Search

Background: Accurate planning of the duration of stays at intensive care units is of utmost importance for resource planning. Currently, the discharge date used for resource management is calculated only at admission time and is called length of stay. However, the evolution of the treatment may be different from one patient to another, so a recalculation of the date of discharge should be performed, called days to discharge. The prediction of days to discharge during the stay at the ICU with statistical and data analysis methods has been poorly studied with low-quality results. This study aims to improve the prediction of the discharge date for any patient in intensive care units using artificial intelligence techniques. Methods: The paper proposes a hybrid method based on group-conditioned models obtained with machine learning techniques. Patients are grouped into three clusters based on an initial length of stay estimation. On each group (grouped by first days of stay), we calculate the group-conditioned length of stay value to know the predicted date of discharge, then, after a given number of days, another group-conditioned prediction model must be used to calculate the days to discharge in order to obtain a more accurate prediction of the number of remaining days. The study is performed with the eICU database, a public dataset of USA patients admitted to intensive care units between 2014 and 2015. Three machine learning methods (i.e., Random Forest, XGBoost, and lightGBM) are used to generate length of stay and days to discharge predictive models for each group. Results: Random Forest is the algorithm that obtains the best days to discharge predictors. The proposed hybrid method achieves a root mean square error (RMSE) and mean average error (MAE) below one day on the eICU dataset for the last six days of stay. Conclusions: Machine learning models improve quality of predictions for the days to discharge and length of stay for intensive care unit patients. The results demonstrate that the hybrid model, based on Random Forest, improves the accuracy for predicting length of stay at the start and days to discharge at the end of the intensive care unit stay. Implementing these prediction models may help in the accurate estimation of bed occupancy at intensive care units, thus improving the planning for these limited and critical health-care resources.

Abstract Background Usually, arterial oxygenation in patients with the acute respiratory distress syndrome (ARDS) improves substantially by increasing the level of positive end-expiratory pressure (PEEP). Herein, we are proposing a novel variable [PaO2/(FiO2xPEEP) or P/FPE] for PEEP ≥ 5 to address Berlin’s definition gap for ARDS severity by using machine learning (ML) approaches. Methods We examined P/FPE values delimiting the boundaries of mild, moderate, and severe ARDS. We applied ML to predict ARDS severity after onset over time by comparing current Berlin PaO2/FiO2 criteria with P/FPE under three different scenarios. We extracted clinical data from the first 3 ICU days after ARDS onset (N = 2738, 1519, and 1341 patients, respectively) from MIMIC-III database according to Berlin criteria for severity. Then, we used the multicenter database eICU (2014–2015) and extracted data from the first 3 ICU days after ARDS onset (N = 5153, 2981, and 2326 patients, respectively). Disease progression in each database was tracked along those 3 ICU days to assess ARDS severity. Three robust ML classification techniques were implemented using Python 3.7 (LightGBM, RF, and XGBoost) for predicting ARDS severity over time. Results P/FPE ratio outperformed PaO2/FiO2 ratio in all ML models for predicting ARDS severity after onset over time (MIMIC-III: AUC 0.711–0.788 and CORR 0.376–0.566; eICU: AUC 0.734–0.873 and CORR 0.511–0.745). Conclusions The novel P/FPE ratio to assess ARDS severity after onset over time is markedly better than current PaO2/FiO2 criteria. The use of P/FPE could help to manage ARDS patients with a more precise therapeutic regimen for each ARDS category of severity.

Fuel moisture content (FMC) of live vegetation is a crucial wildfire risk and spread rate driver. This study presents the first daily FMC product at a global scale and 500 m pixel resolution from the Moderate Resolution Imaging Spectroradiometer (MODIS) and radiative transfer models (RTMs) inversion techniques. Firstly, multi-source information parameterized the PROSPECT-5 (leaf level), 4SAIL (grass and shrub canopy level) and GeoSail (tree canopy level) RTMs to generate three look-up tables (LUTs). Each LUT contained the most realistic model inputs range and combination, and the corresponding simulated spectra. Secondly, for each date and location of interest, a global landcover map classified fuels into three classes: grassland, shrubland and forest. For each fuel class, the best LUT-based inversion strategy based on spectral information, cost function, percentage of solutions, and central tendency determined the optimal model for the global FMC product. Finally, 3,034 FMC measurements from 120 worldwide sites validated the statistically significant results (R2 = 0.62, RMSE = 34.57%, p

