Ratoon rice mapping based on Sentinel-1 and Sentinel-2 imagery Vidya Nahdhiyatul Fikriyah a,b,*, Roshanak Darvishzadeh a, Alice Laborte c, Andrew Nelson a a Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, Enschede, the Netherlands b Faculty of Geography, Universitas Muhammadiyah Surakarta, Surakarta, Indonesia c International Rice Research Institute (IRRI), Los Banos, Laguna, Philippines A R T I C L E I N F O Keywords: Ratoon rice Second harvest Food security Cropping practices Sustainable agriculture A B S T R A C T Rice ratooning has gained increasing interest in Asia as a way to boost rice production by allowing two rice harvests from a single growing season. Accurate mapping of this practice can improve rice production estimates. However, current efforts have mainly relied on optical sen sors, which are limited by cloud cover, especially during the wet season when ratooning is common. This study systematically assessed the use of optical Sentinel-2, Synthetic Aperture Radar (SAR) Sentinel-1 data and their combination to map ratoon rice crops. Field data were collected in four provinces of the Philippines in 2018–19. Backscatter intensity from Sentinel-1, spectral information, and six commonly used vegetation indices (VIs) from Sentinel-2 were analysed using the Mann-Whitney U significance test to examine differences between the main and ratoon rice crops. Next, we compared the classification performance of decision tree (DT), support vector machine (SVM), and random forest (RF) classifiers. Results show that ratoon and main rice crop significantly differed in VV and VH polarisations, red edge and near-infrared bands, and all VIs. The highest accuracy was achieved with selected features in an RF classifier (overall accuracy of 92 %), compared to SVM (87 %) and DT (81 %). Classification using features from both Sentinel-1 and 2 consistently yielded higher accuracy than using features from one sensor alone. The total planting of ratoon rice was estimated at approximately 223 km2 (±4 % of the wet season rice area). This study demonstrates the value of combining SAR Sentinel-1 and optical Sentinel-2 for ratoon rice mapping. 1. Introduction Rice is one of the world’s most consumed staple foods, providing approximately 50 % of the caloric supply for millions of people (Maclean et al., 2002). The majority of the world’s rice is cultivated in Asia (Bandumula, 2018), (Isa et al., 2021), where increased rice production targets to meet future demands require higher yields on existing or even shrinking cropland due to urbanisation, land use conversion and limited opportunities to expand rice cultivation into other areas (Asaba et al., 2024), (Rostiana and Hillah, 2016). As a result, various efforts in rice cropping practices have been carried out to increase rice production using limited land resources, including ratooning. Rice ratooning – the management practice of growing a second crop from the stubble of the harvested main crop - has become a * Corresponding author. Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, Enschede, the Netherlands. E-mail addresses: v.n.fikriyah@utwente.nl, vidya.n.fikriyah@ums.ac.id (V.N. Fikriyah), r.darvish@utwente.nl (R. Darvishzadeh), a.g.laborte@ irri.org (A. Laborte), a.nelson@utwente.nl (A. Nelson). Contents lists available at ScienceDirect Remote Sensing Applications: Society and Environment journal homepage: www.elsevier.com/locate/rsase https://doi.org/10.1016/j.rsase.2025.101592 Received 24 January 2025; Received in revised form 6 May 2025; Accepted 12 May 2025 Remote Sensing Applications: Society and Environment 38 (2025) 101592 Available online 14 May 2025 2352-9385/© 2025 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). mailto:v.n.fikriyah@utwente.nl mailto:vidya.n.fikriyah@ums.ac.id mailto:r.darvish@utwente.nl mailto:a.g.laborte@irri.org mailto:a.g.laborte@irri.org mailto:a.nelson@utwente.nl www.sciencedirect.com/science/journal/23529385 https://www.elsevier.com/locate/rsase https://doi.org/10.1016/j.rsase.2025.101592 https://doi.org/10.1016/j.rsase.2025.101592 http://creativecommons.org/licenses/by/4.0/ popular way of increasing crop yields, sometimes as an alternative to having two rice crops per year, also known as double cropping (Jiang et al., 2021). The advantage of ratooning over double cropping is that it achieves a second yield with limited resources, including water, seeds, fertiliser, labour, time, and money (Dong et al., 2017). Ratooning does not require additional time to prepare the land and establish the crop, as rice tillers will form from existing stubble nodes (Oad et al., 2002). Ratoon rice has a shorter maturity duration when compared to the main rice crop and can contribute about 50 % of the main crop yield with proper management (Saito et al., 2024). Remote sensing technology, specifically optical data, has been successfully used for rice crop mapping since such data reflects the crop’s biological and physical features, such as vegetation cover, Leaf Area Index (LAI), and biomass (Mansaray et al., 2022). However, relying on optical data is challenging due to high cloud cover, particularly during the wet season. In contrast, Synthetic Aperture Radar (SAR) data has been effectively used in rice mapping, given its sensitivity to soil surface moisture and crop biophysical parameters (Chang et al., 2021). Although SAR observation is unaffected by cloud disturbance, it inherently has granular speckle noise, reducing the interpretation capability in cropland classification (Robertson et al., 2020), (Chauhan et al., 2020). Studies have reported that ratoon rice crops have lower height and fewer productive tillers; hence, they typically produce approximately half of the average main rice crop’s yield (Saito et al., 2024), (IRRI, 1988). Under ratooning, rice crops also need less water than the main rice (Oad et al., 2002). The particular canopy architecture and soil moisture condition of ratoon rice compared to the main rice crop could cause significant differences in the crop’s reflectance and radar signal. Because of this, we propose that a synergistic use of both SAR and optical data is likely to outperform optical or SAR-based classification alone. Although many studies have demonstrated the utilisation of remote sensing satellite data to map rice crop areas (Zhao et al., 2021), (Cauba et al., 2025), studies focusing on ratoon rice crop mapping are limited, and most of them extensively used vegetation indices (VIs) from optical data. A prior study has shown the method of mapping ratoon rice using the Normalized Difference Vegetation Index (NDVI) from the 8-daily composite MODIS data, highlighting the phenological difference between ratoon and main rice crop (Li et al., 2022). Since ratoon rice matures faster, image acquisition with a shorter revisit time is required to fully capture the whole growth cycle. The launch of the Sentinel constellations, particularly Sentinel-1 and 2, has opened the possibility of observing the same region at high resolution within a short time interval (of five days). Liu (Liu et al., 2020) mapped ratoon rice in China using high spatio temporal Sentinel-2, showing the relevance of Enhanced Vegetation Index (EVI) and Land Surface Water Index (LSWI) temporal profiles for identifying ratoon rice during crop maturity. Chen (Chen et al., 2023) utilised Landsat and Sentinel-2 data for ratoon rice mapping at 30 m resolution and built a phenology-based ratoon rice vegetation index (PRVI), showing the spectral variability of ratoon rice compared to single and double rice cultivation at harvest time. While these studies showed that phenology-based ratoon rice mapping is feasible, the issue of pervasive cloud cover in rice-growing areas is a challenge. Recently, Li (Li et al., 2024) successfully mapped ratoon rice areas based on multitemporal SAR Sentinel-1 data, revealing the double-peak characteristics of fields with ratoon rice, but their testing only covered irrigated rice fields. In our previous work, we compared the Sentinel-1 backscatter dynamics of main and ratoon rice throughout the whole growing phases in both irrigated and rainfed rice ecosystems. We showed promising discrimination between ratoons and the main rice crop during the peak of the growing period, considering their differences in bio physical characteristics and water regimes for cultivation (Fikriyah et al., 2025). In light of prior study limitations of mapping ratoon rice with only optical data and the potential benefit from data integration, this time, we aim to investigate ratoon rice mapping using both optical and SAR data. Although the benefits of NDVI, LSWI, and EVI have been well documented for rice crop observation, further research is needed