Assuming uniform spreading, this corresponds to an average thickness of 18 ± 2.3 cm ha⁻¹, equivalent to roughly 2000 t ha⁻¹, considering a bulk density of 1.2 g cm⁻³ [53]. Although all blocks received termite mound material, a larger proportion was applied to those where plinthite had been more extensively excavated, to ensure uniform leveling of the field and to compensate for the volume loss resulting from subsoiling.

This site is one of the largest farms in the Lubumbashi region, known for its large production of maize grain for human consumption in the form of flour.

The outskirts of Lubumbashi belong to the Cw6 climate (humid subtropical climate with hot summers and dry winters) according to the Köppen classification. The area is characterized by a dry season (May to September), a rainy season (November to March), and two transition months (April and October) [54]. Average annual rainfall is 1270 mm. The average annual temperature is 20.1 °C, with daily minimum and maximum temperatures reaching 8 °C during the coldest month and 32 °C during the hottest month, respectively [55].

2.2. APSIM Model Description

APSMI NextGen version 2024.10.7607.0 was used for scenario analysis in this study. This model was developed to simulate biophysical processes in agricultural systems in response to environmental variations [37]. A wide range of models is available in APSIM for major crop, pasture, and tree species, as well as key agricultural system processes at a daily time step [56]. The APSIM-Maize crop module was used to simulate the growth and yield of grain maize on plinthite soil over three growing seasons. The processes that affect growth and yield in APSIM-Maize are simulated by interactions between daily weather data, cultivars, soil properties, and different management practices [36,37].

The parameters calibrated to simulate maize growth and yield at the study site are detailed below and were integrated using a step-by-step approach. Phenological phase progression is expressed in thermal time, thus expressing values in degree days [57].

2.3. Parametrization and Calibration

2.3.1. Climate Data

Agrometeorological data were obtained from the NASA Prediction of Worldwide Energy Resource (POWER CERES/MERRA-2) satellite database (https://power.larc.nasa.gov/data-access-viewer/, accessed on 10 February 2025) for the period from 1 January 2016 to 31 December 2024, covering the entire study duration. This source was used because no operational local weather stations were available in the study area. The extracted variables included temperature, precipitation, relative humidity, wind speed, and solar radiation (Figure 3). Although NASA POWER data have demonstrated high reliability in several agroclimatic contexts [58,59,60,61], their use may introduce uncertainty associated with the relatively coarse spatial resolution (0.5° × 0.5°, approximately 55 km) and the gridded interpolation process, which does not always capture fine-scale topographic or microclimatic variations [62]. According to Glotter et al. [63] and Mourtzinis et al. [59], this uncertainty typically ranges between 5% and 15% depending on the variable considered but remains acceptable for agroclimatic modeling applications, particularly in tropical regions where in situ observations are scarce. In this context, NASA POWER products provide a reliable and widely validated alternative for crop modeling and reference evapotranspiration estimation [60].

2.3.2. Reference Evapotranspiration

The FAO Penman–Monteith approach [64] was used to calculate evapotranspiration from maize plants during the study period. This method, which requires the use of meteorological variables, is widely recognized for its ability to provide accurate and unambiguous estimates of ET0 in various environments around the world [65,66,67,68]. Figure 4 shows the results of reference evapotranspiration variation from 2016 to 2024 at the study site.

2.3.3. Maize Yield

Maize grain yield was measured in subplots measuring one square meter each, clearly marked out in the different blocks (Figure 5). At least five measurements per block were taken during the first (2022–2023) and second (2023–2024) growing seasons, and the average yield per block was then calculated for each season. The yield for the 2016–2017 growing season was considered the control (B0), corresponding to the period before the implementation of subsoiling and termite mound material spreading. As a result, only blocks B1 to B10 benefited from these interventions.

In terms of crop management, all blocks, including the control block B0, benefited from the same technical itineraries, including mechanical tillage, harrowing, sowing, mineral fertilization, and the application of a pre-emergence herbicide. Maize was sown at the time of the application of NPK (10-20-10) base fertilizer at a rate of 200 kg ha⁻¹, followed by a urea supplement applied 45 days after sowing at a rate of 200 kg ha⁻¹. The SC719 maize variety was grown at a spacing of 0.75 m between rows and 0.25 m within rows, giving a density of 53,333 plants per hectare. Harvesting took place when the plants reached harvest maturity (

Table 1. Sowing and harvest dates in the blocks. B1–B10: blocks that benefited from subsoiling and spreading of termite mound material. B0: control without subsoiling and amendment with termite mound material.

