Damage Estimation Research Articles

Seismic performance and damage assessment of bridges during the 2023 Kahramanmaras, Türkiye earthquakes (Mw = 7.8, Mw = 7.6)

This article presents a summary of the damage observed in bridges in the regions affected by the 6 February 2023 Kahramanmaras, Türkiye earthquake sequence. A bridge database was developed based on the observations from multiple reconnaissance groups that visited the bridges. These reconnaissance groups collectively visited 140 individual bridges that were subjected to various intensities of ground shaking. The severity of the observed damage ranged from no damage to total collapse. The types of damage to bridge components mainly included cracking and shifting of abutments, failure of pier cap shear blocks, shifting or dislodging of bearing pads, cracking of girders and loss of prestress, plastic hinging at pier bases, residual pier drift, and distress to deck surfaces, handrails, and carried utilities. Recorded and estimated seismic intensity measures are presented for each bridge site, and statistical information and correlations were developed considering the intensity of shaking, bridge parameters, and observed damage. Observations from a few visited sites are presented as case studies to illustrate the common failure mechanisms. The bridge database and presented results are expected to serve as a reference for further analysis, such as statistical verification, correlation, or damage estimations, and discussion regarding the mitigation of the observed vulnerabilities of bridges in Türkiye and those with similar construction worldwide.

Rapid damage assessment effectiveness for the 2023 Kahramanmaraş Türkiye earthquake sequence

The recent devastating 2023 Türkiye earthquakes underscored the urgent need for an early estimate of quantitative losses following destructive seismic events. This study aims to propose and validate a Rapid Damage Assessment (RDA) methodology based solely on open data and tools, easily and readily available globally immediately after a seismic event. In this context, the 2023 Türkiye earthquakes are used as a case study. For the application of the proposed methodology there are two prerequisites: (1) ground shaking data, herein obtained from the US Geological Survey (USGS) ShakeMap system, and (2) a predictive seismic risk model appropriate for the area of interest, consisting of an exposure and a vulnerability model, derived from the 2020 European Seismic Risk Model (ESRM20). Scenario-type analyses for the two major events, the M7.8 - Pazarcik and the M7.5 – Elbistan earthquake, are performed, and the results are compared with the officially reported damages. The main goal is to assess the accuracy of RDA immediately after an earthquake, and whether and to what extent the RDA results can be improved as the USGS ShakeMaps are updated with the incorporation of additional data, and with disaggregation of the exposure model at a finer resolution. The results indicate that the exposure model's disaggregation reduces the damage estimates' standard deviation. At the same time, the effect of the ShakeMap version is minimized when the complete fault rupture is incorporated into the model. The estimated damages are in good agreement with the officially announced data at large scale, which corroborates the proposed methodology, as well as the adopted components of ESRM20.

Spinal cord lesion MRI and behavioral outcomes in a miniature pig model of spinal cord injury: exploring preclinical potential through an ad hoc comparison with human SCI.

prospective case series of Yucatan miniature pig spinal cord contusion injury model with comparison to human cases of spinal cord injury (SCI). to describe magnetic resonance imaging (MRI) measures of spinal cord lesion severity along with estimates of lateral corticospinal tracts spared neural tissue in both a less severe and more severe contusion SCI model, as well as to describe their corresponding behavioral outcome changes. University laboratory setting. Following a more severe and less severe SCI, each pig underwent spinal cord MRI to measure lesion characteristics, along with locomotor and urodynamics outcomes testing. In the pig with more severe SCI, locomotor and urodynamic outcomes were poor, and both the spinal cord lesion volume and damage estimates to the lateral corticospinal tracts were large. Conversely, in the pig with less severe SCI, locomotor and urodynamic outcomes were favorable, with the spinal cord lesion volume and damage estimates to the lateral corticospinal tracts being less pronounced. For two human cases matched on estimates of damage to the lateral corticospinal tract regions, the clinical presentations were similar to the pig outcomes, with more limited mobility and more limited bladder functional independence in the more severe case. Our initial findings contribute valuable insights to the emergent field of MRI-based evaluation of spinal cord lesions in pig models, offering a promising avenue for understanding and potentially improving outcomes in spinal cord injuries.

