Model validation was performed using 50 independent ground truth samples representing settle- ment, paddy field, and non-paddy vegetation classes (Figure 1.b). The Random Forest (RF) clas- sification achieved an overall accuracy of 84% and a Kappa coefficient of 0.71, indicating subs- tantial agreement between classified and observed data. Minor errors occurred in transitional zones where grasslands and fallow plots spectrally resembled paddy fields, a common feature in tropical mosaics. Overall, the RF model showed strong class separability and reliability for tem- poral modeling. The field-based confusion matrix Figure 3 confirms this performance, with high producer and user accuracies across major classes and minimal cross-class misclassification.

Further robustness evaluation was conducted using both area- and probability-based transition matrices for two temporal periods: 2015–2019 and 2019–2024 Figure 4. The 2015–2019 matrix indicated strong class stability, dominated by settlements (3,954 ha; 0.973), paddy fields (7,256 ha; 0.853), and non-paddy vegetation (11,353 ha; 0.903). Reverse transitions of built-up areas were minimal, with only 82 ha (0.020) reclassified as paddy fields and 30 ha (0.007) as non-paddy fields, confirming that classification noise was negligible and that the model retained spatial logic. During 2019–2024, class stability remained high for built-up (4,700 ha; 0.933) and non-paddy fields (11,395 ha; 0.932), while the conversion of paddy fields to settlements increased substan- tially from 411 ha (0.048) to 1,103 ha (0.140), reflecting genuine urban expansion. Reverse tran- sitions from settlements to paddy (320 ha; 0.064) and to non-paddy fields (19 ha; 0.004) were again minor and spatially scattered. Collectively, the two matrices confirm that the RF classifier maintains temporal robustness and spatial coherence, ensuring that the CA–ANN simulation for 2029 accurately reflects fundamental land use dynamics rather than algorithmic artifacts.

Temporally, land use dynamics between 2015 and 2024 reveal a pronounced and systematic trans- formation in the PNAR of Purwokerto, marked by a substantial expansion of settlement areas and a consistent contraction of paddy-field extent. According to the statistical recapitulation Figure 5 built-up areas increased sharply from 4,068.87 ha in 2015 to 5,960.98 ha in 2024, representing an overall growth of approximately 46.5 percent within less than a decade. In contrast, paddy- field coverage exhibited a continuous decline, dropping from 8,506.19 ha in 2015 to 7,893.43 ha in 2019 (−612.76 ha), and further to 7,040.78 ha in 2024 (−852.65 ha). The cumulative loss of 1,465.41 ha over the ten years underscores an accelerating land conversion process primarily driven by rapid urban growth and infrastructure development. Meanwhile, the non-paddy-field category remained relatively stable, fluctuating only slightly from 12,577.80 ha in 2015 to 12,151.10 ha in 2024, suggesting that paddy-field loss occurred predominantly within the rural– urban transition belt rather than through large-scale vegetation clearance.

Spatially, the expansion of settlement areas follows a concentric diffusion pattern radiating outward from the Purwokerto urban core Figure 6 . The most significant outward trajectories are observed toward the east (Kembaran), south (Purwokerto Selatan), north (Baturaden), southwest (Karanglewas), and northwest (Sumbang), reflecting both clustered and linear growth forms. These patterns indicate an ongoing process of urban sprawl that extends beyond the limits of formal spatial planning, in which productive agricultural land, particularly paddy fields, consti- tutes the most vulnerable target for conversion. Two distinct morphologies of expansion can be discerned: (1) compact or clustered development within the central urban core, driven by residen- tial densification and infill; and (2) linear sprawl along primary transportation corridors connec- ting Purwokerto to its surrounding subdistricts. This configuration aligns with the findings of (Munibah et al., 2024), confirming the strong influence of accessibility and spatial hierarchy in accelerating conversion pressures.

Building upon these regional transitions, subdistrict-level evaluation Figure 7 provides a more detailed understanding of paddy-field dynamics by isolating the pure agricultural extent regardless of conversion destination. The data show an uneven but alarming decline across the PNAR. The most significant reduction occurred in Sumbang (516.3 ha), followed by Kembaran (243.7 ha), Kedungbanteng (177.5 ha), and Karanglewas (98.7 ha), all of which historically served as the agrarian buffer of Purwokerto. The persistent contraction of active paddy-field areas across these outer subdistricts signifies that development pressure has advanced beyond the city limits, frag- menting what was once a contiguous rice-producing landscape into a mosaic of residential, com- mercial, and semi-urban land uses.

