Lowest Root Mean Square Error Research Articles

Evaluation of air temperature estimated by ERA5-Land reanalysis using surface data in Pernambuco, Brazil.

This study aimed to analyze the average air temperature data estimated by ERA5-Land reanalysis over 10years (01/01/2011 to 12/31/2020), based on data from 12 automatic weather stations located in different cities in the state of Pernambuco, northeast of Brazil. For more careful evaluation, the average air temperature data were stratified by mesoregions of the state. ERA5-Land reanalysis data were validated using statistical indices that evaluated the accuracy. The average air temperature estimated by ERA5-Land reanalysis agree well with weather stations in almost the entire state of Pernambuco. The highest accuracy of the average air temperature estimated by ERA5-Land occurred in the city of Ibimirim (R2 = 0.98), while the lowest accuracy was measured in the city of Caruaru (R2 = 0.57). Root-mean-square error (RMSE) generated by the ERA5-Land reanalysis was lower than 0.60°C in most Pernambuco. The highest RMSE of ERA5-Land for average air temperature was calculated using data from the city of Caruaru (1.11°C), whereas the lowest RMSE was obtained with data from the city of Recife (0.41°C). According to mean bias errors, (MBE) the ERA5-Land underestimated the average air temperature data compared to automatic weather stations data, especially in Ouricuri, which had the highest underestimation (-0.80°C). On the other hand, among the municipalities where ERA5-Land overestimated the temperature values, the highest overestimation was identified in Garanhuns (0.35°C). Therefore, based on the results of this study, ERA5-Land reanalysis successfully estimated the average air temperature for the state of Pernambuco.

Prediction of geoid undulations: Random forest versus classic interpolation techniques

AbstractLocal geoid determination studies are commonly carried out today to establish the relationship between the ellipsoidal height determined by satellite geodesy methods and the orthometric height found using geoid undulation . The aim of this study was to determine a local geoid using the kriging, local polynomial (LP), and inverse distance to a power (IDP) interpolation methods along with the random forest (RF) regression method and to compare the performance. For the application, 193 Global Navigation Satellite System (GNSS)/leveling points homogeneously distributed over the study area were selected as reference data, and 70 GNSS/leveling points were selected as test data. The results were compared in terms of accuracy using well‐known performance metrics, namely, the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2) values. According to the results, the highest R2 and the lowest RMSE and MAE values were obtained with the RF regression method. These findings demonstrated the superiority of the RF regression over the classic interpolation methods applied in the local geoid determination for the study area.

A comparative analysis of data mining techniques for agricultural and hydrological drought prediction in the eastern Mediterranean

Drought is a natural hazard which affects ecosystems in the eastern Mediterranean. However, limited historical data for drought monitoring and forecasting are available in the eastern Mediterranean. Thus, implementing machine learning (ML) algorithms could allow for the prediction of future drought events. In this context, the main goals of this research were to capture agricultural and hydrological drought trends by using the Standardized Precipitation Index (SPI) and to assess the applicability of four ML algorithms (bagging (BG), random subspace (RSS), random tree (RT), and random forest (RF)) in predicting drought events in the eastern Mediterranean based on SPI-3 and SPI-12. The results reveal that hydrological drought (SPI-12, −24) was more severe over the study area, where most stations showed a significant (p < 0.05) negative trend. The accuracy of ML algorithms in drought prediction varied in relation to the implementation stage. In the training stage, RT outperformed the other algorithms (Root mean square error (RMSE) = 0.3, Correlation Coefficient (r) = 0.97); the performance of the algorithms can be ranked as follows: RT > RF > BG > RSS for both SPI-3 and SPI-12. In the testing stage, both the BG and RF algorithms had the highest correlation r(observed vs. predicted) (0.58–0.64) and lowest RMSE (0.68–0.88). In contrast, the lowest correlation r(observed vs. predicted) (0.3–0.41) and highest RMSE (0.94–1.10) was calculated for the RT algorithm. However, BG was more dynamic in drought capturing, with the lowest RMSE and highest correlation. In the validation stage, the BG performance was satisfactory (RMSE = 0.62–0.83, r = 0.58–0.79). The output of this research will help decision-makers with drought mitigation plans by using the new four machine learning algorithms.