Probabilistic model for fatigue damage estimation of wind turbines with hidden markov model and neural network

The growing demand for sustainable and affordable energy sources has led to rapid development in wind energy systems. Ensuring the long-term performance and reliability of wind turbines requires accurate estimation of fatigue damage under diverse environmental conditions. By incorporating probabilistic methods, this paper presents a comprehensive framework for long-term fatigue damage evaluation for wind turbines with high accuracy and efficiency. The proposed framework adopts the multivariate kernel density estimation to construct the high-dimensional joint probability distribution of environmental variables. A hidden Markov model serves as the driver to build the connection between historical records and future events. The short-term fatigue damage is estimated by the deep neural network based on future environmental parameters. The long-term fatigue damage is then achieved by accumulating the short-term fatigue damage evaluations with sufficient event instances. To demonstrate the applicability and effectiveness of the proposed framework, a case study is conducted with a wind turbine in the North Sea region. The fatigue damage analysis results highlight the necessity of regular inspections and maintenance for wind turbines to ensure their long-term service life and optimal performance. The framework is capable of considering a wide range of environmental factors and accounting for the inherent uncertainties and stochastic nature, providing a robust and comprehensive approach for estimating fatigue damage of wind turbines as well as other structures under long-term environmental conditions.

Defoliator outbreaks track with warming across the Pacific coastal temperate rainforest of North America

The biogeography of irruptive insect herbivores is determined by host availability and climate conditions. As such, outbreak distributions are sensitive to climatic change, especially across large latitudinal gradients. Here, we investigate the outbreak distributions of two understudied defoliators, hemlock sawfly Neodiprion tsugae (Hymenoptera) and western blackheaded budworm Acleris gloverana (Lepidoptera), that have both recently impacted the greatest land area recorded across the Pacific coastal temperate rainforest since the establishment of aerial survey programs. We compiled polygon‐based estimates of insect damage collected by aerial observers, forest inventory, and downscaled climatic data to develop gridded estimates of bioclimatic conditions across the extent of the Pacific coastal temperate rainforest, including the continental United States, British Columbia and Alaska. We leveraged these data to develop ensemble machine learning models with the goal of predicting the outbreak distribution of each insect. In this manuscript we: 1) describe the historical patterns of defoliator outbreaks, 2) identify and describe climatic conditions associated with outbreaks in both species and 3) assess whether historic outbreaks have tracked geographic shifts in climate conditions across the region. We demonstrate that outbreaks of hemlock sawfly and western blackheaded budworm have been observed across the Pacific coastal temperature rainforests of North America in each decade since the establishment of the Canadian and United States aerial survey programs. The distribution of outbreaks by both insects were best explained by host availability, a limited range of spring, summer, and winter temperatures, and minimum precipitation. Finally, we demonstrate that outbreaks have tracked the poleward shift in suitable climate over the last century. This study establishes a baseline understanding of the climatic constraints and biogeographic patterns of historic sawfly and budworm outbreaks across the Pacific coastal temperate rainforest and emphasizes the overarching importance of climate in driving the irruptive dynamics of these defoliator species.Keywords: climate envelope, defoliators, hemlock sawfly, Pacific coastal temperate rainforest, population dynamics, western blackheaded budworm

A comparative study of SPT-based liquefaction potential estimation methods from three sites in India

ABSTRACT In the present study, four different SPT-based methods for liquefaction assessment namely Seed and Idriss (S–I), Tokimatsu and Yoshimi (T–Y), Japan Road Association (JRA), and Chinese code for Seismic Design of Buildings (CSDB) method has been discussed in detail. The present study compares the liquefaction potential estimation of S–I, T–Y, and JRA methods, respectively. The S–I method predicts 9.29 and 3.42 times higher liquefaction occurrence compared to JRA and T–Y methods respectively. Different damage indices such as liquefaction potential index (LPI), liquefaction risk index (LRI), and ground settlement have been evaluated for an earthquake magnitude of 7.5 on the Richter scale and peak ground accelerations varying in the range of 0.1–0.36 g respectively. LRI provides a more precise estimation of the liquefaction damages compared to LPI. The gradation of liquefiable soil has been presented and compared with the established limits of liquefiable soils.