Within the inner urban subdistricts of Purwokerto Selatan, Barat, Timur, and Utara, the remaining paddy-field area has declined to critically low levels, each representing less than 4 percent of its administrative territory. By 2024, Purwokerto Selatan retained only 287. Spatially, these dynamics reflect a dual morphological trajectory, urban densification within the 186.18 ha (2.23 %), Barat 205.64 ha (2.61 %), and Utara 318.42 ha (3.29 %), highlighting the near disappearance of agrarian land within the city’s core. Overall, paddy-field coverage across the PNAR declined from 8,506.19 ha (33.82 %) in 2015 to 7,040.78 ha (27.99 %) in 2024, an unmistakable sign of spatial attrition that precedes direct conversion. This pattern illustrates that the depletion of agricultural land itself has reached a critical stage, providing a quantitative basis for analyzing paddy-field-to-settlement conversion dynamics in the following section.

Building on previous findings, the transition analysis Figure 8 quantifies the extent to which this agricultural decline corresponds directly to settlement conversion, isolating the urbanization- driven component of land transformation. The conversion rate of paddy fields to settlements rose dramatically from 410.39 ha (102.6 ha yr⁻¹ or 1.63 %) during 2015–2019 to 1,105.03 ha (221.0 ha yr⁻¹ or 4.39 %) during 2019–2024, a 115 percent increase in annual conversion intensity. This acceleration coincides with the surge in post-2019 residential and infrastructure projects, particu- larly along the Purwokerto–Kembaran–Sumbang growth corridor, now functioning as the domi- nant axis of urban expansion. In relative terms, conversion shares more than doubled across all subdistricts, from 0.75–4.04 % in the first period to 2.73–7.54 % in the second, confirming both temporal and spatial intensification of land transformation.

Spatially, these dynamics reflect a dual morphological trajectory, urban densification within the core and peripheral sprawl in the outer subdistricts. In the core area, Purwokerto Selatan increased from 56.65 to 101.62 ha (equivalent to a rise from 4.04 to 7.25 percent), Purwokerto Utara from 26.63 to 72.90 ha (2.75 to 7.54%), and Purwokerto Timur from 28.79 to 41.76 ha (3.44 to 5.00%), indicating rapid infill growth driven by the exhaustion of vacant plots and the conversion of re- maining paddy patches into residential areas. In the outer fringe, Sumbang expanded from 74.24 to 301.56 ha (1.35 to 5.48%), Kembaran from 61.49 to 143.79 ha (2.42 to 5.66%), Kedungbanteng from 41.31 to 163.84 ha (0.75 to 2.96%), and Karanglewas from 57.17 to 105.82 ha (1.88 to 3.48 percent), marking the outward progression of low-density urban sprawl into former agricultural hinterlands. Baturraden, constrained by steeper slopes, experienced more limited conversion, in- creasing from 39.44 to 124.45 ha (0.87 to 2.73%). Collectively, these patterns emphasize an in- tensified core–periphery gradient of urbanization: densification now dominates the urban center. At the same time, outward expansion extends across peri-urban transition belts, progressively eroding the agrarian base and heightening ecological vulnerability.

In several hotspot subdistricts such as Sumbang, Kembaran, Baturraden, and Purwokerto Selatan, field validation confirms that many new residential clusters have emerged directly atop former paddy fields Figure 9 . These findings highlight the tangible encroachment of settlement cons- truction on productive agricultural land and reinforce the spatial correlation between urban growth and agricultural loss. Meanwhile, core subdistricts: Purwokerto Timur, Barat, and Utara, though smaller in absolute area, remain strategically critical due to their high population density and in- creasing land fragmentation, which collectively accelerate infill and reduce the remaining agri- cultural resilience.

Overall, paddy field conversion over the past decade has not only increased in volume but also driven urban dispersal into peripheral zones. Without more progressive and spatially informed land use policies, the loss of productive land risks becoming irreversible, threatening local food security and undermining the agrarian foundation of transitional regions such as Purwokerto.

