Conference Agenda

Global International Panel on Climate Change (IPCC) models highlight increasing incidence and intensity of rainfall events due to climate change in the Caribbean. However, more granular regional models are needed to guide sector-specific actions. This paper addresses this data gap by using recent remotely sensed data and other geographic information system layers to model flood susceptibility for Trinidad from 2010 to 2100. Data came from the Climate Hazards Group InfraRed Precipitation datasets, and IPCC’s mean average rainfall layers (assessed in five-year intervals, generating 19 files). Six layers of elevation, road density, drainage density, land cover and rainfall were integrated into a composite flood susceptibility GIS layer that covered the two main seasons- dry (December to May) and wet (June to November). The composite flood susceptibility layer was subdivided into five classes (very high, high, moderate, low, and very low). Results showed a consistent pattern of high and very high flood susceptibility worsening across both wet and dry seasons, in primarily urban/semi-urban communities. These flooding events, driven primarily by low topography, land cover/land use (urban development decisions), and (inadequate) built drainage infrastructure, will be exacerbated by climate change and short-term weather variations. Policy implications underscore the urgent need for radical mitigation strategies across current flood-prone urban areas in Trinidad, including climate-resilience building designs, redesign of drainage infrastructure, enforceable zoning regulations, community-led warning systems and the design of appropriate adaptation interventions that specifically target these currently flood-prone urban areas.

This study aims to monitor the temporal development of soybean crops using Sentinel-2 multispectral imagery and the Normalized Difference Vegetation Index (NDVI). Five image acquisition dates were selected to represent key phenological stages of the crop: emergence, early vegetative stage, flowering, pod formation, and maturation. NDVI maps and difference analyses between consecutive dates were generated to evaluate variations in vegetation vigor over time. The results showed an expected pattern for irrigated soybean: NDVI values increased during early growth and peaked during flowering, followed by a gradual decline toward maturation. The spatial resolution of the images allowed the identification of field-level variations, including planting row differences and the effects of machinery tracks on plant development. A key contribution of this study is the establishment of a reference multitemporal NDVI pattern for soybean under irrigated conditions. This reference can serve as a baseline for comparing non-irrigated fields, supporting the detection of anomalies caused by stress factors such as water scarcity, pests, or diseases. The method stands out for being low-cost, accessible, and easy to interpret, which makes it valuable for both large-scale and smallholder farmers. Although NDVI is effective in identifying variations in plant vigor, it does not indicate the health loss of plants compared to a good health one and should be used in combination with other agronomic information. The approach presented here reinforces the potential of NDVI-based temporal analysis as a practical tool for crop monitoring and precision agriculture in regions with irregular climatic conditions.

This study evaluates the performance of topographic correction models applied to multispectral imagery acquired by Unmanned Aerial Vehicles (UAVs) in an area with complex relief. Although well established for orbital data, such corrections remain underexplored for UAV imagery due to their unique acquisition geometry and high spatial resolution. Seven models were tested—Simple Ratio, Cosine, Minnaert, C Correction, SCS+C, Empirical-Statistical, Empirical-Rotational—and the empirical line approach. The illumination factor (IL) was calculated based on the interaction between solar zenith angle, terrain slope, and aspect. Data were collected with a DJI Mavic 3 Multispectral over Presidente Prudente, Brazil. Photogrammetric processing was conducted in Agisoft Metashape and analysis in QGIS. Model performance was assessed via the standard deviation of reflectance values before and after correction, aiming to quantify improvements in spectral homogeneity. Preliminary results suggest that some models effectively mitigate topography-induced radiometric distortions, contributing to more accurate vegetation indices and land cover analyses. The final version will include quantitative model comparisons.

The mass balance of a glacier is used to quantify the accumulation and melting processes, which affect its mass. With this information, it is possible to infer the variation of water reserves, which will support the decision-making based supply and demand of water. The objective of this work is to show the results of a mass balance model applied to the Zongo Glacier in Bolivia. The model used is a semi-distributed conceptual model (Schaefli et al., 2005). The monthly sensitivity of the prediction of mass balance and discharge was evaluated for the hydrological years 2004-2006.