An enhanced spatiotemporal fusion method – Implications for DNN based time-series LAI estimation by using Sentinel-2 and MODIS

The consequent and accurate monitoring of the seasonal dynamics of crop leaf area index (LAI) is critical to yield estimation and agriculture policy development. It is difficult for a single sensor to balance spatial and temporal resolution. Spatiotemporal fusion is an effective way to meet the need for high spatial and temporal applications. Among the fusion methods, regression model Fitting, spatial Filtering and residual Compensation (Fit-FC) may be recommended for vegetation dynamic monitoring because of its outperformance for the cases with considerable phenological changes. However, it is not good at capturing image structure and textures. To overcome the limitations, an enhanced version of Fit-FC, referred as Enhanced-Fit-FC (EFF), was developed. The EFF method can be applied with one (unidirectional prediction, Uni-EFF) or two (bidirectional prediction, Bi-EFF) coarse-fine image pairs as input for near real-time or post-growth applications. The visual and quantitative assessments indicated that EFF mitigated the blurring effect of Fit-FC and generated accurate reflectance, especially Bi-EFF contributed to capturing land cover changes. Compared with Fit-FC, the correlation coefficient (CC) and quality index (QI) of EFF increased by more than 0.12, and the root mean square error (RMSE) decreased by 0.16 at the maximum. Further, we identified the robustness and adaptability of deep neural network (DNN) model for time-series LAI estimation. The results substantiated the effectiveness of DNN in dealing with nonlinear problems and alleviating spectral saturation with higher CC and lower RMSE and relative RMSE (rRMSE) at whole growth-stages (CC= 0.91, RMSE= 0.28, and rRMSE= 7.18%) and vegetative stage (CC=0.94, RMSE=0.24, and rRMSE= 5.81%). In conclusion, the EFF method proposed in this study is competent in constructing time-series synthetic images for near real-time or post-growth applications. Moreover, the DNN model shows the potential for LAI estimation at whole growth-stages. This research not only contributes to dense time-series remote sensing-based applications (e.g., land cover change monitoring, yield forecast), but also is valuable for high-spatial-precision farmland management. • A framework to enhance spatiotemporal fusion methods was constructed for near real-time or post-growth applications. • The Enhanced-Fit-FC (EFF) method provided more accurate reflectance and spatial details compared with Fit-FC. • The bidirectional prediction of EFF (Bi-EFF) contributed to capturing land cover changes. • The DNN model was potential in alleviating spectral saturation for LAI estimation at whole growth-stages.

Toward an adaptive procedure for multi-frequency categorical loudness scaling: A Monte Carlo simulation study

To capture a listener’s loudness perception profile, categorical loudness scaling (CLS) is typically repeated at various frequencies. The current study aims to develop psychophysical procedures that enable simultaneous estimation of loudness growth across frequencies. For these procedures, the listener hears a pure-tone stimulus provides a categorical rating (“Soft,” “Loud,” etc.) on each trial. After a response is collected, the procedures update a model of the loudness profile and leverage the model to optimize the stimulus (i.e., level and frequency) for the next trial. The modified slope-adaptive procedure selects the stimulus from a uniform distribution spanning the model-predicted dynamic range, while the modified maximum expected information (MEI) procedure optimizes the stimulus based on an entropy metric. Monte Carlo simulations were conducted to evaluate the two procedures using a database that consists of CLS data collected from 148 listeners at Boys Town National Research Hospital. For each listener, the two procedures were run based on responses simulated using their known loudness profiles (i.e., the ground truth). Both procedures were able to estimate the loudness profile close to the ground truth, with a root-mean-square error (RMSE) of about 6 dB after 100 trials. Below 100 trials, the modified MEI procedure showed a lower RMSE.

