Relative Root Mean Square Error Research Articles

Crown Information Extraction and Annual Growth Estimation of a Chinese Fir Plantation Based on Unmanned Aerial Vehicle–Light Detection and Ranging

Forest biomass dynamics are important indicators of forest productivity and carbon sinks, which are useful for evaluating forest ecological benefits and management options. Rapid and accurate methods for monitoring forest biomass would serve this purpose well. This study aimed at measuring aboveground biomass (AGB) and stand growth from tree crown parameters derived using unmanned aerial vehicle–light detection and ranging (UAV–LiDAR). We focused on 17-year-old Chinese fir plantations in a subtropical area in China and monitored them using UAV–LiDAR from February 2019 to February 2020. Two effective crown height (ECH) detection methods based on drone discrete point clouds were evaluated using ground survey data. Based on the evaluation results, the voxel method based on point cloud segmentation (root-mean-squared error (RMSE) = 0.62 m, relative RMSE (rRMSE) = 4.26%) was better than the tree crown boundary pixel sum method based on canopy height segmentation (RMSE = 1.26 m, rRMSE = 8.63%). The effective crown area (ECA) of an individual tree extracted using ECH was strongly correlated with the annual biomass growth (coefficient of determination (R2) = 0.47). The estimation of annual growth of individual tree crowns based on annual tree height increase (ΔH) derived from LiDAR was statistically significant (R2 = 0.33, p < 0.01). After adding the crown projection area or ECA, the model accuracy R2 increased to 0.57 or 0.63, respectively. As the scale increased to the plot level, the direct model with ECA (RMSE = 1.59 Mg∙ha−1∙a−1, rRMSE = 15.02%) had a better performance than the indirect model using tree height and crown diameter (RMSE = 1.81 Mg∙ha−1∙a−1, rRMSE = 17.10%). The mean annual growth rate of AGB per middle-aged Chinese fir tree was determined to be 8.45 kg∙a−1 using ECA and ΔH, and the plot-level growth rate was 11.47 Mg∙ha−1∙a−1. We conclude that the rapid and accurate monitoring of the annual growth of Chinese fir can be achieved based on multitemporal UAV–LiDAR and effective crown information.

Radar Sensing of Displacement Motions With High Robustness Against Additive Noise

Doppler radar uses phase demodulation to extract the target displacement motion contained in the phase of the echo signal. The demodulation performance is vulnerable to the additive Gauss noise. Especially when the signal-to-noise ratio (SNR) is extremely low in the scenarios of long-range or subtle displacement motion sensing, it may be impossible to extract the correct phase information. In this article, a phase demodulation method with high robustness against additive noise is proposed for accurate displacement motion reconstruction. Mathematical derivations are performed to analyze the noise effect on phase demodulation, and it shows that the influence of phase noise is much smaller than the additive noise. Based on this finding, the key idea of the proposed method is to transform the additive noise of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$I$</tex-math> </inline-formula> / <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$Q$</tex-math> </inline-formula> signals into phase noise. The simulation results validate the noise analysis and verify the high robustness of the proposed method against additive noise. Furthermore, experiments were carried out to evaluate the proposed method under different radial distances and displacement amplitudes. The experimental results show that the proposed method enables accurate motion sensing even in low-SNR scenarios. In our experiments, we observe that the relative root-mean-square error (RRMSE) of the reconstructed displacement motion can be greatly reduced with the proposed method.

LXGB: a machine learning algorithm for estimating the discharge coefficient of pseudo-cosine labyrinth weir

