Mapping 30 m fractional tree cover using PlanetScope-3B images and Landsat-8 spectral-texture data for a case study of different forest types in Indonesia

Deep learning-based state estimation of lithium batteries is widely used in battery management system (BMS) design. However, due to the limitation of on-board computing resources, multiple single-state estimation models are more difficult to deploy in practice. Therefore, this paper proposes a multi-task learning network (MTL) combining a multi-layer feature extraction structure with separated expert layers for the joint estimation of the state of charge (SOC) and state of energy (SOE) of Li-ion batteries. MTL uses a multi-layer network to extract features, separating task sharing from task-specific parameters. The underlying LSTM initially extracts time-series features. The separated expert layer, consisting of task-specific and shared experts, extracts features specific to different tasks and shared features for multiple tasks. The information extracted by different experts is fused through a gate structure. Tasks are processed based on specific and shared information. Multiple tasks are trained simultaneously to improve performance by sharing the learned knowledge with each other. SOC and SOE are estimated on the Panasonic dataset, and the model is tested for generalization performance on the LG dataset. The Mean Absolute Error (MAE) values for the two tasks are 1.01% and 0.59%, and the Root Mean Square Error (RMSE) values are 1.29% and 0.77%, respectively. For SOE estimation tasks, the MAE and RMSE values are reduced by 0.096% and 0.087%, respectively, when compared with single-task learning models. The MTL model also achieves reductions of up to 0.818% and 0.938% in MAE and RMSE values, respectively, compared to other multi-task learning models. For SOC estimation tasks, the MAE and RMSE values are reduced by 0.051% and 0.078%, respectively, compared to single-task learning models. The MTL model also outperforms other multi-task learning models, achieving reductions of up to 0.398% and 0.578% in MAE and RMSE values, respectively. In the process of simulating online prediction, the MTL model consumes 4.93 ms, which is less than the combined time of multiple single-task learning models and almost the same as that of other multi-task learning models. The results show the effectiveness and superiority of this method.

The aim of this study was to model the surface water quality of the Broad River Basin, South Carolina. The most suitable two monitoring stations numbered as USGS 02156500 (Near Carlisle) and USGS 02160991 (Near Jenkinsville) were selected for the reason that the river water temperature (WT), pH, and specific conductance (SC), as well as dissolved oxygen (DO) concentration, were simultaneously monitored and recorded at these sites. The monitoring period from September 2016 to August 2017 was taken into account for the modeling studies. The electrical conductivity (EC) values corresponding to the river SC values were calculated. First, the conventional regression analysis (CRA) was applied to three regression forms, i.e., linear, power, and exponential functions, to estimate the river DO concentration. Then, the multivariate adaptive regression splines (MARS) and TreeNet gradient boosting machine (TreeNet) techniques were employed. Three performance statistics, i.e., root means square error (RMSE), mean absolute error (MAE), and Nash-Sutcliffe coefficient of efficiency (NS), were used to compare the estimation capabilities of these techniques. The TreeNet technique, which was used for the first time in the modeling of DO concentration, had higher estimation success with the RMSE, MAE, and NS values of 0.182mg/L, 0.123mg/L, and 0.990, respectively, for the Carlisle station and 0.313mg/L, 0.233mg/L, and 0.965, respectively, for the Jenkinsville station in the training phase. The MARS technique, which had limited availability of its application in the modeling of DO concentration, had higher estimation success with the RMSE, MAE, and NS values of 0.240mg/L, 0.195mg/L, and 0.981, respectively, for the Carlisle station and 0.527mg/L, 0.432mg/L, and 0.980, respectively, for the Jenkinsville station in the testing phase. Considering the RMSE and MAE values being lower, as well as NS values being higher for the model having an input combination of WT, pH, and EC, the Carlisle station came into prominence. It was concluded that international researchers, who have engaged in the river water quality modeling studies, can favor the MARS and TreeNET techniques without any hesitation and estimate the river DO concentration successfully. The models developed for the Carlisle station were tested with the data sets for the monitoring period from September 2017 to August 2018 at the same station. Similarly, the models developed for the Jenkinsville station were tested with the data sets for the monitoring period from September 2017 to August 2018 at the same station. It was concluded that the models could estimate the river DO concentrations very close to in situ measurements at the same site but for the different monitoring periods, too. Furthermore, the models developed for the Carlisle station were tested with the data sets from the Jenkinsville station for the same monitoring period. Similarly, the models developed for the Jenkinsville station were tested with the data sets from the Carlisle station for the same monitoring period. It was also concluded that the developed models could estimate the river DO concentrations very close to in situ measurements at different monitoring sites but for the same monitoring period on the same river, too. It can be asserted that the models developed for any monitoring site on a river can be employed for another monitoring site on the same river, too, as in the case of the Broad River, South Carolina.