Maize Blocks	Season 2016–2017		Season 2022–2023		Season 2023–2024
	Sowing	Harvest	Sowing	Harvest	Sowing	Harvest
B1			22 November 2022	23 June 2023	23 November 2023	23 June 2024
B2			24 November 2022	21 June 2023	23 November 2023	22 June 2024
B3			30 November 2022	25 June 2023	30 November 2023	25 June 2024
B4			2 December 2022	27 June 2023	30 November 2023	27 June 2024
B5			5 December 2022	24 June 2023	2 December 2023	24 June 2024
B6			7 December 2022	27 June 2023	30 December 2023	30 June 2024
B7			3 December 2022	24 June 2023	30 November 2023	26 May 2024
B8			4 December 2022	26 June 2023	2 December 2023	26 June 2024
B9			25 November 2022	22 June 2023	22 November 2023	24 June 2024
B10			27 November 2022	24 June 2023	22 November 2023	24 June 2024
B0	9 December 2016	24 June 2017

), i.e., when the seeds had a moisture content of 13%.

2.3.4. Soil Properties

A detailed description of a soil profile of ±1 m in each block was carried out in each block following the FAO soil description guidelines [69]. Composite soil samples were collected in 2022 from the diagnostic horizons, air-dried for five days, sieved to 2 mm, for chemical analyses. Analyses were performed at the Soil Chemistry Laboratory, Gembloux Agro-Bio Tech (Gembloux, Belgium). Soil pH was determined pH in water; organic C and total N were determined using the Walkley–Black and Kjeldahl methods, respectively. Particle size distribution was obtained using the pipette method following the XFX 31-107 protocol. Undisturbed samples were also collected using Kopecky cylinders to measure bulk density and assess soil water dynamics at the Soil Physics and Mechanics Platform (PhyMeSol), Gembloux Agro-Bio Tech, University of Liège, Gembloux, Belgium. Saturated hydraulic conductivity (Ksat) was measured with the constant-head permeameter, and water retention was determined using Richard’s apparatus.

For the control block (B0), soil analyses previously conducted in 2016 in Belgium were used. The same analytical protocols were applied, except for hydraulic parameters, which were estimated using the pedotransfer functions of Saxton & Rawls [70]. The results relating to water content, physicochemical properties (

Table 2. Soil properties of the site used in the simulations: saturation water content (SAT), drained upper limit (DUL), lower limit of available soil water (LL), plant available water capacity (PAWC), bulk density (BD), saturated hydraulic conductivity (Ksat), and total organic carbon (TOC). Blocks 1 to 10: blocks that benefited from subsoiling and spreading of termite mound materials. Block 0: control without subsoiling or amendment with termite mound materials.

Soil Depth	SAT	DUL	LL	PAWC	BD	Ksat	pH	TOC
(cm)	(m3 m−3)				(g cm−3)	(mm day−1)	Water	(%)
Block 1
0–26	0.49	0.26	0.16	0.10	1.4	7780	8.2	0.9
26–50	0.45	0.18	0.08	0.11	1.5	86	8.0	0.4
Block 2
0–27	0.42	0.22	0.16	0.06	1.5	51.7	6.9	0.8
27–79	0.43	0.24	0.20	0.04	1.5	664	5.6	0.2
Block 3
0–30	0.41	0.19	0.11	0.08	1.6	86	7.9	1.1
30–43	0.30	0.19	0.08	0.07	1.9	6130	6.0	0.6
43–80	0.28	0.14	0.08	0.06	1.9	7780	5.5	0.4
Block 4
0–20	0.45	0.23	0.18	0.05	1.5	125	8.4	1.2
20–35	0.55	0.26	0.19	0.07	1.2	126	8.0	1.8
Block 5
0–46	0.55	0.30	0.24	0.06	1.2	625	6.1	2.0
46–92	0.36	0.24	0.20	0.04	1.7	276	7.2	0.4
92–150	0.33	0.27	0.23	0.04	1.8	63.9	7.8	0.2
Block 6
0–25	0.48	0.21	0.14	0.08	1.4	333	5.9	1.2
25–132	0.38	0.19	0.15	0.047	1.7	32.9	6.1	0.2
Block 7
0–20	0.46	0.26	0.176	0.10	1.4	1210	8.0	1.2
20–30	0.37	0.18	0.09	0.10	1.7	4320	6.5	0.2
30–75	0.28	0.15	0.09	0.060	1.9	864	5.6	0.2
Block 8
0–35	0.45	0.16	0.10	0.07	1.5	44.7	6.0	0.7
35–110	0.34	0.12	0.06	0.05	1.7	400	5.8	0.2
Block 9
0–30	0.56	0.26	0.09	0.17	1.8	1810	7.1	1.4
30–70	0.45	0.19	0.09	0.10	1.5	892	5.3	0.7
Block 10
0–27	0.45	0.18	0.13	0.06	1.5	1410	7.5	2.3
27–130	0.39	0.21	0.16	0.05	1.6	767	7.0	0.3
Block 0
0–27	0.30	0.16	0.09	0.07	1.9	398	5.3	0.7
27–44	0.28	0.11	0.07	0.04	1.9	91.5	5.0	0.3