Prediction of carbon sequestration of biochar produced from biomass pyrolysis by artificial neural network

Biochar has demonstrated its potential in carbon sequestration to mitigate climate change. The fact that biochar can be produced from the slow pyrolysis of numerous potential biomasses has also attracted attention. However, research on predicting the effect of different pyrolysis conditions and a wide variety of biomass on carbon sequestration potential of biochar is extremely limited. In order to estimate the feasibility and performance of biomass to produce biochar with good carbon sequestering potential, it is crucial to predict the quantity and surface property of the resulting biochar based on composition of different biomass and pyrolysis conditions. In this study, the optimum feedforward artificial neural network (FANN) models were developed for 3 analytical cases based on pyrolysis conditions, proximate and ultimate analysis of biomass feedstocks to predict yield and BET surface area of the resulting biochar. These aspects can reflect the carbon sequestration potential of the resulting biochar. Data collected from literature were used to train, validate and test the ANN model for all analytical cases in this study. Levenberg-Marquardt (LM) algorithm was found to be the most suitable training algorithm for each analytical case, as it had shown satisfactory performance with low mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean square error (RMSE). The ANN models showed good performances with correlation coefficient (R) more than 0.96, coefficient of determination (R2) more than 0.92 and low MAPE, MAE and RMSE, thus indicating good alignment between the predicted and actual biochar yield and BET surface area. The relative importance of input was also determined using Garson’s equation and connection weight approach to investigate the impact of the input variables on biochar yield and BET surface area. The proximate analysis of biomass was found to have significant impact on biochar yield whereas ultimate analysis of biomass had large influence on biochar BET surface area.

Deep Neural Networks Can Accurately Detect Blood Loss and Hemorrhage Control Task Success From Video.

Deep neural networks (DNNs) have not been proven to detect blood loss (BL) or predict surgeon performance from video. To train a DNN using video from cadaveric training exercises of surgeons controlling simulated internal carotid hemorrhage to predict clinically relevant outcomes. Video was input as a series of images; deep learning networks were developed, which predicted BL and task success from images alone (automated model) and images plus human-labeled instrument annotations (semiautomated model). These models were compared against 2 reference models, which used average BL across all trials as its prediction (control 1) and a linear regression with time to hemostasis (a metric with known association with BL) as input (control 2). The root-mean-square error (RMSE) and correlation coefficients were used to compare the models; lower RMSE indicates superior performance. One hundred forty-three trials were used (123 for training and 20 for testing). Deep learning models outperformed controls (control 1: RMSE 489 mL, control 2: RMSE 431 mL, R2 = 0.35) at BL prediction. The automated model predicted BL with an RMSE of 358 mL (R2 = 0.4) and correctly classified outcome in 85% of trials. The RMSE and classification performance of the semiautomated model improved to 260 mL and 90%, respectively. BL and task outcome classification are important components of an automated assessment of surgical performance. DNNs can predict BL and outcome of hemorrhage control from video alone; their performance is improved with surgical instrument presence data. The generalizability of DNNs trained on hemorrhage control tasks should be investigated.

Comparison of machine learning methods for photovoltaic power forecasting based on numerical weather prediction

The increase of the worldwide installed photovoltaic (PV) capacity and the intermittent nature of the solar resource highlights the importance of power forecasting for the grid integration of the technology. This study compares 24 machine learning models for deterministic day-ahead power forecasting based on numerical weather predictions (NWP), tested for two-year-long 15-min resolution datasets of 16 PV plants in Hungary. The effects of the predictor selection and the benefits of the hyperparameter tuning are also evaluated. The results show that the two most accurate models are kernel ridge regression and multilayer perceptron with an up to 44.6% forecast skill score over persistence. Supplementing the basic NWP data with Sun position angles and statistically processed irradiance values as the inputs of the learning models results in a 13.1% decrease of the root mean square error (RMSE), which underlines the importance of the predictor selection. The hyperparameter tuning is essential to exploit the full potential of the models, especially for the less robust models, which are prone to under or overfitting without proper tuning. The overall best forecasts have a 13.9% lower RMSE compared to the baseline scenario of using linear regression. Moreover, the power forecasts based on only daily average irradiance forecasts and the Sun position angles have only a 1.5% higher RMSE than the best scenario, which demonstrates the effectiveness of machine learning even for limited data availability. The results of this paper can support both researchers and practitioners in constructing the best data-driven techniques for NWP-based PV power forecasting.