Influence of stiffeners on acoustic emission monitoring of ship structures

Naval vessels are valuable assets that are expensive to build and maintain. Predictive maintenance is crucial to enhance efficiency and operability and to extend the service life of these structures. Fatigue is regarded as one of the main damage modes affecting the structural integrity of ship structures. Fatigue cracks can develop at the intersections of stiffeners and other structural elements, without being detected until they substantially grow, introducing a degree of uncertainty in the estimation of the fatigue damage. Acoustic emission (AE) is a passive ultrasound method for the detection and localization of different types of damage. AE is sensitive to crack propagation and can cover large areas, making it suitable for structural health monitoring of ship hulls. However, for accurate fatigue crack detection via AE it is crucial to understand how ultrasound waves interact with structural members i.e. stiffeners and longitudinal frames. Dispersion, scattering, reflection, and multimodality have so far hindered the application of AE in this context. This study aims to assess ultrasound wave behavior in ship structures in the presence of structural elements such as stiffeners and longitudinal/transversal frames. It investigates the ultrasound wave transmission dependence on source frequency and angle of incidence using spectral finite element (SEM) simulations and experimental measurements. The experimental setup includes a 10mm thick steel plate with a stiffener and a longitudinal beam (Figure 1). Pairs of sensors are placed on each side of the stiffener to record the incoming and transmitted waves, and to determine the transmission coefficient. By combining the results of both simulation and experiments, an expression for the transmission coefficient as a function of different frequencies and angles is presented. The results of this study can be used to maximize the coverage of AE monitoring systems and enhance their sensitivity for detection of fatigue cracks on ship structures.

Analysis of safety and impact probability of subsea Christmas tree impacted by falling objects

At present, there is no clear data to analyze the probability and impact of falling objects on the subsea Christmas tree. However, during the long-term operation in the complex deep-sea environment, objects may fall and hit the tree due to the anchored platform, fishing boat operation or other reasons, the impact of large heavy objects onto the tree can potentially give rise to the development of internal forces higher than those for which the tree was designed to. Therefore, aiming at the risk problem of falling objects, the paper presents the development process and results of an improved impact probability numerical method. And the safety model of falling object impact on the tree is established. Preliminary results indicate that the method offers valuable insight including and beyond a probability estimation of third-party damage, and the safety results were obtained. At the same time, the influencing factors of the tree impacted by falling objects are analyzed, and the serious influencing factors are obtained, which provides theoretical basis for the design and optimization of the protection frame of subsea square equipment.

The Role of Political Uncertainty in Climate-Related Disaster Impacts on Financial Markets

This research presents a new model for analyzing the effects of government policies on climatic disasters on financial markets. Using Fama–MacBeth rolling regressions and the construction of model-proposed risk factors, three major risk factors are found to be significant in explaining stock returns. First, there is the risk of climate disasters. Second, there is the risk of uncertainty regarding government actions. Third, there is the risk of government response to climatic disasters. Through the increase in the cost of capital from climate disasters and the uncertainty of government response, the future cost of capital is higher, leading to less investment and lower productivity. However, the government’s actions to compensate for losses due to climate damage help offset the damages from disasters. This implies that the previous estimates of economic damages due to climate risk have been underestimated. This work adds to the literature by providing a fuller estimate of the economic implications of climate change

AI-enabled damage estimation of hurricane-impacted residential communities through CFD simulations and stratified sampling

This study introduces a novel framework for estimating hurricane-induced damage to individual buildings by integrating AI-based 3D building modeling, Computational Fluid Dynamics (CFD), and stratified sampling. By leveraging the precise geometric characteristics of individual buildings extracted from digital data, the framework enables the accurate computation of pressure distributions for individual buildings through high-fidelity CFD simulations. The subsequent integration of an AI-based building component detection technique with vulnerability modeling and enhanced stratified sampling enables the rapid computation of the component and building-level failure probabilities. Applied to a case study in Atlantic City, NJ, the research underscores the effectiveness of realistic geometric characterizations and the inclusion of neighboring buildings in estimating individual-building level damage. Unlike conventional archetype-based approaches, the proposed methodology offers individual-building level risk assessment, reflecting the crucial role played by geometric and aerodynamic variables. The framework offers a promising step towards automated high-fidelity assessments of hurricane risks, leveraging detailed modeling and simulation to contribute to more resilient communities.