Our results show that the model presents discharge and mass balance values with an MAE error of 2.4 and a BIAS of 1.3 concerning the data observed in the glacier. Simple conceptual models can be a valuable tool to project the behavior of a glacial basin, but only if it has sufficient information for the calibration and validation of the model parameters.

The use of Unmanned Aerial Vehicles (UAVs) equipped with multispectral and hyperspectral sensors has become fundamental in precision agriculture applications. The images acquired by these sensors are commonly represented in Digital Numbers (DNs), which are influenced by various external conditions and therefore do not directly reflect the true surface reflectance. In order to make this data usable, radiometric calibration is necessary to preserve the spectral properties of the scene. This study evaluates the radiometric accuracy of images obtained with the DJI Mavic 3M UAV by comparing three sets of reflectance data: (i) the orthomosaics generated from automatic calibration using the DJI Mavic 3M UAV's sunlight sensor; (ii) the orthomosaic calibrated using the empirical line method (ELM); and (iii) the reflectance values obtained in the field with the ASD FieldSpec spectroradiometer. Ten radiometric targets with different physical properties and coloring were distributed over the study area to determine the linear regression coefficients, based on the selection of the most suitable targets for vegetation prediction. The geometric processing of the multispectral image block was carried out using Agisoft Metashape software, while the radiometric adjustment via ELM was conducted using Excel and QGIS. The results show that the ELM produced reflectance estimates for vegetation that were highly consistent with field measurements (MAE ≤ 0.009), while the automatic method overestimated the values in all bands. The findings highlight the limitations of automatic calibration and reinforce the effectiveness of ELM for applications that require high accuracy in estimating the reflectance of the target of interest.

Extreme flooding events increasingly threaten urban settlements, making advanced remote sensing approaches essential for effective monitoring and assessment. This study presents a novel methodology that combines multiple spectral indices with K-Means clustering to map flood extent and intensity using Sentinel-2 imagery, focusing on the extreme flooding event that affected Porto Alegre, Brazil, in April-May 2024. The approach integrates the water index MNDWI with complementary indices, including NDVI, NDWI, NIR, and SWIR bands. Additionally, terrain slope and size filtering constraints are applied to minimise false positives while maintaining high detection accuracy. K-Means clustering was employed for automatic threshold determination, creating binary water/non-water classifications across the 1,400 km² study area. The method demonstrated its ability to monitor temporal flood dynamics, revealing a maximum water coverage of 428.2 km² during the peak flooding period in May-June 2024. Statistical analysis indicated consistently negative mean values for MNDWI (-0.44 to -0.46) across all periods. Although the combined approach achieved a high recall rate (84%) for actual water detection, precision was moderate (31-32%) due to spectral confusion arising from wet soil, shadows, and urban features. The enhanced multi-index methodology demonstrated superiority over single-index approaches in complex mixed land-cover environments, offering robust flood intensity mapping capabilities that are crucial for disaster management and risk assessment in vulnerable riverbank communities.

Este artigo apresenta uma abordagem baseada em aprendizado profundo para a caracterização geométrica de câmeras, com o objetivo de estimar parâmetros intrínsecos a partir de uma única imagem, sem a necessidade de alvos de calibração. Para aumentar a precisão e a eficiência, o método integra o framework Optuna para otimização bayesiana de hiperparâmetros, permitindo uma convergência mais rápida e menor complexidade do modelo. Os resultados experimentais demonstram uma redução superior a 95\% no Erro Quadrático Médio (MSE) e 82\% no Desvio Padrão dos Erros (SDE) em comparação a modelos não otimizados. A solução proposta oferece uma alternativa escalável e automatizada aos métodos tradicionais de calibração, com aplicação direta em sistemas de navegação autônoma e posicionamento baseado em imagens.

