Relative Root Mean Square Error Research Articles

Sequency encoding single pixel spectroscopy based on Hadamard transform.

Single pixel spectroscopy based on Hadamard transform (SPS-HT) has been applied widely because of its capability of wavelength multiplexing and associated advantage in signal-to-noise ratio. In this paper, we propose a sequency encoding single pixel spectroscopy (SESPS) based on two-dimensional (2D) masks for concurrent coding of all Hadamard coefficients instead of one-dimensional (1D) Hadamard masks (only coding one coefficient at a time) widely used in the traditional SPS-HT. Moreover, each Hadamard coefficient is coded along the time dimension with a different sequency value such that the alternating current (AC) measurements of the time-domain signal can be used to reconstruct all Hadamard coefficients simultaneously, which reduces the influence of noise and dramatically speeds up data acquisition. We demonstrate that the SESPS with 32 spectral channels can accelerate spectral measurements from white light sources and fluorescence particles by around 14 times and 70 times, respectively, compared to measurements using a commercial spectrometer when the relative root mean square error (RMSE) is around 3% or smaller. The acceleration factors can be boosted by an extra 4 times when only eight spectral channels are used to achieve a compression ratio of 4:1, in which the relative RMSEs change only marginally. Compared to our previous SPS-HT, this new scheme can increase the speed by three orders of magnitude. This technique is expected to be useful in applications requiring high-speed spectral measurements such as the spectral flow cytometry and on-site medical diagnosis using fluorescence or Raman spectroscopy.

Investigation of Two Prediction Models of Maximum Usable Frequency for HF Communication Based on Oblique- and Vertical-Incidence Sounding Data

As one of the key technologies of HF communication, the maximum usable frequency (MUF) prediction method has been widely discussed. To experimentally confirm the reliability of commonly used MUFs prediction models for high-frequency communication, we have compared maximum observed frequencies (MOFs) and predicted MUFs to assess the accuracy of two typical prediction models. The root-mean-square error (RMSE) and relative RMSE (RRMSE) between oblique sounding MOFs and the predicted MUFs were used to assess the model’s accuracy. The oblique sounding path was from Changchun to Jinyang, and the vertical-sounding ionosonde was located in Beijing, which was approximately the midpoint of the oblique sounding circuit. The statistical analysis results show that: (a) the trend of prediction results from the Lockwood and the Istituto Nazionale di Geofisica e Vulcanologia (INGV) model are in good agreement with the observations: the mean RMSE and RRMSE of the INGV model are less than those of the Lockwood model; (b) in the four different periods (sunrise, daytime, sunset, and nighttime) of the whole day, the maximum difference of RMSE between the Lockwood and INGV model is 0.14 MHz (the INGV performs better than the LWM), with the corresponding differences of RRMSE being 0.31% at sunrise and 0.68% at daytime; (c) in the four seasons of spring, summer, autumn, and winter, the minimum RMSE values of the Lockwood and INGV models are 1.51 MHz and 1.37 MHz, respectively, which are obtained in winter, and the corresponding RRMSEs are 11.47% and 11.79%, respectively; (d) in the high and low solar activity epochs, the mean RMSEs of the Lockwood and INGV models are 1.63 MHz, and 1.54 MHz, with corresponding mean RRMSE values of 11.47% and 11.55%. In conclusion, the INGV model is more suitable for MUF prediction over Beijing and its adjacent mid-latitude regions from the RMSE comparison of the two models.

Above-ground biomass estimation in a Mediterranean sparse coppice oak forest using Sentinel-2 data

Implementing a scheduled and reliable estimation of forest characteristics is important for the sustainable management of forests. This study aimed at evaluating the capability of Sentinel-2 satellite data to estimate above-ground biomass (AGB) in coppice forests of Persian oak (Quercus brantii var. persica) located in Western Iran. To estimate the AGB, field data collection was implemented in 80 square plots (40×40 m, area of 1600 m2). Two diameters of the crown were measured and used to calculate the AGB of each tree based on allometric equations. Then, the performance of satellite data in estimating the AGB was evaluated for the area of study using the field-based AGB (dependent variable) as well as the spectral band values, spectrally-derived vegetation indices (independent variables) and four machine learning (ML) algorithms: Multi-Layer Perceptron Artificial Neural Network (MLPNN), k-Nearest Neighbor (kNN), Random Forest (RF), and Support Vector Regression (SVR). A five-fold cross-validation was used to verify the effectiveness of models. Examination of the Pearson’s correlation coefficient between AGB and the extracted ‎values showed that IPVI and NDVI vegetation indices had the highest correlation with AGB (r = 0.897). The results indicated that the MLPNN algorithm was the best ML option (RMSE = 1.71 t ha-1; MAE = 1.37 t ha-1; relative RMSE = 24.75%; R2 = 0.87) in estimating the AGB, providing new insights on the capability of remotely sensed-based AGB modeling of sparse Mediterranean forest ecosystems in an area with limited number of field sample plots.