Land use prediction for the year 2029 in the PNAR of Purwokerto was developed using a spatial modeling approach that integrates Cellular Automata–Artificial Neural Network (CA–ANN). This approach combines historical land use data from 2019 and 2024 as inputs, along with five driving factors that determine the direction of spatial change. These variables include slope (Figure 10.a), distance to roads (Figure 10.b), population density (Figure 10.c), distance to regional activity centers (Figure 10.d), and distance to subdistrict activity centers (Figure 10.e), each weighted based on its vulnerability to paddy-field conversion into settlement areas. The integra- tion process was conducted using a weighted overlay technique, resulting in a land conversion vulnerability map that served as the basis for predicting land use changes in 2029.

Specific biophysical and socio-economic characteristics play a pivotal role in driving land con- version dynamics within peri-urban landscapes. Areas with low slope gradients (0–3%) are more prone to conversion due to ease of access and high construction feasibility. Likewise, locations situated within a 100-meter buffer from major roads are prioritized for development, benefiting from superior transport connectivity. Furthermore, zones with high population density, particu- larly those in proximity to regional and local growth centers (regional center and subdistrict cen- ter), face intense spatial pressure on remaining agricultural lands. The convergence of favorable topographic conditions, infrastructural accessibility, demographic stress, and nearness to growth hubs positions these areas as potential hotspots for future land use transformation and ((Pratami et al., 2019); (Sugandhi et al., 2022)) .

Using the Tabulate Area analysis in QGIS, the 2024 paddy-field extent was intersected with re- classified driving-factor layers to compute the total area (in hectares) and the percentage distribu- tion by class. Results show that approximately 3,920 ha (49%) of paddy fields are located on slopes below 3%, 3,425 ha (43%) lie within 100 m of major roads, and 4,476 ha (56%) are situated in high-density settlement zones, confirming that a large share of agricultural land already over- laps with zones of elevated conversion pressure. In contrast, only 934 ha (12%) occur on steeper terrain (>15%), and less than 1,440 ha (18%) are located beyond 200 m from main roads, sugges- ting that spatial refuges from urban encroachment are becoming increasingly limited. The detailed hierarchical structure of these driving factors and their suitability classes is illustrated in the sun- burst diagram Figure 12 , which visualizes the relative contribution of each factor to the overall spatial conversion potential.

According to the CA–ANN simulation results (Figure 11b), the extent of paddy fields is projected to decrease from 7,040.78 ha in 2024 to 6,511.94 ha in 2029, marking a loss of about 528.84 ha, or 7.5% within five years. Although the rate of decline appears slower than in the previous period, this moderation primarily reflects the diminishing availability of easily convertible land. Most low-lying, highly accessible paddy fields near major road corridors and urban centers have already been transformed in earlier phases, leaving fewer plots within the current urbanization frontier. Consequently, recent expansion tends to occur in smaller, fragmented patches dispersed across peri-urban areas, which slows the aggregate conversion rate while still indicating persistent spatial pressure on productive agricultural land (Figure 13a).

The total conversion share decreases from 4.39% to 2.09% of the PNAR’s total area, equivalent to approximately 526.52 ha of paddy fields converted into settlement areas. Sumbang remains the dominant conversion hotspot, with 119.99 ha (2.18%) of its territory projected to transition to settlement areas, thereby maintaining its role as the principal expansion front. Kembaran follows with 77.75 ha (3.06%), while Purwokerto Utara (42.97 ha; 4.44%), Purwokerto Barat (31.03 ha; 3.95%), Purwokerto Timur (31.37 ha; 3.75%), and Purwokerto Selatan (48.52 ha; 3.46%) conti- nue to experience significant infill within the urban core. In contrast, outer subdistricts such as Karanglewas (57.92 ha; 1.90%), Kedungbanteng (56.80 ha; 1.03%), and Baturraden (60.17 ha; 1.32%) show relatively minor but spatially scattered conversion patches, reflecting a shift toward fragmented development along secondary road networks. These projections suggest that while the pace of overall conversion may appear to decelerate, the spatial extent of settlement expansion continues to encroach upon the remaining agricultural land, particularly along mixed urban–rural interfaces. By 2029, under the Business-as-Usual (BaU) scenario, the total paddy-field area is expected to decline to just 6,511.94 ha, equivalent to only 25.9% of the PNAR’s total land area, signaling a critical reduction in the region’s agrarian base and intensifying challenges for local rice self-sufficiency.

