Estimating Gross Primary Production via Recurrent Neural Networks: A comparative analysis

An attempt to introduce atmospheric CO2 concentration data to estimate the gross primary production by the terrestrial biosphere and analyze its effects

ABSTRACTVegetation gross primary production (GPP), the photosynthetic yields by green plants per unit area per unit time, is a key metric of carbon flux in understanding the land–atmosphere interactions and terrestrial carbon cycles. Satellite-based light use efficiency (LUE) models are valuable methods to retrieve large-scale terrestrial GPP using remote sensing data. As studies have reported that maximum light use efficiency, a key parameter that is often assumed to be constant in the LUE models, there is a need to explore the effects of LUE seasonality on GPP simulation and ways for correction. This study proposes a method based on leaf area index to account for LUE seasonality and applies it to four different light use efficiency models (i.e., the MOD17 algorithm, the vegetation photosynthesis model, the radiation partitioning model, and the vegetation index model) for comparisons. Based on 59 site-years flux tower data from deciduous broadleaf forest sites in the United States, the results show that all models could simulate daily GPP time series well and explain more than 85.0% variance of tower-based GPP. There is, however, a tendency to overestimate GPP during the non-growing season but underestimate GPP during the growing season. By applying the correction function, GPP simulation using the LUE models improved in all experiments as indicated by increased correlation coefficients, the index of agreement and decreased root-mean-square errors. Among all models, the radiation partitioning model achieves the highest correlation coefficients between modelled and observed daily GPP likely because it considers the influences of direct and diffuse radiation partitioning on daily canopy photosynthesis. Our study indicates that satellite-based light use efficiency models could be successfully applied for deriving daily vegetation GPP and potentially producing daily routine satellite products, while considering the effects of LUE seasonality on canopy could help improve significantly the simulation accuracy of daily GPP in phenology.

The quantification of forest Gross Primary Productivity (GPP) has been the focus of many scientific studies (e.g. carbon cycle, climate change, etc.). Current remote sensing-based models (i.e., the MODIS MOD_17 model), rely on the accurate meteorological data, specific vegetation parameter, the applicability and explicability of remote sensing data. In this study, the original MODIS GPP products were validated and showed significant underestimation compared to the eddy covariance measurements of the four forest sites over China. Thus the strategy of simple yet accurate and quantitative simulation of carbon fluxes was improved by using Sims TG model which was termed the Temperature and Greenness (TG) model and included the land surface temperature (LST) product and enhanced vegetation index (EVI) product from MODIS. The results indicated that Sims TG model was poor adaptive to tropical and subtropical evergreen forest in China. The model precision of deciduous forest was high but the GPP of evergreen forest are underestimated in Qianyanzhou and Xishuangbanna station in summer. After the parameter was optimized in Sims TG model, the estimation accuracy of evergreen forest GPP was improved to a certain extent and the model better adapt to dynamic change of forest GPP in China.

Maize is one of the most important crops cultivated on the global scale. Accurate estimation of maize Gross Primary Production (GPP) can provide valuable information for regional and global carbon budget studies. From site level to regional/global scales, GPP estimation depends on remote sensing or eddy covariance flux data. In this research, the 8-day composite GPP of maize was estimated by Moderate Resolution Imaging Spectroradiometer (MODIS) and flux tower data at eight study sites using a Regional Production Efficiency Model (REG-PEM). The performance of the model was assessed by analyzing the linearly regression of GPP estimated from the REG-PEM model (GPPEST) with the GPP predicted from the eddy covariance data (GPPEC). The coefficient of determination, root mean squared error and mean absolute error of the regression model were calculated. The uncertainties of the model are also discussed in this research. The seasonal dynamics (phases and magnitudes) of the GPPEST reasonably agreed with those of GPPEC, indicating the potential of the satellite-driven REG-PEM model for up-scaling the GPP in maize croplands. Furthermore, the maize GPP estimated by this model is more accurate than the MODIS GPP products (MOD17A2). In particular, MOD17A2 significantly underestimated the GPP of maize croplands. The uncertainties in the REG-PEM model are mostly contributed by the maximum light use efficiency and the fraction of photosynthetically active radiation. Keywords: gross primary production, maize, light use efficiency, MODIS, remote sensing, flux data DOI: 10.3965/j.ijabe.20160902.2245 Citation: Li S H, He P, Liu B S, Ni P, Han X. Modeling of maize gross primary production using MODIS imagery and flux tower data. Int J Agric & Biol Eng, 2016; 9(2): 110－118.

