Artículos científicos

SATELLITE CHLOROPHYLL–A (CHLA) PRODUCTS ARE WIDELY USED TO STUDY COASTAL PRODUCTIVITY, BUT THEIR PERFORMANCE OFTEN DEGRADES IN RIVER-INFLUENCED AND OPTICALLY COMPLEX WATERS. WE EVALUATED MODIS-AQUA CHLA RETRIEVALS IN THE COASTAL UPWELLING AREA OFF CENTRAL-SOUTHERN CHILE, A REGION STRONGLY AFFECTED BY SEASONAL RIVER PLUMES, USING MONTHLY IN SITU CHLA AND HYDROGRAPHIC OBSERVATIONS FROM STATION 18 (AUGUST 2002 TO SEPTEMBER 2011), DAILY MODIS PRODUCTS, AND MATCHUP ANALYSES BASED ON 3 × 3 PIXEL WINDOWS AND 1-, 3-, 5-, AND 7-DAY COMPOSITES. MODIS CHLA AND NORMALIZED FLUORESCENCE LINE HEIGHT (NFLH) REPRODUCED THE BROAD SEASONAL CYCLE, WITH MAXIMA DURING SPRING–SUMMER, BUT DEFAULT MODIS CHLA SYSTEMATICALLY EXCEEDED IN SITU OBSERVATIONS, PARTICULARLY DURING PERIODS OF ENHANCED TURBIDITY AND RIVER-INFLUENCED OPTICAL COMPLEXITY. AMONG THE RAW SATELLITE PRODUCTS, 1-DAY MODIS CHLA MATCHUPS SHOWED THE STRONGEST AGREEMENT WITH IN SITU CHLA (R = 0.77, RMSE = 8.5 MG M−3), WHEREAS 5-DAY COMPOSITES INCREASED MATCHUP AVAILABILITY TO 95% BUT REDUCED THE CORRELATION (R = 0.46, RMSE = 10.5 MG M−3). IN CONTRAST, NFLH SHOWED MORE STABLE PERFORMANCE ACROSS COMPOSITE LENGTHS, ALTHOUGH IT UNDERESTIMATED HIGH CHLA VALUES AND SHOULD THEREFORE BE INTERPRETED AS A COMPLEMENTARY FLUORESCENCE-BASED DIAGNOSTIC RATHER THAN AS A DIRECT SUBSTITUTE FOR LOCALLY VALIDATED CHLA RETRIEVALS. A GRADIENT BOOSTING MODEL TRAINED WITH MODIS REMOTE-SENSING REFLECTANCES IMPROVED THE CORRESPONDENCE BETWEEN SATELLITE AND IN SITU CHLA RELATIVE TO THE DEFAULT MODIS PRODUCT WITHIN THE AVAILABLE STATION 18 MATCHUP DATASET. BECAUSE THIS MODEL WAS EVALUATED USING CROSS-VALIDATION RATHER THAN AN INDEPENDENT REGIONAL VALIDATION DATASET, THE MACHINE-LEARNING RESULTS SHOULD BE INTERPRETED AS A LOCAL PROOF OF CONCEPT RATHER THAN A FULLY VALIDATED REGIONAL ALGORITHM. THESE RESULTS INDICATE THAT STANDARD MODIS ALGORITHMS OVERESTIMATE CHLA IN THIS RIVER-INFLUENCED UPWELLING SYSTEM AND HIGHLIGHT THE VALUE OF LOCAL CORRECTION APPROAC