The proposed study focuses on comparing analyses applied to Tectona grandis stands managed by the company 4M Agroflorestal Ltda, located on the Rancho Alegre Farm in the municipality of São José dos Quatro Marcos, Mato Grosso, Brazil. The farm is situated approximately 300 km from Cuiabá and spans a total area of 483 hectares, with 284 hectares covered by teak stands divided into 10 plots. Data were collected in October 2022 through aerial surveys using the Micasense Altum multispectral camera integrated into a DJI Matrice 100 RPAS, as well as imagery from the Planet SuperDove satellite. Subsequent data processing was carried out to obtain the Leaf Area Index (LAI) for each plot, aiming to quantify the percentage of leaf area and, consequently, compare these percentages, taking into account the statistical relationship derived from the fact that both data sources (Altum and Planet) imaged the study area on the same date (October 4, 2022). Data analysis confirmed that in the five plots located in the northern part of the farm, Planet data underestimated the foliar cover, whereas in the five southern plots, it overestimated the leaf area. On the other hand, data from the Micasense Altum camera, integrated with a UAV, demonstrated greater accuracy and precision in the spatial measurement of individual canopies, mainly due to its centimeter-level spatial resolution, thus enabling advancements in the forest management of planted stands.

With the rising amount of small satellite Earth observation missions, robust model transferability and generalization is becoming more important in satellite remote sensing image processing pipelines to enable a faster and more efficient processing setup. In this work, the transferability and generalization capabilities of two multiclass cloud masking approaches are tested on the tropical rainforest biome in the Amazon, that is unknown to the models, and on two new satellite systems, Platero and Sentinel-2. This is developed as an examplary test, if the models can be used as a baseline for transfer learning and finetuning for new satellite mission cloud masking processors. The evaluation is on a qualitative level due to the lack of ground truth data and small sample size. The results are promising for the new biome but show that regionally occurring phenomena like snow can mislead the origin networks. Furthermore, the experiments show the importance of finetuning on datasets for new sensor systems, especially when facing high discrepancies in the spectral channels.

This study evaluates landscape dynamics in the piedmont and escarpment front of the Sierras de Comechingones (San Luis, Argentina) between 1984 and 2025, using remote sensing (Landsat) and landscape metrics to quantify land cover changes and fragmentation patterns. The research addresses two key questions: (1) How has land cover evolved over time (1984, 2005, 2025) at a 1:250,000 scale? and (2) What is the degree of landscape disturbance? The results indicate a significant reduction in native forests (from 35.2% to 17.4%) and an expansion of grasslands (from 19.6% to 48.4%), driven by agricultural intensification, urban sprawl, and recurrent wildfires. Landscape metrics reveal increased fragmentation during the first period (1984–2005), followed by homogenization in the second period (2005–2025), with a decrease in Shannon diversity index values from 1.317 (1984) to 1.413 (2005) and 1.268 (2025), and an increase in the contagion index from 56% (1984) to 48% (2005) and 59% (2025). These metrics suggest a simplification of landscape complexity.The study highlights the role of anthropogenic pressures (agricultural and urban expansion) in transforming the piedmont ecosystem into a more homogeneous scenario, dominated by grasslands and croplands. It underscores the urgent need for integrated landscape planning to mitigate environmental degradation.

Orbital imagery acquired in the optical spectrum is inherently subject to geometric and radiometric errors. These inaccuracies stem from instrumental imperfections, imaging limitations, and various effects that influence the signal from the information source (targets) to the sensor. Such errors can be mitigated through radiometric calibration followed by atmospheric correction. Rigorous atmospheric correction requires data that allow for the estimation of scattering and absorption by atmospheric constituents at different wavelengths. Considering this context and aiming to enable the effective use of Landsat 8 satellite imagery (Level 1 processing) for spectral index-based change assessment, this study proposes a method based on radiometric data calibration followed by atmospheric correction using the 6S radiative transfer model, as implemented in the i.atcorr tool of the Geographic Resources Analysis Support System (GRASS GIS). Results, evaluated qualitatively and quantitatively, indicate that atmospheric correction using i.atcorr provides reliable outcomes only for the visible-red, near infrared, and shortwave infrared bands when assessing changes between two distinct epochs. For applications requiring the use of bands at the shortest wavelengths (blue and visible-green), which are more susceptible to atmospheric scattering effects, it is recommended to use Level 2 corrected images, processed by the LaSRC algorithm and provided by the U.S. Geological Survey (USGS).