One of the practical and financial solutions to increase the efficiency of weirs is to modify the geometry of the plan and increase the length of the weir to a specific width. This increases the discharge coefficient (Cd) of the weir. In this study, a new weir referred to pseudo-cosine labyrinth weir (PCLW) was introduced. A hybrid machine learning LXGB algorithm was introduced to estimate the Cd of the PCLW. The LXGB is a combination of the linear population size reduction history-based adaptive differential evolution (LSHADE) and extreme gradient boosting (XGB) algorithm. Seven different input scenarios were presented to estimate the discharge coefficient of the PCLW weir. To train and test the proposed method, 132 data series, including geometric and hydraulic parameters from PCLW1 and PCLW2 models were used. The root mean square error (RMSE), relative root mean square error (RRMSE), and Nash–Sutcliffe model efficiency coefficient (NSE) indices were used to evaluate the proposed approach. The results showed that the input variables were the ratio of the radius to the weir height (R/W), the ratio of the length of the weir to the weir height (L/W), and the ratio of the hydraulic head to the weir height (H/W), with the average values of RMSE = 0.009, RRMSE = 0.010, and NSE = 0.977 provided better results in estimating the Cd of PCLW1 and PCLW2 models. The improvement compared to SAELM, ANFIS-FFA, GEP, and ANN in terms of R2 is 2.06%, 3.09%, 1.03%, and 5.15%. In general, intelligent hybrid approaches can be introduced as the most suitable method for estimating the Cd of PCLW weirs.

Performance of Panel-Estimated GFR Among Hospitalized Older Adults

Older adults represent nearly half of all hospitalized patients and are vulnerable to inappropriate dosing of medications eliminated through the kidneys. However, few studies in this population have evaluated the performance of equations for estimating the glomerular filtration rate (GFR)-particularly those that incorporate multiple filtration markers. Cross-sectional diagnostic test substudy of a randomized clinical trial. Adults≥65 years of age presenting to the emergency department of Copenhagen University Hospital Amager and Hvidovre in Hvidovre, Denmark, between October 2018 and April2021. Measured GFR (mGFR) determined using 99mTc-DTPA plasma clearance compared with estimated GFR (eGFR) calculated using 6 different equations based on creatinine; 3 based on creatinine and cystatin C combined; and 2 based on panels of markers including creatinine, cystatin C, β-trace protein (BTP) and/or β2-microglobulin (B2M). The performance of each eGFR equation compared with mGFR with respect to bias, relative bias, inaccuracy (1-P30), and root mean squared error (RMSE). We assessed eGFR performance for 106 patients (58% female, median age 78.3 years, median mGFR 62.9mL/min/1.73m2). Among the creatinine-based equations, the 2009 CKD-EPIcr equation yielded the smallest relative bias (+4.2%). Among the creatinine-cystatin C combination equations, the 2021 CKD-EPIcomb equation yielded the smallest relative bias (-3.4%), inaccuracy (3.8%), and RMSE (0.139). Compared with the 2021 CKD-EPIcomb, the CKD-EPIpanel equation yielded a smaller RMSE (0.136) but larger relative bias (-4.0%) and inaccuracy (5.7%). Only White patients were included; only a subset of patients from the original clinical trial underwent GFR measurement; and filtration marker concentration can be affected by subclinical changes in volume status. The 2009 CKD-EPIcr, 2021 CKD-EPIcomb, and CKD-EPIpanel equations performed best and notably outperformed their respective full-age spectrum equations. The addition of cystatin C to creatinine-based equations improved performance, while the addition of BTP and/or B2M yielded minimal improvement. Grants from public sector industry (Amgros I/S) and government (Capital Region of Denmark). Registered at ClinicalTrials.gov with registration number NCT03741283. Inaccurate kidney function assessment can lead to medication errors, a common cause of hospitalization and early readmission among older adults. Several novel methods have been developed to estimate kidney function based on a panel of kidney function markers that can be measured from a single blood sample. We evaluated the accuracy of these new methods (relative to a gold standard method) among 106 hospitalized older adults. We found that kidney function estimates combining 2 markers (creatinine and cystatin C) were highly accurate and noticeably more accurate than estimates based on creatinine alone. Estimates incorporating additional markers such as β-trace protein and β2-microglobulin did not further improve accuracy.