The relationship between paddy-field area and rice production in the PNAR of Purwokerto exhi- bits a robust, consistent correlation over time. Based on linear regression analysis, the coefficient of determination (R²) values were 0.9093 in 2015, 0.8518 in 2019, and 0.8327 in 2024. The ave- rage R² value of 0.8647 indicates that changes in paddy-field area can explain more than 86% of the variation in rice production. This positive correlation confirms that paddy fields remain the primary and decisive input in maintaining regional rice production capacity. The decline in paddy- field area from 8,506.19 ha (2015) to 7,040.78 ha (2024) is reflected in the corresponding drop in rice production, from 95,148.10 tons to 80,148.52 tons. This indicates that land conversion di- rectly weakens local food security by reducing production potential.

To assess the sustainability of the food system, a rice availability balance was calculated using a paddy-to-rice conversion rate of 51.24% (BPS) and a per capita rice consumption rate of 79.08 kg/year based on national food standards (Bapanas). The results show a surplus of 3,791.05 tons in 2015, decreasing to 2,180.41 tons in 2019, and shifting to a significant deficit of −7,140.57 tons in 2024 Figure 14 . This decline highlights an alarming trend in which the population growth, reaching 609,619 people in 2024, is not matched by adequate rice supply capacity. What was once a self-sufficient production system is no longer sufficient to meet local needs, with a wide deficit margin indicating increasing dependence on external supply sources.

Furthermore, spatial disparities between rice supply and demand needs are most evident in urban districts such as the four districts of Purwokerto. Although these districts serve as administrative and economic hubs, each has fewer than 400 ha of paddy fields and a rapidly growing population. As a result, they face significant food deficits. In 2024, Purwokerto Selatan recorded a deficit of−4,989 tons, Purwokerto Barat −3,603 tons, Purwokerto Timur −3,945 tons, and Purwokerto Utara −3,434 tons. These disparities reveal that an urbanization-driven development model that neglects the protection of food-producing areas has created spatial dysfunction in the local food security system.

Meanwhile, agrarian areas such as Sumbang, Kedungbanteng, and Kembaran remain the main contributors to rice supply, with surpluses of 2,224 tons, 3,639 tons, and 1,802 tons, respectively, in 2024. However, signs of declining production are also emerging in these regions, driven by increasing pressure from land conversion and population growth. If this trend is not controlled, the PNAR in Purwokerto will likely rely entirely on external food supplies, undermining the auto- nomy and sustainability of the local food system.

These findings affirm that regional food security depends directly on the spatial protection of paddy fields. The persistent rice deficit amid rapid population growth and land conversion signals a structural crisis requiring urgent intervention. Policies should prioritize the designation of Sus- tainable Food Agricultural Land, stricter spatial regulation of conversion, and the promotion of intensive small-scale urban agriculture. Without such integrative measures, dependence on exter- nal food sources will deepen, undermining long-term regional resilience.

To anticipate future food security challenges in the PNAR of Purwokerto, a simulation of the 2029 rice availability balance was conducted based on three spatial and demographic management sce- narios: BaU, Moderate, and Optimistic. These scenarios are based on three key variables: the rate of paddy-field conversion, population growth, and the increase in the cropping index (CI), as shown in Table 2.

The BaU scenario represents a no-intervention condition, assuming the CI remains at 200, the paddy-field area declines to 6,511.94 ha, and the population increases to 648,490. The Moderate scenario assumes controlled urban expansion, with CI increased to 250, a 2% reduction in paddy- field conversion (to 6,642.18 ha), and a 4% reduction in population growth, resulting in a popu- lation of 622,550. The Optimistic scenario represents maximum intervention: CI is raised to 300, paddy-field conversion is reduced by 4% (to 6,772.42 ha), and population growth is further redu- ced by 5%, resulting in 616,066 people.