Identification of Flexural Modulus and Poisson’s Ratio of Fresh Femoral Bone Based on a Finite Element Model

Finite element analysis (FEA) is increasingly applied to medicine because it could increase accuracy and rapid outcomes. However, there is a lack of the method to determine Young’s modulus and Poisson’s ratio for fresh femoral bone and the mathematical principle’s optimization for calculating nonuniform configuration. This study aimed to investigate the surrogate model for the optimization method to determine Young’s modulus and Poisson’s ratio of the fresh femoral bone. Young’s modulus and Poisson’s ratio obtained 20 ranked pairs by the Latin hypercube sampling method. The values ​​were calculated in the finite element for root mean square error (RMSE) and were then used for solutions by a quadratic function, radial basis function (RBF), and Kriging (KG). The lowest RMSE value was 0.1518 for the RBF method, with the young’s modulus at 304.4756 and the Poisson’s ratio at 0.3334. The current study identified the RBF technique to determine the properties of the femoral bone. Moreover, the RBF procedure might apply to other long bones because of the comparable nonuniform configuration.

Spatial analysis of seasonal precipitation using various interpolation methods in the Euphrates basin, Turkey

It is vital to accurately map the spatial distribution of precipitation, which is widely used in many fields such as hydrology, climatology, meteorology, ecology, and agriculture. This study aimed to reveal the spatial distribution of seasonal, long-term average precipitation in the Euphrates Basin with various interpolation methods. For this reason, Simple Kriging, Ordinary Kriging, Universal Kriging, Ordinary CoKriging, Empirical Bayesian Kriging, Radial Basis Functions (Completely Regularized Spline, Thin Plate Spline, Multiquadratic, Inverse Multiquadratic, Spline with Tensor), Local Polynomial Interpolation, Global Polynomial Interpolation, and Inverse Distance Weighting methods have been applied in the Geographical Information Systems environment. Long-term seasonal precipitation averages between 1966 and 2017 are presented as input for predicting precipitation maps. The accuracy of the precipitation prediction maps was based on linear regression analysis, root mean square error (RMSE), mean absolute error (MAE), correlation coefficient (R), and determination coefficient (R2) values obtained from the cross-validation tests. The most suitable method was chosen for the interpolation method that gives the lowest RMSE, MAE, and the largest R and R2. As a result of the study, Ordinary CoKriging in spring and winter precipitation, Local Polynomial Interpolation in summer precipitation, and Ordinary Kriging in autumn precipitation were the most appropriate estimation methods.

Reservoir Performance Prediction in Steam Huff and Puff Injection Using Proxy Modelling

Steam huff and puff injection is one of the thermal EOR methods in which steam is injected in a cyclical manner alternating with oil production. The cost and time inefficiency problem of reservoir simulation persists in the design of a steam huff and puff injection scheme. Building predictive proxy models is a suitable solution to deal with this issue. In this study, predictive models of the steam huff and puff injection method were developed using two machine learning algorithms, comprising conventional polynomial regression and an artificial neural network algorithm. Based on a one-well cylindrical synthetic reservoir model, 6043 experiment cases with 28 input parameter values were generated and simulated. Outputs from the results such as cumulative oil production, maximum oil production rate and oil rate at cycle end were extracted from each simulation case to build the predictive model. Reservoir properties that could change after an injection cycle were also modeled. The developed models were evaluated based on the fitting performance from the R-square value, the mean absolute error (MAE) value and the root mean square error (RMSE) value. Then, Sobol analysis was conducted to determine the significance of each parameter in the model. The results show that neural network models have better performance compared to the polynomial regression models. Neural network models have an average R-square value of over 0.9 and lower MAE and RMSE values than the polynomial regression model. The result of applying the Sobol analysis also indicates that initial reservoir water saturation and oil viscosity are the most important parameters for predicting reservoir production performance.