Estimation of Aboveground Biomass Stock in Tropical Savannas Using Photogrammetric Imaging

The use of photogrammetry technology for aboveground biomass (AGB) stock estimation in tropical savannas is a challenging task and is still at a preliminary stage. This work aimed to use metrics derived from point clouds, constructed using photogrammetric imaging obtained by an RGB camera on board a remotely piloted aircraft (RPA), to generate a model for estimating AGB stock for the shrubby-woody stratum in savanna areas of Central Brazil (Cerrado). AGB stock was estimated using forest inventory data and an allometric equation. The photogrammetric digital terrain model (DTM) was validated with altimetric field data, demonstrating that the passive sensor can identify topographic variations in sites with discontinuous canopies. The inventory estimated an average AGB of 18.3 (±13.3) Mg ha−1 at the three sampled sites. The AGB model selected was composed of metrics used for height at the 10th and 95th percentile, with an adjusted R2 of 93% and a relative root mean squared error (RMSE) of 16%. AGB distribution maps were generated from the spatialization of the metrics selected for the model, optimizing the visualization and our understanding of the spatial distribution of forest AGB. The study represents a step forward in mapping biomass and carbon stocks in tropical savannas using low-cost remote sensing platforms.

A data-driven framework for forecasting transient vehicle thermal performances

Transient vehicle dynamics are deeply affected by the spatial distribution of tire temperature, heat generation, and heat dissipation capacity. However, the tire testing process involves numerous parameters, adding the complexity to monitoring the tire’s thermal properties variation. To address this issue, this paper proposes a data-driven framework that combines the well-known Magic Formula with scaling factors and a long- and short-term memory (LSTM) neural network. The framework forecasts tire thermal properties and functionality under given conditions by learning from time-series data on tire operating conditions and mechanical characteristics. It predicts future spatially varying thermal properties of the tire, including temperatures, heat generation, and heat dissipation. This is achieved with a high degree of forecasting ability quantified by train size/test size ratios equaling 22 and 38% with an impressive Relative Root Mean Square Error (RRMSE) around 1−2%. Here, this framework is applied to quantify the impacts of various operating conditions, including vertical load, slip angle, and vehicle speed, on heat transfer for three crucial radial positions on the tire. Our results show that the vertical load is the primary influencing parameter among all operating conditions. The knowledge about transient tire dynamics will be advantageous to their performance optimization.

Performance of two complementary machine-learned potentials in modelling chemically complex systems

Chemically complex multicomponent alloys possess exceptional properties derived from an inexhaustible compositional space. The complexity however makes interatomic potential development challenging. We explore two complementary machine-learned potentials—the moment tensor potential (MTP) and the Gaussian moment neural network (GM-NN)—in simultaneously describing configurational and vibrational degrees of freedom in the Ta-V-Cr-W alloy family. Both models are equally accurate with excellent performance evaluated against density-functional-theory. They achieve root-mean-square-errors (RMSEs) in energies of less than a few meV/atom across 0 K ordered and high-temperature disordered configurations included in the training. Even for compositions not in training, relative energy RMSEs at high temperatures are within a few meV/atom. High-temperature molecular dynamics forces have similarly small RMSEs of about 0.15 eV/Å for the disordered quaternary included in, and ternaries not part of training. MTPs achieve faster convergence with training size; GM-NNs are faster in execution. Active learning is partially beneficial and should be complemented with conventional human-based training set generation.

A Novel Bio-Inspired Optimization Algorithm Design for Wind Power Engineering Applications Time-Series Forecasting.

Wind patterns can change due to climate change, causing more storms, hurricanes, and quiet spells. These changes can dramatically affect wind power system performance and predictability. Researchers and practitioners are creating more advanced wind power forecasting algorithms that combine more parameters and data sources. Advanced numerical weather prediction models, machine learning techniques, and real-time meteorological sensor and satellite data are used. This paper proposes a Recurrent Neural Network (RNN) forecasting model incorporating a Dynamic Fitness Al-Biruni Earth Radius (DFBER) algorithm to predict wind power data patterns. The performance of this model is compared with several other popular models, including BER, Jaya Algorithm (JAYA), Fire Hawk Optimizer (FHO), Whale Optimization Algorithm (WOA), Grey Wolf Optimizer (GWO), and Particle Swarm Optimization (PSO)-based models. The evaluation is done using various metrics such as relative root mean squared error (RRMSE), Nash Sutcliffe Efficiency (NSE), mean absolute error (MAE), mean bias error (MBE), Pearson's correlation coefficient (r), coefficient of determination (R2), and determination agreement (WI). According to the evaluation metrics and analysis presented in the study, the proposed RNN-DFBER-based model outperforms the other models considered. This suggests that the RNN model, combined with the DFBER algorithm, predicts wind power data patterns more effectively than the alternative models. To support the findings, visualizations are provided to demonstrate the effectiveness of the RNN-DFBER model. Additionally, statistical analyses, such as the ANOVA test and the Wilcoxon Signed-Rank test, are conducted to assess the significance and reliability of the results.