), and particle size distribution (

Table 3. Soil particle size distribution of blocks used for simulations. B1–B10: blocks that benefited from subsoiling and spreading of termite mound material. B0: control without subsoiling or amendment with termite mound material.

Blocks	Horizons	Depth	Clay	Silt	Sand
		(cm)		(%)
B1	Ap	0–26	19.4	33.2	47.4
	AB	26–50	18.0	50.7	31.3
B2	Ap	0–27	21.1	40.4	38.5
	AB	27–79	28.4	37.7	33.9
B3	Ap	0–30	22.2	47.6	30.2
	AB	30–43	20.0	45.2	34.8
	Bs	43–80	30.6	44.5	24.9
B4	Ap	0–20	24.1	57.9	18.0
	AB	20–35	16.1	52.3	31.6
B5	Ap	0–46	18.6	56.0	25.4
	AB	46–92	37.6	30.2	32.2
	Bs	92–150	44.0	27.2	28.8
B6	Ap	0–25	19.7	32.9	47.4
	AB	25–132	40.2	28.1	31.7
	Bs	132–201	34.6	38.0	27.4
B7	Ap	0–20	20.6	50.5	28.9
	AB	20–30	18.0	51.7	30.3
	Bs	30–75	24.3	51.3	24.4
B8	Ap	0–35	18.8	41.2	40.0
	Bcs	35–110	34.1	45.0	20.9
B9	Ap	0–30	22.2	53.2	24.6
	Bcs1	30–70	27.8	49.6	22.6
B10	Ap	0–27	12.0	43.5	44.5
	Bs	27–130	21.8	46.4	31.8

) were incorporated into the APSIM model for the simulations.

2.3.5. Leaf Area Index

The leaf area index (LAI) was determined for maize crops using Sentinel-2 satellite images, available via Google Earth Engine (GEE, https://code.earthengine.google.com, accessed on 22 February 2025). LAI is a key parameter for quantifying leaf area per unit of soil surface area and is crucial for assessing crop health and productivity. In the APSIM model, LAI directly controls the solar radiation intercepted by the canopy and, consequently, the ability to capture light for photosynthesis. Remote sensing is an effective tool for studying the seasonal dynamics of LAI, particularly in the absence of field measurements [71,72]. Sentinel-2 imagery is particularly suitable for this purpose due to its high spatial (10 m) and temporal (5-day) resolution, enabling precise monitoring throughout the growing season [73]. The study area was defined by a shapefile delineating the boundaries of the experimental fields to ensure accurate geospatial analysis. Sentinel-2 images were atmospherically corrected using the Surface Reflectance (Level-2A) product available in GEE, which applies the Sen2Cor algorithm to remove atmospheric and scattering effects. Only cloud-free scenes (cloud cover < 20%) were retained for analysis (

Table 4. Number of Sentinel images used per month and per year in the LAI estimate.

Year	Jan.	Feb.	Mar.	Apr.	May	Jun.	Jul.	Aug.	Sep.	Oct.	Nov.	Dec.
2016												2
2017	2	2	2	2	3	4
2022											3	3
2023	2	2		2	7	7					3	3
2024		4	3	3	7	5					4	2
					No image		Image

Blue cells (with white font) indicate months in which usable Sentinel-2 images (cloud cover < 20%) were available. White cells indicate months without usable images, or months outside the maize growing period (sowing to harvest).

).

The Simple Ratio (SR) index, computed from the near-infrared (B8) and red (B4) bands of Sentinel-2, was selected to derive the LAI (SR; Equation (1)). The choice of SR was guided by its strong linear relationship with canopy density, its low sensitivity to soil background variations, and its robust performance under moderate vegetation cover—conditions typical of maize fields in southern DRC [73,74,75]. Previous studies have shown that SR maintains a higher dynamic range than NDVI in dense canopies and exhibits strong correlations with measured LAI (R² > 0.70, RMSE < 0.5) [76,77]. Although more complex indices such as EVI2 or MSAVI can reduce saturation effects in very dense vegetation, their performance advantage remains limited under moderate canopy conditions and in environments with variable atmospheric effects. Consequently, SR was considered the most suitable and operationally efficient index for this study, providing a good balance between simplicity, sensitivity, and proven accuracy.