CONTINUOUS MEDIUM-DENSITY FIBERBOARD (MDF) PRODUCTION PRESENTS A PERSISTENT QUALITY-ASSURANCE PROBLEM: DESTRUCTIVE LABORATORY TESTS RETURN RESULTS TOO LATE TO PREVENT OFF-SPECIFICATION MATERIAL FROM ACCUMULATING BEFORE A CORRECTIVE RESPONSE IS POSSIBLE. THIS STUDY DEVELOPS AN INSPECTION-ORIENTED PREDICTIVE QUALITY FRAMEWORK USING INDUSTRIAL DISTRIBUTED CONTROL SYSTEM (DCS) DATA AND AUTOMATED MACHINE LEARNING, TREATING THE PRODUCTION LINE AS AN INTEGRATED NINE-STAGE SYSTEM IN WHICH UPSTREAM PROCESS DISTURBANCES PROPAGATE THROUGH COUPLED THERMOMECHANICAL AND CHEMICAL OPERATIONS BEFORE BECOMING VISIBLE IN FINAL PANEL PROPERTIES. TWO QUALITY TARGETS WERE MODELED ACROSS ULTRALIGHT (UL) AND STANDARD THIN (STD) PANELS USING 3365 PRODUCTION BATCHES AND 327 DCS PROCESS VARIABLES. THE PIPELINE COMBINED RANDOM FOREST IMPUTATION, PEARSON COLLINEARITY FILTERING (|𝑟|≥0.8 ), TARGET-SPECIFIC FEATURE SELECTION, AND STACKED ENSEMBLE REGRESSION VIA H2O AUTOML. THE VERTICAL DENSITY PROFILE INDEX (VSC), A PLANT-REPORTED SCALAR DERIVED FROM X-RAY DENSITY PROFILING, WAS PREDICTED ACCURATELY IN BOTH PRODUCT FAMILIES (TEST RMSE: 1.39 AND 1.78, INDEX UNITS FOR UL AND STD RESPECTIVELY), REFLECTING ITS CLOSE COUPLING TO DRYING STABILITY, RESIN DOSING, AND THERMAL CONDITIONS. INTERNAL BOND STRENGTH (IB) WAS HARDER TO PREDICT, ESPECIALLY FOR THIN STD PANELS (TEST RMSE: 78.93 KPA VS. 27.41 KPA FOR UL), AS CORE-LAYER BONDING MECHANISMS ARE ONLY INDIRECTLY OBSERVABLE THROUGH STANDARD DCS INSTRUMENTATION. MODEL-AGNOSTIC FEATURE IMPORTANCE RANKINGS WERE PHYSICALLY COHERENT ACROSS BOTH PRODUCT FAMILIES, WITH DOMINANT PREDICTORS CONCENTRATED IN DRYING, RESIN APPLICATION, FORMING, AND HOT PRESSING, CONSISTENT WITH THE COUPLED-SUBSYSTEM NATURE OF MDF QUALITY FORMATION. THE HISTORICAL DATASET WAS DOMINATED BY ACCEPTABLE AND OVER-QUALITY IB PRODUCTION, WHICH PRECLUDED CONFORMITY CLASSIFICATION AND SAMPLING-REDUCTION ANALYSIS; A PROSPECTIVE DATASET WITH NEAR-THRESHOLD OBSERVATIONS IS REQUIRED FOR THOSE EVA

ESTIMATING VERTICAL HYDRAULIC CONDUCTIVITY (KV) IN SEMI-ARID BASINS IS CHALLENGING BECAUSE LITHOLOGICAL HETEROGENEITY LIMITS THE REPRESENTATIVENESS OF SPARSE HYDRAULIC MEASUREMENTS. THIS STUDY EVALUATES A SURFACE-BASED SEISMOELECTRIC WORKFLOW FOR INFERRING DEPTH-RESOLVED K V PROFILES WITHOUT USING DRILLING INFORMATION DURING MODEL DEVELOPMENT. BOREHOLE DATA WERE RESERVED FOR EXTERNAL VALIDATION. THE WORKFLOW INTEGRATES FIELD ACQUISITION, GEOSPATIAL DATA SYNTHESIS, SIGNAL CONDITIONING, PHYSICALLY GUIDED FEATURE ENGINEERING, SUPERVISED REGRESSION BENCHMARKING, AND HYDROGEOLOGICAL EXPERT APPRAISAL. ITS MAIN CONTRIBUTION IS A DRILLING-INDEPENDENT MODELING STRATEGY THAT TRANSLATES SURFACE SEISMOELECTRIC ATTRIBUTES INTO VERTICAL K V PROFILES. THIS PROVIDES PRE-DRILLING INFORMATION FOR WELL SITING, SCREEN-INTERVAL DESIGN, AND PLANNING OF DRILLING AND COMPLETION COSTS. A HIGH-DENSITY SURVEY WAS CONDUCTED IN THE ÑUBLE REGION OF CENTRAL CHILE USING A DUAL-CHANNEL SEISMOELECTRIC SYSTEM WITH INVESTIGATION DEPTHS OF UP TO 200 M. MORE THAN 500 GEOREFERENCED STATIONS WERE ACQUIRED. SEISMOELECTRIC ATTRIBUTES DERIVED FROM THE DUAL-CHANNEL RESPONSE, TOGETHER WITH TOPOGRAPHIC DESCRIPTORS, WERE USED TO TRAIN AND COMPARE MULTIPLE REGRESSION MODELS. COMPARED WITH A BASELINE CONFIGURATION BASED ONLY ON RAW VOLTAGE AND DEPTH, THE AUGMENTED MODEL SHOWED STRONGER CROSS-VALIDATED PERFORMANCE. THE COEFFICIENT OF DETERMINATION INCREASED FROM R 2 = 0.711 TO R 2 = 0.968 , RMSE DECREASED FROM 3.534 TO 1.305 M/D, AND MAE DECREASED FROM 1.368 TO 0.546 M/D. THE PREDICTED K V PROFILES BETTER PRESERVED PERSISTENT HIGH- AND LOW-CONDUCTIVITY INTERVALS AND ABRUPT VERTICAL TRANSITIONS THAN THE BASELINE MODEL. THEY WERE ALSO CONSISTENT WITH INDEPENDENT LITHOSTRATIGRAPHIC EVIDENCE AND POST-DRILLING YIELD VERIFICATION AT THE VALIDATION SITE. PROFILE-LEVEL SMOOTHING WITH SAVITZKY–GOLAY AND WAVELET FILTERS IMPROVED VISUAL CONTINUITY WHILE PRESERVING THE DOMINANT CONDUCTIVITY CONTRASTS. THESE RESULTS