Comparative study of advanced computational techniques for estimating the compressive strength of UHPC

The effect of raw materials on the compressive strength of concrete is a complex process, especially in the case of ultra-high-performance concrete (UHPC), where a higher number of inter-dependent parameters are involved in the strength development. In this era of digitalization, advanced machine learning methods are used to predict the material's mechanical characteristics because of their superior performance compared to conventional and nonlinear statistical regression models. Thus, the goal of the current study is to estimate the compressive strength of UHPC from the designed raw materials using advanced machine learning techniques. The compressive strength of UHPC is predicted from the 14 input parameters, i.e., cement, fly ash, slag, silica fume, nano-silica, limestone powder, sand, coarse aggregate, quartz powder, water, superplasticizer, PE fiber, steel fiber, and curing time. A total of eight machine learning models were compared that include multi-layer perceptron neural network (MLPNN), MLPNN Bootstrap aggregating (MLPNN-BA), MLPNN adaptive boosting (MLPNN-AB), Gradient boosting (GB), Decision tree (DT), DT Bootstrap aggregating (DT-BA), DT adaptive boosting (DT-AB) and Random Forest (RNF). The validation and performance evaluation of the above models were checked by using K-fold cross-validation, mean absolute error (MAE), root mean square error (RSME), coefficient of determination (R2), relative root mean square error (RRMSE), performance index (PI), and Nash Sutcliffe efficiency (NSE). The optimal model was selected based on the results of all statistical checks. It was found the ensembled machine learning models especially decision tree-based models outperform the neural network-based models with higher accuracy and low error. Thus, the recommended machine learning model is random forest having superior prediction capacity followed by DT Bootstrap aggregating and DT adaptive boosting.

The electricity consumption forecast: Adopting a hybrid approach by deep learning and ARIMAX-GARCH models

Modeling and forecasting electricity consumption (EC) help industry managers make better strategic decisions. In this study, a hybrid approach for predicting EC is proposed which first EC is decomposed into approximate and detail parts based on wavelet transfer (WT). Next, the modeling of wavelet components is accomplished based on the adaptive WT (AWT)-long short-term memory (LSTM) and autoregressive integrated moving average with explanatory variable (ARIMAX)-generalized autoregressive conditional heteroscedasticity (GARCH) type models. The key concept of AWT–LSTM is adding an optimal set of adaptive layers through which AWT–LSTM can automatically concentrate on various frequency components based on the dynamic evolution of the input sequence revealing the multi-frequency pattern of EC. Then, the components are predicted separately and eventually, the results are combined nonlinearly through random forest (RF) algorithm. Besides the lag calendar setting, the affecting exogenous variables are applied on the electricity consumption in the structure of hybrid models which include humidity, water precipitation, solar radiation, effective temperature, wind direction, and wind speed. To measure the predictive power, the proposed models were tested using a 24 h ahead forecasting horizon using a sample rate of 30 min in two different seasons (winter and summer). The empirical results indicate the hybrid model of AWT–LSTM-ARIMAX-exponential (EGARCH) in mean-RF with root mean square error (RMSE) = 1.536, mean absolute percentage error (MAPE) = 1.276, normalized mean absolute error (NMAE) = 0.0116, normalized RMSE (NRMSE) = 0.0143, relative RMSE (RRMSE) = 0.1906, Theil’s inequality coefficient (TIC) = 0.01025, sum of squared errors (SSE) = 113.2 values in summer test and RMSE = 1.249, MAPE = 0.869, NMAE = 0.0078, NRMSE = 0.0092, RRMSE = 0.1223, TIC = 0.00533, SSE = 74.9 values in winter test outperforms the benchmark models (WT-feedforward neural network (FFNN)-RF, WT-LSTM-RF, WT-exponential smoothing (ETS)-RF, and WT-seasonal ARIMA with explanatory variable (SARIMAX)-GARCH types models) and alternative filtering methods (Curvelet, Ridgelet, and Contourlet). The results suggest that employing the WT for the decomposition of data and RF for the nonlinear aggregation of sub-signals (predicted values of detail and approximate parts) can significantly improve the prediction power of predictive models.