Assessing the health estimation capacity of air pollution exposure prediction models

BackgroundThe era of big data has enabled sophisticated models to predict air pollution concentrations over space and time. Historically these models have been evaluated using overall metrics that measure how close predictions are to monitoring data. However, overall methods are not designed to distinguish error at timescales most relevant for epidemiologic studies, such as day-to-day errors that impact studies of short-term health associations.MethodsWe introduce frequency band model performance, which quantifies health estimation capacity of air quality prediction models for time series studies of air pollution and health. Frequency band model performance uses a discrete Fourier transform to evaluate prediction models at timescales of interest. We simulated fine particulate matter (PM2.5), with errors at timescales varying from acute to seasonal, and health time series data. To compare evaluation approaches, we use correlations and root mean squared error (RMSE). Additionally, we assess health estimation capacity through bias and RMSE in estimated health associations. We apply frequency band model performance to PM2.5 predictions at 17 monitors in 8 US cities.ResultsIn simulations, frequency band model performance rates predictions better (lower RMSE, higher correlation) when there is no error at a particular timescale (e.g., acute) and worse when error is added to that timescale, compared to overall approaches. Further, frequency band model performance is more strongly associated (R2 = 0.95) with health association bias compared to overall approaches (R2 = 0.57). For PM2.5 predictions in Salt Lake City, UT, frequency band model performance better identifies acute error that may impact estimated short-term health associations.ConclusionsFor epidemiologic studies, frequency band model performance provides an improvement over existing approaches because it evaluates models at the timescale of interest and is more strongly associated with bias in estimated health associations. Evaluating prediction models at timescales relevant for health studies is critical to determining whether model error will impact estimated health associations.

Gaming behavior and brain activation using functional near-infrared spectroscopy, Iowa gambling task, and machine learning techniques.

IntroductionThe current study investigates the utilization and performance of machine learning (ML) algorithms in the cognitive task of finding the correlation between numerical parameters of the human brain activation during gaming. We hypothesize that our integrated feature extraction platform is able to distinguish between different psychosomatic conditions in the gaming process as measured by the functional near‐infrared brain imaging technique.MethodsFor demonstration, the decision‐making process was constructed in the experiment environment that combined gaming simulator, such as the Iowa Gaming Task (IGT), with functional near‐infrared spectroscopy (fNIRS) as the neuroimaging technique. Features of fNIRS levels were extracted, averaged, and synchronized by time with the IGT dataset to predict the task score inside ML algorithms, such as multiple regression, classification and regression trees, support vector machine, artificial neural network, and random forest. For findings validation, the experiment data were resampled by training and testing sets. Further, a training dataset was used to train the ML algorithms, and prediction accuracy was estimated by repeated cross‐validation methods and compared by R squared and root mean square error (RMSE). The model with the best accuracy was used with the testing dataset and finalized the experiment.ResultsDuring the experiment, the highest correlation was identified in the fourth block between the oxy‐hemoglobin signal and IGT score in average value (0.24) and signal feature (0.57). Such relationship is due to block 4 characterization as “conceptual” period when participants task experience reaches the maximum, and rewards raise accordingly. Simultaneously, ML algorithms, constructed based on training data set, demonstrate acceptable performance, and RMSE as the primary performance metric dynamically increases from block 1 to block 5, from the state of uncertainty and unknown to the certainty and risky. In contrast, R squared decreases during the same transition. In most IGT blocks, the best fitted model was determined as support vector machine with radial bases function kernel, and predictions were made with the highest accuracy (lowest RMSE) than in training models.ConclusionObtained findings showed the applicability and capability of ML models as a powerful technique to evaluate the cognitive neuroimaging task result. Moreover, in terms of features it was identified that the hemodynamic response reacts to the acceleration decision‐making process and raises more significance than it was observed before.