to fully understand the use of vegetation indices on ratoon rice mapping. Compared to NDVI, the Soil Adjusted Vegetation Index (SAVI) is also commonly used for crop analysis as it is sensitive to crop biophysical and biochemical parameters while limiting soil brightness influence (Huete, 1988). The Normalized Difference Red Edge (NDRE) (Kross et al., 2015) and Green Normalized Difference Vege tation Index (GNDVI) (Gitelson et al., 1996) have also been used to extract and analyse rice areas in comparison to other land cover types, providing distinct spectral responses among crop canopy closure (open or closed) and background reflectance. However, there is a lack of understanding of the spectral observation and the potential capability of the VIs mentioned above to distinguish ratoon rice from the main rice crops. Particularly, there is notable oversight of the significance of the red-edge band index, which is closely related to nitrogen content, chlorophyll content, and biomass (Kanke et al., 2016). Ratoon rice crops grow from the stubble of the main rice crop and are often left without proper fertiliser management, leading to lower levels of plant nitrogen content and yield. This spectral range, therefore, could be critical to improving the main and ratoon rice classification. Machine learning-based classifiers are often used with remote sensing data for discriminating rice fields from other land covers and mapping rice crop extent (Zhao et al., 2021), (Fernández-Urrutia et al., 2023). Those classifiers include decision tree (DT), support vector machine (SVM), random forest (RF), and convolutional neural network (CNN). DT is one of the most common learning algo rithms for crop mapping; it can formulate a tree-like set of classification rules that are easy to interpret (Liakos et al., 2018). A previous study by (dela Torre et al., 2021), for example, attempted to identify rice extent in the Philippines using a DT classification, resulting in an overall accuracy of 68 % for the dry season but a higher score of 75 % for the wet season. The SVM classifier is an advantageous approach, mainly for non-linear data and when only small samples are available (Kang et al., 2018). As a result, SVM has been effectively applied to rice and non-rice classification using satellite images from various sensors, achieving >80 % overall accuracy (Son et al., 2018), (Xu et al., 2016). In addition, the RF classifier has emerged as an exceptionally well-performing classifier for rice crop mapping. Liu (Liu et al., 2022) developed a rice field identification framework using an RF classifier with Sentinel-2 data, producing 91 % accuracy, while Fiorillo (Fiorillo et al., 2020) successfully mapped rice areas with RF with high accuracy (87 %). Despite the broad implementation of those three classifiers for rice mapping, no study has confirmed their potential application for ratoon rice mapping. The few previously mentioned studies on ratoon rice mapping have identified some promising opportunities to detect and V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 2 distinguish it from the main crop. Still, they have not fully exploited the available and commonly used SAR and optical data sources for rice mapping and monitoring. Nor have they evaluated the skill of different classification methods. With recent advances in Earth observation satellites, further opportunities to map rice ratooning practices have arisen. Therefore, we used Sentinel-1 and Sentinel-2 data in this study and assessed which combination of polarisations, spectral bands, and VIs was suitable for mapping and discrimi nating ratoon rice from the main rice crop. Ratoon discrimination from the main rice crop was performed using DT, SVM, and RF algorithms. The mapping was tested in the Philippines since it is one of the major rice-growing countries where increasing rice pro duction is needed, and the practice of rice ratooning has been reported. Given the rising interest in rice ratooning adaptation, this study contributes to improving our understanding of how to exploit multisource remote sensing data for ratoon rice mapping, especially in regions and seasons where cloud cover is pervasive. 2. Materials and methods 2.1. Study site description The Philippines produced 20 million metric tons of rice in 2023, making it among the top ten largest rice cultivation countries worldwide (PhilRice, 2024a). This study was conducted in four rice-producing provinces (Iloilo, Agusan del Sur, Leyte, and Pan gasinan) (Fig. 1), where rice ratooning was reported. The average monthly rainfall ranges from 115 to 308 mm, and the average temperature ranges between 25.07 and 27.51 ◦C (WorldBank, 2021). Rice is planted in the wet and dry seasons, and rice ratooning is practised in both seasons. However, it is more frequently adopted in the wet season because of the longer duration of water availability. Fig. 2 presents the crop calendar for the main and ratoon rice cultivation based on our survey in the wet and dry seasons (2018–2019). Fig. 1. The four studied provinces: Pangasinan, Leyte, Agusan Del Sur, and Iloilo. The location of rice fields where ratooning was practiced is shown in red (n = 58), while those without are shown in black (n= 70). Rice field extents were based on the Philippine Rice Information System (PRISM) data (PhilRice, 2024b). V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 3 2.2. Data 2.2.1. Ground reference collection and farmer survey Data from ground surveys and farmers’ interviews were obtained from the International Rice Research Institute (IRRI). The field data collection was conducted during the wet and dry seasons of 2018–2019 under the Pest Risk Identification and Management (PRIME) project. Field surveys and farmer interviews contained data on field boundaries, crop calendar, crop type, cropping pattern (number of crops), variety, crop establishment methods (transplanting or direct seeding), maturity duration, fertiliser use, water source (irrigated or rainfed field), and ratooning. Crop calendar dates were recorded for land preparation, flooding, irrigation (in irrigated fields), and harvesting (Fig. 2). Each of these events was documented to the nearest week and month, according to what farmers could recall, and we used the mid-week date as the specific date for the next step of analysis. Our survey collected data from 30 fields per class (for fields with and without ratooning). Field sizes are 1.5 ha on average, ranging between 0.5 and 3.5 ha. Within those fields, sample points were randomly located with a minimum distance of 40 m (twice the satellite image pixel size) to prevent reflectance and backscatter values from being extracted from neighbouring pixels. In total, we had 58 reference data points from rice fields where ratooning was practised and a further 70 points for rice fields where there was no ratooning. The period of the reproductive to ripening phases of the main rice and the ratoon rice crops was estimated and used as a basis for discriminating between the two. For the growth of the main rice crop, we used a growth duration that was composed of the vegetative phase (30–65 days, depending on the rice variety), plus the reproductive phase (35 days), plus the ripening phase (30 days) (IRRI, 2007). The estimated period of each phase was counted forward from the reported rice establishment time. Calculation adjustments for the vegetative period were made for transplanted rice using the seedling’s age information. For ratoon rice crops, the calculation started from the main rice harvest period. According to farmers surveyed, the growth duration of ratoon rice ranged between 30 and 80 days. The growing period of the ratoon rice crop is reduced to almost half that of the regular rice crop (Dong et al., 2017). This is due to Fig. 2. (A) Image footprints for Sentinel-1 and 2. (B) Rice crop calendar based on survey data in the four provinces and Sentinel-1 and 2 image dates. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 4 the shorter vegetative phase, which can begin as soon as the main crop is harvested, due to the residual carbohydrates found in the crop’s roots and stubble (IRRIa). According to (Oad et al., 2002), the first tiller emerges one to ten days after harvest, and for ratoon rice that is grown for longer than sixty days, the growth period is comparable to that of the main crop (Jiang et al., 2021). Based on that, we assume the ratoon rice growth period is as follows. For the short − 30-day- growth period, the estimation duration for each phase is five days (vegetative), 15 days (reproductive), and eight days (ripening). For the medium duration (i.e., 60 days), we assumed the vegetative phase lasts for ten days, the reproductive phase for 25 days, and the ripening phase for 20 days. Moreover, for the long ratoon growth period (i.e., 80 days), the assumption is 15 days for vegetative, 30 days for the reproductive, and 30 days for the ripening phase. Based on our survey, most ratoon rice samples have a medium duration (53 %), followed by a short growth duration (40 %). Only 7 % of the samples had a long ratoon period. 