With the continuous improvement of urbanization, the pollution of NO₂ in urban atmosphere has become increasingly severe. In order to accurately predict the concentration of NO₂ in the atmosphere of Urumqi, this study utilized the SARIMA model, Holt - Winters model, and the intervention model to forecast the NO₂ concentration in the atmosphere of Urumqi from April 2024 to March 2025. Additionally, the fitting performance and prediction accuracy of the three models were compared. Mean Squared Error (MSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) were selected as the evaluation indicators for model fitting and accuracy. The results showed that for the optimal SARIMA model, the values of MAE, RMSE (Root Mean Squared Error) and MAPE were , , and , respectively. For the optimal Holt - Winters model, the values of MAE, RMSE, and MAPE were , , and respectively. And for the intervention model, the values of MAE, RMSE, and MAPE were , , and . By comparing the prediction results of the three models, it was concluded that the modified optimal Holt - Winters model had the best prediction performance, followed by the intervention model, while the optimal SARIMA model had the worst prediction performance. Finally, the optimal Holt - Winters model was used to predict the NO₂ concentration in the atmosphere of Urumqi in the coming year, providing rational suggestions for the local government's policy - making and residents' travel arrangements.

The COVID-19 virus has also caused shocks to the Bangka Belitung Islands Province in various sectors, especially the economy. To overcome this problem, of course the government has prepared responsive policies, both fiscal and monetary policies to prevent post-COVID-19 risks, especially in the economic recession. To prevent a post-COVID-19 economic recession, a prediction or time series forecast is needed on four variables that influence the economic recession, namely the number of tin exports, population, poverty and labor force in the Bangka Belitung Islands Province so that economic growth is maintained. This research aims to predict the four research variables by comparing the Moving Average and Exponential Smoothing methods. This research also uses Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) as indicators of model accuracy. Based on the results of the accuracy indicators of this model, it was found that the Exponential Smoothing method was better than the Moving Average method. The predicted results for the value of tin exports in 2024 are -3.3645811 with The RMSE value is 42293770, MAE is 29558091, and MAPE is 84.46131. The negative value in the tin export prediction means that the decline in the value of tin exports in 2024 will not have a significant effect because it is still within a reasonable figure. The total labor force in 2024 will be 11057.23 with RMSE value is 16536.48, MAE value is 14194.02, and MAPE is 112.8078. Then for population the predicted result is 21241.92 with RMSE is 19537.82, MAE is 11548.41, and MAPE is 37.51894. Then for the predicted results the number of poverty is 70.22749 with RMSE, MAE, and MAPE respectively of 3992.146, 3205.528, and 139.1129. The alpha value used is 0.0183.

The aim of this paper was to compare effects of organic and mineral fertilizers on greenhouse gas (GHG) emissions from legume grasslands in Finland. We invoke DNDC, a process-based model that integrates effects of agricultural practices, soil characteristics, nitrogen mass balance and climate change on GHG emissions from soil-plant ecosystems. Data measured in the field were collected from 2017 to 2020 using an eddy covariance site cultivated with legume grass species (Phleum pratense L., Festuca pratensis Huds, Trifolium pratense L., Hordeum vulgare L.) at Anttila, Maaninka, eastern Finland. The focus of the modelling was to evaluate the performance of DNDC heat exchange version under two distinct management practices: organic input, utilizing digestate residue (slurry), and mineral input (NPK) with chemical fertilizer. The primary emphasis was on understanding the model's accuracy in simulating greenhouse gas emissions and comparing the total annual greenhouse gas exchanges between these two management approaches. The DNDC heat exchange model version was calibrated and validated for key processes, including Gross Primary Productivity (GPP), Net Ecosystem Exchange (NEE), Ecosystem Respiration (Reco), Soil Temperature, and Water-Filled Pore Space (WFPS) at 5 cm and 20 cm depths. The model demonstrated satisfactory performance in estimating the total annual GHG exchanges during validation years under both management practices. For the mineral treatment, the model demonstrated fair performance (Spearman's correlation (ρ) for GPP (0.81), NEE (0.72), and Reco (0.85)). Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) values indicated reasonable agreement between model predictions and measured data. Notably, soil temperature simulations demonstrated an excellent correlation (ρ=0.99) with low RMSE and MAE. Water-Filled Pore Space (WFPS) at both 5 cm and 20 cm depths exhibited good correlations, with acceptable RMSE and MAE values. Similarly, for organic inputs, the DNDC model had fair correlation (ρ) for GPP (0.81), NEE (0.72), and Reco (0.85). Soil temperature and WFPS at 5 cm presented high positive correlations (ρ=0.98 and 0.55), accompanied by low RMSE and MAE. WFPS at 20cm, while exhibiting good correlation (ρ=0.065), displayed a slightly elevated RMSE and MAE. Overall, we conclude that the model offered valuable insights into GHG dynamics associated with organic and mineral fertilization practices. Overestimation of biomass yield for some of the data by DNDC suggests that future work would be well placed targeting physiology determinants of biomass in the model.