Estimating Leaf Chlorophyll Content of Moso Bamboo Based on Unmanned Aerial Vehicle Visible Images

Leaf chlorophyll content is an important indicator of the physiological and ecological functions of plants. Accurate estimation of leaf chlorophyll content is necessary to understand energy, carbon, and water exchange between plants and the atmosphere. The leaf chlorophyll content index (CCI) of 109 Moso bamboo samples (19 for training data, 19 for validation data, and 71 for extrapolation data) was measured from December 2019 to May 2021, while their corresponding red–green–blue (RGB) images were acquired using an unmanned aerial vehicle (UAV) platform. A method for estimating leaf CCI based on constructing relationships between field leaf CCI measurements and UAV RGB images was evaluated. The results showed that a modified excess blue minus excess red index and 1.4 × H-S in the hue–saturation–value (HSV) color space were the most suitable variables for estimating the leaf CCI of Moso bamboo. No noticeable difference in accuracy between the linear regression model and backpropagation neural network (BPNN) model was found. Both models performed well in estimating leaf CCI, with an R2 > 0.85 and relative root mean square error (RMSEr) < 15.0% for the validation data. Both models failed to accurately estimate leaf CCI during the leaf-changing period (April to May in off-year), with the problems being overestimation in low leaf CCI and underestimation in high leaf CCI values. At a flight height of 120 m and illumination between 369 and 546 W/m2, the CCI for an independent sample dataset was accurately estimated by the models, with an R2 of 0.83 and RMSEr of 13.78%. Flight height and solar intensity played a role in increasing the generality of the models. This study provides a feasible and straightforward method to estimate the leaf CCI of Moso bamboo based on UAV RGB images.

Inversion of Coniferous Forest Stock Volume Based on Backscatter and InSAR Coherence Factors of Sentinel-1 Hyper-Temporal Images and Spectral Variables of Landsat 8 OLI

Forest stock volume (FSV) is a basic data source for estimating forest carbon sink. It is also a crucial parameter that reflects the quality of forest resources and forest management level. The use of remote sensing data combined with a support vector regression (SVR) algorithm has been widely used in FSV estimation. However, due to the complexity and spatial heterogeneity of the forest biological community, in the FSV high-value area with dense vegetation, the optical re-mote sensing variables tend to be saturated, and the sensitivity of synthetic aperture radar (SAR) backscattering features to the FSV is significantly reduced. These factors seriously affect the ac-curacy of the FSV estimation. In this study, Landsat 8 (L8) Operational Land Imager multispectral images and C-band Sentinel-1 (S1) hyper-temporal SAR images were used to extract three re-mote sensing feature datasets: spectral variables (L8), backscattering coefficients (S1), and inter-ferometric SAR factors (S1-InSAR). We proposed a feature selection method based on SVR (FS-SVR) and compared the FSV estimation performance of FS-SVR and stepwise regression analysis (SRA) on the aforementioned three remote sensing feature datasets. Finally, an estima-tion model of coniferous FSV was constructed using the SVR algorithm in Wangyedian Forest Farm, Inner Mongolia, China, and the spatial distribution map of coniferous FSV was predicted. The experimental results show the following: (1) The coherence amplitude and DSM data ob-tained based on S1 images contain information relat-ed to forest canopy height, and the hy-per-temporal S1 image data significantly enrich the diversity of S1-InSAR feature factors. There-fore, the S1-InSAR dataset has a better FSV response than remote sensing factors such as the S1 backscattering coefficient and L8 vegetation index, and the corresponding root mean square er-ror (RMSE) and relative RMSE (rRMSE) values reached 47.6 m3/ha and 20.9%, respectively. (2) The integrated dataset can provide full play to the synergy of the L8, S1, and S1-InSAR remote sensing data. Its RMSE and rRMSE values are 44.3 m3/ha and 19.4% respectively. (3) The proposed FS-SVR method can better select remote sensing variables suitable for FSV estimation than SRA. The average value of the rRMSE (23.17%) based on the three datasets was 13.8% lower than that of the SRA method (26.87%). This study provides new insights into forest FSV retrieval based on active and passive multisource remote sensing joint data.