Modelling the spectral shape of continuous-wave lidar measurements in a turbulent wind tunnel

Abstract. This paper describes the development of a theoretical model for the turbulence spectrum measured by a short-range, continuous-wave lidar (light detection and ranging). The lidar performance was assessed by measurements conducted with two WindScanners in an open-jet wind tunnel equipped with an active grid, for a range of different turbulent wind conditions. A hot-wire anemometer is used as reference to assess the lidar's measured statistics, time series and spectra. In addition to evaluating the statistics, the correlation between the time series and the root-mean-square error (RMSE) on the wind speed, the turbulence spectrum measured by the lidar is compared with a modelled spectrum. The theoretical spectral model is applied in the frequency domain, using a Lorentzian filter in combination with Taylor's frozen turbulence hypothesis for the probe length averaging effect and an added white noise term, evaluated by qualitatively matching the lidar measurement spectrum. High goodness-of-fit coefficients and low RMSE values between the hot wire and WindScanner were observed for the measured time series. The correlation showed an inverse relationship with the prevalent turbulence intensity in the flow for cases with a comparable power spectrum shape. Larger flow structures can be captured more accurately by the lidar, whereas small-scale turbulent flow structures are partly filtered out as a result of the lidar's probe volume averaging effect. It is demonstrated that an accurate way to define the cut-off frequency at which the lidar's power spectrum starts to deviate from the hot-wire reference spectrum is the frequency at which the coherence drops below 0.5. This coherence-based cut-off frequency increases linearly with the mean wind speed and is generally an order of magnitude lower than the probe length equivalent cut-off frequency, estimated according to a simple model based on the full width at half maximum (FWHM) of the laser beam intensity along the line of sight and assuming Taylor's frozen turbulence hypothesis. A convincing match between the modelled and the measured WindScanner power spectrum was found for various different cases, which confirmed that the deviation of the lidar's measured power spectrum in the higher frequency range can be analytically explained and modelled as a combination of a Lorentzian-shaped intensity function and white noise in the lidar measurement. Although the models were developed on the basis of wind tunnel measurements, they should be applicable to atmospheric boundary layer field measurements as well.

An ensemble approach to deep‐learning‐based wireless indoor localization

The authors investigate the use of deep learning in wireless indoor localization to address the shortcomings of the existing range-based (e.g. trilateration and triangulation) and range-free (e.g. fingerprinting) localization. Instead of relying on geometric models and hand-picked features, deep learning can automatically extract the relationship between the observed data and the target's location. Nevertheless, a deep neural network (DNN) model providing a satisfactory accuracy might perform differently when it is retrained in the deployment. To mitigate this issue, the authors propose an ensemble method where DNN models obtained from multiple training sessions are combined to locate the target. In the authors' evaluation, several DNN models are trained on the data, which consists of the received signal strength (RSS), angle of arrival (AOA), and channel state information (CSI), used in the existing hybrid RSS/AOA and RSS/CSI fingerprinting, and their root-mean-square error (RMSE) values are compared accordingly. The results show that the proposed method achieves the lower RMSE than the existing methods, and the RMSE can be lowered by up to 1.47 m compared with the ones obtained from a single model. Moreover, for some DNN models, the RMSE values are even lower than the minimum RMSE obtained by their single-model counterparts.

Prediction of Multivariate Air Quality Time Series Data using Long Short-Term Memory Network

Malaysia often suffers from haze problems almost every year. Therefore, there is a need for good air quality forecasting model for monitoring and management purposes. In this study, the air quality model based on the Long Short-Term Memory Network (LSTM) and Auto-Regressive Integrated Moving Average (ARIMA) was developed. The prediction of the particulate matter 10 micrometres or less in diameter (PM10) in Malaysia could be made from both models, and their performance was compared. The purpose of comparison between the two models was to determine the most suitable model to use in predicting PM10 since it is the dominant pollutant in Malaysia most of the time, especially during the haze period. This study used air quality data obtained from the Department of Environment Malaysia from July 2017 to June 2019. The results showed that forecasting for PM10 using multivariate LSTM model was better than the univariate LSTM model and univariate ARIMA model with the lowest root mean square error (RMSE) for those selected stations. The model with a lower RMSE value means better models and provide higher accuracy in forecasting for PM10.