Fuel types are one of the key variables that drive wildfire ignition and propagation. A new method is proposed to automatically classify and map fuel types from LiDAR data. The 3-dimensional Discrete Anisotropic Radiative Transfer (DART) model generated a fuel type LiDAR signature library. These simulations provided reference endmembers and additional data to demonstrate the feasibility to classify fuel types using spectral matching algorithms, like multiple endmember spectral mixture analysis (MESMA) and spectral angle mapper (SAM). When choosing a single endmember per fuel type, MESMA outperformed SAM with 63.3% and 48.9% agreement, respectively. Multiple endmembers per fuel type improved the classification results to 85.3% in SAM and 86.5% in MESMA. Endmembers need to identify different scan angles that account for the variability in height and number of trees for better results. Contrary to empirical models, a fuel type LiDAR signature library provides a comprehensive suite of solutions to classify fuel types from LiDAR data that is less study site dependent and applicable to multiple sensors.

This paper develops an algorithm to estimate vegetation canopy gap fraction (GF), taking advantage of the full Terrestrial Laser Scanner (TLS) resolution. After calculating the TLS angular resolution, the algorithm identifies the missing laser hits (gaps) within an angular grid in the azimuthal and zenithal directions. The algorithm was first tested on angular data simulations with random (R), cluster (C) and random and cluster together (RC) gap pattern distributions. Noise introduced in the simulations as a percentage of the resolution accounted for the effect of TLS angular uncertainty. The algorithm performs accurately if angular noise is Quercus ilex L. trees, the algorithm rendered a GF between 0.26 and 0.40. TLS had an angular noise

Live Fuel Moisture Content (LFMC) contributes to fire danger and behavior, as it affects fire ignition and propagation. This paper presents a two layered Landsat LFMC product based on topographically corrected relative Spectral Indices (SI) over a 2000–2011 time series, which can be integrated into fire behavior simulation models. Nine chaparral sampling sites across three Landsat-5 Thematic Mapper (TM) scenes were used to validate the product over the Western USA. The relations between field-measured LFMC and Landsat-derived SIs were strong for each individual site but worsened when pooled together. The Enhanced Vegetation Index (EVI) presented the strongest correlations (r) and the least Root Mean Square Error (RMSE), followed by the Normalized Difference Infrared Index (NDII), Normalized Difference Vegetation Index (NDVI) and Visible Atmospherically Resistant Index (VARI). The relations between LFMC and the SIs for all sites improved after using their relative values and relative LFMC, increasing r from 0.44 up to 0.69 for relative EVI (relEVI), the best predictive variable. This relEVI served to estimate the herbaceous and woody LFMC based on minimum and maximum seasonal LFMC values. The understory herbaceous LFMC on the woody pixels was extrapolated from the surrounding pixels where the herbaceous vegetation is the top layer. Running simulations on the Wildfire Analyst (WFA) fire behavior model demonstrated that this LFMC product alone impacts significantly the fire spatial distribution in terms of burned probability, with average burned area differences over 21% after 8 h burning since ignition, compared to commonly carried out simulations based on constant values for each fuel model. The method could be applied to Landsat-7 and -8 and Sentinel-2A and -2B after proper sensor inter-calibration and topographic correction.

The thermal-based two-source energy balance (TSEB) model has accurately simulated energy fluxes in a wide range of landscapes with both remote and proximal sensing data. However, tree-grass ecosystems (TGE) have notably complex heterogeneous vegetation mixtures and dynamic phenological characteristics presenting clear challenges to earth observation and modeling methods. Particularly, the TSEB modeling structure assumes a single vegetation source, making it difficult to represent the multiple vegetation layers present in TGEs (i.e., trees and grasses) which have different phenological and structural characteristics. This study evaluates the implementation of TSEB in a TGE located in central Spain and proposes a new strategy to consider the spatial and temporal complexities observed. This was based on sensitivity analyses (SA) conducted on both primary remote sensing inputs (local SA) and model parameters (global SA). The model was subsequently modified considering phenological dynamics in semi-arid TGEs and assuming a dominant vegetation structure and cover (i.e., either grassland or broadleaved trees) for different seasons (TSEB-2S). The adaptation was compared against the default model and evaluated against eddy covariance (EC) flux measurements and lysimeters over the experimental site. TSEB-2S vastly improved over the default TSEB performance decreasing the mean bias and root-mean-square-deviation (RMSD) of latent heat (LE) from 40 and 82 W m−2 to −4 and 59 W m−2, respectively during 2015. TSEB-2S was further validated for two other EC towers and for different years (2015, 2016 and 2017) obtaining similar error statistics with RMSD of LE ranging between 57 and 63 W m−2. The results presented here demonstrate a relatively simple strategy to improve water and energy flux monitoring over a complex and vulnerable landscape, which are often poorly represented through remote sensing models.