2.2.2. Sentinel-1 and 2 time-series images Multi-temporal Sentinel-1A and 1B Ground Range Detected (GRD) data in Interferometric Wide (IW) swath mode were obtained from the Google Earth Engine (GEE) platform. The majority of the data was in the descending orbit. Due to separate pre-processing requirements, data in the ascending orbit was only collected when the former type was not available. This product was chosen based on its applicability to rice crop studies (Fikriyah et al., 2019), (Rudiyanto et al., 2019). The data has undergone several pre-processing steps: removal of thermal noise, radiometric, and terrain correction. An additional step of speckle filter with a 50m radius smoothing function was also implemented in GEE, based on (Podest et al., 2020). A 10-day median composite was extracted during the peak of the growing time of each sample field (during the estimated period of the reproductive and ripening phases) (Fig. 2). This median composite approach has been shown to be helpful in time series crop analysis to filter noise (Rudiyanto et al., 2019). Data was acquired for dual polarisations (VV and VH) at a 20m pixel size, and backscatter values were converted to decibels (dB). Sentinel-2A and 2B MSI (Multispectral Instrument) data level-1C were obtained due to the unavailability of Level-2A data in the desired area and period in the GEE data catalogue. Like Sentinel-1 data, A 10-day median composite image was acquired during the peak growth periods. The same median composite is applied for both Sentinel-1 and Sentinel-2 for the computational efficiency in the image collection process and to maintain the same temporality in both types of satellite data. Since there is no atmospheric correction model fully facilitated in the GEE environment at present, we applied an image correction using the Dark Object Subtraction (DOS) technique in GEE (Zuspan, 2020) to improve the image quality, which performs comparably to other correction methods (Lantzanakis et al., 2017). All images were also screened and masked for opaque and cirrus clouds using the ‘QA60’ band containing clouds and cirrus polygons (with a 50 % cloud cover probability threshold). Bands B1 (aerosol), B9 (water vapour), and B10 (cirrus) were irrelevant to the study and excluded. All remaining bands were resampled (where needed) to 20m resolution to facilitate the data combination between the visible-NIR band (10 m), the red edge (20m), the SWIR band (20m), and the Sentinel-1 data. Fig. 3. Methodological flowchart for mapping the ratoon rice distribution in this study. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 5 2.3. Methods We first estimated each sample’s reproductive to ripening phases, then extracted the reflectance and backscatter values of images in the corresponding period. This step was to create the classification model. Once the best-performing model was identified, we implemented the model on the image when ratooning was mainly practiced based on our survey data to predict the ratoon rice dis tribution across the entire study area. The current approach is beneficial for computational efficiency. The testing on the backscatter difference between main and ratoon rice using a high-temporal-density Sentinel-1 throughout the whole growing period has been conducted (Fikriyah et al., 2025), showing that the reproductive to ripening phases become the critical time window for classification between main and ratoon rice. A similar approach has also been demonstrated in previous studies. For instance, multitemporal Landsat data were used by Singla et al. (2018) for ratoon and non-ratooning discrimination in sugarcane, and this study demonstrated that the discrimination was most effective during the tillering and peak growth phases. Another study (Mahlayeye et al., 2024) shows how well Sentinel-2 data can detect intercropping agricultural practices, particularly during the vegetative and flowering-yield formation crop phase. Fig. 3 shows the general workflow for mapping ratoon rice using Sentinel-1 and -2 data. The following sections explain the details of the method.Table. 1 2.3.1. Features extraction from Sentinel-1, -2, and vegetation indices Three features, VV, VH, and VH/VV, were generated from Sentinel-1 data, while ten spectral features were extracted from Sentinel- 2 data (Bands 2, 3, 4, 5, 6, 7, 8, 8A, 11, and 12). We also considered six commonly used VIs for rice crop studies using the all-optical spectral range from visible, red-edge, near-infrared, and shortwave infrared bands (Table 2). These features were statistically tested to identify significant differences and correlations. 2.3.2. Statistical test Feature selection by means of a statistical test was conducted to select meaningful features, filter out redundant data and ultimately optimise classifier performance (Georganos et al., 2018). We carried out two stages of feature selection using several statistical tests. First, we compared the extracted ratoon rice spectral and backscatter responses to those of the main rice crop using the Mann-Whitney U test. This nonparametric test was used to investigate the mean differences between two groups-ratoon and main rice crops-since the normality assumption was not met in our data based on a Shapiro-Wilk normality test. Only features with significant differences in separating the ratoon and main rice crops were included in the next step. Second, we used Spearman’s rank correlation coefficient to assess the potential multicollinearity among features (Zhang et al., 2024). We omitted highly correlated features (r > 0.8). Both tests used a 95 % confidence interval. 2.3.3. Main and ratoon rice crop classification Three machine learning-based classifiers, DT, SVM, and RF, were used for the main and ratoon rice classification. All classification procedures were implemented in R Studio 2024.04.02 (R 4.4.1). A 10-fold cross-validation was used for the model training and validation. This is to allow the use of all sample observations for training and validation during the modelling iteration, leading to a more reliable model (Ramezan et al., 2019). The DT, SVM, and RF models were generated using the caret package version 6.0–94 (Kuhnaut et al., 2023). We used the default setting for model training in all classifiers since the classifiers will automatically select the most optimal parameter tuning that maximises accuracy. A tuning parameter for tree splitting in DT, the complexity parameter (cp), was set to the default. Two tuning parameters in RF classification, the number of trees (ntree) and splitting nodes (mtry), were also set to default. The SVM modelling used the linear kernel type, which is preferred for its simplicity in the hyperparameter tuning and im proves computational efficiency. This kernel type is also optimal for binary classification tasks (Fang et al., 2020). The ‘cost’ control parameter was one by default. For the mapping purpose, we used the rice extent (PhilRiceb) to mask out non-rice fields. We also computed the relative variable importance for both models to determine highly relevant features for each classification. Classification modelling was made for five Table 1 List of vegetation indices for spectral analysis of main and ratoon rice crops. No Index Wavelength Formula Related to 1 Normalized Difference Vegetation Index (NDVI) ( Rouse et al., 1974) Visible - NIR NDVI = NIR − Red NIR + Red LAI and biomass (Cohen et al., 2003) 2 Green Normalized Difference Vegetation Index (GNDVI) (Gitelson et al., 1996) Visible - NIR GNDVI = NIR − Green NIR + Green Rice canopy coverage (Liu et al., 2024) 3 Soil Adjusted Vegetation Index (SAVI) (Huete, 1988) Visible - NIR SAVI = NIR − Red NIR + Red + L (1 + L) Reduce soil background reflectance ( Kimura et al., 2004) 4 Enhanced Vegetation Index (EVI) (Huete et al., 2002) Visible - NIR EVI = 2.5 NIR − Red NIR + 6(Red) − 7.5(Blue) + 1 Rice crop biomass (Mansaray et al., 2020) 5 Normalized Difference Red Edge (NDRE) (Kross et al., 2015) NIR - RedEdge NDRE = NIR − RE NIR + RE Rice cop biomass (Kanke et al., 2016) 6 Land Surface Water Index (LSWI) (Xiao et al., 2005) NIR- SWIR LSWI = NIR − SWIR NIR + SWIR Soil wetness and crop water content ( Xiang et al., 2020) V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 6 sets of input features (Table 3). 