Recurrent Neural Networks (RNNs), which are a powerful scheme for modeling temporal and sequential data need to capture long-term dependencies on datasets and represent them in hidden layers with a powerful model to capture more information from inputs. For modeling long-term dependencies in a dataset, the gating mechanism concept can help RNNs remember and forget previous information. Representing the hidden layers of an RNN with more expressive operations (i.e., tensor products) helps it learn a more complex relationship between the current input and the previous hidden layer information. These ideas can generally improve RNN performances. In this paper, we proposed a novel RNN architecture that combine the concepts of gating mechanism and the tensor product into a single model. By combining these two concepts into a single RNN, our proposed models learn long-term dependencies by modeling with gating units and obtain more expressive and direct interaction between input and hidden layers using a tensor product on 3-dimensional array (tensor) weight parameters. We use Long Short Term Memory (LSTM) RNN and Gated Recurrent Unit (GRU) RNN and combine them with a tensor product inside their formulations. Our proposed RNNs, which are called a Long-Short Term Memory Recurrent Neural Tensor Network (LSTMRNTN) and Gated Recurrent Unit Recurrent Neural Tensor Network (GRURNTN), are made by combining the LSTM and GRU RNN models with the tensor product. We conducted experiments with our proposed models on word-level and character-level language modeling tasks and revealed that our proposed models significantly improved their performance compared to our baseline models.

Estimation of crop gross primary production (GPP): I. impact of MODIS observation footprint and impact of vegetation BRDF characteristics

This paper examines and evaluates the effectiveness of recurrent neural network (RNN) architectures for natural language knowledge. The paper starts by imparting historical data on recurrent neural networks and their software to various responsibilities. It then goes on to talk about diverse RNN architectures, such as lengthy quick-time period reminiscence (LSTM), Gated Recurrent Unit (GRU) and Bonior Recurrent Neural network (BONN). The paper then focuses on the efficacy of each RNN structure for textual content expertise obligations. A sequence of experiments is carried out to degree the overall performance of the one-of-a-kind fashions, the use of 3 preferred datasets-Wiki text, Squad and IMDB, as well as two newly created datasets. The paper analyses the empirical consequences of the various RNN architectures and finds that LSTM fashions are normally satisfactory performers, although BONN architectures can also be aggressive in some cases. The paper additionally looks at numerous techniques to enhance performance, which include pre-processing and hyper-parameter tuning. The paper concludes that, whilst RNN architectures are powerful for herbal language know-how responsibilities, the exact architecture and parameters can also need to be chosen cautiously, relying on the project and dataset to hand.

Changes in tropical forest carbon sink strength during El Niño Southern Oscillation (ENSO) events can indicate future behavior under climate change. Previous studies revealed ˜6MgCha-1 yr-1 lower net ecosystem production (NEP) during ENSO year 1998 compared with non-ENSO year 2000 in a Costa Rican tropical rainforest. We explored environmental drivers of this change and examined the contributions of ecosystem respiration (RE) and gross primary production (GPP) to this weakened carbon sink. For 1998-2000, we estimated RE using chamber-based respiration measurements, and we estimated GPP in two ways: using (1) the canopy process model MAESTRA, and (2) combined eddy covariance and chamber respiration data. MAESTRA-estimated GPP did not statistically differ from GPP estimated using approach 2, but was ˜28% greater than published GPP estimates for the same site and years using eddy covariance data only. A 7% increase in RE (primarily increased soil respiration) and a 10% reduction in GPP contributed equally to the difference in NEP between ENSO year 1998 and non-ENSO year 2000. A warming and drying climate for tropical forests may yield a weakened carbon sink from both decreased GPP and increased RE. Understanding physiological acclimation will be critical for the large carbon stores in these ecosystems.