Seasonal Maize yield forecasting in South and East African Countries using hybrid Earth observation models

Climate change still adversely affects agriculture in the sub-Saharan Africa. There is need to strengthen early action to bolster livelihoods and food security. Most governments use pre- and post-harvest field surveys to capture statistics for National Food Balance Sheets (NFBS) key in food policy and economic planning. These surveys, though accurate, are costly, time consuming, and may not offer rapid yield estimates to support governments, emergency organizations, and related stakeholders to take advanced strategic decisions in the face of climate change. To help governments in Kenya (KEN), Zambia (ZMB), and Malawi (MWI) adopt digitally advanced maize yield forecasts, we developed a hybrid model based on the Regional Hydrologic Extremes Assessment System (RHEAS) and machine learning. The framework is set-up to use weather data (precipitation, temperature, and wind), simulations from RHEAS model (soil total moisture, soil temperature, solar radiation, surface temperature, net transpiration from vegetation, net evapotranspiration, and root zone soil moisture), simulations from DSSAT (leaf area index and water stress), and MODIS vegetation indices. Random Forest (RF) machine learning model emerged as the best hybrid setup for unit maize yield forecasts per administrative boundary scoring the lowest unbiased Root Mean Square Error (RMSE) of 0.16 MT/ha, 0.18 MT/ha, and 0.20 MT/ha in Malawi's Karonga district, Kenya's Homa Bay county, and Zambia's Senanga district respectively. According to relative RMSE, RF outperformed other hybrid models attaining the lowest score in all countries (ZMB: 25.96%, MWI: 28.97%, and KEN: 27.54%) followed by support vector machines (ZMB: 26.92%, MWI: 31.14%, and KEN: 29.50%), and linear regression (ZMB: 29.44%, MWI: 31.76%, and KEN: 47.00%). Lastly, the integration of VI and RHEAS information using hybrid models improved yield prediction. This information is useful for NFBS bulletins forecasts, design and certification of maize insurance contracts, and estimation of loss and damage in the advent of climate justice.

Hedging the climate change risks of China's brown assets: Green assets or precious metals?

A large discrepancy exists between the dire impact of climate change projected by most natural scientists and the modest estimates of damages calculated by mainstream economists. In this article, we first measure the climate risk, including the transition and physical risks, in China using textual analysis. Then, considering the heterogeneity of climate change's effect, we investigate the impact of climate risks on the brown and green assets of China (including price, volatility, and correlation) for the period from February 2012 to April 2022. In addition, the performance of China's green assets and international precious metal assets is examined in hedging the climate risk exposure of China's brown assets. The results reveal that transition and physical risks lead to contrary changes in the prices of brown and hedging assets. Transition risk causes the volatility of brown and green assets to change in the opposite direction, while physical risk causes the volatility of the two types of assets to change in the same direction. Under extreme climate risk conditions, the correlation between brown and hedging assets decreases. Finally, the findings show that the performance of international precious metals in hedging climate risk exposure is better than that of China's green equity assets.

Can a simple 0–10 RheuMetric physician estimate of inflammatory activity (DOCINF) depict a detailed swollen joint count (SJC) as accurately as a DAS28 or CDAI in patients with rheumatoid arthritis?

ObjectiveTo compare a 0–10 physician subglobal estimate of inflammatory activity (DOCINF) on a RheuMetric checklist to a formal swollen joint count (SJC) and other rheumatoid arthritis (RA) Core data set measures in a disease activity score 28 (DAS28), clinical disease activity index (CDAI), and simplified disease activity index (SDAI) in patients with RA, recognizing that RA measures, index scores and physician global assessment (DOCGL) may be elevated by joint damage and patient distress, independent of inflamamtory activity, and that formal joint counts are not recorded at most routine care visits. MethodsA cross-sectional study at a routine care visit included a RheuMetric checklist completed by a rheumatologist, with four 0–10 visual numeric scales (VNS) for DOCGL, and three sub-global estimates for inflammatory activity (DOCINF), joint damage (DOCDAM), and patient distress (DOCDIS), e.g., anxiety, depression, and/or fibromyalgia, etc. Variation in SJC according to other individual measures in the DAS28, CDAI, and SDAI, and in the indices was analyzed using Spearman correlation coefficients and regressions with and without DOCINF as an independent variable. ResultsIn 173 patients with long disease duration, regressions which included individual DAS28, CDAI or SDAI measures and added DOCINF as an independent variable explained 46 % of variation in SJC, compared to 23 % if DOCINF was not included. DOCINF was more explanatory of SJC than even the DAS28 or CDAI indices themselves, although SJC is a component of these indices. ConclusionIn routine care RA patients with long disease duration, DOCINF depicts SJC as effectively as RA indices which require 90–100 seconds to record, and may provide a feasible, informative quantitative clinical measure without recording formal joint counts.