2.3.4. Accuracy assessment In the 10-fold cross-validation, the available datasets are divided into ten sections at random, each containing around 10 % of the samples from each ratoon and main rice class. For model training, 90 % of the data was used as reference, and the remaining 10 % of the data was used as a validation set. Results were then aggregated and reported after ten iterations. Four widely used metrics for accuracy assessment were computed: overall accuracy (OA), user’s accuracy (UA), producer’s accuracy (PA), and F1-score. OA, PA, and UA are statistics that reflect the consistency between the predicted and reference data. An additional F1-score was calculated to measure the average precision (ratio of true positive to positive prediction) and recall or sensitivity (ratio of true positive to the actual object) of the model (Maxwell et al., 2021). Accuracy assessment was done for the five input data sets and three machine learning models. 3. Results 3.1. Main and ratoon rice spectral reflectance and backscatter signatures The spectral signatures of main and ratoon rice crops in the optical spectrum range from blue to the SWIR band of Sentinel-2 are presented in Fig. 4. The reflectance of main and ratoon rice crops showed similarity in the blue (490 nm) to RE1 (705 nm). Distinct spectral variations were observed from the RE2 (740 nm) to NIR2 (865 nm) range, although ratoon rice had consistently lower reflectance than main rice crops. No reflectance difference was apparent between the main and ratoon rice within the two SWIR bands (1610–2190 nm). From the extracted backscatter in Sentinel-1 data, ratoon rice’s backscatter distribution is skewed to the right in both VV and VH (Fig. 5). As also observed, the distribution patterns of backscatter response in the ratio of VH/VV exhibited an overlapping data range for main rice and ratoon rice. The Mann-Whitney U later identified that the backscatter values of these groups were insignificant. The backscatter and reflectance of the main and ratoon rice crops during the reproductive to ripening phase are presented in Fig. 6. Ratoon rice crops had higher backscatter than the main rice crops in both polarisations, VV and VH. On the contrary, lower reflectance of ratoon rice crops was shown, especially prominent in the following bands: RE2, RE3, NIR, NIR2, and all of the VIs (GNDVI, EVI, NDVI, NDRE, LSWI, and SAVI). The reflectance of the main and the ratoon rice crops was similar within the spectral regions of blue, green, red, and RE1. Table 2 Features included from Sentinel-1 and 2 data for the five combinations. Combination Description Features 1 S1 (bands + ratio) VV, VH, VH/VV 2 S2 bands blue, green, red, RE1, RE2, RE3, NIR, NIR2, SWIR1, SWIR2 3 S2 (bands + VIs) blue, green, red, RE1, RE2, RE3, NIR, NIR2, SWIR1, SWIR2, NDVI, GNDVI, SAVI, EVI, NDRE, LSWI 4 S1+S2 (bands + ratio + VIs) VV, VH, VH/VV, blue, green, red, RE1, RE2, RE3, NIR, NIR2, SWIR1, SWIR2, NDVI, GNDVI, SAVI, EVI, NDRE, LSWI 5 Selected features Selected features based on significant differences and correlation test Table 3 Summary of p-values from the Mann-Whitney U test between main rice (n =70) and ratoon rice crops (n = 58) in the reproductive phase. *Significant, p < 0.05; **Significant, p < 0.01; ***Significant, p < 0.001. Data Feature p-value Data Feature p-value Sentinel-1 VV 1.2e-08*** Sentinel-2 Blue 0.14 ​ VH 7.5e-08*** ​ Green 0.57 ​ VH/VV 0.056 ​ Red 0.21 ​ ​ ​ ​ RE1 0.60 ​ ​ ​ ​ RE2 9.2e-06*** ​ ​ ​ ​ RE3 2.0e-07*** ​ ​ ​ ​ NIR 5.6e-08*** ​ ​ ​ ​ NIR2 2.4e-07*** ​ ​ ​ ​ SWIR1 0.44 ​ ​ ​ ​ SWIR2 0.28 ​ ​ ​ ​ NDVI 7.9e-05*** ​ ​ ​ ​ GNDVI 0.00035*** ​ ​ ​ ​ SAVI 2.5e-08*** ​ ​ ​ ​ EVI 3.3e-08*** ​ ​ ​ ​ NDRE 1.8e-06*** ​ ​ ​ ​ LSWI 7.7e-09*** V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 7 Fig. 4. Mean reflectance signatures for main and ratoon rice crops during the reproductive to ripening phase. The number of wavelengths represents the central wavelength of each Sentinel-2 band. n main rice = 70, and n ratoon rice = 58. Fig. 5. Scatter plot (A) and density distribution of main and ratoon rice’s backscatter during the reproductive to ripening phase in VV, VH, and VH/ VV (B, C, and D, respectively). n main rice = 70, and n ratoon rice = 58. Fig. 6. Backscatter (Sentinel-1) and reflectance (Sentinel-2) values of main (yellow) and ratoon rice crops (blue) during the reproductive to ripening growth phases. n main rice = 70, and n ratoon rice = 58. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 8 3.2. Feature selection for main and ratoon rice classification Probability values (p-value) from the Mann-Whitney U test conducted for all the features from Sentinel-1 and 2 are presented in Table 4. Backscatter differences between main and ratoon rice were significant in VV and VH. Differences in spectral reflectance were also evident in bands of RE2, RE3, NIR, and NIR2, except for the visible spectral bands (blue, green, red), RE1, and shortwave bands (SWIR1 and SWIR2). All of the VIs showed significant differences in reflectance between the main and the ratoon rice crops. Possible multicollinearity among the 12 features showing significant differences (Table 4) was assessed using Spearman’s corre lation coefficient (Fig. 7). The correlation coefficient revealed that the red-edge bands and several VIs were highly correlated. This information was used to select the optimum features for classification. Considering the strong correlation between different features, only five features were chosen for the classification in the DT, SVM, and RF models in combination 5: VV, VH, NIR2, NDRE, and LSWI. 3.3. DT, SVM, and RF classification performance for discriminating ratoon from the main rice crops We conducted a comparative classification test using the five input data and verified their accuracies (Table 4). In DT, a moderate accuracy was obtained from input using bands of Sentinel-1 only (OA = 70 %). The accuracy was low when using Sentinel-2 bands data (OA = 64 %). Adding the VIs to the Sentinel-2 band gained a better overall accuracy of 73 %. When all Sentinel-1 and 2 data were used as features, the accuracy increased to 78 %. The model obtained the highest accuracy using the selected features (OA = 81 %). In the SVM classifier, the model with Sentinel-1 data produced the lowest overall accuracy (76 %). Using VIs, the classification performs slightly better than those in the model with Sentinel-2 bands only (UA of ratoon 79 % and 78 %, respectively). The highest accuracy was derived from selected Sentinel-1 and 2 features as input (87 %). In RF, the lowest accuracy was obtained using Sentinel-1 data only (OA = 67 %). The accuracies progressively increased with the model using selected features registering the highest accuracy (OA = 92 %). All classifiers and models produced an overall accuracy of at least 70 %, except DT using Sentinel-2 data and RF using Sentinel-1 data only. In our observations, higher accuracy was achieved when using Sentinel-1 and -2 features compared with individual Sentinel data as classification input. The same pattern was found in the better accuracy for the selected feature combination compared to the accuracies when all features were included in DT, SVM, and RF. 3.4. Feature importance for main and ratoon rice classification DT, SVM, and RF classifiers facilitated the extraction of feature importance rank among the total 19 analysed features from Sentinel-1 and -2 data, aiding in identifying the most influential variables to discriminate main and ratoon rice crops (Fig. 8). From Sentinel-1 features, VV and VH had high classification contributions determined by the three classifiers. In Sentinel-2, the rank revealed that visible bands (blue, green, and red) and the SWIR bands were of low importance. On the other hand, NIR, NIR2, red-edge (RE3), and VIs (SAVI, EVI, and NDRE) were significant contributor features in the main rice and ratoon classification models. We also observed that the results from the substantial difference test (Table 4) correspond well with those from the feature importance rank, where less classification contribution was gained from the visible and SWIR regions. 