Gross primary productivity (GPP) quantifies the rate at which plants convert atmospheric carbon dioxide into organic matter through photosynthesis, playing a vital role in the terrestrial carbon cycle. Machine learning (ML) techniques excel in handling spatiotemporally complex data, facilitating accurate spatial-scale inversion of forest GPP by integrating limited ground flux measurements with Remote Sensing (RS) observations. Enhancing ML algorithm performance for precise GPP estimation is a key research focus. This study introduces the Random Grid Search Algorithm (RGSA) for hyperparameters tuning to improve Random Forest (RF) and eXtreme Gradient Boosting (XGB) models across four major forest regions in China. Model optimization progressed through three stages: the Unoptimized (UO) XGB model achieved R2 = 0.77 and RMSE = 1.42 g Cm−2 d−1; the Hyperparameter Optimized (HO) XGB model using RGSA improved performance by 5.19% in R2 (0.81) and reduced RMSE by 9.15% (1.29 g Cm−2 d−1); the Hyperparameter and Variable Combination Optimized (HVCO) XGB model with selected variables (LAI, Temp, NR, VPD, and NDVI) further enhanced R2 to 0.83 and decreased RMSE to 1.23 g Cm−2 d−1. The optimized GPP estimates exhibited high spatial consistency with existing high-quality products like GOSIF GPP, GLASS GPP, and FLUXCOM GPP, validating the model’s reliability and effectiveness. This research provides crucial insights for improving GPP estimation accuracy and optimizing ML methodologies for forest ecosystems in China.

We explore the architecture of recurrent neural networks (RNNs) by studying the complexity of string sequences it is able to memorize. Symbolic sequences of different complexity are generated to simulate RNN training and study parameter configurations with a view to the network's capability of learning and inference. We compare Long Short-Term Memory (LSTM) networks and gated recurrent units (GRUs). We find that an increase in RNN depth does not necessarily result in better memorization capability when the training time is constrained. Our results also indicate that the learning rate and the number of units per layer are among the most important hyper-parameters to be tuned. Generally, GRUs outperform LSTM networks on low-complexity sequences while on high-complexity sequences LSTMs perform better.

MOOCs (Massive Open Online Courses) are definitely one of the best approach to support the international agenda about inclusive and equitable education and lifelong learning opportunities for all (SDG4) [1]. A great deal universities and institutions offer valuable free courses to their numerous students and to people around the word through MOOC platforms. However, because of huge number of learners and data generated, learner’s behaviour in those platforms remain a kind of black box for learners themselves and for courses instructors who are supposed to tutor or monitor learners in the learning process. Therefore, learner do not receive sufficient support from instructors and from their peers, during the learning process [2]. This is one the main reasons that lead to high dropout, low completion and success rate observed in the MOOCs. Many research work have suggested descriptive, predictive and prescriptive models to address this issue, but most of these models focus on predicting dropout, completion and/or success, and do not generally pay enough attention to one of the key step (learner behaviour), that comes before, and can explain dropping out and failure. Our research aims to develop a deep learning model to predict learner behaviour (learner interactions) in the learning process, in order to equip learners and course instructors with insight understanding of the learner behaviour in the learning process. This specific paper will focus on analysing relevant algorithms to develop such model. For this analysis, we used data from UNESCO-IICBA (International Institute for Capacity Building in Africa) MOOC platform, designed for teacher training in Africa, and then we examine many types of features: geographical, social behavioural and learning behavioural features. Learner’s behaviour being a time series Big data, we built the predictive model using Deep Learning algorithms for better performance and accuracy (Thanks to the power of deep learning) compared to baseline Machine learning algorithms. Time series data is best handled by recurrent neural networks (RNN) [3], so, we choose RNN and implemented/tested the three main architectures of RNN: Simple RNNs, GRU (Gated Recurrent Unit) RNNs and LSTM (Long short-term memory) RNNs. The models were trained using L2 Regularization, based on the predictions results, we unexpectedly found model with simple RNNs produced the best performance and accuracy on the dataset used than the other RNN architectures. We had couple of observations, example: we saw a correlation between video viewing and quiz behaviour and the participation of the learner to the learning process. This observation could allow teachers to provide meaningful support and guidance to at risk or less active students. We also observed that, the shorter the video or the quiz, the more the viewer. We conclude that we could use learner video or quiz viewing behaviour to predict his behaviour concerning other MOOC contents. These results suggest the need of deeper research on educational video and educational quiz designing for MOOCs.