Under the BaU scenario, the rice availability balance is projected to face a significant deficit of −12,520.54 tons Figure 15. This is due to a combination of shrinking paddy-field area and stagnant productivity (CI remains at 200), producing only 75,648.03 tons of paddy or 38,762.05 tons of rice. Meanwhile, rice demand reaches 51,282.59 tons due to population growth. Spatially, all central urban districts face severe deficits, including Purwokerto Selatan (−5,651.25 tons), Purwokerto Barat (−3,636.65 tons), and Purwokerto Timur (−4,171.02 tons), reflecting a structural imbalance between consumption centers and production areas. Even agriculturally based districts such as Karanglewas and Baturraden are under pressure, with small surpluses of only 23.27 and 276.13 tons, respectively.

The Moderate scenario yields much more stable results. With moderate controls on housing expansion and population growth, combined with a higher CI of 250, paddy production increases to 96,451.24 tons, or 49,421.61 tons of rice. With a projected rice demand of 49,231.29 tons, the overall rice availability balance is nearly even, with a slight surplus of 190.34 tons. Spatial disparities remain in core urban areas such as Purwokerto Selatan (−5,219.85 tons), Purwokerto Barat (−3,199.89 tons), and Purwokerto Timur (−3,832.65 tons). However, these deficits are offset by large surpluses from agrarian areas such as Sumbang (+4,624.46 tons), Kedungbanteng (+3,826.65 tons), and Kembaran (+2,955.41 tons). These results indicate that even limited interventions can stabilize the rice balance when implemented in a planned, integrated manner.

The Optimistic scenario offers the most promising outcome. With CI increased to 300, paddy-field conversion limited to 6,772.42 ha, and population capped at 616,066, paddy production surges to 118,010.93 tons, equivalent to 60,468.80 tons of rice. At the same time, rice demand is only 48,718.46 tons. Consequently, the PNAR of Purwokerto has the potential to achieve a rice surplus of +11,750.36 tons by 2029. Agrarian districts emerge as the key buffers, with significant surpluses from Sumbang (+7,535.05 tons), Kedungbanteng (+5,911.05 tons), Kembaran (+5,202.18 tons), Karanglewas (+3,633.56 tons), and Baturraden (+3,253.26 tons).

Managing population growth and intensifying production can effectively reverse the trend from food deficits to balance or surplus. Therefore, the Moderate and Optimistic scenarios are not mere idealizations but provide a concrete foundation for long-term policy decisions aimed at achieving sustainable regional food security.

This study demonstrates that paddy-field conversion in the PNAR of Purwokerto has evolved systematically and intensified over the past decade, reflecting the classic pattern of peri-urban sprawl consistent with (Munibah et al., 2024). Although Purwokerto has not yet been formally designated as a new autonomous region, its functional urban expansion already exhibits the characteristics of an emerging city region. However, evidence from the PNAR highlights a critical trade-off: while infrastructure development and modernization have accelerated, they have simultaneously driven extensive farmland conversion, undermining regional food security. This dynamic mirrors global experiences in peri-urban areas, where development fosters opportunity yet deepens vulnerability ((Allen, 2003); (Simon, 2008)). Urban expansion, though an inevitable aspect of modernization, can either strengthen resilience or amplify systemic risk, depending on how it is governed. The convergence of urban and agricultural land uses thus creates contested spaces where competing priorities collide. Without robust spatial regulation and food-sensitive planning, transitional zones risk becoming structurally fragile, reinforcing urban bias and exacerbating food insecurity ((Lipton, 1976); (Seto et al., 2012)).

Beyond area loss, land conversion signifies a degradation in spatial quality and production capacity. Between 2015 and 2024, paddy fields declined from 8,506 ha to 7,041 ha. Projections indicate a further reduction to 6,512 ha by 2029. Every 100 hectares converted corresponds to approximately 767 tons of lost rice output Figure 13, underlining the direct link between spatial transformation and systemic vulnerability. As land conversion continues, the regional rice balance deteriorates, shifting from a surplus in 2015 (+3,791 tons) to a deficit in 2024 (−7,141 tons), and potentially reaching a deficit of −12,520 tons by 2029. These results support the food–land nexus framework (Cumming et al., 2014) which emphasizes that spatial integrity, resource governance, and production systems underpin food security. Comparable trade-offs have been documented in China, the Philippines, and Sub-Saharan Africa, where rapid peri-urban growth has undermined agricultural resilience, increasing reliance on imports and exposure to market fluctuations ((Dumdumaya & Cabrera, 2023); (Molla et al., 2024); (Park et al., 2024)).