3.5. Ratoon rice area distribution Fig. 9 presents an example of ratoon rice crop distribution predicted by the DT, SVM, and RF classifications. The best-performing model with the selected features (VV, VH, NIR2, NDRE, and LSWI) was applied. The dark brown polygons highlight the ratoon rice fields. The results showed that the prediction by DT overestimated the ratoon rice class. This could be because the model only relied on one feature for classification, the LSWI index (see the supplementary material- Figure S1). Based on the models’ performance comparison in Table 4, we implemented an RF classifier using selected features to map the Table 4 Summary of classification accuracies in the decision tree (DT), support vector machine (SVM), and random forest (RF) for five combinations. MR: main rice, RR: ratoon rice. UA: user’s accuracy, PA: producer’s accuracy, OA: overall accuracy, F1: F1-score. main rice (n =70) and ratoon rice crops (n = 58). Combination Input Class DT SVM RF UA PA OA F1 UA PA OA F1 UA PA OA F1 1 S1 all (bands + ratio) MR 0.69 0.81 0.70 0.62 0.77 0.80 0.76 0.73 0.68 0.76 0.67 0.61 RR 0.71 0.55 0.75 0.71 0.66 0.57 2 S2 bands MR 0.65 0.62 0.64 0.65 0.79 0.77 0.78 0.78 0.75 0.77 0.76 0.75 ​ RR 0.63 0.67 0.78 0.79 0.76 0.74 3 S2 bands and VIs MR 0.70 0.79 0.73 0.71 0.78 0.79 0.78 0.78 0.77 0.77 0.77 0.77 RR 0.76 0.67 0.79 0.77 0.77 0.77 4 S1+ S2 (bands + ratio + VIs) MR 0.78 0.79 0.78 0.78 0.78 0.82 0.79 0.79 0.91 0.82 0.87 0.88 ​ RR 0.79 0.77 0.81 0.77 0.84 0.92 5 Selected features MR 0.84 0.80 0.81 0.80 0.87 0.89 0.87 0.86 0.94 0.91 0.92 0.91 RR 0.78 0.83 0.87 0.85 0.90 0.94 V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 9 distribution of ratoon rice in all four provinces. Ratoon rice was predicted for the wet season since ratooning was highly reported during that season compared to the dry season. Brown polygons show the location of ratoon rice crops (Fig. 10A). We also estimated the ratoon rice areas relative to the rice planting areas in each province (Fig. 10B). The lowest ratoon rice area percentage was in Pangasinan (2.5 %), while the highest proportion areas were estimated in the Province of Leyte (7.1 %). All provinces’ total ratoon rice area was 223 km2, or about 4.4 % of the mapped wet season rice area. 4. Discussion 4.1. Differences between main and ratoon rice crops in spectral and backscatter response Accurate ratoon rice detection and mapping require a thorough understanding of the spectral behaviour of ratoon rice crops. For the first time, this study demonstrates the potential use of optical Sentinel-2 and SAR Sentinel-1 data in identifying ratoon rice crops. Based on the results, our study underscores the importance of co- and cross-SAR polarisations, red-edge, NIR band, and VIs in optimal Fig. 7. Spearman’s correlation coefficient between each feature pair for feature selection in the classification and mapping model. Fig. 8. Feature importance rank for classifying main and ratoon rice crops based on DT, SVM, and RF models. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 10 main and ratoon rice crop classification and mapping. Our analysis showed a significantly higher backscatter (in VV and VH polarisation) of ratoon rice than the main rice crops (Fig. 6). The higher VV backscatter could be linked to the double-bounce scatters between the ratoon rice canopy and the underlying soil because the ratoon rice crop is not as tall as the main rice crop and has fewer productive tillers (Oad et al., 2002), (IRRIa). The variation in crop spacing is also more visible in ratoon rice, resulting in more double-bounce scattering between crop and soil. VH backscatter is highly correlated with rice biomass (Inoue et al., 2014). However, we observed a higher VH backscatter in the ratoon rice than in the main rice crop. This disagreement could be attributed to the lower plant density in ratoon rice, which creates a more scattering contribution from the ground and ground-stem layer to the total backscatter (Zhang et al., 2014). A lower spectral reflectance of ratoon rice was observed across the red-edge (RE740 and RE783) and NIR spectrum (NIR842 and NIR865). This difference is probably because of the lower canopy percentage and biomass of the ratoon rice (IRRI, 1988). Ratoon rice also has fewer and smaller leaves per stem (Huang et al., 2023), resulting in a lower leaf area index (LAI) than the main crop. This is supported by prior studies that explained the strong correlation between reflectance at the NIR band and rice crop leaf area index (Kimura et al., 2004). Unlike the single spectral bands, distinguishable differences were observed between main and ratoon rice crops in all VIs (GNDVI, NDVI, EVI, LSWI, NDRE, and SAVI) (Fig. 6). The different ratoon rice canopy characteristics from the main rice could explain this finding. NDVI is important for assessing the cover percentage and LAI (Cohen et al., 2003), while EVI and NDRE are good indicators of rice biomass (Kanke et al., 2016), (Mansaray et al., 2020). SAVI was mainly designed to minimise the soil background interference with low leaf area crops’ reflectance (such as rice crops), for which other indices are difficult to interpret (Kimura et al., 2004). Incorporating the shortwave infrared reflectance, LSWI is known for its sensitivity to water content and has been used as an indicator of soil moisture to identify rice fields (Xiang et al., 2020). As such, ratoon rice captured a lower LSWI value since water flooding is not required for ratooning compared to the main rice cultivation. 4.2. Feature importance for ratoon and main rice crop classification Based on the findings, VV, VH, NIR, NIR2, RE3, and VIs were determined as valuable features for ratoon rice classification. Unlike the single-band observation approach, VIs are acknowledged for their ability to strengthen the sensitivity of spectral bands to the crop’s biophysical and biological characteristics. This is primarily beneficial for the identification of crops with relatively low LAI, like rice and even ratoon rice (Xiao et al., 2002). The Spearman’s rank correlation showed strong correlations among VIs between NIR and red-edge band (RE2, RE3) (Fig. 7), indicating the redundancy of spectral information from them. The NDRE and LSWI indices had less correlation with other features. Therefore, this result corresponds to the potential use of those indices in effectively characterising ratoon rice compared to other indices. 4.3. Classification of main and ratoon rice crops Main and ratoon rice classification was done using DT, SVM, and RF algorithms. The best results were achieved using selected features, reaching 81 % (DT), 87 % (SVM), and 92 % (RF). The validation result indicates that the RF classification model performed better than DT and SVM among all combinations, with OA = 92 % and F1-score = 0.91 (Table 4). The classification results also aligned with the surveyed ratoon rice distribution (Fig. 9). Choosing features is an essential step in classification, as a large number of pre dictors may be prone to overfitting risk and capturing random variation of data (Georganos et al., 2018). Although it does not always give the highest classification accuracy, a lower amount of data increases the computation process efficiency (Pratama et al., 2024). A poor result was produced by the DT model (Fig. 9) based on the default tuning parameters. As shown in Fig. 8, some features did not contribute to the classification, resulting in small tree sizes. It seems that a single decision tree might not capture complex relationships present in the data, requiring an ensemble tree method for better performance. A customisation of the parameters can also be Fig. 9. Distribution of ratoon rice in Leyte province based on the a) DT, b) SVM, and c) RF models with selected features. Prediction used the median composite images between October and November 2018 (reproductive to ripening phases). V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 11 implemented to optimise the result (Qian et al., 2015), (Seyedzadeh et al., 2019), although it can be time-consuming and computa tionally intensive due to its iterative process. The assessment results show that the model with individual optical bands yielded 64–78 % accuracy, while the model using backscatter bands resulted in 67–76 % accuracy. In comparison, the models’ performances with backscatter and reflectance infor mation ranged between 78 % and 87 %. Accuracy scores using selected features, which include backscatter (VV, VH), NIR2 reflectance, and VIs (NDRE, LSWI), ranged between 81 % and 92 %. These findings indicate that integrating Sentinel-1 and Sentinel-2 enhances classification accuracy for ratoon rice crop mapping. Similar results were also obtained from the case of cropland and ricefield clas sification in prior studies (Fiorillo et al., 2020), (Fikriyah et al., 2023), wherein the model accuracy improved by adding SAR to optical data. The VIs also improved the classifier’s ability to detect main and ratoon rice based on the feature performance rank of the three models. This analysis, therefore, provides insight into the valuable feature combinations for rice and ratoon rice classification. 