In the present investigation, the usefulness and capabilities of four artificial intelligence (AI) models, namely feedforward neural networks (FFNNs), gene expression programming (GEP), adaptive neuro-fuzzy inference system with grid partition (ANFIS-GP) and adaptive neuro-fuzzy inference system with subtractive clustering (ANFIS-SC), were investigated in an attempt to evaluate their predictive ability of the phycocyanin pigment concentration (PC) using data from two stations operated by the United States Geological Survey (USGS). Four water quality parameters, namely temperature, pH, specific conductance and dissolved oxygen, were utilized for PC concentration estimation. The four models were evaluated using root mean square errors (RMSEs), mean absolute errors (MAEs) and correlation coefficient (R). The results showed that the ANFIS-SC provided more accurate predictions in comparison with ANFIS-GP, GEP and FFNN for both stations. For USGS 06892350 station, the R, RMSE and MAE values in the test phase for ANFIS-SC were 0.955, 0.205 μg/L and 0.148 μg/L, respectively. Similarly, for USGS 14211720 station, the R, RMSE and MAE values in the test phase for ANFIS-SC, respectively, were 0.950, 0.050 μg/L and 0.031 μg/L. Also, using several combinations of the input variables, the results showed that the ANFIS-SC having only temperature and pH as inputs provided good accuracy, with R, RMSE and MAE values in the test phase, respectively, equal to 0.917, 0.275 μg/L and 0.200 μg/L for USGS 06892350 station. This study proved that artificial intelligence models are good and powerful tools for predicting PC concentration using only water quality variables as predictors.

To improve the accuracy and efficiency of tool wear predictions, this study proposes a tool wear prediction model called LSTM_ResNet which is based on the long short-term memory (LSTM) network and the Residual Network (ResNet). The model utilizes LSTM layers for processing, where the first block and loop blocks serve as the core modules of the deep residual network. The model employs a series of methods including convolution, batch normalization (BN), and Rectified Linear Unit (ReLU) to enhance the model’s expression and prediction capabilities. The performance of the LSTM_ResNet model was evaluated using experimental data from the PHM2010 datasets and two different depths (64 and 128 layers), training both LSTM_ResNet models for 200 epochs. The 64-layer model’s root mean square error (RMSE) values are 3.36, 4.35, and 3.59, and the mean absolute error (MAE) values are 2.42, 2.85, and 2.21; using 128 layers, the RMSE values are 3.66, 3.99, and 3.77, and the MAE values are 2.49, 2.73, and 3.01. The results indicate that the 64-layer LSTM has smaller average errors, suggesting that compared to other common network structures, the LSTM_ResNet network has a higher performance. This research provides an effective solution for tool wear prediction and helps to improve the technical level of tool wear prediction in China.