Predicting Canopy Chlorophyll Content in Sugarcane Crops Using Machine Learning Algorithms and Spectral Vegetation Indices Derived from UAV Multispectral Imagery

The use of satellite-based Remote Sensing (RS) is a well-developed field of research. RS techniques have been successfully utilized to evaluate the chlorophyll content for the monitoring of sugarcane crops. This research provides a new framework for inferring the chlorophyll content in sugarcane crops at the canopy level using unmanned aerial vehicles (UAVs) and spectral vegetation indices processed with multiple machine learning algorithms. Studies were conducted in a sugarcane field located in Sugarcane Research Institute (SRI, Uda Walawe, Sri Lanka), with various fertilizer applications over the entire growing season from 2020 to 2021. An UAV with multispectral camera was used to collect the aerial images to generate the vegetation indices. Ground measurements of leaf chlorophyll were used as indications for fertilizer status in the sugarcane field. Different machine learning (ML) algorithms were used ground-truthing data of chlorophyll content and spectral vegetation indices to forecast sugarcane chlorophyll content. Several machine learning algorithms such as MLR, RF, DT, SVR, XGB, KNN and ANN were applied in two ways: before feature selection (BFS) by training the algorithms with all twenty-four (24) vegetation indices with five (05) spectral bands and after feature selection (AFS) by training algorithms with fifteen (15) vegetation indices. All the algorithms with both BFS and AFS methods were compared with an estimated coefficient of determination (R2) and root mean square error (RMSE). Spectral indices such as RVI and DVI were shown to be the most reliable indices for estimating chlorophyll content in sugarcane fields, with coefficients of determination (R2) of 0.94 and 0.93, respectively. XGB model shows the highest validation score (R2) and lowest RMSE in both methods of BFS (0.96 and 0.14) and AFS (0.98 and 0.78), respectively. However, KNN and SVR algorithms show the lowest validation accuracy than other models. According to the results, the AFS validation score is higher than BFS in MLR, SVR, XGB and KNN. Even though, validation score of the ANN model is decreased in AFS. The findings demonstrated that the use of multispectral UAV could be utilized to estimate chlorophyll content and measure crop health status over a larger sugarcane field. This methodology will aid in real-time crop nutrition management in sugarcane plantations by reducing the need for conventional measurement of sugarcane chlorophyll content.

Prediction of nickel concentration in peri-urban and urban soils using hybridized empirical bayesian kriging and support vector machine regression

Soil pollution is a big issue caused by anthropogenic activities. The spatial distribution of potentially toxic elements (PTEs) varies in most urban and peri-urban areas. As a result, spatially predicting the PTEs content in such soil is difficult. A total number of 115 samples were obtained from Frydek Mistek in the Czech Republic. Calcium (Ca), magnesium (Mg), potassium (K), and nickel (Ni) concentrations were determined using Inductively Coupled Plasma Optical Emission Spectroscopy. The response variable was Ni, while the predictors were Ca, Mg, and K. The correlation matrix between the response variable and the predictors revealed a satisfactory correlation between the elements. The prediction results indicated that support vector machine regression (SVMR) performed well, although its estimated root mean square error (RMSE) (235.974 mg/kg) and mean absolute error (MAE) (166.946 mg/kg) were higher when compared with the other methods applied. The hybridized model of empirical bayesian kriging-multiple linear regression (EBK-MLR) performed poorly, as evidenced by a coefficient of determination value of less than 0.1. The empirical bayesian kriging-support vector machine regression (EBK-SVMR) model was the optimal model, with low RMSE (95.479 mg/kg) and MAE (77.368 mg/kg) values and a high coefficient of determination (R2 = 0.637). EBK-SVMR modelling technique output was visualized using a self-organizing map. The clustered neurons of the hybridized model CakMg-EBK-SVMR component plane showed a diverse colour pattern predicting the concentration of Ni in the urban and peri-urban soil. The results proved that combining EBK and SVMR is an effective technique for predicting Ni concentrations in urban and peri-urban soil.