To improve classification precision and minimize background noise, a supervised classification was applied using the Support Vector Machine (SVM) algorithm. The SVM classifier was trained with 120 reference samples (60 vegetation and 60 non-vegetation) distributed across the study area. The overall classification accuracy reached 94%, with a Kappa coefficient of 0.89, indicating an “almost perfect” agreement according to the thresholds proposed by Landis and Koch [78] and consistent with recent Sentinel-2 classification results [79]. This approach allowed for a better separation of vegetated and non-vegetated areas, reducing noise and improving the reliability of the LAI estimates [80].

$$\mathit{S}\mathit{R}={\displaystyle \frac{\mathit{B}\mathbf{8}}{\mathit{B}\mathbf{4}}}$$

(1)

This calculation yields values ranging from 0 (bare soil) to higher values for more densely vegetated areas. The SR is particularly relevant in this study, as this ratio is sensitive to the leaf area present in the canopy, a key factor in LAI estimation.

LAI was then estimated from SR using an empirical relationship specific to maize (LAI; Equation (2)), where a factor of 0.32 and a bias of −0.06 are empirical coefficients specific to maize cultivation and can sometimes be adjusted according to the region [81]. These coefficients have been specifically adjusted for maize cultivation based on agronomic research and are widely used to estimate LAI from SR in studies on this crop. Although no direct in-field calibration was performed, the empirical SR–LAI relationship was locally validated by comparing SR-derived LAI values with APSIM estimates on matching dates, confirming consistent trends and no systematic bias.

$$\mathit{L}\mathit{A}\mathit{I}=\mathbf{0.32}\times \mathit{R}\mathit{S}-\mathbf{0.06}$$

(2)

LAI values were extracted for each image based on the previously defined crop blocks. For each image in the collection, the average LAI was calculated for each block using the ‘reduceRegion’ function in GEE. This operation made it possible to determine the average LAI for each crop area on each image capture date. These results enabled a visual and temporal assessment of crop health, tracking the progression of leaf development from sowing to harvest (Supplementary Figure S1), which was important in calibrating the APSIM model. In addition, all LAI values for each block and each image were exported as a CSV file, allowing for detailed analysis and comparison with APSIM-simulated values and performance statistics.

2.3.6. Crop Parameters and LAI Calibration

The parameter ranges used for APSIM calibration were defined based on values from the literature and local agronomic observations to realistically represent maize growth under the edapho-climatic conditions of Lubumbashi. Phenological parameters (juvenile, flowering, and grain-filling phases) were adjusted according to degree-day observations from the field and technical data provided by SEED-CO Zambia for the SC719 variety. These values are consistent with those reported by Holzworth et al. [37], Sheng et al. [82], and Chauhdary et al. [83] for maize hybrids cultivated in similar tropical environments.

Morphological parameters (plant height, maximum grains per cob, and potential grain size) were adapted from field measurements, while parameters related to canopy dynamics (LAI) were calibrated using SR-derived values for each block and acquisition date from Sentinel-2 imagery. Parameters that could not be directly observed were kept within the default APSIM ranges, following the recommendations of Keating et al. [56].

Each parameter was thus adjusted within a biologically and agronomically plausible range, ensuring the internal consistency of the model. For LAI calibration, the values estimated on each Sentinel-2 image acquisition date were used to represent leaf growth dynamics throughout the crop cycle, using a cross-validation approach (2/3 of the data for calibration and 1/3 for validation) similar to that applied for maize grain yield (

Table 5. Input parameters used for calibrating the APSIM model for the SC719 maize variety and fertilization, accompanied by their descriptions and values. °Cd: degree day.

Cultivar Parameters	Description	Unit	Values	Source
Density		Plants m−2	6	Adjusted
Juvenile. Target	Development time of the juvenile phase	°Cd	170	Adjusted
FloweringToGrainFilling. Target	Time required to transition from flowering to grain filling	°Cd	175	Adapted
FlagLeafToFlowering. Target	Time from flag leaf appearance to flowering	°Cd	50	Adjusted
GrainFilling. Target	Time required for grain filling	°Cd	860	Default
MaturityToHarvestRipe	Time from maturity to harvest	°Cd	10	Default
Photosensitive. Target.	Photoperiod sensitivity	-	0; 12.5; 24	Default
Height	Height crop	cm	243	Adjusted
MaximumGrainsPerCob	Maximum number of grains per ear	number	1050	Adjusted
MaximumPotentialGrainSize	Maximum theoretical grain size	g	0.80	Adjusted
Root. SpecificRootLength	Specific root length	Cm g−1	100	Default
Proportion of plant mortality	Proportion of plant mortality (dimensionless, between 0 and 1)	-	0.02	Adapted
LAI	Leaf area index	m2 leaf m−2 soil	X a	Calibrated
Fertilizer
N Fertilization	Urea (45% N)	kg ha−1	200	Adapted

X ^a: LAI values obtained after calculating the SR were used for each acquisition date and block individually.