Recurrent neural networks (RNN) are ubiquitous computing systems for sequences and multivariate time-series data. While several robust RNN architectures are known, it is unclear how to relate RNN initialization, architecture, and other hyperparameters with accuracy for a given task. In this work, we propose treating RNN as dynamical systems and correlating hyperparameters with accuracy through Lyapunov spectral analysis, a methodology designed explicitly for nonlinear dynamical systems. To address the fact that RNN features go beyond the existing Lyapunov spectral analysis, we propose to infer relevant features from the Lyapunov spectrum with an Autoencoder and an embedding of its Latent representation (AeLLE). Our studies of various RNN architectures show that AeLLE successfully correlates RNN Lyapunov spectrum with accuracy. Furthermore, the Latent representation learned by AeLLE is generalizable to novel inputs from the same task and is formed early in the process of RNN training. The latter property allows for predicting the accuracy to which RNN would converge when training is complete. We conclude that the representation of RNN through the Lyapunov spectrum, along with AeLLE, provides a novel method for the organization and interpretation of variants of RNN architectures.

Forecasting longevity is essential in multiple research and policy areas, including the pricing of life insurance contracts, the valuation of capital market solutions for longevity risk management, and pension policy. This paper empirically investigates the predictive performance of Recurrent Neural Networks (RNN) with Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) architectures in jointly modeling and multivariate time series forecasting of age-specific mortality rates at all ages. We fine-tune the three hidden layers GRU and LSTM model’s hyperparameters for time series forecasting and compare the model’s forecasting accuracy with that produced by traditional Generalised Age-Period-Cohort (GAPC) stochastic mortality models. The empirical results suggest that the two RNN architectures generally outperform the GAPC models investigated in both the male and female populations, but the results are sensitive to the accuracy criteria. The empirical results also show that the RNN-GRU network slightly outperforms the RNN with an LSTM architecture and can produce mortality schedules that capture relatively well the dynamics of mortality rates across age and time. Further investigations considering other RNN architectures, calibration procedures, and sample datasets are necessary to confirm the superiority of RNN in forecasting longevity.

Improving tropical forest current biomass estimates can help more accurately evaluate ecosystem services in tropical forests. The Global Ecosystem Dynamics Investigation (GEDI) lidar provides detailed 3D forest structure and height data, which can be used to improve above‐ground biomass estimates. However, there is still debate on how best to predict tropical forest biomass using GEDI data. Here we compare stand biomass predicted by GEDI data with the observed data of 2,102 inventory plots in tropical forests and find that adding a remotely sensed (RS) trait map of leaf mass area (LMA) significantly (P < 0.001) improves field biomass predictions, but by only a small amount (r2 = 0.01). However, it may also help reduce the bias of the residuals because there was a negative relationship between both LMA (r2 of 0.34) and percentage of phosphorus (%P, r2 = 0.31) and residuals. Leaf spectral data (400–1,075 nm) from 523 individual trees along a Peruvian tropical forest elevation gradient predicted Diameter at Breast height (DBH) (the critical measurement underlying plot biomass) with an r2 = 0.01 and LMA predicts DBH with an r2 = 0.04. Other data sets may offer further improvements and max temperature (Tmax) predicts Amazonian biomass residuals with an r2 of 0.76 (N = 66). Finally, for a network of net primary production (NPP) and gross primary production (GPP) plots (N = 21), leaf traits predicted with remote sensing are better at predicting fluxes than structure variables. Overall, trait maps, especially future improved ones produced by Surface Biology Geology, may improve biomass and carbon flux predictions by a small but significant amount.

A new land surface temperature fusion strategy based on cumulative distribution function matching and multiresolution Kalman filtering