A non-destructive approach to estimate buttress volume using 3D point cloud data

Buttressed trees provide mechanical support for themselves and offer essential ecological functions, such as nutrient acquisition, while being one of the largest sources of volume or biomass estimation variation in tropical forests. In this study, we collected 51 buttressed trees from (33) Democratic Republic of Congo, (12) Indonesia, and (6) Costa Rica, including (48) point clouds, and (3) destructive measurement. Specifically, we compared the performance of the Alpha Shape Algorithm (ASA) and the Slice Triangulation (ST) method on buttress volume estimation based on 30 point clouds with two species. Six point clouds from Costa Rica were used to validate the 3D surface reconstruction method. Meanwhile, we developed three allometric models based on 36 point clouds: a diameter above the buttress-based (DAB, 39 to 203 cm) model, a diameter computed from the non-convex area (Darea130) model, and the convex hull perimeter (Dconvex130) of the breast height model. The developed models were validated with independent data, including (6) point clouds and (3) destructive measurements, to highlight the broader contextualization and application of these methods. Volume estimated by the ASA and ST showed a high agreement with the reference volume acquired using the Smalian formula (relative RMSE of 0.07 and 0.11, respectively, regardless of species effect). ASA was also robust when modeling trees with more and shallower horizontal buttresses. Darea130 was the most accurate predictor to estimate buttress volume, with a lower Akaike information criterion (−66.25) than DAB (−59.55) and Dconvex130 (30.56); however, DAB and Darea130 (relative RMSE of 0.21 and 0.23, respectively) showed similar performance when validated with independent datasets. Our results indicate that the ASA approach performs better than both the ST and allometric models used in this study. Furthermore, the ASA method can help correct the bias in the present and past estimates of volume and biomass of large trees, which are foundational components for understanding biomass allocation and dynamics in tropical forests contemporary fields.

A hybrid machine learning framework for forecasting house price

House price prediction is one of the most important factors affecting national real estate policies. However, developing an accurate housing price prediction model is a significant challenge for the real estate market. This study presents a framework of house price prediction models that address this issue by improving forecasting performance, explicitly demonstrating the novelty in the Hybrid Bayesian Optimization (HBO) models combined with Stacking (HBOS), Bagging (HBOB), and Transformer (HBOT) techniques. These hybrid models employ Bayesian Optimization for hyperparameter tuning, leading to superior prediction accuracy and stability. Additionally, the proposed framework can assess a statistical and accurate assessment of the predictive performance of house price forecasting models in different scenarios. Furthermore, we constructed a multi-source dataset containing 1,898,175 transactions of the Hong Kong real estate market covering a period from January 2, 1996, to May 13, 2021. This dataset, another major contribution to the field, enables comprehensive model testing and could be a valuable resource for future research. Then, the proposed hybrid models are compared with 18 benchmark models. Thirteen evaluation metrics are used to evaluate the predictive performance, while the non-parametric testing, including Friedman, Iman–Davenport, and Nemenyi post-hoc tests methods, are adopted to assess the significance of differences in the predictive performance of each model. The experimental results show that the HBOS models are superior to the other benchmark models for application in the house price prediction problem. The HBOS-CatBoost model showed superior performance in terms of RMSE compared to both the HBOB-XGBoost and HBOT-ConvLSTM models, with relative RMSE reductions of 5.11% and 25.56%, respectively. The main contributions of this work are the creation of a rich multi-source dataset, the proposal of novel hybrid models for improved house price prediction, and a comprehensive performance evaluation framework. These findings offer a significant step forward in the housing price prediction field.

Can meteorological data and normalized difference vegetation index be used to quantify soil pH in grasslands?