Prediction of the Nitrogen, Phosphorus and Potassium Contents in Grape Leaves at Different Growth Stages Based on UAV Multispectral Remote Sensing

The rapid and accurate acquisition of nitrogen, phosphorus and potassium nutrient contents in grape leaves is critical for improving grape yields and quality and for industrial development. In this study, crop growth was non-destructively monitored based on unmanned aerial vehicle (UAV) remote sensing technology. Three irrigation levels (W1, W2 and W3) and four fertilization levels (F3, F2, F1 and F0) were set in this study, and drip irrigation fertilization treatments adopted a complete block design. A correlation analysis was conducted using UAV multispectral image data obtained from 2019 to 2021 and the field-measured leaf nitrogen content (LNC), leaf potassium content (LKC) and leaf phosphorus content (LPC) values; from the results, the vegetation indices (VIs) that were sensitive to LNC, LKC and LPC were determined. By combining spectral indices with partial least squares (PLS), random forest (RF), support vector machine (SVM) and extreme learning machine (ELM) machine-learning algorithms, prediction models were established. Finally, the optimal combinations of spectral variables and machine learning models for predicting LNC, LPC and LKC in each grape growth period were determined. The results showed that: (1) there were high demands for nitrogen during the new shoot growth and flowering periods, potassium was the main nutrient absorbed in the fruit expansion period, and phosphorus was the main nutrient absorbed in the veraison and maturity periods; (2) combining multiple spectral variables with the RF, SVM and ELM models could result in improved LNC, LPC and LKC predictions. The optimal prediction model determination coefficient (R2) derived during the new shoot growth period was above 0.65, and that obtained during the other growth periods was above 0.75. The relative root mean square error (RRMSE) of the above models was below 0.20, and the Willmott consistency index (WIA) was above 0.88. In conclusion, UAV multispectral images have good application effects when predicting nutrient contents in grape leaves. This study can provide technical support for accurate vineyard nutrient management using UAV platforms.

Motion Intention Prediction and Joint Trajectories Generation Toward Lower Limb Prostheses Using EMG and IMU Signals

Powered intelligent lower limb prostheses have been gaining interest as they provide functionality for walking on different terrains. This study proposes a hierarchical planner for intelligent lower limb prostheses based on sensor fusion and a central pattern generator (CPG). Electromyographic (EMG) and inertial measurement unit (IMU) signals were recorded and fused in the feature and decision levels. The high-level planner consists of cascade classifiers with a gait phase dependence. A secondary classifier for each stand phase and swing phase was developed to recognize five gait patterns. The mid-level planner was designed as a walking frequency estimator based on the Newton method. The low-level planner incorporates the CPG models, achieving the planning of lower limb joint trajectories. The proposed layered planner can recognize the users’ walking intention and gait speed in real-time, ensuring coordination between the joints of both legs. Eight healthy subjects were recruited, and the average accuracy of motion pattern recognition reached 99.13% and 99.39% for the standing and swing phases, respectively. The relative root mean square error (RMSE) of the walking frequency estimate was 2.3% under approximately 3.5 m/s gait speed. The promising results indicate that this method effectively predicts the continuous joint angle for lower limb prostheses.

Regional thermal analysis approach: A management tool for predicting water temperature metrics relevant for thermal fish habitat