TOFU FROM SIX DIFFERENT LANDRACES OF CHILEAN COMMON BEANS (ARAUCANO, CIMARR & OACUTE;N, MAGNUM, PEUMO, SAPITO, AND TORTOLA) WAS PREPARED AND ANALYZED FOR PROXIMATE AND LIPID COMPOSITION, ANTIOXIDANT CAPACITY, AND PHENOLIC CONTENT. TOFU HAS HIGHER PROTEIN AND LIPID CONTENT, LOWER CARBOHYDRATE AND PHENOLIC CONTENT, AND SHOWS ANTIOXIDANT CAPACITY. THE HIGHEST TOTAL PROTEIN WAS FOUND FOR TOFU PREPARED FROM CIMARR & OACUTE;N AND SAPITO BEANS. THE MAIN PHENOLICS IN THE SECONDARY METABOLITE-ENRICHED EXTRACT OF THE TOFU WERE KAEMPFEROL 3-O-GLUCOSIDE AND KAEMPFEROL. NUCLEAR MAGNETIC RESONANCE (NMR) ANALYSES SHOWED THAT MOST LIPIDS FROM THE BEANS AND TOFU WERE UNSATURATED FATTY ACIDS TRIGLYCERIDES.

THE WOOD MANUFACTURING INDUSTRY INCREASINGLY REQUIRES AUTOMATED AND INTELLIGENT SYSTEMS FOR DEFECT DETECTION TO ENSURE CONSISTENT AND RELIABLE QUALITY CONTROL. TRADITIONALLY, THIS PROCESS HAS RELIED ON VISUAL INSPECTION BY HUMAN OPERATORS, WHICH INTRODUCES VARIABILITY AND LIMITS PERFORMANCE. THIS STUDY ADDRESSES THIS CHALLENGE BY EVALUATING CONVOLUTIONAL NEURAL NETWORKS FOR AUTOMATIC DEFECT CLASSIFICATION IN MELAMINE-FACED BOARDS. MULTISPECTRAL IMAGES IN THE VISIBLE (VIS) AND NEAR-INFRARED (NIR) BANDS WERE CAPTURED UNDER REAL PRODUCTION CONDITIONS USING AN INDUSTRIAL IMAGING SYSTEM. THE RESIDUAL NETWORK 18 AND VISUAL GEOMETRY GROUP 16 MODELS WERE TESTED ON THE DATASET AND ACHIEVED ACCURACY LEVELS COMPARABLE TO THOSE OF EXPERT HUMAN INSPECTORS. THE PROPOSED METHOD CONSISTENTLY REACHED OVER 92% ACCURACY ACROSS ALL CLASSIFICATION TASKS, INDICATING ITS PRACTICAL POTENTIAL FOR INDUSTRIAL QUALITY CONTROL APPLICATIONS.