Quantifying soil pH at manifold spatio-temporal scales is critical for examining the impacts of global change on soil quality. It is still unclear whether meteorological data and normalized difference vegetation index (NDVI) can be used to quantify soil pH in grasslands. Here, nine methods (i.e., RF: random-forest, GLR: generalized-linear-regression, GBR: generalized-boosted-regression, MLR: multiple-linear-regression, ANN: artificial-neural-network, CIT: conditional-inference-tree, SVM: support-vector-machine, eXGB: eXtreme-gradient-boosting, RRT: recursive-regression-tree) were applied to quantify soil pH. Three independent variables (i.e., AP: annual precipitation, AT: annual temperature, ARad: annual radiation) were used to quantify potential soil pH (pHp), and four independent variables (i.e., AP, AT, ARad and NDVImax: maximum NDVI during growing season) were applied to quantify actual soil pH (pHa). Overall, the developed eXGB models performed the worst (linear regression slope &amp;lt; 0.60; R2 = 0.99; relative deviation ≤ –43.54%; RMSE ≥ 3.14), but developed RF models performed the best (linear regression slope: 0.99–1.01; R2 = 1.00; relative deviation: from –1.26% to 0.65%; RMSE ≤ 0.28). The linear regression slope, R2, absolute value of relative deviation and RMSE between modelled and measured soil pH were 0.96–1.03, 0.99–1.00, ≤ 3.87% and ≤ 0.88 for the other seven methods, respectively. Accordingly, except the developed eXGB approach, the developed other eight methods can have relative greater accuracies in quantifying soil pH. However, the developed RF had the uppermost quantification accuracy for soil pH. Whether or not meteorological data and normalized difference vegetation index can be used to quantify soil pH was dependent on the chosen models. The RF developed by this study can be used to quantify soil pH from measured meteorological data and NDVImax, and may be conducive to scientific studies related to soil quality and degradation (e.g., soil acidification and salinization) at manifold spatial-temporal under future globe change.

Dynamic Prediction Model for Progressive Surface Subsidence Based on MMF Time Function

It is imperative to timely and accurately predict the progressive surface subsidence caused by coal mining in the context of precision coal mining. However, the existing dynamic prediction methods that use time functions still have limitations, especially in the description of the moments of initiation and maximum subsidence velocity, which hinder their wide application. In this study, we proposed the MMF (Morgan–Mercer–Flodin) time function for predicting progressive surface subsidence based on the model assumptions and formula derivations. MMF time function can resolve the limitations in the description of the moments of initiation and maximum subsidence velocity perfectly. Afterward, we established the dynamic prediction model by combining the probability integral method with the MMF time function. Finally, using the measured subsidence data of working panel 22101 as an example, the accuracy and reliability of the dynamic prediction model was verified. The average RMSE and average relative RMSE (RRMSE) of prediction progressive subsidence using MMF time function are 46.65 mm and 4.63%, respectively. The accuracy is optimal compared with other time functions (for the average RMSE, Logistic time function is 80.57 mm, Gompertz time function is 79.77 mm, and Weibull time function is 90.61 mm; for the average RRMSE, Logistic time function is 7.66%, Gompertz time function is 7.73%, and Weibull time function is 8.62%). The results show that the method proposed in this paper can fully meet the requirements of practical engineering applications, achieve accurate dynamic prediction during the coal mining process, and provide good guidance for surface deformation and building protection.

Can meteorological data improve the short-term prediction of individual milk yield in dairy cows?

Many farms document daily milk yields of individual cows because these are a good indicator of cow well-being. It is established that extreme meteorological conditions influence the milk yields by causing heat and cold stress, whereas less is known about the effects of moderate changes in meteorological conditions. Thus, the aim of the present study was to evaluate whether individual daily milk yield predictions can be improved by considering such changes. We evaluated 8 years of milking and meteorological data from Eastern Switzerland with a total of 33,938 daily milkings from 145 Brown Swiss and 64 Swiss Fleckvieh cows. The cows were aged between 1.9 and 13.5 years at parturition. The data set was split into 7 periods according to the days in milk (DIM) and subsequently filtered into subsets by breed and parity. We applied Gaussian process regression to predict individual daily milk yield. We compared different models including DIM, lagged milk yield, and meteorological variables as features and found that models including the lagged milk yield performed best. Within the period of 5 to 90 DIM, we were able to predict individual next-day milk yield from the cow's last milkings with a root mean squared error (RMSE) of 2.1 kg. In contrast, without information on the previous milk yield, accuracy of milk yield prediction was lower, with an RMSE close to 8 kg. The models holding information about previous milk yields showed a substantial increase in performance. Within a more homogeneous data subset filtered by breed or parity or both, predictions were even better, with a relative RMSE of 4.3% for first-parity Fleckvieh cows. However, we found that including meteorological features, such as temperature, rainfall, wind speed, temperature humidity index, cooling degree, and barometric pressure, did not improve the predictions in any of the evaluated periods. This finding indicates that considering meteorological features in daily milk yield prediction models is not useful in moderate climates; considering lagged milk yield is sufficient. We hypothesize that this meteorological information, among other influences, is indirectly contained in the lagged milk yield.