The field of water resource management, including fisheries, is facing new challenges associated with climate change. This study sheds light on the modeling of water temperature indices (metrics) that describe critical thermal maxima of the Atlantic salmon (salmo salar). These thermal metrics include MaxWaterTmax (interannual mean of maximum summer temperature), MaxNumDay (interannual mean of the number of consecutive days with maximum water temperature > 25 °C and minimum water temperature > 20 °C). The latter is an important indicator to evaluate thermal variability. Three other parameters of a Gaussian function fitted to the interannual daily mean temperatures characterizing the thermal regime of 146 stations located in Eastern Canada were estimated. These three parameters are Gaussian_a (maximum of interannual daily mean temperature), Gaussian_b (mean duration of the warm period), and Gaussian_c (date of occurrence of the interannual maximum temperature). The classical Multiple linear regression model (MLR) and the non-linear Generalized additive model (GAM) were tested and compared to estimate the five thermal metrics. The regression-based approaches involve the identification of thermally homogeneous regions based on three approaches: hierarchical clustering analysis (HCA), regions of influence (ROI) as well as canonical correlation analysis (CCA). Then, the regional MLR and GAM models were applied within the delineated homogenous regions. Also, the regional models were compared to models encompassing all stations (i.e., one region). For each regional estimation model and each thermal metric, a set of optimal explanatory variables were selected using a forward stepwise procedure. The database consisted of 22 environmental predictors related to physiography, topography, climate, land cover and surface deposits. To assess performance of the models, the following statistical metrics were used: coefficient of determination R2, root mean square error (RMSE), bias, relative root mean square error (RRMSE) and percent bias (PBias). The results demonstrate that the non-linear GAM model was consistently better than the simpler MLR model for estimating the five thermal metrics. Results also show that the best practice consists in delineating homogeneous regions before applying the regional GAM model. According to all performance criteria, delineation of regions with the HCA approach is considered to be more flexible and to lead to better performances than neighborhood-based approaches (CCA and ROI).

Modeling site index of selected poplar clones using airborne laser scanning data

Accurate growth and yield projection for plantations is critical for evaluating management decisions and anticipating future yields. Development of site index (SI) models is often costly and can be problematic when new, short-rotation species are introduced, for example, hybrid poplar plantations, which are increasingly common due to their very fast growth and high productivity. Airborne laser scanning (ALS) allows accurate measurement of tree and stand height and is increasingly being used to develop top height models. In this paper, we demonstrate an approach to develop SI models from ALS data across hybrid poplar plantations in Quebec, Canada. We exploit a single time step ALS acquisition to generate top height estimates at 10 m grid level. Using existing information on planting date and management practices, we developed top height models for unique classes of fertilization treatment and clone. The generic models for unfertilized and fertilized stands showed good fit statistics, with R2 of 0.71 and 0.82, respectively. Clone-specific models showed similar goodness of fit, with the best model resulting in an R2 of 0.89 and relative root mean square error (RMSE) of 16.6%. Analysis of variance (ANOVA) results showed that clone, fertilization status, and the interaction term between clone and fertilization were significant. Our results confirm the development of top height models from a chronosequence of ALS data was successful and offers a new approach to derive SI models in single-species plantation sites.

Predicting Antibiotic Effect of Vancomycin Using Pharmacokinetic/Pharmacodynamic Modeling and Simulation: Dense Sampling versus Sparse Sampling.

This study aimed to investigate the effect of a structural pharmacokinetic (PK) model with fewer compartments developed following sparse sampling on the PK parameter estimation and the probability of target attainment (PTA) prediction of vancomycin. Two- and three-compartment PK models of vancomycin were used for the virtual concentration–time profile simulation. Datasets with reduced blood sampling times were generated to support a model with a lesser number of compartments. Monte Carlo simulation was conducted to evaluate the PTA. For the two-compartment PK profile, the total clearance (CL) of the reduced one-compartment model showed a relative bias (RBias) and relative root mean square error (RRMSE) over 90%. For the three-compartment PK profile, the CL of the reduced one-compartment model represented the largest RBias and RRMSE, while the steady-state volume of distribution of the reduced two-compartment model represented the largest absolute RBias and RRMSE. A lesser number of compartments corresponded to a lower predicted area under the concentration–time curve of vancomycin. The estimated PK parameters and predicted PK/PD index from models built with sparse sampling designs that cannot support the PK profile can be significantly inaccurate and unprecise. This might lead to the misprediction of the PTA and selection of improper dosage regimens when clinicians prescribe antibiotics.