Prediction of atmospheric air temperature (AAT) time series is an important issue as it gives information to society and sustainability for future planning. In this study, a deep learning method, namely, long short-term memory (LSTM) network, based on one-step-ahead prediction approach was proposed to predict AAT using the actual time series data. For the proposed prediction method, a set of measurement data in 10-min, hourly, and daily intervals obtained from Mersin and Belen stations located in the Eastern Mediterranean Region of Turkey was used. Mean absolute percentage error (MAPE), root mean square error (RMSE), correlation coefficient (R), mean absolute error (MAE), and average bias were considered as evaluation criteria. According to the testing process, the RMSE, MAPE, MAE, R, and bias values for the 10-min interval AAT prediction were calculated as 0.35 °C, 1.40%, 0.25 °C, 0.995, and 0.074 °C, respectively. Considering the prediction results of the hourly AAT predication, the above statistical metrics with the same order were obtained as 0.61 °C, 1.85%, 0.43 °C, 0.945, and −0.013 °C. Concerning the daily AAT prediction results with LSTM, the above statistical metrics with the same order were computed as 1.33 °C, 3.27%, 0.99 °C, 0.97, and −0.116 °C. Compared to the hourly and daily AAT predictions, LSTM provided better accuracy results in predicting 10-min interval AAT. The prediction results from the three different time series data show that the prediction of AATs using LSTM can provide high accuracy results for short-term prediction using data with a long period time. On the other hand, adaptive neuro-fuzzy inference system with fuzzy C-means (ANFIS-FCM) method and autoregressive moving average (ARMA) model were also used to compare the results of LSTM method. Both LSTM and ANFIS-FCM network model showed high accuracy for the prediction of 10-min interval, hourly, and daily AAT data with RMSE values between 0.31 and 1.52 °C, while ARMA model failed to provide high accuracies for all predictions.

Water is a prime necessity for the survival and sustenance of all living beings. Over the past few years, the water quality of rivers is adversely affected due to harmful wastes and pollutants. This ever-increasing water pollution is a big matter of concern as it deteriorating the water quality, making it unfit for any type of use. Recently, water quality modelling using machine learning techniques has generated a lot of interest and can be very beneficial in ecological and water resources management. However, they suffer many times from high computational complexity and high prediction error. The good performance of a deep neural network like long short-term memory network (LSTM) has been exploited for the time-series data. In this paper, a deep learning-based Bi-LSTM model (DLBL-WQA) is introduced to forecast the water quality factors of Yamuna River, India. The existing schemes do not perform missing value imputation and focus only on the learning process without including a loss function pertaining to training error. The proposed model shows a novel scheme which includes missing value imputation in the first phase, the second phase generates the feature maps from the given input data, the third phase includes a Bi-LSTM architecture to improve the learning process, and finally, an optimized loss function is applied to reduce the training error. Thus, the proposed model improves forecasting accuracy. Data comprising monthly samples of different water quality factors were collected for 6 years (2013-2019) at several locations in the Delhi region. Experimental results reveal that predicted values of the model and the actual values were in a close agreement and could reveal a future trend. The performance of our model was compared with various state of the art techniques like SVR, random forest, artificial neural network, LSTM, and CNN-LSTM. To check the accuracy, metrics like root mean square errors (RMSE), the mean absolute error (MAE), mean square error (MSE), and mean absolute percentage error (MAPE) have been used. Experimental analysis is carried out by measuring the COD and BOD levels. COD analysis reveals the MSE, RMSE, MAE, and MAPE values as 0.015, 0.117, 0.115, and 20.32, respectively, for the Palla region. Similarly, BOD analysis indicates the MSE, RMSE, MAE, and MAPE values as 0.107, 0.108, 0.124, and 18.22, respectively. A comparative analysis reveals that the proposed model outperforms all other models in terms of the best forecasting accuracy and lowest error rates.

Aim:This study aimed to compare the performance of two deep learning algorithms, attention-based gated recurrent unit (GRU), and the artificial neural networks (ANNs) algorithm for coloring silicone maxillofacial prostheses.Settings and Design:This was an in vitro study.Materials and Methods:A total of 21 silicone samples in different colors were produced with four pigments (white, yellow, red, and blue). The color of the samples was measured with a spectrophotometer, then the L*, a*, and b* values were recorded. The relationship between the L*, a*, and b* values of each sample and the amount of each pigment in the compound of the same sample was used as the training dataset, entered into each algorithm, and the prediction models were obtained. While generating the prediction model for each sample, the data of the corresponding sample assigned as the target color were excluded. L*, a*, and b* values of each target sample were entered into the obtained models separately, and recipes indicating the ratios for mixing the four pigments were predicted. The mean absolute error (MAE) and root mean square error (RMSE) values between the original recipe used in the production of each silicone and the recipe created by both prediction models for the same silicone were calculated.Statistical Analysis Used:Data were analyzed with the Student t-test (α=0.05).Results:The mean RMSE values and MAE values for the ANN algorithm (0.029 ± 0.0152 and 0.045 ± 0.0235, respectively) were found significantly higher than the attention-based GRU model (0.001 ± 0.0005 and 0.002 ± 0.0008, respectively) (P < 0.001).Conclusions:Attention-based GRU model provided better performance than the ANN algorithm with respect to the MAE and RMSE values.