A dynamic shelf-life prediction method considering actual uncertainty: Application to fresh fruits in long-term cold storage

Shelf-life is an important tool for visually conveying remaining food quality. However, the traditional printed shelf-life is a fixed value based on a stable environment, whereas food supplies undergo long-term dynamic conditions. Moreover, the accumulating effects of real-world uncertainty will lead to stochastic changes in remaining shelf-life in a probabilistic manner. Therefore, we proposed a novel dynamic shelf-life prediction method integrating the kinetic reaction and stochastic process. First, to predict practical changes in food quality indices (FQIs), we established the stochastic kinetic model, which combines basic deterministic modeling with zero-order reaction and the Arrhenius equation, and stochastic factor modeling with the Wiener process. Second, we conducted real-time probability analysis of remaining shelf-life to quantify the potential degeneration of food quality based on the established model. By using datasets monitoring firmness and Vitamin C of Kiwifruit in long-term cold storage to verify the performance of our proposed integrated model, we showed that our method was more accurate in modeling stochastic changes in FQIs than the traditional reaction model, resulting in mean absolute error, mean absolute percentage error, and root mean squared error (RMSE) less than 0.2136, 0.0162, and 0.7402, respectively. Furthermore, the shelf-life probability analysis was efficient, with relative RMSE 5.26% and computing time 0.047 s. This valuable food quality information can be provided to managers or customers to reduce food loss and waste.

The Performance of Sensitivity-Maps Method in Reconstructing Low Contrast and Multi-Contrast Objects for Microwave Imaging Applications

The microwave imaging system for breast tumor/cancer detection requires high sensitivity to detect abnormal tissue that has little contrast in high-density breasts. This paper proposes a qualitative microwave imaging system simulation based on inverse scattering using the sensitivity-maps method. This method utilizes two measurement types for system calibration: a reference object as a scatterer-free background and a calibration object to obtain the system's impulse response. The object under test (OUT) consists of an object with low dielectric contrast and a phantom with multiple low dielectric contrasts (multi-contrast). Reconstruction is carried out on three types of S-parameter measurement data, namely S_11,〖 S〗_21, and a combination of both. S-parameters are measured at several frequencies, which are 3, 10, 14, 15, 16, 20 GHz, and the combination of all those frequencies (multifrequency). Reconstructed images show that the system is capable of reconstructing dielectric objects accurately. Quantitatively, the results show that the multifrequency S_21 measurement yields the best image quality with relative root mean squared error (RRMSE) values of 0.1272 and structural similarity index (SSIM) of 0.9076. The designed imaging system also successfully reconstructs multi-contrast phantom accurately with RRMSE of 0.1434 and SSIM of 0.4609.

Comparison between deep learning architectures for the 1 km, 10/15-min estimation of downward shortwave radiation from AHI and ABI