4.4. Ratoon rice area estimation and future mapping improvement The best-performing model, which used selected features in RF classification, estimated the total ratoon rice area to be 223 km2 (4.4 % of the rice field in the four provinces included in this study). The highest ratoon rice percentage area was in the Province of Leyte, covering 55 km2 of the 774 km2 rice areas (7.1 %). On the other hand, the lowest percentage was in the Province of Pangasinan, Fig. 10. (A) Ratoon rice distribution in rice fields based on Sentinel-1 and -2 images from October to November 2018 (peak period of wet season). The selected features in RF were used for mapping. (B) Estimates of ratoon rice area in the four provinces from the RF model. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 12 which had only 2.5 % of ratoon rice. In comparison to other countries, the prediction of the ratooning practice in this study area is lower than what has been estimated in China, with studies reporting 5 % (Li et al., 2022) and 10 % (Liu et al., 2020). This might be due to the higher adaptation of ratooning practice in that region, as confirmed by prior studies on mapping ratoon rice crops (Li et al., 2022), (Liu et al., 2020), (Chen et al., 2023), (Li et al., 2024). The approach for mapping ratoon rice crops with optical and SAR data during the reproductive to ripening phases showed good results. Nevertheless, there are several challenges and limitations. Although we see spectral differences between mature ratoon rice and mature main rice crops, misclassification might still occur between ratooned fields and those that remain stubble until further land preparation. Selecting the correct periods of the year to accurately map ratooning practice for the dry and wet seasons is still chal lenging and requires accurate and timely crop calendar information. It is also noteworthy that the estimation of growth stages in this study relied on prior cropping calendar information from the farmers. This approach can present a challenge when such information is lacking or when implemented in a study area with a large variation in crop calendars. A multitemporal analysis applying the maximum values of vegetation index, for example, maximum NDVI, would be a potential alternative to identify the growth phases’ time, like heading-stage information, in such cases (dela Torre et al., 2021), (Panuju et al., 2021). The pixel-based classification approach could also introduce errors due to mixed pixels, especially for ratoon rice in small fields that are potentially undetected. Additional research on field-based classification methods is suggested for better ratoon rice area estimation, although this requires larger datasets than that used here. In future work, a detailed study assessing the biophysical characteristics of ratoon rice will help monitor the growth of ratoon rice. We also recommend further testing in both the wet and dry seasons to assess the robustness of any ratoon rice classifier. 5. Conclusions This study is the first attempt to systematically assess the difference in backscatter and reflectance between the main and ratoon rice crops using optical and radar spectral bands alone and in combination. We also demonstrated the DT, SVM, and RF classifiers implemented for main and ratoon rice classification with an optimal overall accuracy of 81 %, 87 %, and 92 %, respectively. The observation was made using SAR Sentinel-1 and optical Sentinel-2 data during the rice peak growing season. Our findings support the conclusion that ratoon rice had a significantly higher backscatter than the main rice crops in VV and VH. Inspection of ratoon and main rice crops in red-edge, near-infrared spectrum, and vegetation indices (NDVI, GNDVI, SAVI, EVI, NDRE, and LSWI) showed distinct lower reflectance of ratoon rice. This study calculated the ratoon rice area of 223 km2, accounting for about 4 % of the total rice planting areas in the four provinces studied. Optical and SAR data can enhance the accuracy of ratoon and main rice crop classification. Further research and development in the modelling algorithm are expected to improve the accuracy of the mapping. As rice ratooning practice becomes more widely adopted for agricultural input efficiency purposes, our study contributes to the potential use of remote sensing technologies in estimating its adoption and improvement in estimating rice production. CRediT authorship contribution statement Vidya Nahdhiyatul Fikriyah: Writing – review & editing, Writing – original draft, Validation, Methodology, Formal analysis, Conceptualization. Roshanak Darvishzadeh: Writing – review & editing, Supervision, Methodology, Conceptualization. Alice Laborte: Writing – review & editing, Resources, Data curation. Andrew Nelson: Writing – review & editing, Supervision, Resources, Conceptualization. Ethical statements Authors declare that the manuscript is an original work and follows the ethical standards of research. Funding V.N.F. was supported by the Indonesia Endowment Fund for Education Agency (LPDP). Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments We thank all farmers who participated in the farm survey and IRRI for providing the survey data. We acknowledge the Philippine Department of Agriculture National Rice Program through the Bureau of Agriculture Research for funding the PRIME Project and the collection of field survey data used in this study. We acknowledge the Philippine Rice Research Institute for the rice extent map from PRISM. V N. Fikriyah is also grateful for the support from the Indonesia Endowment Fund for Education Agency (LPDP) in conducting this research. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 13 Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.rsase.2025.101592. Data availability The authors do not have permission to share data. References Asaba, J., Mukwaya, P.I., Lwasa, S., Bamutaze, Y., Omolo, F., 2024. Spatial differentiation of the land and nutrient footprints for kampala: implications for urban food sustainability. Forum Geografi 38 (2). https://doi.org/10.23917/forgeo.v38i2.2421. Art. no. 2. Bandumula, N., 2018. Rice production in Asia: key to global food security. Proc. Natl. Acad. Sci., India, Sect. B Biol. Sci. 88 (4), 1323–1328. https://doi.org/10.1007/ s40011-017-0867-7. Cauba, A.G., Darvishzadeh, R., Schlund, M., Nelson, A., Laborte, A., 2025. Estimation of transplanting and harvest dates of rice crops in the Philippines using Sentinel- 1 data. Remote Sens. Appl.: Society and Environment 37, 101435. https://doi.org/10.1016/j.rsase.2024.101435. Chang, L., Chen, Y.-T., Wang, J.-H., Chang, Y.-L., 2021. Rice-field mapping with sentinel-1A SAR time-series data. Remote Sens. 13 (1). https://doi.org/10.3390/ rs13010103. Art. no. 1. Chauhan, S., Darvishzadeh, R., Boschetti, M., Nelson, A., 2020. Discriminant analysis for lodging severity classification in wheat using RADARSAT-2 and Sentinel-1 data. ISPRS J. Photogrammetry Remote Sens. 164, 138–151. https://doi.org/10.1016/j.isprsjprs.2020.04.012. Chen, Y., et al., 2023. A phenology-based vegetation index for improving ratoon rice mapping using harmonized Landsat and Sentinel-2 data. J. Integr. Agric. https:// doi.org/10.1016/j.jia.2023.05.035. Cohen, W.B., Maiersperger, T.K., Gower, S.T., Turner, D.P., 2003. An improved strategy for regression of biophysical variables and Landsat ETM+ data. Rem. Sens. Environ. 84 (4), 561–571. https://doi.org/10.1016/S0034-4257(02)00173-6. dela Torre, D.M.G., Gao, J., Macinnis-Ng, C., Shi, Y., 2021. Phenology-based delineation of irrigated and rain-fed paddy fields with Sentinel-2 imagery in Google Earth Engine. Geo-Spatial Inf. Sci. 24 (4), 695–710. https://doi.org/10.1080/10095020.2021.1984183. Dong, H., et al., 2017. The growth and yield of a wet-seeded rice-ratoon rice system in central China. Field Crops Res. 208, 55–59. https://doi.org/10.1016/j. fcr.2017.04.003. Fang, P., et al., 2020. The classification performance and mechanism of machine learning algorithms in winter wheat mapping using Sentinel-2 10 m resolution imagery. Appl. Sci. 10 (15). https://doi.org/10.3390/app10155075. Art. no. 15. Fernández-Urrutia, M., Arbelo, M., Gil, A., 2023. Identification of paddy croplands and its stages using remote sensors: a systematic review. Sensors 23 (15). https:// doi.org/10.3390/s23156932. Art. no. 15. Fikriyah, V.N., Darvishzadeh, R., Laborte, A., Khan, N.I., Nelson, A., 2019. Discriminating transplanted and direct seeded rice using Sentinel-1 intensity data. Int. J. Appl. Earth Obs. Geoinf. 