The fourth most consumed food in the world, potatoes are a crop of great importance for food security. To fully explore their potential, it is essential to expand productivity studies, enabling yield to be predicted more efficiently. This study aimed to integrate remote sensing data and machine learning algorithms to develop yield prediction models for potatoes across two cultivars. For this purpose, yield samples were collected in a commercial field cultivated with two cultivars: Asterix and Markies. The sampling grid consisted of 160 points, with 80 points assigned to each cultivar. For sampling, a frame was used within which all tubers were collected and weighed. Satellite imagery from the PlanetScope platform, captured 25 days before crop desiccation, was used to calculate vegetation indices: NDVI, SAVI, and GNDVI. To develop the models, machine learning algorithms were employed: K-Nearest Neighbors, Support Vector Regression, Ridge Regression, Random Forest, XGBoost, Gradient Boosting, and Decision Tree — all of which are supervised learning methods. Given the differences between cultivars, we opted to generate separate models for each, as well as a combined model using both cultivars, in order to assess which dataset would yield more accurate predictions. The dataset was split into 80% for model training and 20% for testing. The Random Forest algorithm stood out by producing the most accurate models when using the combined cultivar dataset, with a mean absolute error (MAE) of 4.58 t·ha⁻¹. These results highlight the potential of integrating remote sensing (RS) and machine learning algorithms — particularly Random Forest — for predicting potato yield. Further studies aimed at expanding the dataset will be essential for obtaining more precise and reliable models.

This study developed a novel combinatorial approach for flood susceptibility modeling in urban areas with limited historical flood inventory data. The methodology was based on the application of the Analytic Hierarchy Process-Gaussian (AHP-G) for the generation of training samples, which were subsequently used in machine learning algorithms. The study area was the city of Belém, located in the eastern Brazilian Amazon. A spatial database was constructed using a range of geographic conditioning factors, including elevation, slope, aspect, stream power index, topographic wetness index (TWI), height above the nearest drainage (HAND), distance to drainage channels, and flow accumulation area. The Random Forest (RF), Support Vector Machine (SVM), and Generalized Linear Model (GLM) algorithms were applied to produce flood susceptibility maps. Model validation was conducted using in-situ flood occurrence data collected between 2010 and 2023. The reliability of the models was assessed using various statistical performance metrics, with particular emphasis on the area under the receiver operating characteristic curve (AUC-ROC). The results demonstrated that the Random Forest (RF) algorithm achieved the highest predictive accuracy, with an AUC exceeding 90%. The most influential factors in the modeling process were elevation, HAND, soil characteristics, distance to drainage channels, and precipitation. Overall, the study provides robust and high-quality results that can support decision-making processes regarding future flood risk management and urban planning.

Isoprene is a biogenic volatile organic compound with significant relevance in atmospheric chemistry, as it participates in the formation of tropospheric ozone and secondary aerosols. Argentina lacks a detailed inventory of biogenic emissions, as well as continuous measurements of isoprene or volatile organic compounds. This work aims to validate a simulation of isoprene emissions carried out with the MEGAN model for one summer week, using satellite observations of formaldehyde provided by Sentinel-5P. To do this, the spatial coherence between both datasets was mainly analyzed, seeking to verify whether areas with higher modeled isoprene emissions coincide with zones of high formaldehyde concentrations. Consequently, the results show elevated values in the northeast of the country, where vegetation, solar radiation, and temperature favor the emission of volatile organic compounds, especially isoprene. This preliminary analysis helps to evaluate the quality of the simulation and highlights the usefulness of satellite data for studies in locations with scarce field information.

Over the past few decades, harmful algal blooms (HABs) have become increasingly frequent in the fjords and channels of southern Chile. However, knowledge of the environmental factors that trigger them remains limited in remote areas. This study uses a combination of in situ observations and satellite remote sensing (Sentinel-3A/B) to analyse the physical, chemical and biological conditions during a phytoplankton bloom in spring (November–December 2021) in the Magellan region. Fourteen stations were sampled during the EXOFAN cruise to evaluate the hydrographic structure, nutrient availability and phytoplankton composition, paying particular attention to potentially harmful species such as Phaeocystis spp. and the Pseudonitzschia cf. pseudodelicatissima complex. The results revealed spatial heterogeneity in the distributions of salinity, temperature and nutrients, with freshwater inputs and solar heating generating stratified water masses in interior fjords. High cell densities of the target taxa were found in these stable environments, which may act as retention zones favouring bloom persistence. Satellite-derived Sea surface temperature and chlorophyll-a data corroborated the in-situ findings, revealing seasonal warming and productivity hotspots despite frequent cloud cover. This study highlights the importance of combining traditional oceanographic methods with remote sensing to understand and monitor bloom dynamics in complex subpolar coastal systems. The observed patterns offer valuable insights into the ecological conditions that drive HAB development. This information is crucial for the development of future prediction models, early warning systems and sustainable aquaculture management strategies, particularly in the face of ongoing climate-driven changes.