Mapping crop leaf area index at the parcel level via inverting a radiative transfer model under spatiotemporal constraints: A case study on sugarcane

Crop leaf area index (LAI) mapping from remote sensing observations is highly demanded for regional agricultural applications, such as crop health monitoring and crop yield prediction. However, the popularly used model inversion method inevitably presents an ill-posed problem, which leads to unstable and inaccurate retrieval results. An agricultural parcel is a relatively homogeneous object due to uniform agricultural practices and similar environmental conditions. Crops inside a parcel generally present the same growth stage and similar growth status. In this study, a new method was proposed by adopting spatiotemporal constraints at the parcel level in the model inversion process for LAI retrieval. Firstly, phenology information of parcels was utilized to constrain the LAI ranges based on the established prior knowledge, which described the temporal variation of LAI during the life cycle of crop. Subsequently, spatial constraint was adopted in the model inversion through a proposed novel cost function, which assumed the spatial autocorrelation of parameters inside a parcel according to the first law of geography. Sugarcane was taken as an example to evaluate the proposed method. The method was applied to Sentinel-2 data and validated using ground-measured LAI data. The retrieved model parameters exhibited smoother spatial patterns and lower intra-parcel spatial variations through the proposed parcel-level inversion method, compared to the conventional pixel-level inversion method. Evaluations of LAI retrieval accuracy showed that the parcel-level inversion method yielded more accurate results (root mean square error (RMSE): 0.34 m2/m2; relative root mean square error (RRMSE): 20.89%), compared to the pixel-level inversion method (RMSE: 0.56 m2/m2; RRMSE: 34.53%). The spatiotemporal constraint strategy presented the ability to prevent severe overestimation and underestimation. Among the validation data set, the accuracy of the severely overestimated samples of the pixel-level method (RMSE: 0.93 m2/m2) was improved via the new method (RMSE: 0.43 m2/m2); the accuracy of the severely underestimated samples of the pixel-level method (RMSE: 0.66 m2/m2) was improved via the new method (RMSE: 0.26 m2/m2). Finally, the proposed method was applied to obtain sugarcane LAI mapping. The study demonstrates that the proposed parcel-level spatiotemporal constraint strategy improves the accuracy of LAI retrieval, has the capability of producing a highly reliable crop LAI mapping, and shows good potential for agricultural applications at regional scales.

Modelling of Modulus of Elasticity of Low-Calcium-Based Geopolymer Concrete Using Regression Analysis

Despite the unremitting efforts to model the modulus of elasticity of low-calcium-based geopolymer concrete, the state-of-the-art models need much improvement to reduce the error signals and increase the reliability. This study represents a comprehensive regression analysis to model the modulus of elasticity of low-calcium-based geopolymer concrete in terms of its compressive strength. The proposed model’s assumptions are based on taking into account the chemical composition and compressive strength class and considering the normal density of concrete. The modelling is based on 67 different mix-design samples collected from peer-reviewed literature, which are divided into two groups. The first group consists of 59 samples that are used to construct the proposed model, while the second group consists of 8 samples that are used to test the validity of the proposed model. The analysis showed that the proposed model gives the root mean squared error value (RMSE) of 3.122 GPa and the mean absolute percentage error value (MAPE) of 15.0%. Therefore, the proposed model gives 41% and 52.2% reductions in RMSE and MAPE, respectively, from the state-of-the-art model in the literature. Furthermore, other statistical parameters to evaluate the goodness of fitness of the proposed model have been considered, such as the relative root mean squared error (RRMSE) (15.5%), the coefficient of determination (R-squared) (0.773), and the coefficient of correlation (R) (0.88); all of which indicated that the goodness of fitness is good and the proposed model has a high correlation to the actual values. Applying the proposed model in future applications will help reduce the time and cost of geopolymer production, as the proposed model has significantly reduced the error signals.

BIOMASS ESTIMATION ALONG A CLIMATIC GRADIENT USING MULTI-FREQUENCY POLARIMETRIC RADAR VEGETATION INDEX

Abstract. Monitoring biomass changes in the Mediterranean ecosystems is important for better understanding their responses to climatic and anthropogenic changes. Multi-frequency SAR data (L-band PASLAR and C-band Sentinel-1) is investigated for developing a biomass estimation model along a climatic gradient in Israel. First, the relationships between biomass and transmissivity at various frequencies are discussed. Multi-frequency polarimetric radar vegetation index (MPRVI) is then proposed utilizing the ensemble average of degree of polarization and cross-polarized backscattering coefficients. After randomly partitioning the data for training set and testing set, the new MPRVI-based biomass model is evaluated. It shows a good agreement with reference biomass data with an r-square of 0.899 and a root-mean-squared error (RMSE) of 0.381 kg/m2 and with a relative RMSE (RRMSE) of 10.8%.