The climatic reanalysis datasets represent a crucial form of data that can help to address the shortage of data observations. This study evaluates the accuracy of NASA POWER reanalysis data by comparing it with AMMAN Environmental Department in-situ observations from four weather stations in West Sumbawa, collected from 2013 to 2022. The reanalysis data includes surface daily average temperatures, average wind speed, wind direction, relative humidity, and rainfall. The statistical analysis used in this research includes Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) to measure the differences between the observed and predicted values, both of which expressed in the same unit as the variable being measured. The comparison of 10 years of historical data revealed that the temperature parameter had the lowest RMSE and MAE values of 0.003 and 0.18, respectively, while the highest values were 0.46 and 5.11. For humidity, the lowest RMSE and MAE values were 0.001 and 3.6, respectively, and the highest values were 1.3 and 14.2. The RMSE and MAE values for rainfall were 0.02 and 0.9 (lowest) and 2.2 and 5.7 (highest). Regarding wind speed, the lowest RMSE and MAE values were 0.001 and 0.27, while the highest values were 0.07 and 1. Finally, the lowest RMSE and MAE values for wind direction were 0.1 and 6.6, and the highest values were 15.7 and 67.6. The comparison between the predicted and observed values showed a relatively high level of similarity for temperature, rainfall and wind speed parameters. However, parameters such as humidity and wind direction resulted varying degrees of deviation between the weather station data and the NASA POWER reanalysis data. These results highlight both the accuracy and discrepancy in the reanalysis data, emphasizing the importance of considering the limitations of such datasets when addressing the shortage of observational data.

The Christiansen Uniformity Coefficient (CUC) describes the distribution of water in a sprinkler system. In this study, two types of models were developed to predict the Christiansen Uniformity Coefficient (CUC) of sprinkler irrigation systems: Artificial Neural Network (ANN), specifically the feed-forward neural networks, and multiple linear regression (MLR) models. The models were trained on a dataset of published research on the CUC of sprinkler irrigation systems, which included data on a variety of design, operating, and meteorological condition variables. In order to build the predictive model of CUC, 10 input parameters were used including sprinkler height (H), working pressure (P), nozzle diameter (D and da), sprinkler line spacing (SL), sprinkler spacing (SS), wind speed (WS), wind direction (WD), temperature (T), and relative humidity (RH). Fifty percent (50%) of the data was used to train ANN models and the remaining data for cross-validation (25%) and for testing (25%). Multiple linear regression models were built using the training data. Four statistical criteria were used to evaluate the model’s predictive quality: the correlation coefficient (R), the index of agreement (d), the root mean square error (RMSE), and the mean absolute error (MAE). Statistical analysis demonstrated that the best predictive ability was obtained when the models (ANN and MLR) utilized all the input variables. The results demonstrated that the accuracy of ANN models, predicting the CUC of sprinkler irrigation systems, is higher than that of the MLR ones. During the training stage, the ANN models were more accurate in predicting CUC than MLR, with higher R (0.999) and d (0.999) values and lower MAE (0.167) and RMSE (0.456) values. The R values of the MLR model fluctuated between 0.226 and 0.960, the d values oscillated from 0.174 to 0.979, the MAE values were in the range of 2.458% and 10.792%, and the RMSE values fluctuated from 2.923% to 13.393%. Furthermore, the study revealed that WS and WD are the most influential climatic parameters. The ANN model can be used to develop more accurate tools for predicting the CUC of sprinkler irrigation systems. This can help farmers to design and operate their irrigation systems more efficiently, which can save them time and money.