76, 143–153. https://doi.org/10.1016/j.jag.2018.11.007. Fikriyah, V.N., Anggani, N.L., Kiat, U.E.I., Khikmah, F., Arroyan, W.A., Rizki, M.F., 2023. “Integrated Use of Optical and Radar Data for Cropland Mapping over the Mountain Slope Area in Boyolali, Indonesia,” GT 18 (1), 108–122. https://doi.org/10.21163/GT_2023.181.08, 2023. Fikriyah, V.N., Darvishzadeh, R., Laborte, A., Rathore, J., Nelson, A., 2025. Temporal backscatter characterisation of ratoon rice crops based on Sentinel-1 intensity data. GIScience Remote Sens. 62 (1), 2455081. https://doi.org/10.1080/15481603.2025.2455081. Fiorillo, E., Di Giuseppe, E., Fontanelli, G., Maselli, F., 2020. Lowland rice mapping in sédhiou region (Senegal) using Sentinel 1 and Sentinel 2 data and random forest. Remote Sens. 12 (20), 3403. https://doi.org/10.3390/rs12203403. Georganos, S., et al., 2018. Less is more: optimizing classification performance through feature selection in a very-high-resolution remote sensing object-based urban application. GIScience Remote Sens. 55 (2), 221–242. https://doi.org/10.1080/15481603.2017.1408892. Gitelson, A.A., Kaufman, Y.J., Merzlyak, M.N., 1996. Use of a green channel in remote sensing of global vegetation from EOS-MODIS. Rem. Sens. Environ. 58 (3), 289–298. https://doi.org/10.1016/S0034-4257(96)00072-7. Huang, X., et al., 2023. Comparison of the source–sink characteristics between main season and ratooning in rice (oryza sativa L.). Agronomy 13 (7). https://doi.org/ 10.3390/agronomy13071731. Art. no. 7. Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Rem. Sens. Environ. 25 (3), 295–309. https://doi.org/10.1016/0034-4257(88)90106-X. Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Rem. Sens. Environ. 83 (1), 195–213. https://doi.org/10.1016/S0034-4257(02)00096-2. Inoue, Y., Sakaiya, E., Wang, C., 2014. Capability of C-band backscattering coefficients from high-resolution satellite SAR sensors to assess biophysical variables in paddy rice. Rem. Sens. Environ. 140, 257–266. https://doi.org/10.1016/j.rse.2013.09.001. IRRI. RICE RATOONING - IRRI books - International Rice Research Institute [Online]. Available: https://www.yumpu.com/en/document/read/43526653/rice- ratooning-irri-books-international-rice-research-institute. (Accessed 30 May 2023). IRRI. 0.2 Growth stages of the rice plant [Online]. Available: http://www.knowledgebank.irri.org/ericeproduction/0.2._Growth_stages_of_the_rice_plant.htm. (Accessed 7 May 2024). Isa, M., Wajdi, M.F., Mabruroh, M., Hayati, S.F.N., Kamarulzaman, N.H., 2021. Sustainability of rice business in flood-prone areas. Environ. Res. Eng. Manag. 77 (4). https://doi.org/10.5755/j01.erem.77.4.28096. Art. no. 4. Jiang, P., et al., 2021. Impact of tillage and crop establishment methods on rice yields in a rice-ratoon rice cropping system in Southwest China. Sci. Rep. 11 (1). https://doi.org/10.1038/s41598-021-98057-x. Art. no. 1. Kang, J., Zhang, H., Yang, H., Zhang, L., 2018. Support vector machine classification of crop lands using sentinel-2 imagery. In: 2018 7th International Conference on Agro-Geoinformatics (Agro-Geoinformatics), pp. 1–6. https://doi.org/10.1109/Agro-Geoinformatics.2018.8476101. Kanke, Y., Tubaña, B., Dalen, M., Harrell, D., 2016. Evaluation of red and red-edge reflectance-based vegetation indices for rice biomass and grain yield prediction models in paddy fields. Precision Agric 17 (5), 507–530. https://doi.org/10.1007/s11119-016-9433-1. Kimura, R., Okada, S., Miura, H., Kamichika, M., 2004. Relationships among the leaf area index, moisture availability, and spectral reflectance in an upland rice field. Agric. Water Manag. 69 (2), 83–100. https://doi.org/10.1016/j.agwat.2004.04.009. Kross, A., McNairn, H., Lapen, D., Sunohara, M., Champagne, C., 2015. Assessment of RapidEye vegetation indices for estimation of leaf area index and biomass in corn and soybean crops. Int. J. Appl. Earth Obs. Geoinf. 34, 235–248. https://doi.org/10.1016/j.jag.2014.08.002. Kuhn, M., aut, et al., 2023. Caret: classification and regression training [Online]. Available: https://cran.r-project.org/web/packages/caret/index.html. (Accessed 7 August 2024). Lantzanakis, G., Mitraka, Z., Chrysoulakis, N., 2017. Comparison of physically and image based atmospheric correction methods for sentinel-2 satellite imagery. In: Karacostas, T., Bais, A., Nastos, P.T. (Eds.), Perspectives on Atmospheric Sciences. Springer International Publishing, Cham, pp. 255–261. https://doi.org/ 10.1007/978-3-319-35095-0_36. V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 14 https://doi.org/10.1016/j.rsase.2025.101592 https://doi.org/10.23917/forgeo.v38i2.2421 https://doi.org/10.1007/s40011-017-0867-7 https://doi.org/10.1007/s40011-017-0867-7 https://doi.org/10.1016/j.rsase.2024.101435 https://doi.org/10.3390/rs13010103 https://doi.org/10.3390/rs13010103 https://doi.org/10.1016/j.isprsjprs.2020.04.012 https://doi.org/10.1016/j.jia.2023.05.035 https://doi.org/10.1016/j.jia.2023.05.035 https://doi.org/10.1016/S0034-4257(02)00173-6 https://doi.org/10.1080/10095020.2021.1984183 https://doi.org/10.1016/j.fcr.2017.04.003 https://doi.org/10.1016/j.fcr.2017.04.003 https://doi.org/10.3390/app10155075 https://doi.org/10.3390/s23156932 https://doi.org/10.3390/s23156932 https://doi.org/10.1016/j.jag.2018.11.007 https://doi.org/10.21163/GT_2023.181.08 https://doi.org/10.1080/15481603.2025.2455081 https://doi.org/10.3390/rs12203403 https://doi.org/10.1080/15481603.2017.1408892 https://doi.org/10.1016/S0034-4257(96)00072-7 https://doi.org/10.3390/agronomy13071731 https://doi.org/10.3390/agronomy13071731 https://doi.org/10.1016/0034-4257(88)90106-X https://doi.org/10.1016/S0034-4257(02)00096-2 https://doi.org/10.1016/j.rse.2013.09.001 https://www.yumpu.com/en/document/read/43526653/rice-ratooning-irri-books-international-rice-research-institute https://www.yumpu.com/en/document/read/43526653/rice-ratooning-irri-books-international-rice-research-institute http://www.knowledgebank.irri.org/ericeproduction/0.2._Growth_stages_of_the_rice_plant.htm https://doi.org/10.5755/j01.erem.77.4.28096 https://doi.org/10.1038/s41598-021-98057-x https://doi.org/10.1109/Agro-Geoinformatics.2018.8476101 https://doi.org/10.1007/s11119-016-9433-1 https://doi.org/10.1016/j.agwat.2004.04.009 https://doi.org/10.1016/j.jag.2014.08.002 https://cran.r-project.org/web/packages/caret/index.html https://doi.org/10.1007/978-3-319-35095-0_36 https://doi.org/10.1007/978-3-319-35095-0_36 Li, B., et al., 2022. Development of a new index for automated mapping of ratoon rice areas using time-series normalized difference vegetation index imagery. Pedosphere 32 (4), 576–587. https://doi.org/10.1016/S1002-0160(21)60053-X. Li, Y., Zhao, R., Wang, Y., 2024. Mapping ratoon rice fields based on SAR time series and phenology data in cloudy regions. Remote Sens. 16 (15). https://doi.org/ 10.3390/rs16152703. Art. no. 15. Liakos, K.G., Busato, P., Moshou, D., Pearson, S., Bochtis, D., 2018. Machine learning in agriculture: a review. Sensors 18 (8). https://doi.org/10.3390/s18082674. Art. no. 8. Liu, S., Chen, Y., Ma, Y., Kong, X., Zhang, X., Zhang, D., 2020. Mapping ratoon rice planting area in Central China using sentinel-2 time stacks and the phenology- based algorithm. Remote Sens. 12 (20). https://doi.org/10.3390/rs12203400. Art. no. 20. Liu, Y., Xiao, D., Yang, W., 2022. An algorithm for early rice area mapping from satellite remote sensing data in southwestern Guangdong in China based on feature optimization and random Forest. Ecol. Inform. 72, 101853. https://doi.org/10.1016/j.ecoinf.2022.101853. Liu, L., Xie, Y., Zhu, B., Song, K., 2024. Rice leaf chlorophyll content estimation with different crop coverages based on Sentinel-2. Ecol. Inform. 81, 102622. https:// doi.org/10.1016/j.ecoinf.2024.102622. Maclean, J.L., Dawe, D.C., Hardy, B., Hettel, G.P., 2002. Rice Almanac: source book for the most important economic activity on earth, 3rd ed. CABI https://doi.org/ 10.1079/9780851996363.0000. Mahlayeye, M., Darvishzadeh, R., Nelson, A., 2024. Characterising maize and intercropped maize spectral signatures for cropping pattern classification. Int. J. Appl. Earth Obs. Geoinf. 