Data-driven surrogate model with latent data assimilation: Application to wildfire forecasting

The large and catastrophic wildfires have been increasing across the globe in the recent decade, highlighting the importance of simulating and forecasting fire dynamics in near real-time. This is extremely challenging due to the complexities of physical models and geographical features. Running physics-based simulations for large wildfire events in near real-time are computationally expensive, if not infeasible. In this work, we develop and test a novel data-model integration scheme for fire progression forecasting, that combines Reduced-order modelling, recurrent neural networks (Long-Short-Term Memory), data assimilation, and error covariance tuning. The Reduced-order modelling and the machine learning surrogate model ensure the efficiency of the proposed approach while the data assimilation enables the system to adjust the simulation with observations. We applied this algorithm to simulate and forecast three recent large wildfire events in California from 2017 to 2020. The deep-learning-based surrogate model runs around 1000 times faster than the Cellular Automata simulation which is used to generate training data-sets. The daily fire perimeters derived from satellite observation are used as observation data in Latent Assimilation to adjust the fire forecasting in near real-time. An error covariance tuning algorithm is also performed in the reduced space to estimate prior simulation and observation errors. The evolution of the averaged relative root mean square error (R-RMSE) shows that data assimilation and covariance tuning reduce the RMSE by about 50% and considerably improves the forecasting accuracy. As a first attempt at a reduced order wildfire spread forecasting, our exploratory work showed the potential of data-driven machine learning models to speed up fire forecasting for various applications.

RZWQM2 Simulated Drip Fertigation Management to Improve Water and Nitrogen Use Efficiency of Maize in a Solar Greenhouse

The drip fertigation technique is a modern, efficient irrigation method to alleviate water scarcity and fertilizer surpluses in crop production, while the precise quantification of water and fertilizer inputs is difficult for drip fertigation systems. A field experiment of maize (Zea mays L.) in a solar greenhouse was conducted to meet different combinations of four irrigation rates (I125, I100, I75 and I50) and three nitrogen (N) fertilizer rates (N125, N100 and N75) under surface drip fertigation (SDF) systems. The Root Zone Water Quality Model (RZWQM2) was used to assess the response of soil volumetric water content (VWC), leaf area index (LAI), plant height and maize yield to different SDF managements. The model was calibrated by the I100N100 scenario and validated by the remaining five scenarios (i.e., I125N100, I75N100, I50N100, I100N125 and I100N75). The predictions of VWC, LAI and plant height were satisfactory, with relative root mean square errors (RRMSE) < 9.8%, the percent errors (PBIAS) within ±6%, indexes of agreement (IoA) > 0.85 and determination of coefficients (R2) > 0.71, and the relative errors (RE) of simulated yields were in the range of 1.5–7.2%. The simulation results showed that both irrigation and fertilization had multiple effects on water and N stresses. The calibrated model was subsequently used to explore the optimal SDF scenarios for maximizing yield, water use efficiency (WUE) or nitrogen use efficiency (NUE). Among the SDF managements of 21 irrigation rates × 31 N fertilizer rates, the optimal SDF scenarios were I120N130 for max yield (10516 kg/ha), I50N70 for max WUE (47.3 kg/(ha·mm)) and I125N75 for max NUE (30.2 kg/kg), respectively. The results demonstrated that the RZWQM2 was a promising tool for evaluating the effects of SDF management and achieving optimal water and N inputs.

Further application of using a personal laser scanner and simultaneous localization and mapping technology to estimate the log’s volume and its comparison with traditional methods