).

2.4. Cross-Validation

Cross-validation was conducted by pooling data from the three growing seasons (2016–2017, 2022–2023, and 2023–2024), resulting in n = 21 observations. After random sampling, approximately two-thirds of the dataset (12 blocks from the 2022–2023 and 2023–2024 seasons) were used for model calibration, while the remaining one-third (8 blocks from the same seasons and 1 block from the 2016–2017 season) were used for validation. This randomized cross-validation approach ensured that both calibration and validation datasets captured the variability observed across seasons and management conditions within the same experimental site.

For LAI, the cross-validation followed the same structure as that used for grain yield, with identical calibration and validation blocks and seasons. However, the number of available observations (n) depended on the number of cloud-free Sentinel-2 images providing valid Simple Ratio (SR) values for each block. Each SR-derived LAI estimate corresponded to one image acquisition date, so n varied according to the temporal frequency of usable images within each growing season. This approach ensured methodological consistency with yield calibration while accounting for the variability in Sentinel-2 data availability across blocks and seasons.

This approach ensured a rigorous separation between model training and evaluation while testing the model’s ability to generalize within the same pedoclimatic context but on unseen observations. Commonly applied in agro-environmental modeling, this method provides a robust assessment of APSIM’s stability and predictive performance under realistic spatial and temporal variability [37]. Commonly used in agro-environmental modeling, this method allows the robustness of the APSIM model to be evaluated by testing its ability to predict yields on blocks not used during calibration. The model was evaluated using the following statistical metrics:

-

Coefficient of determination (R²), which measures the proportion of variance explained by the model (R²; Equation (3)). A coefficient of determination (R²) greater than 0.70 is generally considered indicative of good model performance in crop growth simulations [84]. A value close to 1 indicates a strong correlation between simulated and observed values, reflecting good model performance.

$${\mathit{R}}^{\mathbf{2}}={\displaystyle \frac{\sum ({\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\overline{\mathbf{Y}}}_{\mathit{o}\mathit{b}\mathit{s}})({\mathit{Y}}_{\mathit{s}\mathit{i}\mathit{m}}-{\overline{\mathbf{Y}}}_{\mathit{s}\mathit{i}\mathit{m}})}{\sqrt{\sum {({\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\overline{\mathbf{Y}}}_{\mathit{o}\mathit{b}\mathit{s}})}^{\mathbf{2}}}\sum {({\mathit{Y}}_{\mathit{s}\mathit{i}\mathit{m}}-{\overline{\mathbf{Y}}}_{\mathit{s}\mathit{i}\mathit{m}})}^{\mathbf{2}}}}$$

(3)

-

RMSE (root mean square error), quantifying the average standard deviation between simulated and observed yields (RMSE; Equation (4)). Lower RMSE values (<0.5 for LAI or <1 t ha⁻¹ for yield) indicate higher model accuracy.

$$\mathit{R}\mathit{M}\mathit{S}\mathit{E}=\sqrt{{\displaystyle \frac{\mathbf{1}}{\mathit{n}}}\sum {({\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\mathit{Y}}_{\mathit{s}\mathit{i}\mathit{m}})}^{\mathbf{2}}}$$

(4)

-

NSE (Nash–Sutcliffe efficiency), which evaluates the accuracy of the model by comparing it to the average of the observations (NSE; Equation (5)). A model is considered satisfactory for NSE > 0.50, good for NSE > 0.65, and very good for NSE > 0.75 [84].

$$\mathit{N}\mathit{S}\mathit{E}=\mathbf{1}-{\displaystyle \frac{\sum {({\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\mathit{Y}}_{\mathit{s}\mathit{i}\mathit{m}})}^{\mathbf{2}}}{{({\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}})}^{\mathbf{2}}}}$$

(5)

-

MAE (Mean Absolute Error), which represents the average of the absolute differences between simulated and observed values, providing a complementary error measure less sensitive to outliers (MAE; Equation (6)). In crop yield simulations, MAE values below 0.5 t ha⁻¹ are generally considered indicative of acceptable model accuracy [84].

$$\mathit{M}\mathit{A}\mathit{E}={\displaystyle \frac{\mathbf{1}}{\mathit{n}}}\sum \left|{\mathit{Y}}_{\mathit{o}\mathit{b}\mathit{s}}-{\mathit{Y}}_{\mathit{s}\mathit{i}\mathit{m}}\right|$$

(6)

where Y_obs represents the observed value, Y_sim represents the simulated value, $\overline{\mathbf{Y}}$ represents the mean of the values, and n is the number of observations.