This regional vulnerability mirrors Indonesia’s broader food security challenges, as dependence on imports of national rice has surged in recent years. In 2024, Indonesia imported over 4.5 million tons of rice, an elevenfold increase from 2019 (444,508 tons), with a CIF value reaching US$2.71 billion, the highest in eight years. The escalation stems from domestic production shortfalls driven by farmland loss and climate disruptions. Continued conversion of productive paddy fields, such as in the PNAR Purwokerto, risks compounding local deficits into systemic national shortages and deepening dependence on volatile global markets. This feedback loop, in which shrinking domestic capacity fuels greater reliance on imports, underscores that the spatial fragility observed in PNAR reflects a structural national risk. Safeguarding peri-urban agricultural land is therefore not merely a regional issue but a strategic imperative for sustaining Indonesia’s long-term food resilience.

Underlying this trajectory is a structural governance deficit. Although Indonesian Law No. 41/2009 mandates the protection of agricultural land to ensure food security, enforcement in transitional regions remains weak. Fragmented institutional authority and poorly coordinated spatial plans create a regulatory gap that allows speculative development to outpace policy control. As a result, farmland protection becomes nominal, mainly with regulations existing in form but not in effect. According to peri-urban resilience theory, such governance fragility reflects low adaptive capacity, where misaligned policies and weak institutional responsiveness expose regions to systemic risks ((Allen, 2003); (Woltjer, 2014)). The implications extend beyond food insecurity to socio-economic fragility, as rapid land conversion without livelihood alternatives deepens agrarian distress and widens rural–urban inequality. Similar urban-biased development patterns in Africa and South Asia (Simon, 2008) show that unless food security objectives are embedded within spatial governance, the PNAR of Purwokerto risks reproducing unsustainable and inequitable growth trajectories.

Scenario modeling illustrates that without intervention, the PNAR’s rice deficit could deepen to −12,520 tons by 2029. While modest gains in CI offer partial relief, they cannot offset ongoing farmland loss. A moderate scenario combining CI 250, a 2% reduction in land conversion, and a 4% slowdown in population growth stabilizes the rice balance near equilibrium (+190 tons), whereas a surplus under the optimistic case would demand major governance and behavioral reforms.

These findings highlight that technological improvements alone are insufficient; effective land use control and demographic management are essential to curb agricultural decline. Global experiences from Shanghai, Tokyo, and Havana show that enforceable zoning, fiscal incentives, and food-sensitive spatial planning can harmonize urban growth with farmland preservation (Skar et al., 2020).

The implications for governance are far-reaching. Addressing the trade-offs in the PNAR of Purwokerto requires a coordinated, proactive, and integrated policy framework that aligns spatial planning with food system resilience. Farmland protection must become a core governance priority. This involves enforcing Law No. 41/2009 through binding agricultural zones in the Regional Spatial Plan, strengthened by tax incentives for active farming, penalties for illegal conversion, and GIS-based monitoring to identify high-risk conversion areas.

Additionally, production intensification must complement regulatory efforts. Boosting CI to 250 using high-yield, short-duration rice varieties (e.g., Inpari 32, Mekongga) and mechanized rotations can improve yields. These practices require reliable input subsidies, extension services, and efficient micro-irrigation to ensure viability. Population pressure also plays a role. Community-based family planning programs (e.g., Kampung KB) and reproductive health education can help slow demographic growth in high-risk peri-urban districts. Finally, economic diversification can reduce conversion incentives. Agrotourism and high-value horticulture can provide alternative income streams for rural households, easing pressure on land resources. Together, these approaches represent a comprehensive response to the multidimensional risks facing the PNAR.

Ultimately, the PNAR of Purwokerto exemplifies the broader urban–agrarian paradox observed across Indonesia’s secondary cities, where decentralization stimulates economic growth yet simultaneously undermines agrarian resilience. The integrative modeling framework, combining remote sensing, land use simulation, and rice balance analysis, serves as both a diagnostic and planning instrument for evidence-based policy. Its approach is transferable to other rapidly urbanizing cities in the Global South, particularly those undergoing decentralization and facing similar tensions between development and food system sustainability.