The retrieval of downward shortwave radiation (DSR) with high spatiotemporal resolution and short latency is critical. It is the fundamental driving force of surface energy, carbon, and hydrological circulations, and a key energy source for photovoltaic electricity. However, existing methods face significant challenges owing to cloud heterogeneity and their reliance on other satellite-derived products, which hinder the retrieval of accurate and timely DSR with high spatiotemporal resolution. In addition to the spectral features used in traditional approaches, deep learning (DL) can incorporate the spatial and temporal features of satellite data. This study developed and compared three DL methods, namely the DenseNet, the bidirectional gated recurrent unit without surface albedo as inputs (BiGRUnor), and the convolutional neural network with gated recurrent unit without surface albedo as inputs (CNNGRUnor). These methods were used to estimate DSR at 1 km and 10/15 min resolutions directly from top-of-atmosphere reflectance over the Advanced Himawari Imager (AHI) onboard Himawari-8 and the Advanced Baseline Imager (ABI) onboard GOES-16 coverage, achieving high accuracies. The instantaneous root mean square error (RMSE) and relative RMSE for the three models were 68.4 (16.1%), 69.4 (16.3%), and 67.1 (15.7%) W/m2, respectively, which are lower than the baseline machine learning method, the multilayer perceptron model (MLP), with RMSE at 76.8 W/m2 (18.0%). Hourly accuracies for the three DL methods were 58.6 (14.1%), 57.8 (14.0%), and 57.3 (13.8%) W/m2, which are within the DSR RMSEs that we estimated for existing datasets of the Earth's Radiant Energy System (CERES) (88.8 W/m2, 21.4%) and GeoNEX (77.8 W/m2, 18.8%). The study illustrates that DL models that incorporate temporal information can eliminate the need for surface albedo as an input, which is crucial for timely monitoring and nowcasting of DSR. Incorporating spatial information can enhance retrieval accuracy in overcast conditions, and incorporating infrared bands can further improve the accuracy of DSR estimation.

Boosting the prediction accuracy of a process-based greenhouse climate-tomato production model by particle filtering and deep learning

By generating high quality data without the big time investment and economic cost of real experiments, dynamic greenhouse climate and crop simulation models can support decisions on greenhouse climate control, crop management and greenhouse design. The reliability of simulation-based decisions depends on both the prediction accuracy and interpretability of simulation models. The prediction accuracy of these simulation models can be increased by: 1) improving mechanisms in process-based models; 2) calibrating process-based model parameters; 3) deriving black-box relationships from data. Considering the descending interpretability from (1) to (3), this study presents a knowledge-based data-driven modelling approach where firstly a process-based model is selected and modified based on domain knowledge, then data-driven improvement is applied including two steps: parameter value estimation by particle filter (PF) and further black-box improvement by deep neural networks (DNN). The approach was tested with an example of greenhouse climate-tomato production system modelling. Modules from GreenLight (Katzin et al., 2020) and TOMSIM (Heuvelink, 1995, 1996) were selected, modified and integrated into a process-based greenhouse climate-tomato model. Validation showed that PF-calibration of five greenhouse parameters decreased the seasonal relative root mean squared error (RRMSE) of indoor air vapor pressure predictions from 40.7% of that before PF-calibration to 16.4%, while it did not decrease the RRMSE of indoor air temperature predictions. Combining the PF-calibrated model with a DNN trained on a season of data decreased the RRMSE of indoor air temperature from 15.0% without DNN to 6.7%, and decreased the RRMSE of indoor air vapor pressure to 12.6%. The knowledge-based data-driven greenhouse climate-tomato model had a relative error of 0.9% for seasonal total fresh yield, and an RRMSE of 6.6% for the cumulative yield throughout the season. If process-based model parameters were not calibrated before combining the model with DNNs, the required amount and diversity of DNN training data increased because more information needed to be learnt from data by the DNNs. Without PF-calibration, combining a DNN trained on 50 days of data with the process-based model resulted in RRMSEs of 44.8% and 31.8% for indoor air temperature and vapor pressure prediction, respectively; with PF-calibration, the RRMSEs were decreased to 13.1% and 17.9%. The proposed three-step knowledge-based data-driven approach can not only improve the model prediction accuracy, but can also help to track and interpret the improvements.

A Ground Point Fitting Method for Winter Wheat Height Estimation Using UAV-Based SfM Point Cloud Data