This study examines socio-environmental issues in the La Silla River, situated in the Metropolitan Area of Monterrey (MAM), Nuevo León, Mexico, utilizing advanced technologies such as unmanned aerial vehicles (UAVs), elevation profiles, and a population analysis based on a national database. Like other urban rivers, the La Silla River faces significant challenges, including illegal dumpsites, wastewater discharges, irregular settlements, and solid waste accumulation, all of which are exacerbated by accelerated urban growth. To document these issues, UAVs equipped with RGB and LiDAR cameras captured high-resolution data, enabling the generation of orthophotos and point clouds, which were then processed for detailed spatial analysis. This information enabled the photo-interpretation mapping of pollution sources and the study of their spatial distribution. Ground-based observations were also collected to compare the effectiveness of field-level data with that of UAV-acquired imagery. A population analysis is presented to understand the social context of the residents living in the river basin, many of whom are vulnerable. This integrated approach strengthens the link between technology and society, offering a valuable and replicable tool for environmental monitoring in urban areas.

This study presents the application of UAV-based LiDAR and aerial photogrammetry for the detailed mapping and morphometric analysis of karst features in the Sete Lagoas region, Brazil. Thirteen karst depressions were identified, including eight caprock dolines and five karst lagoons, based on high-resolution elevation models and structural geological field data. The integration of geomorphological and structural information enabled a precise characterization and volumetric estimation of the features using GIS-based methods. The results provide essential insights into the karst system’s morphology and contribute to the development of thematic cartographic products. These products may support future investigations on groundwater recharge and enhance regional strategies for the sustainable management of water resources in an area strongly reliant on a karst aquifer.

The effective discrimination and combat of land degradation, principally in arid, semi-arid, and dry sub-humid environments, relies on the efficient application of detection and monitoring Land Use and Land Cover Changes (LULCC) techniques. In this way, this study evaluated the decadal trends of environmental changes in Cabrobó’s Desertification Nucleus (CDN), Pernambuco, Brazilian semiarid region, between 2001 and 2021. For this, we manipulated CHIRPS data, MODIS products (MOD13A1, MCD43A3, and MOD16A2), and MapBiomas data in the Google Earth Engine (GEE) digital platform and the QGIS software (version 3.10.9). We performed a space-temporal trend assessment for all biophysical parameters via non-parametric tests Mann-Kendall (MK) and Sen’s Slope Estimator (SSE). The results obtained showed the effectiveness of using these orbital products for long-term environmental analysis, as well as highlighting the municipalities of Belém do São Francisco, Itacuruba, and Floresta with significant trends at the 1% probability level of the biophysical indices (from decrease to potential evapotranspiration and NDVI, and addition to albedo surface) due to a more change in land use and cover. This methodology proved useful for studies on the discrimination of space-temporal trends in other Caatinga areas after the appropriate methodological adjustments.

This paper presents a university extension project aimed at improving the spatial documentation of the Martin Luther Church cemetery in Ibirama, Brazil, through the integration of UAV-based photogrammetry, high-precision GNSS surveying, and GIS mapping. A multi-altitude flight strategy (30 m, 50 m, and 80 m) combined with a network of nine ground control points enabled the generation of high-resolution orthophotos and a digital surface model. Accuracy assessment using 24 independent checkpoints confirmed compliance with Class A of the Brazilian Cartographic Accuracy Standard (PEC) for 1:250 scale mapping. The highest-resolution orthophoto (1.4 cm GSD) was used to manually digitize 1,112 graves, linked to a database containing 1,594 burial records with names, birth/death dates, and photographs. The final deliverables included digital files and a large-format map, which replaced paper records and are now used by the church for cemetery management. This study demonstrates a replicable workflow that combines open-source tools and low-cost geospatial technologies to deliver precise cartographic outputs for community use. The project also highlights the educational and social value of integrating geospatial innovation into university extension programs.