Land Surface Temperature (LST) is one of the key inputs for Soil-Vegetation-Atmosphere transfer modeling in terrestrial ecosystems. In the frame of BIOSPEC (Linking spectral information at different spatial scales with biophysical parameters of Mediterranean vegetation in the context of global change) and FLUXPEC (Monitoring changes in water and carbon fluxes from remote and proximal sensing in Mediterranean “dehesa” ecosystem) projects LST retrieved from Landsat data is required to integrate ground-based observations of energy, water, and carbon fluxes with multi-scale remotely-sensed data and assess water and carbon balance in ecologically fragile heterogeneous ecosystem of Mediterranean wooded grassland (dehesa). Thus, three methods based on the Radiative Transfer Equation were used to extract LST from a series of 2009–2011 Landsat-5 TM images to assess the applicability for temperature input generation to a Landsat-MODIS LST integration. When compared to surface temperatures simulated using MODerate resolution atmospheric TRANsmission 5 (MODTRAN 5) with atmospheric profiles inputs (LSTref), values from Single-Channel (SC) algorithm are the closest (root-mean-square deviation (RMSD) = 0.50 °C); procedure based on the online Radiative Transfer Equation Atmospheric Correction Parameters Calculator (RTE-ACPC) shows RMSD = 0.85 °C; Mono-Window algorithm (MW) presents the highest RMSD (2.34 °C) with systematical LST underestimation (bias = 1.81 °C). Differences between Landsat-retrieved LST and MODIS LST are in the range of 2 to 4 °C and can be explained mainly by differences in observation geometry, emissivity, and time mismatch between Landsat and MODIS overpasses. There is a seasonal bias in Landsat-MODIS LST differences due to greater variations in surface emissivity and thermal contrasts between landcover components.

The 3-D Radiative Transfer Model (RTM) FLIGHT can represent scattering in open forest or savannas featuring underlying bare soils. However, FLIGHT might not be suitable for multilayered tree-grass ecosystems (TGE), where a grass understory can dominate the reflectance factor (RF) dynamics due to strong seasonal variability and low tree fractional cover. To address this issue, we coupled FLIGHT with the 1-D RTM PROSAIL. The model is evaluated against spectral observations of proximal and remote sensing sensors: the ASD Fieldspec® 3 spectroradiometer, the Airborne Spectrographic Imager (CASI) and the MultiSpectral Instrument (MSI) onboard Sentinel-2. We tested the capability of both PROSAIL and PROSAIL+FLIGHT to reproduce the variability of different phenological stages determined by 16-year time series analysis of Moderate Resolution Imaging Spectroradiometer-Normalized Difference Vegetation Index (MODIS-NDVI). Then, we combined concomitant observations of biophysical variables and RF to test the capability of the models to reproduce observed RF. PROSAIL achieved a Relative Root Mean Square Error (RRMSE) between 6% to 32% at proximal sensing scale. PROSAIL+FLIGHT RRMSE ranged between 7% to 31% at remote sensing scales. RRMSE increased in periods when large fractions of standing dead material mixed with emergent green grasses —especially in autumn—; suggesting that the model cannot represent the spectral features of this material. PROSAIL+FLIGHT improves RF simulation especially in summer and at mid-high view angles.

A Bayesian classifier mapped the Burned Area (BA) in the Northeastern Siberian boreal forest (70°N 120°E–60°N 170°E) from 1982 to 2015. The algorithm selected the 0.05° (~5 km) Long-Term Data Record (LTDR) version 3 and 4 data sets to generate 10-day BA composites. Landsat-TM scenes of the entire study site in 2002, 2010, and 2011 assessed the spatial accuracy of this LTDR-BA product, in comparison to Moderate-Resolution Imaging Spectroradiometer (MODIS) MCD45A1 and MCD64A1 BA products. The LTDR-BA algorithm proves a reliable source to quantify BA in this part of Siberia, where comprehensive BA remote sensing products since the 1980s are lacking. Once grouped by year and decade, this study explored the trends in fire activity. The LTDR-BA estimates contained a high interannual variability with a maximum of 2.42 million ha in 2002, an average of 0.78 million ha/year, and a standard deviation of 0.61 million ha. Going from 6.36 in the 1980s to 10.21 million ha BA in the 2010s, there was a positive linear BA trend of approximately 1.28 million ha/decade during these last four decades in the Northeastern Siberian boreal forest.