128, 103699. https://doi.org/10.1016/j.jag.2024.103699. Mansaray, L.R., Kanu, A.S., Yang, L., Huang, J., Wang, F., 2020. Evaluation of machine learning models for rice dry biomass estimation and mapping using quad- source optical imagery. GIScience Remote Sens. 57 (6), 785–796. https://doi.org/10.1080/15481603.2020.1799546. Mansaray, L.R., Kabba, V.T.S., Yang, L., 2022. Rice biophysical parameter retrieval with optical satellite imagery: a comparative assessment of parametric and non- parametric models. Geocarto Int. 37 (26), 13561–13578. https://doi.org/10.1080/10106049.2022.2082546. Maxwell, A.E., Warner, T.A., Guillén, L.A., 2021. Accuracy assessment in convolutional neural network-based deep learning remote sensing studies—Part 2: recommendations and best practices. Remote Sens. 13 (13). https://doi.org/10.3390/rs13132591. Art. no. 13. Oad, F.C., Samo, M.A.S., Oad, N.L., Chandio, G.Q., Cruz, P. Sta, 2002. Relationship of physiological, growth and yiedl contributing parameters of locklodged rice ratoon crop. J. Appl. Sci. 2 (4). https://doi.org/10.3923/jas.2002.429.432. Panuju, D.R., Paull, D.J., Griffin, A.L., Trisasongko, B.H., 2021. Mapping rice growth stages employing MODIS NDVI and ALOS AVNIR-2. In: Kumar, P., Sajjad, H., Chaudhary, B.S., Rawat, J.S., Rani, M. (Eds.), Remote Sensing and GIScience : Challenges and Future Directions. Cham: Springer International Publishing, pp. 185–203. https://doi.org/10.1007/978-3-030-55092-9_11. PhilRice. State of the rice sector in the Philippines [Online]. Available: https://www.philrice.gov.ph/ricelytics/. (Accessed 13 March 2024). PhilRice, “Data products | Philippine Rice Information System,” Philippine Rice Information System (PRiSM). Accessed: July. 8, 2024. [Online]. Available: https:// prism.philrice.gov.ph/dataproducts/. Podest, E., McCullum, A., Perez, J.L.T., McCartney, S., 2020. Land cover classification with radar and optical DataForest mapping and monitoring with SAR data [Online]. Available: https://appliedsciences.nasa.gov/sites/default/files/2020-11/SAR_Part2.pdf. Pratama, B.A.S., Danoedoro, P., Arjasakusuma, S., 2024. Exploring optimal integration schemes for Sentinel-1 SAR and Sentinel-2 multispectral data in land cover mapping across different atmospheric conditions. Remote Sens. Appl.: Society and Environment 34, 101185. https://doi.org/10.1016/j.rsase.2024.101185. Qian, Y., Zhou, W., Yan, J., Li, W., Han, L., 2015. Comparing machine learning classifiers for object-based land cover classification using very high resolution imagery. Remote Sens. 7 (1). https://doi.org/10.3390/rs70100153. Art. no. 1. Ramezan, C.A., Warner, T.A., Maxwell, A.E., 2019. Evaluation of sampling and cross-validation tuning strategies for regional-scale machine learning classification. Remote Sens. 11 (2). https://doi.org/10.3390/rs11020185. Art. no. 2. Robertson, L.D., et al., 2020. Synthetic Aperture Radar (SAR) image processing for operational space-based agriculture mapping. Int. J. Rem. Sens. 41 (18), 7112–7144. https://doi.org/10.1080/01431161.2020.1754494. Rostiana, E., Hillah, Y.F., 2016. Internal and external factors affecting the size of paddy field sold. Jurnal Ekonomi Pembangunan: Kajian Masalah Ekonomi dan Pembangunan 17 (2). https://doi.org/10.23917/jep.v17i2.1680. Art. no. 2. Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation systems in the great plains with ERTS [Online]. Available: https://ntrs.nasa.gov/ citations/19740022614. (Accessed 21 March 2024). Rudiyanto, Minasny, B., Shah, R.M., Che Soh, N., Arif, C., Indra Setiawan, B., 2019. Automated near-real-time mapping and monitoring of rice extent, cropping patterns, and growth stages in southeast Asia using sentinel-1 time series on a Google Earth engine platform. Remote Sens. 11 (14). https://doi.org/10.3390/ rs11141666. Art. no. 14. Saito, K., Dossou-Yovo, E.R., Ibrahim, A., 2024. Ratoon rice research: review and prospect for the tropics. Field Crops Res. 314, 109414. https://doi.org/10.1016/j. fcr.2024.109414. Seyedzadeh, S., Pour Rahimian, F., Rastogi, P., Glesk, I., 2019. Tuning machine learning models for prediction of building energy loads. Sustain. Cities Soc. 47, 101484. https://doi.org/10.1016/j.scs.2019.101484. Singla, S.K., Garg, R.D., Dubey, O.P., 2018. Sugarcane ratoon discrimination using LANDSAT NDVI temporal data. Spatial Information Research 26 (4), 415–425. https://doi.org/10.1007/s41324-018-0184-0. Son, N.-T., Chen, C.-F., Chen, C.-R., Minh, V.-Q., 2018. Assessment of Sentinel-1A data for rice crop classification using random forests and support vector machines. Geocarto Int. 33 (6), 587–601. https://doi.org/10.1080/10106049.2017.1289555. WorldBank. Philippines - summary | climate change knowledge portal [Online]. Available: https://climateknowledgeportal.worldbank.org/country/philippines. (Accessed 21 October 2024). Xiang, K., Yuan, W., Wang, L., Deng, Y., 2020. An LSWI-based method for mapping irrigated areas in China using moderate-resolution satellite data. Remote Sens. 12 (24). https://doi.org/10.3390/rs12244181. Art. no. 24. Xiao, X., et al., 2002. Quantitative relationships between field-measured leaf area index and vegetation index derived from VEGETATION images for paddy rice fields. Int. J. Rem. Sens. 23 (18), 3595–3604. https://doi.org/10.1080/01431160110115799. Xiao, X., et al., 2005. Mapping paddy rice agriculture in southern China using multi-temporal MODIS images. Rem. Sens. Environ. 95 (4), 480–492. https://doi.org/ 10.1016/j.rse.2004.12.009. Xu, X., Cheng, T., Yao, X., Tian, Y., Zhu, Y., Cao, W., 2016. Mapping rice planting area from Landsat 8 imagery using one-class support vector machine. In: 2016 Fifth International Conference on Agro-Geoinformatics (Agro-Geoinformatics), pp. 1–5. https://doi.org/10.1109/Agro-Geoinformatics.2016.7577693. Zhang, Y., Liu, X., Su, S., Wang, C., 2014. Retrieving canopy height and density of paddy rice from Radarsat-2 images with a canopy scattering model. Int. J. Appl. Earth Obs. Geoinf. 28, 170–180. https://doi.org/10.1016/j.jag.2013.12.005. Zhang, S., Zhao, J., Yang, J., Xie, J., Sun, Z., 2024. Feature selection and regression models for multisource data-based soil salinity prediction: a case study of minqin oasis in arid China. Land 13 (6). https://doi.org/10.3390/land13060877. Art. no. 6. Zhao, R., Li, Y., Ma, M., 2021. Mapping paddy rice with satellite remote sensing: a review. Sustainability 13 (2). https://doi.org/10.3390/su13020503. Art. no. 2. Zuspan, A.. aazuspan/geeTools [Online]. Available: https://github.com/aazuspan/geeTools. (Accessed 20 June 2024). V.N. Fikriyah et al. Remote Sensing Applications: Society and Environment 38 (2025) 101592 15 https://doi.org/10.1016/S1002-0160(21)60053-X https://doi.org/10.3390/rs16152703 https://doi.org/10.3390/rs16152703 https://doi.org/10.3390/s18082674 https://doi.org/10.3390/rs12203400 https://doi.org/10.1016/j.ecoinf.2022.101853 https://doi.org/10.1016/j.ecoinf.2024.102622 https://doi.org/10.1016/j.ecoinf.2024.102622 https://doi.org/10.1079/9780851996363.0000 https://doi.org/10.1079/9780851996363.0000 https://doi.org/10.1016/j.jag.2024.103699 https://doi.org/10.1080/15481603.2020.1799546 https://doi.org/10.1080/10106049.2022.2082546 https://doi.org/10.3390/rs13132591 https://doi.org/10.3923/jas.2002.429.432 https://doi.org/10.1007/978-3-030-55092-9_11 https://www.philrice.gov.ph/ricelytics/ https://prism.philrice.gov.ph/dataproducts/ https://prism.philrice.gov.ph/dataproducts/ https://appliedsciences.nasa.gov/sites/default/files/2020-11/SAR_Part2.pdf https://doi.org/10.1016/j.rsase.2024.101185 https://doi.org/10.3390/rs70100153 https://doi.org/10.3390/rs11020185 https://doi.org/10.1080/01431161.2020.1754494 https://doi.org/10.23917/jep.v17i2.1680 https://ntrs.nasa.gov/citations/19740022614 https://ntrs.nasa.gov/citations/19740022614 https://doi.org/10.3390/rs11141666 https://doi.org/10.3390/rs11141666 https://doi.org/10.1016/j.fcr.2024.109414 https://doi.org/10.1016/j.fcr.2024.109414 https://doi.org/10.1016/j.scs.2019.101484 https://doi.org/10.1007/s41324-018-0184-0 https://doi.org/10.1080/10106049.2017.1289555 https://climateknowledgeportal.worldbank.org/country/philippines https://doi.org/10.3390/rs12244181 https://doi.org/10.1080/01431160110115799 https://doi.org/10.1016/j.rse.2004.12.009 https://doi.org/10.1016/j.rse.2004.12.009 https://doi.org/10.1109/Agro-Geoinformatics.2016.7577693 https://doi.org/10.1016/j.jag.2013.12.005 https://doi.org/10.3390/land13060877 https://doi.org/10.3390/su13020503 https://github.com/aazuspan/geeTools Ratoon rice mapping based on Sentinel-1 and Sentinel-2 imagery 1 Introduction 2 Materials and methods 2.1 Study site description 2.2 Data 2.2.1 Ground reference collection and farmer survey 2.2.2 Sentinel-1 and 2 time-series images 2.3 Methods 2.3.1 Features extraction from Sentinel-1, -2, and vegetation indices 2.3.2 Statistical test 2.3.3 Main and ratoon rice crop classification 2.3.4 Accuracy assessment 3 Results 3.1 Main and ratoon rice spectral reflectance and backscatter signatures 3.2 Feature selection for main and ratoon rice classification 3.3 DT, SVM, and RF classification performance for discriminating ratoon from the main rice crops 3.4 Feature importance for main and ratoon rice classification 3.5 Ratoon rice area distribution 4 Discussion 4.1 Differences between main and ratoon rice crops in spectral and backscatter response 4.2 Feature importance for ratoon and main rice crop classification 4.3 Classification of main and ratoon rice crops 4.4 Ratoon rice area estimation and future mapping improvement 5 Conclusions CRediT authorship contribution statement Ethical statements Funding Declaration of competing interest Acknowledgments Appendix A Supplementary data Data availability References