Terrestrial laser scanning (TLS) systems generate dense point clouds for many applications. These systems generate point clouds in a local three-dimensional (3D) coordinate system (x, y, z) and measure the intensity of the backscattered signal. Although the intensity data collected can be useful for various applications, the received signals are affected by distance and angle of incidence. The objective of this study is to assess existing methods for correcting intensity values using polynomial approximation to compensate for the effect of distance and to evaluate the impact on intensity data collected under different lighting conditions. In this study, it was possible to obtain a good approximation of the ideal case through polynomial approximation, with a low R². In addition to the model and, the coefficient of variation RMS were evaluated to evaluate the adjustment accuracy

This study investigated the relationship between spectral reflectance, derived from a thermal sensor mounted on an Unmanned Aerial Vehicle (UAV), and soil moisture gradients. The research was based on the premise that soil moisture significantly influences its spectral response, particularly in the visible and infrared regions, impacting the absorption and reflection of electromagnetic radiation. High-spatial-resolution remote sensing data were used to monitor variations in soil moisture. The methodology involved acquiring thermal and multispectral imagery over an experimental area with laboratory-identified moisture gradients. An analysis of the importance of moisture predictive variables was performed using machine learning techniques, such as the Random Forest algorithm, due to its robustness against noise and its ability to non-parametrically assess variable relevance. The performance of the soil moisture prediction model was evaluated using statistical metrics such as predictive correlation and Root Mean Square Error (RMSE). The results indicated a significant correlation between thermal sensor readings and soil moisture levels, demonstrating the approach's capability to distinguish different water saturation conditions. The analysis of spectral variable importance confirmed the sensitivity of specific electromagnetic spectrum bands to moisture variations. The investigation highlighted the importance of precise sensor calibration to ensure the consistency and comparability of data acquired at different times or environmental conditions, a critical factor for analyzing temporal changes and evaluating the effectiveness of management practices.

This study analyses the ionospheric scintillations over the Brazilian territory, focusing on their magnitude, seasonality, and impact on Global Navigation Satellite Systems (GNSS) signal quality, specifically GPS (Global Positioning System), GLONASS (Global'naya Navigatsionnaya Sputnikovaya Sistema / Global Navigation Satellite Systems) and GALILEO. The main objective is to quantify and evaluate how such phenomena affect positioning integrity by establishing relationships between scintillation intensity, expressed by the S4 index, and signal loss events (Loss of Lock). Additionally, a statistical and temporal analysis of satellite availability per hour for different GNSS constellations is conducted, assessing their variability and reliability. Data are obtained using the ISMR (Ionospheric Scintillation Monitor Receivers) Query Tool platform, which provides information from several GNSS stations over Brazil. The analysis was carried out using SQL queries and a MATLAB-developed algorithm. The results contribute to a better understanding of ionospheric behavior in tropical regions and aim to improve GNSS-based applications, especially in environments subject to signal degradation. A significant correlation was observed between scintillation intensity and periods of high solar activity, particularly in 2014. Furthermore, although most loss of lock events were expected during strong scintillation, approximately 83% occurred during low S4 index intervals (< 0.29), suggesting other contributing factors or limitations in S4 data availability during signal interruptions.