The socio-economic factors are of key importance during all phases of wildfire management that include prevention, suppression and restoration. However, modeling these factors, at the proper spatial and temporal scale to understand fire regimes is still challenging. This study analyses socio-economic drivers of wildfire occurrence in central Spain. This site represents a good example of how human activities play a key role over wildfires in the European Mediterranean basin. Generalized Linear Models (GLM) and machine learning Maximum Entropy models (Maxent) predicted wildfire occurrence in the 1980s and also in the 2000s to identify changes between each period in the socio-economic drivers affecting wildfire occurrence. GLM base their estimation on wildfire presence-absence observations whereas Maxent on wildfire presence-only. According to indicators like sensitivity or commission error Maxent outperformed GLM in both periods. It achieved a sensitivity of 38.9% and a commission error of 43.9% for the 1980s, and 67.3% and 17.9% for the 2000s. Instead, GLM obtained 23.33, 64.97, 9.41 and 18.34%, respectively. However GLM performed steadier than Maxent in terms of the overall fit. Both models explained wildfires from predictors such as population density and Wildland Urban Interface (WUI), but differed in their relative contribution. As a result of the urban sprawl and an abandonment of rural areas, predictors like WUI and distance to roads increased their contribution to both models in the 2000s, whereas Forest-Grassland Interface (FGI) influence decreased. This study demonstrates that human component can be modelled with a spatio-temporal dimension to integrate it into wildfire risk assessment.

[es] La conjunción entre desarrollo y sostenibilidad del medio ambiente se torna un aspecto de singular importancia en nuestra sociedad. La incidencia de la actividad humana en la aparición de incendios forestales supone un ejemplo nítido de esa afirmación, ya que los cambios recientes en los patrones de utilización del territorio van acompañados de un notable aumento de los incendios forestales, con la consiguiente degradación del suelo y la cubierta vegetal. En este contexto, resulta de suma importancia mejorar las herramientas de prevención de incendios y evaluación de sus efectos, de cara a reducir la ocurrencia o paliar los efectos de este fenómeno. En este artículo analizamos el papel de la teledetección espacial para obtener cartografía de modelos de combustible. Estos mapas aportan una información básica en los sistemas de prevención, predicción y modelado de incendios forestales, ya que permiten analizar las características y estructura de la vegetación de cara a la propagación del fuego. Hemos utilizado como área de estudio el Parque Nacional de Cabañeros (Ciudad Real). La cartografía de combustible se ha basado en el análisis de dos imágenes de satélite Landsat- TM, junto con información auxiliar (modelos digitales de elevación y texturas). Los resultados de la clasificación ofrecieron una precisión promedio del 83 %. Las mayores dificultades que encontramos hacen referencia a la discriminación entre las diferentes alturas del matorral y la identificación del matorral bajo el arbolado. [fr] La cartographie des modèles de combustible avec télédétection: apports pour développement environnemental durable. La conjonction entre développement et soutenance de l´environnement devient un aspect de singulière importance dans notre société. Llncidence de l'activité humaine dans l´apparition des incendies forestiers suppose un exemple net de cette affirmation, puis que les récents changes des modèles d'utilisation du territoire vont accompagnés d'une notable augmentation des incendies forestiers, avec la conséquente dégradation du sol et la couverture végétale. Dans ce contexte suite de somme importance améliorer les outils de prévention des incendies et l´évaluation de ses effets, pour réduire l'incidence ou pallier les effets de ce phénomène. Dans cet article, on présente un exemple de ce tipe de vises. En concret, nous analysons le papier de la télédétection spatiale pour obtenir cartographie des modèles de combustible. Ces cartes apportent une information basique dans les systèmes de prévention, prédiction et modelage des incendies forestiers, puis que permettent analyser les caractéristiques et les structures de la végétation vers la propagation du feu. Nous avons utilisé comme surface d'étude le Parc National de Cabañeros (Ciudad Real). La cartographie de combustible s'est appuyée sur l'analyse de deux images de satellite Landsat-TM et renseignement auxiliaire (modèles digitales d'élévation et textures). Les résultats de la classification offrent une précision d'un 83%. Les difficultés les plus grandes que présente cette technologie sont la discrimination entre les hauteurs différentes de l'arbuste et l'identification de ceux-ci sous des arbres.