Height is a key factor in monitoring the growth status and rate of crops. Compared with large-scale satellite remote sensing images and high-cost LiDAR point cloud, the point cloud generated by the Structure from Motion (SfM) algorithm based on UAV images can quickly estimate crop height in the target area at a lower cost. However, crop leaves gradually start to cover the ground from the beginning of the stem elongation stage, making more and more ground points below the canopy disappear in the data. The terrain undulations and outliers will seriously affect the height estimation accuracy. This paper proposed a ground point fitting method to estimate the height of winter wheat based on the UAV SfM point cloud. A canopy slice filter was designed to reduce the interference of middle canopy points and outliers. Random Sample Consensus (RANSAC) was applied to obtain the ground points from the valid filtered point cloud. Then, the missing ground points were fitted according to the known ground points. Furthermore, we achieved crop height monitoring at the stem elongation stage with an R2 of 0.90. The relative root mean squared error (RRMSE) of height estimation was 5.9%, and the relative mean absolute error (RMAE) was 4.6% at the stem elongation stage. This paper proposed the canopy slice filter and fitting missing ground points. It was concluded that the canopy slice filter successfully optimized the extraction of ground points and removed outliers. Fitting the missing ground points simulated the terrain undulations effectively and improved the accuracy.

A copula-supported Bayesian framework for spatial downscaling of GRACE-derived terrestrial water storage flux

The GRACE and GRACE-FO satellite missions provide mass variations as a fundamentally new observation type for a broad spectrum of novel applications in Earth science disciplines, including oceanography, geophysics, hydrology, and hydrometeorology. Despite all the key findings in hydrology, the utility of GRACE-derived Terrestrial Water Storage Anomaly (TWSA) and its time derivative Terrestrial Water Storage Flux (TWSF) have mainly been limited to large catchments due to their coarse spatial resolution. Here, we propose a method to downscale TWSF by incorporating available finer-resolution data. We determine the downscaled TWSF and its uncertainty within a proposed Bayesian framework by incorporating the fine-scale data of TWSF and Soil Moisture Change (SMC) from different available sources. For the Bayesian ingredients, we rely on GRACE data to obtain the prior and rely on copula models to obtain nonparametric likelihood functions based on the statistical relationship between GRACE TWSF with fine-scale TWSF data and SMC. We apply our method to the Amazon Basin and assess the performances of our products from various fine-scale input datasets of TWSFs and SMCs. Given the lack of ground truth for TWSF, we validate our results against space-based Surface Water Storage Change (SWSC) in the Amazon river system and also against the Vertical Crustal Displacements Rate (VCDR) observed by the Global Positioning System (GPS). Overall, the results show that the proposed method is able to estimate a downscaled TWSF, which is informed by GRACE and fine-scale data. Validation shows that our downscaled products are better anticorrelated with VCDR (−0.81) than fine-scale TWSF (−0.73) and show a mean relative RMSE of 26% with SWSC versus 70% for fine-scale TWSF. The proposed methodology can be extended to other coarse scale datasets, which are crucial for hydrological application at regional and local scales.

A theoretical approach for water saturation estimation in shaly sandstones

The process of quantitative reservoir characterization is a crucial component spanning from the initial estimation of in-place reserves to the optimization of production design. Water saturation is a critical parameter in quantitative reservoir characterization that directly impacts the in-place reserve estimates. In the case of shale-free sandstones, the computation of water saturation using Archie's equation yields fairly accurate results. However, in shaly sandstone reservoirs, the presence of multiple conductive entities in the rock poses additional complexities. Currently, various empirical, semi-empirical and theoretical equations are available in literature to estimate water saturation; nevertheless, none of these models is such that it is both theoretically sound and practically applicable.In this study, we present a novel approach that is both theoretically sound and practically applicable in computing water saturation in shaly sandstones. The fluid-saturated rocks are treated as a system consisting of two interlocked lattice-like structures, namely grains and fluid in the interconnected pores. The composite fluid comprising water and hydrocarbon is modeled using the general mixture rule, while the Hanai-Bruggeman equation is employed to model the electrical conductivity of this system. The resulting electrical conductivity equation is then utilized to compute water saturation in shaly sandstones.The proposed equation was effectively validated using the published core data. Furthermore, we compare the new model with other grain conductivity-based models such as Mixing model and Berg's model. Our findings show that while describing core data the proposed equation gives a lower average Relative Root Mean Squared Error (RRMSE) of 2.05% as compared to 7.56% and 5.58% for the Mixing and Berg's model available in literature, respectively. Moreover, the range of porosity and water conductivity for which the proposed model is applicable, surpasses that of the current models. We also suggest that this equation can be extended to other systems with a similar two interlocked lattice-like structure, such as carbonates.