The aim of this study was the development of a methodical processing line for estimating the log’s volume from scanning logs using a long-range high-speed personal laser scanner (PLS) GeoSLAM ZEB HORIZON and simultaneous localization and mapping (SLAM) technology in an open-source software framework of Cloud Compare. Additionally, the accuracy and precision of using PLS and the suggested method when measuring roundwood volume was examined compared to measurements made using a xylometer and estimations obtained from applying the formulae of Huber, Smalian and Newton. For this purpose, several parameters were measured in 50 logs of Norway spruce with an average length of 2.53 m and a mean diameter of 19.97 cm. Afterwards, the volume of these 50 logs was measured with a xylometer. The results of these measurements served as reference values. The same 50 logs were subsequently scanned with the PLS. The scans were converted into point clouds and were analyzed in Cloud Compare to estimate the volume of the scanned logs. Next, the root mean square error (RMSE) and the mean bias error (MBE) as well as their relative values were calculated for the volumes determined in Cloud Compare and estimated with the above mentioned formulae. The calculated RMSE (and relative RMSE) determined a deviation of the logs' volumes estimated in Cloud Compare from the xylometric volumes of 2.88 dm3 (3.54%) whereas the deviations of the log’s volumes calculated applying the formulae of Huber, Smalian and Newton from the xylometric volumes were respectively 9.63 dm3 (11.83%), 10.33 dm3 (12.69%) and 4.69 dm3 (5.76%). The calculated MBE (and relative MBE) showed that the estimated volumes in Cloud Compare and those using the formulae were overestimated, with the lowest overestimation of 0.10 dm3 (0.12%) in Cloud Compare and the highest, 5.50 dm3 (6.75%), using the Smalian formula. Therefore, it can be stated that results of suggested methodical processing line came closest to the logs' volumes obtained with the xylometer, i.e., they were more accurate and precise compared to the conventionally formulae for log’s volume estimation. The implementation of this method has the potential filling a gap towards a wall-to-wall complete digitization of the roundwood commercialization and ensure transparency and acceptance between the stakeholders involved in the wood supply chain.

Prediction of groundwater table and drought analysis; a new hybridization strategy based on bi-directional long short-term model and the Harris hawk optimization algorithm

Abstract In the present study, a new hybridization strategy for predicting the groundwater table (GWT) and drought analysis is presented. Therefore, a hybrid of the bi-directional long short-term model (BLSTM) and the Harris hawk optimization (HHO) algorithm, namely the BLSTM–HHO algorithm, is applied. In this algorithm, the lagged data of the GWT are used as the input, whereas the current GWT data are used as the output. Additionally, the standalone BLSTM, the long short-term model (LSTM), artificial neural networks (ANN), Seasonal Autoregressive Integrated Moving Average (SARIMA), and the Autoregressive Integrated Moving Average (ARIMA) are employed as benchmark simulating algorithms. The results show that the BLSTM–HHO algorithm has more accuracy than the other investigated simulating algorithms based on the different evaluation criteria such as relative root mean squared error (RRMSE), Nash–Sutcliffe coefficient (NSE), and refined Willmott index (dr). The prediction results (from 2018 to 2022) in all three investigated aquifers show the decline of the GWT (−5.40 m for Brojen aquifer, −7.23 m for Javanmardi aquifer, and −5.81 m for Shahrekord aquifer). Accordingly, the drought analysis by the ground resource index (GRI) in the investigated areas shows that drought is expected to be continued for the next 5 years with an increasing magnitude of severity.

Improving Heterogeneous Forest Height Maps by Integrating GEDI-Based Forest Height Information in a Multi-Sensor Mapping Process

Forests are one of the key elements in ecological transition policies in Europe. Sustainable forest management is needed in order to optimise wood harvesting, while preserving carbon storage, biodiversity and other ecological functions. Forest managers and public bodies need improved and cost-effective forest monitoring tools. Research studies have been carried out to assess the use of optical and radar images for producing forest height or biomass maps. The main limitations are the quantity, quality and representativeness of the reference data for model training. The Global Ecosystem Dynamics Investigation (GEDI) mission (full waveform LiDAR on board the International Space Station) has provided an unprecedented number of forest canopy height samples from 2019. These samples could be used to improve reference datasets. This paper aims to present and validate a method for estimating forest dominant height from open access optical and radar satellite images (Sentinel-1, Sentinel-2 and ALOS-2 PALSAR-2), and then to assess the use of GEDI samples to replace field height measurements in model calibration. Our approach combines satellite image features and dominant height measurements, or GEDI metrics, in a Support Vector Machine regression algorithm, with a feature selection process. The method is tested on mixed uneven-aged broadleaved and coniferous forests in France. Using dominant height measurements for model training, the cross-validation shows 7.3 to 11.6% relative Root Mean Square Error (RMSE) depending on the forest class. When using GEDI height metrics instead of field measurements for model training, errors increase to 12.8–16.7% relative RMSE. This level of error remains satisfactory; the use of GEDI could allow the production of dominant height maps on large areas with better sample representativeness. Future work will focus on confirming these results on new study sites, improving the filtering and processing of GEDI data, and producing height maps at regional or national scale. The resulting maps will help forest managers and public bodies to optimise forest resource inventories, as well as allow scientists to integrate these cartographic data into climate models.