A Suitable Retrieval Algorithm of Arctic Snow Depths with AMSR-2 and Its Application to Sea Ice Thicknesses of Cryosat-2 Data

Arctic sea ice and snow affect the energy balance of the global climate system through the radiation budget. Accurate determination of the snow cover over Arctic sea ice is significant for the retrieval of the sea ice thickness (SIT). In this study, we developed a new snow depth retrieval method over Arctic sea ice with a long short-term memory (LSTM) deep learning algorithm based on Operation IceBridge (OIB) snow depth data and brightness temperature data of AMSR-2 passive microwave radiometers. We compared climatology products (modified W99 and AWI), altimeter products (Kwok) and microwave radiometer products (Bremen, Neural Network and LSTM). The climatology products and altimeter products are completely independent of the OIB data used for training, while microwave radiometer products are not completely independent of the OIB data. We also compared the SITs retrieved from the above different snow depth products based on Cryosat-2 radar altimeter data. First, the snow depth spatial patterns for all products are in broad agreement, but the temporal evolution patterns are distinct. Snow products of microwave radiometers, such as Bremen, Neural Network and LSTM snow depth products, show thicker snow in early winter with respect to the climatology snow depth products and the altimeter snow depth product, especially in the multiyear ice (MYI) region. In addition, the differences in all snow depth products are relatively large in the early winter and relatively small in spring. Compared with the OIB and IceBird observation data (April 2019), the snow depth retrieved by the LSTM algorithm is better than that retrieved by the other algorithms in terms of accuracy, with a correlation of 0.55 (0.90), a root mean square error (RMSE) of 0.06 m (0.05 m) and a mean absolute error (MAE) of 0.05 m (0.04 m). The spatial pattern and seasonal variation of the SITs retrieved from different snow depths are basically consistent. The total sea ice decreases first and then thickens as the seasons change. Compared with the OIB SIT in April 2019, the SIT retrieved by the LSTM snow depth is superior to that retrieved by the other SIT products in terms of accuracy, with the highest correlation of 0.46, the lowest RMSE of 0.59 m and the lowest MAE of 0.44 m. In general, it is promising to retrieve Arctic snow depth using the LSTM algorithm, but the retrieval of snow depth over MYI still needs to be verified with more measured data, especially in early winter.

Modeling Soil Temperature for Different Days Using Novel Quadruplet Loss-Guided LSTM.

Soil temperature (Ts), a key variable in geosciences study, has generated growing interest among researchers. There are many factors affecting the spatiotemporal variation of Ts, which poses immense challenges for the Ts estimation. To enrich processing information on loss function and achieve better performance in estimation, the paper designed a new long short-term memory model using quadruplet loss function as an intelligence tool for data processing (QL-LSTM). The model in this paper combined the traditional squared-error loss function with distance metric learning between the sample features. It can zoom analyze the samples accurately to optimize the estimation accuracy. We applied the meteorological data from Laegern and Fluehli stations at 5, 10, and 15 cm depth on the 1st, 5th, and 15th day separately to verify the performance of the proposed soil temperature estimation model. Meanwhile, this paper inputs the variables into the proposed model including radiation, air temperature, vapor pressure deficit, wind speed, air pressure, and past Ts data. The performance of the model was tested by several error evaluation indices, including root mean square error (RMSE), mean absolute error (MAE), Nash-Sutcliffe model efficiency coefficient (NS), Willmott Index of Agreement (WI), and Legates and McCabe index (LMI). As the test results at different soil depths show, our model generally outperformed the four existing advanced estimation models, namely, backpropagation neural networks, extreme learning machines, support vector regression, and LSTM. Furthermore, as experiments show, the proposed model achieved the best performance at the 15 cm depth of soil on the 1st day at Laegern station, which achieved higher WI (0.998), NS (0.995), and LMI (0.938) values, and got lower RMSE (0.312) and MAE (0.239) values. Consequently, the QL-LSTM model is recommended to estimate daily Ts profiles estimation on the 1st, 5th, and 15th days.