Overall, high R² and NSE values combined with low RMSE and MAE indicate strong agreement between simulated and observed data, demonstrating the satisfactory performance of the APSIM model.

2.5. Simulation Runs

All simulations were carried out using a common set of calibrated parameters. They covered the period from sowing to harvest (Table 1) for each block and season. Additional APSIM simulation outputs showing the temporal evolution of LAI and grain yield for representative blocks are provided in the Supplementary Materials (Figures S2–S7).

Figure 6 below shows the methodological diagram used for the construction and calibration of the APSIM model.

2.6. Data Analysis

All statistical analyses in this study were performed using R software, version 4.4.1.

3. Results

3.1. Performance of the APSIM Model for LAI Simulation

The cross-validation analysis of the APSIM model (

Table 6. Summary of APSIM model performance indicators in cross-validation on LAI. R² coefficient of determination; NSE = Nash–Sutcliffe efficiency; RMSE = root mean square error; MAE = Mean Absolute Error; n = number of blocks.

Set	LAI (m2 m−2)			Metrics
	Observed Mean	Simulated Mean	R2	NSE	RMSE	MAE	n
					(m2 m−2)
Calibration	0.50	0.39	0.87	0.71	0.32	0.25	220
Validation	0.51	0.42	0.85	0.70	0.35	0.27	148
Overall	0.50	0.40	0.86	0.67	0.33	0.26	368

; Figure 7) revealed a strong agreement between simulated and observed LAI values. The model accurately reproduced the temporal dynamics of canopy development across calibration and validation datasets, as illustrated by the clustering of points along the 1:1 line. Overall, the statistical indicators confirm satisfactory model accuracy, with high coefficients of determination and Nash–Sutcliffe efficiencies exceeding commonly accepted thresholds for good model performance (NSE > 0.5). The moderate error values further indicate that APSIM reliably captures leaf area variability between the different experimental blocks.

The simulated leaf dynamics during the crop cycle show that the APSIM model is capable of accurately representing maize growth through the simulation of the leaf area index (Figure 8). The simulated values generally follow the trend of the observations across all blocks, with a peak corresponding to the phase of full leaf cover, followed by a gradual decline at the end of the maize crop cycle (senescence).

For the 2022–2023 growing season (red box), the linear adjustments between observed and simulated values show a strong correlation, confirming that the model is well-suited to the field data. In 2023–2024 (blue box), the trends remain similar, although some blocks show more marked deviations, particularly block B8, which performs less well (NSE = 0.50; RMSE ≥ 0.30 m² m⁻²), suggesting more pronounced heterogeneity in plant structure.

For the 2016–2017 season, the model shows a satisfactory ability to reproduce LAI variations (NSE = 0.67), including under control conditions (B0) without amendment or subsoiling.

3.2. Evaluation of APSIM Model Performance in Cross-Validation of Maize Grain Yield

These results demonstrate that the APSIM model is capable of accurately reproducing maize yields over different years and experimental blocks while maintaining good stability between calibration and validation (

Table 7. Performance metrics of APSIM model: cross-validation with 2/3 calibration and 1/3 validation. R² = coefficient of determination; NSE = Nash–Sutcliffe efficiency; RMSE = root mean square error; MAE = Mean Absolute Error; n = number of measures.

Set	Maize Grain Yield (t ha−1)			Metrics
	Observed Mean	Simulated Mean	R2	NSE	RMSE	MAE	n
					(t ha−1)
Calibration	7.3	7.4	0.92	0.99	0.48	0.47	12
Validation	7.5	7.5	0.89	0.88	0.46	0.44	9
Overall	7.4	7.4	0.91	0.90	0.47	0.45	21

; Figure 8). Slightly more efficient on the training set, the model shows no signs of overfitting, which reinforces its applicability for larger-scale simulations.