Objective:This study aims to develop a mapping algorithm for EORTC QLQ-C30 to EQ-5D-5L which can produce utility values in patients with cancer. Methods:We used a cross sectional study design with 300 cancer patients. The research instruments used were EORTC QLQ-C30 and EQ-5D-5L. Data were collected by interviewing cancer patients who were hospitalized in the Kasuari Installation of Dr Kariadi Hospital Semarang, Indonesia. The Ordinary Least Squares (OLS) regression method was used to predict the utility value of EQ-5D-5L. This study uses two models to predict utility values, namely model 1 with all domains, and model 2 with domains that affect the EQ-5D-5L. The predictive power of regression on the model is evaluated by calculating the mean absolute error (MAE) and root mean square error (RMSE) values. Result:The highest score in the functional domain is the ‘emotional function’ domain (mean: 85.89; SD: 16.04) and the highest symptom domain is ‘weakness’ (mean: 36.21; SD:21.69). The predicted utility values of models 1 and 2 are 0.683. The mean absolute error (MAE) and root mean square error (RMSE) values of model 1 are 0.128 and 0.173, while in model 2 the MAE and RMSE values obtained are 0.125 and 0.168. Conclusion:The development of the mapping algorithm from the EORTC QLQ-C30 to EQ-5D-5L instrument shows a predictive value of utility in a sample of patients with cancer at Dr. Kariadi Hospital, Semarang, Indonesia. The utility prediction in both model is similar, however model 2 involves fewer domains and symptoms.

The arrival of the era of big data has realized the transformation of people’s production and lifestyle. At the same time, it also increases people’s desire to consume, and the feedback behavior of consumers’ comments and ratings is the feedback of users’ experience in merchants’ products, that is, the matching of products to consumer needs and preferences. When the product can reach the user’s satisfaction level, the customer-aware mobile terminal system is constructed and optimized by using the advanced methods and technologies of big data information display and the principles and laws of the collaborative filtering algorithm in cloud computing. It ensures the ecological development of the consumer industry. Among them, in the experimental evaluation of the collaborative filtering recommendation algorithm, the mean absolute error (MAE) and root mean square error (RMSE) values of the SVD++ algorithm are higher than those of the other three algorithm models, indicating that other algorithm models can effectively improve the accuracy of the recommendation algorithm. A cross-sectional comparative analysis of experimental results has shown that, as the number of neighbors increased, the MAE and RMSE values first decreased and then increased. When the number of neighbors N is 25, the MAE and RMSE reach the minimum value, so the optimal number of neighbors is 25. Therefore, it is very important to use the collaborative filtering algorithm to analyze and construct the consumer behavior and customer perception mobile terminal system.

Chlorophyll-a (Chl-a), an important indicator of phytoplankton biomass and eutrophication, is sensitive to water constitutes and optical characteristics. An integrated machine learning method of genetic algorithm and artificial neural networks (GA–ANN) was developed to retrieve the concentration of Chl-a. In situ spectra and simultaneous water quality parameters of 107 samples from two reservoirs (Res) and coastal waters (CW) were used to calibrate GA–ANN and three-band models (TBM) for comparison of Chl-a estimation. Both GA–ANN and TBM methods perform well for the joint dataset (WGD) of Res and CW with the R 2 exceeding 0.90, and the root mean square error (RMSE) of corresponding validation ( N = 35) are 4.40 and 5.23 μ g/L, respectively. Similarly, for independent dataset of Res ( N = 45), GA–ANN and TBM methods show robust performance: the R 2 values are 0.87 and 0.80, respectively; and the corresponding RMSE values are 7.79 and 7.73 μ g/L, respectively. For CW dataset ( N = 62), the R 2 values of two methods are 0.81 and 0.62, respectively; and the corresponding RMSE values are 0.79 and 1.32 μ g/L, respectively. When the GA–ANN and TBM models were applied to retrieve Chl-a concentration from the calibrated Sentinel 2 MSI reflectance data in two Res on October 20, 2019, however, the validated results of MSI-derived Chl-a concentrations using quasi-synchronous in situ data ( N = 36) indicated that the GA–ANN model outperforms TBM with higher R 2 value (0.91 vs. 0.26) and smaller RMSE (4.41 vs. 13.85 μ g/L) and mean absolute errors (3.40 vs. 11.87 μ g/L) values. Although TBM has obvious overestimation of Chl-a concentration when applied to remote sensing image, we still thought that both GA–ANN and TBM are useful methods for Chl-a estimation in case-II waters, and GA–ANN performs marginally better with less deviation to measured Chl-a for multispectral remote sensing data. The ratio of TSS to Chl-a, experimental measurements, abundance of sampling points, and Chl-a concentration range are several important factors affecting the accuracy and robustness of GA–ANN and TBM methods.