Hydrogen Jet Flame Simulation and Thermal Radiation Damage Estimation for Leakage Accidents in a Hydrogen Refueling Station

With the rapid development of hydrogen energy worldwide, the number of hydrogen energy facilities, such as hydrogen refueling stations, has grown rapidly in recent years. However, hydrogen is prone to leakage accidents during use, which could lead to hazards such as fires and explosions. Therefore, research on the safety of hydrogen energy facilities is crucial. In this paper, a study of high-pressure hydrogen jet flame accidents is conducted for a proposed integrated hydrogen production and refueling station in China. The effects of leakage direction and leakage port diameter on the jet flame characteristics are analyzed, and a risk assessment of the flame accident is conducted. The results showed that the death range perpendicular to the flame direction increased from 2.23 m to 5.5 m when the diameter of the leakage port increased from 4 mm to 10 mm. When the diameter of the leakage port is larger than 8 mm, the equipment on the scene will be within the boundaries of the damage. The consequences of fire can be effectively mitigated by a reasonable firewall setup to ensure the overall safety of the integrated station.

Systematic Mapping of Global Research on Disaster Damage Estimation for Buildings: A Machine Learning-Aided Study

Systematic Mapping of Global Research on Disaster Damage Estimation for Buildings: A Machine Learning-Aided Study

Prediction of local site influence on seismic vulnerability using machine learning: A study of the 6 February 2023 Türkiye earthquakes

This study uses machine learning to analyze local seismic features' influence on damage from the 6 February 2023 Türkiye Earthquakes.The input features include Vs30 (the average shear wave velocity to a depth of 30 m), f0 (the predominant frequency of the site), A0 (HVSR ratio for the site), and EBd (engineering bedrock depth), along with the target feature of damage status for 44 locations. Machine learning involves Random Forest (RF), K-nearest Neighbor (KNN), Logistic Regression (LR), Decision Trees (DT), Support Vector Machines (SVM), Stochastic Gradient Descent (SGD), and Multilayer Perceptron (MP) algorithms. Also, five-fold cross-validation is employed to acquire suitable hyperparameters, enhancing its efficacy in modeling small sample sets. RF emerged as the most effective in whole performance metrics, presenting recall scores for damage and no damage conditions respectively by a 94% and 92% ratio and achieving a damage status prediction accuracy of 93%. All remaining algorithms also exhibited remarkable performance, reaching a minimum accuracy of 89% by DT, and recall score for no damage condition with 80% by MP and damage condition with 88% by SVM and SGD. The outcomes definitively designate EBd as the most crucial parameter, attributing 52% importance to its role in building damage occurrence within the study area. In contrast, significance values were determined as 24%, 18%, and 6% for f0, Vs30 and A0 respectively. These findings underscore the importance of demonstrating that initial damage estimation in high seismic hazard zones can be effectively carried out using machine learning approaches through seismic-based local site parameters.

Visual Field Prognosis From Macula and Circumpapillary Spectral Domain Optical Coherence Tomography.

To explore the structural-functional loss relationship from optic-nerve-head- and macula-centred spectral-domain (SD) Optical Coherence Tomography (OCT) images in the full spectrum of glaucoma patients using deep-learning methods. A cohort comprising 5238 unique eyes classified as suspects or diagnosed with glaucoma was considered. All patients underwent ophthalmologic examination consisting of standard automated perimetry (SAP), macular OCT, and peri-papillary OCT on the same day. Deep learning models were trained to estimate G-pattern visual field (VF) mean deviation (MD) and cluster MD using retinal thickness maps from seven layers: retinal nerve fiber layer (RNFL), ganglion cell layer and inner plexiform layer (GCL + IPL), inner nuclear layer and outer plexiform layer (INL + OPL), outer nuclear layer (ONL), photoreceptors and retinal pigmented epithelium (PR + RPE), choriocapillaris and choroidal stroma (CC + CS), total retinal thickness (RT). The best performance on MD prediction is achieved by RNFL, GCL + IPL and RT layers, with R2 scores of 0.37, 0.33, and 0.31, respectively. Combining macular and peri-papillary scans outperforms single modality prediction, achieving an R2 value of 0.48. Cluster MD predictions show promising results, notably in central clusters, reaching an R2 of 0.56. The combination of multiple modalities, such as optic-nerve-head circular B-scans and retinal thickness maps from macular SD-OCT images, improves the performance of MD and cluster MD prediction. Our proposed model demonstrates the highest level of accuracy in predicting MD in the early-to-mid stages of glaucoma. Objective measures recorded with SD-OCT can optimize the number of visual field tests and improve individualized glaucoma care by adjusting VF testing frequency based on deep-learning estimates of functional damage.