Cross-validation results demonstrate that APSIM performs robustly in simulating maize grain yield across seasons and experimental blocks. The model showed a strong correlation between observed and simulated yields, with minimal deviation from the 1:1 line, reflecting good agreement and stable calibration–validation consistency. Error statistics confirm the reliability of APSIM predictions, with low discrepancies and a homogeneous error distribution, indicating no overfitting. Overall, the model successfully reproduces yield variations under contrasting soil management conditions, highlighting its potential for broader application in similar agroecological environments.

The integration of all seasons, including the 2016–2017 control (Figure 9), reinforces the overall consistency of the model, with a very strong relationship between simulated and observed values (Y = 1.00X + 0.01; R² = 0.91; NSE = 0.90), reflecting low residual variability and good generalization capacity of the model across seasons and different agronomic conditions.

3.3. Performance of the APSIM Model in Predicting Yields

Analysis of simulated and observed yields in the different blocks indicates that the APSIM model is capable of accurately reproducing maize grain yield production trends (

Table 8. Measured and simulated maize grain yield in the blocks. B1–B10: blocks that benefited from subsoiling and spreading of termite mound materials. B0: control without subsoiling and amendment with termite mound materials. (-) indicates that no yield measurement or simulation was available for that block during the corresponding growing season.

			Maize Grain Yield (t ha−1)
Block	2022–2023		2023–2024		2016–2017
	Obs	Pred	Obs	Pred	Obs	Pred
B0	(-)	(-)	(-)	(-)	4.1	4.4
B1	7.1	7.6	8.7	8.1	(-)	(-)
B2	8.1	8.5	8.9	9.4	(-)	(-)
B3	7.3	7.9	8.2	7.8	(-)	(-)
B4	8.9	9.4	10.4	10.9	(-)	(-)
B5	6.7	6.1	9.7	9.1	(-)	(-)
B6	6.1	6.5	6.1	6.4	(-)	(-)
B7	8.0	8.4	8.7	8.1	(-)	(-)
B8	5.1	5.0	5.8	5.3	(-)	(-)
B9	7.7	8.1	6.1	5.8	(-)	(-)
B10	6.2	6.7	7.5	7.0	(-)	(-)

). The results show that for the 2022–2023 and 2023–2024 seasons, the simulated values generally remain close to the observations, with moderate differences between the two. The blocks that benefited from subsoiling and amendment with termite mound materials (B1–B10) show higher yields, reaching 10.4 t ha⁻¹ in 2023–2024 for B4, reflecting the positive effect of these interventions on productivity. Conversely, the control block (B0) in 2016–2017, without subsoiling or the addition of termite mound materials, showed a lower yield (4.1 t ha⁻¹ observed versus 4.4 t ha⁻¹ simulated), highlighting the limiting effect of the properties of unamended plinthosols on maize production.

Although the accuracy of the model is generally satisfactory, some differences are observed, particularly in B5 (2022–2023) and B9 (2023–2024), where the differences between observed and simulated yields exceed 0.5 t ha⁻¹, suggesting possible adjustments to the model based on the specific conditions of these blocks.

4. Discussion

4.1. Evaluation of the Accuracy of LAI Derived from Sentinel-2 and Its Simulation with APSIM

The integration of Sentinel-2 imagery in LAI estimation enabled effective monitoring of maize canopy dynamics across the different blocks. The use of the Simple Ratio (SR) index proved relevant, as it is less prone to spectral saturation than other indices (e.g., NDVI, EVI) and is better suited to variable canopy densities [73,74]. The strong correlation between observed and simulated LAI values (R² = 0.86; NSE = 0.67) indicates that the APSIM model accurately reproduced the main growth stages, particularly the expansion phase followed by progressive senescence, consistent with maize phenology [36,85].

The differences observed among blocks nevertheless reflect the influence of local conditions on canopy response. Blocks showing slightly lower LAI values (e.g., B8) may have been affected by differences in compaction, texture, or water availability, which influence root growth and water uptake [80,86]. In this block, although subsoiling and amendment were implemented, soil variability could result from uneven subsoiler penetration depth or an uneven distribution of termite mound material within the plot, leading to locally less improved zones and slightly limited canopy development. This may also suggest an insufficient first pass of the subsoiler in that block.