The stay-green trait in maize has become a key target in breeding programs aiming to enhance grain yield (GY) stability under adverse environmental conditions. Characterized by delayed leaf senescence during grain filling, the stay-green phenotype maintains photosynthetic activity and canopy integrity longer, increasing radiation interception and biomass partitioning to kernels. In tropical environments, where high temperature and erratic rainfall are frequent, the ability of a hybrid to maintain a green canopy during reproductive stages is particularly advantageous. Remote sensing technologies, especially those based on unmanned aerial vehicles (UAVs), provide a rapid, non-destructive, and high-throughput means to quantify canopy dynamics. Vegetation indices (VIs) derived from RGB images, such as the Visible Atmospherically Resistant Index (VARI), offer a cost-effective alternative to multispectral methods. The VARI index has shown promising results in distinguishing green vegetation from soil background, especially under field conditions where spectral contrast may be subtle. This study aimed to quantify green canopy cover (GCC) of maize hybrids using VARI, across different phenological stages, and to investigate its relationship with GY. The objective was to evaluate whether VARI-based green cover could serve as an effective phenotyping tool for selecting high-performing stay-green hybrids adapted to subtropical low-input systems. The experiment was conducted at the Experimental Station Ilha Barra do Pomba, Itaocara–RJ. Eight maize hybrids were used: the interpopulation hybrids UENF 506-11 (H1) and UENF 506-16 (H2), both grain-type; UENF MSV 2210 (H3), suitable for silage and green corn; and UENF MS 2208 (H4), for silage. Commercial hybrids included BM 207 (H5) and AG 1051 (H8) (double hybrids); and LG 6036 (H6) and 30F35R (H7) (single hybrids). Due to replanting, H5 was excluded from statistical analyses. The design was a randomized complete block with four replicates (32 plots). Each plot had four 4-m rows (0.70 m between rows; 0.20 m between plants), totalling 80 plants. A DJI Mavic 2 Pro UAV with a 20 MP RGB camera was used to collect aerial images at 80 m (2 cm GSD). Flights were conducted at 0 (emergence), 19 (V2–V3), 34 (V3–V4), 91 (R2), 98 (R3–R4), 112 (R4–R5), 119 (R5–R6). Figure 2 illustrates the orthophoto captured at 119 DAP. VARI was calculated as (G−R)/(G+R+B). In QGIS, thresholding (≥0.02) defined green vegetation. Zonal Statistics was used to extract green canopy cover (%) per plot. GY (13% moisture) was converted to kg/ha (Gonçalves et al., 2025). ANOVA and Tukey tests identified differences; Pearson correlation assessed GCC-yield relationships. All analyses were performed in GENES. Significant differences among hybrids (p < 0.05) were observed at 19, 98, 112, and 119 DAP. At 19 DAP (V2–V3), early-stage differences reflected genotypic variation in seedling vigor, which is fundamental for biomass accumulation and initial competition with weeds. At 98 DAP (R3–R4), corresponding to the grain formation phase, differences among hybrids in green canopy coverage became more evident. Hybrids LG 6036 (H6) and 30F35R (H7) maintained higher GCC values, suggesting greater leaf longevity and photosynthetic capacity. In contrast, interpopulation UENF hybrids (especially H3 and H4) showed reduced green cover, associated with faster onset of senescence. At 112 and 119 DAP (R4–R6), these differences were accentuated. As shown in Tukey test, H6 and H7 reached the highest GCC values (~50%), indicating superior stay-green behavior, while H3 and H4 presented values below 15%, characterizing rapid senescence. This phenotypic divergence in canopy maintenance reflects physiological differences in resource allocation and stress resilience. The strong positive phenotypic correlations between GCC and grain yield at 112 and 119 DAP (r = 0.88 and 0.96, respectively) confirm that prolonged canopy greenness enhances sink filling and yield. These results demonstrates that extended photosynthetic duration during grain filling contributes significantly to productivity. Commercial hybrids maintained greener canopies longer, which contributes to prolonged photosynthesis and improved yield potential under subtropical conditions. Additionally, your thesis evidences that commercial hybrids with longer reproductive phases preserved canopy integrity longer than UENF hybrids, which have extended vegetative and shorter reproductive stages. This cycle mismatch contributed to early senescence and reduced grain yield in UENF materials. Therefore, using VARI for monitoring GCC over time enables effective identification of stay-green genotypes, especially in climates prone to water stress and thermal fluctuations. Integrating such phenotyping tools into breeding programs supports the selection of hybrids with enhanced adaptability and yield stability under subtropical conditions. VARI-derived green canopy cover successfully differentiated maize hybrids with stay-green traits. Commercial hybrids such as LG 6036 and 30F35R maintained higher GCC at late stages and produced greater yields. The VARI index, applied via UAV-RGB imagery, proved to be a reliable tool for high-throughput phenotyping in breeding programs.