Expected damage assessment of RC half-joints under traffic load

A probabilistic framework for the expected damage estimation of RC half-joints under traffic load is proposed. Taking into account the uncertainties in geometrical features and mechanical properties, and introducing a probabilistic model for the traffic load, the proposed procedure links the structural response of RC half-joints, described by fragility curves, to financial consequences, according to the Loss-Based Design (LBD). In addition to the reliability of RC Gerber bridges, as impacted by the structural response of RC half-joints, the procedure allows to identify proper strengthening interventions focusing on cost-effectiveness and efficiency. The improved level of information regarding the structural performance of bridges can become the basis for more rational decision making as part of risk management strategies. The procedure is applied to the RC Gerber bridge “Annone” (Italy), which is investigated, as case study, in the as-built configuration and assuming a strengthening intervention.

Key drivers of vulnerability to rainfall flooding in New Orleans

IntroductionFuture urban stormwater flood risk is determined by the confluence of both climate-driven changes in precipitation patterns and the effectiveness of flood mitigation systems, such as urban drainage and pump systems. This is especially true in coastal cities protected by levee systems like New Orleans, where even present-day rainfall would be enough to cause serious flooding in the absence of extensive stormwater drainage and pumping. However, while the uncertainties associated with climate change have been well studied, uncertainties in infrastructure performance and operation have received less attention.MethodsWe investigated how these interrelated sets of uncertainties drive flood risk in New Orleans using a Robust Decision Making (RDM) approach. RDM is a framework for Decision Making Under Deep Uncertainty (DMDU) that leverages simulation models to facilitate exploration across many possible futures and the identification of decision-relevant scenarios. For our work, we leveraged a detailed Storm Water Management Model (SWMM) representation of the New Orleans urban stormwater management system to examine flood depths across the city when faced with different levels of future precipitation, sea-level rise, drainage pipe obstruction, and pumping system failure. We also estimated direct flood damage for each neighborhood in the city for this scenario ensemble. These damage estimates were then subjected to vulnerability analysis using scenario discovery—a technique designed to determine which combinations of uncertainties are most stressful to the system in terms of an outcome of interest (excess flood damage).ResultsOur results suggest that key drivers of vulnerability depend on geographic scale. Specifically, we find that possible climate-driven precipitation increases are the most important determinant of vulnerability at the citywide level. However, for some individual neighborhoods, infrastructure operation challenges under present day conditions are a more significant driver of vulnerability than possible climate-driven precipitation increases.

Estimation of flea beetle damage in the field using a multistage deep learning-based solution

Estimation of damage in plants is a key issue for crop protection. Currently, experts in the field manually assess the plots. This is a time-consuming task that can be automated thanks to the latest technology in computer vision (CV). The use of image-based systems and recently deep learning-based systems have provided good results in several agricultural applications. These image-based applications outperform expert evaluation in controlled environments, and now they are being progressively included in non-controlled field applications.A novel solution based on deep learning techniques in combination with image processing methods is proposed to tackle the estimate of plant damage in the field. The proposed solution is a two-stage algorithm. In a first stage, the single plants in the plots are detected by an object detection YOLO based model. Then a regression model is applied to estimate the damage of each individual plant. The solution has been developed and validated in oilseed rape plants to estimate the damage caused by flea beetle.The crop detection model achieves a mean precision average of 91% with a mAP@0.50 of 0.99 and a mAP@0.95 of 0.91 for oilseed rape specifically. The regression model to estimate up to 60% of damage degree in single plants achieves a MAE of 7.11, and R2 of 0.46 in comparison with manual evaluations done plant by plant by experts. Models are deployed in a docker, and with a REST API communication protocol they can be inferred directly for images acquired in the field from a mobile device.