The model also satisfactorily reproduced the LAI dynamics in the control block (B0), confirming its ability to represent foliar growth processes under restrictive soil conditions. The absence of marked differences between the control and some improved blocks could be explained by a relative homogeneity of canopy cover at full growth stage, or by the model’s limited sensitivity to certain fine structural soil variables (e.g., bulk density, macroporosity, micro-aggregation) whose spatial variability is not explicitly represented in APSIM, as well as by the spatial resolution of Sentinel-2 data, which tends to smooth local canopy contrasts and reduce the detection of intra-plot variability [87]. Nevertheless, the results suggest that termite mound amendments contributed to improving soil structure, porosity, and stability, thereby promoting better water infiltration and enhanced rhizosphere aeration. These conditions supported more efficient root expansion, improved light interception, and more stable photosynthetic activity throughout the growing cycle. These findings are consistent with Santos et al. [88] and Dilla et al. [89], confirming that LAI is a sensitive indicator of soil structure, water status, and foliar development interactions.

4.2. Evaluation of the Performance of the APSIM Model for Simulating Maize Grain Yields

The yield simulations obtained with APSIM (R² = 0.91; NSE = 0.90) confirm the robustness of the model and its ability to generalize production trends observed across seasons and blocks. These results are consistent with those of Song et al. [90] and Robertson et al. [91], who demonstrated the reliability of APSIM in predicting maize grain yield across diverse pedoclimatic conditions.

The LAI dynamics described above directly explain the yield patterns obtained. Indeed, a more stable and prolonged leaf development in the improved blocks favored greater light interception, leading to higher cumulative photosynthesis and, consequently, greater carbon assimilation for grain formation [92]. These results confirm the role of LAI as an integrative indicator of maize physiological vigor, linking soil structural conditions to final productivity [85,89].

The productivity gap between the control block (B0) and the improved blocks (B1–B10), although moderate, illustrates the positive effect of subsoiling and termite mound amendments on crop performance. These materials are rich in fine particles, calcium, magnesium, and stable organic matter, which improve cation exchange capacity and availability of essential nutrients [20,93,94,95]. Their incorporation into the cultivated horizon increases porosity and available water capacity while reducing bulk density, thus promoting better root exploration and more efficient carbon translocation to grains [33,96].

In parallel, termite mound materials host an active microflora and microfauna that enhance progressive mineralization of organic matter and stimulate soil biological activity [97,98,99]. These biological dynamics, coupled with improved physical structure, sustain a more continuous nitrogen and phosphorus supply, contributing to stronger vegetative growth and a more complete grain filling [100].

Conversely, unimproved plinthosols are characterized by low hydraulic conductivity, high mechanical resistance, and reduced biological activity, which limit root growth and nutrient uptake [69,101,102].

The moderate discrepancies between simulated and observed values in some blocks (e.g., B5 and B9, >0.5 t ha⁻¹) may arise from uncertainties in input data (meteorology, soil properties), incomplete parameterization of the root profile, or microclimatic variability not captured by gridded data. Future studies could include local meteorological measurements, along with data on root vertical distribution and the active organic matter fraction, to refine calibration and improve the representation of water and nutrient fluxes in the model. Holzworth et al. [37] emphasized that a finer optimization of coefficients related to soil properties and biotic factors can significantly improve the accuracy of APSIM’s growth and yield simulations. Furthermore, the consistency of APSIM performance across contrasting blocks suggests that, with appropriate soil and climate parameterization, the model can be effectively adapted to other plinthic and ferrallitic environments in sub-Saharan Africa, thereby extending its applicability beyond the study site.

Finally, beyond model validation, these results highlight the relevance of combining biophysical modeling and remote sensing for spatial assessment of cropping practices. This integrated approach provides a solid basis for the development of decision support systems (DSSs) and sustainable intensification strategies in sub-Saharan Africa. The integration of APSIM with high-resolution satellite data also opens promising perspectives for precision agriculture, enabling the identification of water stress zones and the targeted management of inputs.

5. Conclusions

This study confirmed the ability of the APSIM model to accurately simulate maize growth and yield on subsoiled plinthosols amended with termite mound materials in Lubumbashi. The integration of APSIM and Sentinel-2 data provided reliable estimates of leaf area index and yield, successfully reproducing the positive effects of soil improvement practices on productivity. However, slight discrepancies between simulated and observed values highlight the need for refined calibration, supported by complementary datasets such as local meteorological observations, measurements of root vertical distribution, belowground biomass dynamics, and the active organic matter fraction. Incorporating these variables would enhance the representation of soil–plant interactions and improve predictive accuracy. From an operational perspective, the APSIM-Sentinel-2 framework offers strong potential as a decision support tool for crop monitoring and the optimization of management practices in near-real time. Overall, this modeling framework provides a robust scientific basis for promoting sustainable intensification, that is, increasing productivity while preserving soil quality and environmental integrity, and for strengthening the climate resilience of maize production systems on degraded or plinthic soils in southern DRC and similar sub-Saharan agroecological zones.