Improving the prediction performance of leaf water content by coupling multi-source data with machine learning in rice (Oryza sativa L.)

BACKGROUND

Leaf water content (LWC) significantly affects rice growth and development. Real-time monitoring of rice leaf water status is essential to obtain high yield and water use efficiency of rice plants with precise irrigation regimes in rice fields. Hyperspectral remote sensing technology is widely used in monitoring crop water status because of its rapid, nondestructive, and real-time characteristics. Recently, multi-source data have been attempted to integrate into a monitored model of crop water status based on spectral indices. However, there are fewer studies using spectral index model coupled with multi-source data for monitoring LWC in rice plants. Therefore, 2-year field experiments were conducted with three irrigation regimes using four rice cultivars in this study. The multi-source data, including canopy ecological factors and physiological parameters, were incorporated into the vegetation index to accurately predict LWC in rice plants.

RESULTS

The results presented that the model accuracy of rice LWC estimation after combining data from multiple sources improved by 6-44% compared to the accuracy of a single spectral index normalized difference index (ND). Additionally, the optimal prediction accuracy of rice LWC was produced using a machine algorithm of gradient boosted decision tree (GBDT) based on the combination of ND(1287,1673) and crop water stress index (CWSI) (R2 = 0.86, RMSE = 0.01).

CONCLUSIONS

The machine learning estimation model constructed based on multi-source data fully utilizes the spectral information and considers the environmental changes in the crop canopy after introducing multi-source data parameters, thus improving the performance of spectral technology for monitoring rice LWC. The findings may be helpful to the water status diagnosis and accurate irrigation management of rice plants.

A method to experimentally clamp leaf water content to defined values to assess its effects on apoplastic pH

BACKGROUND

Leaf hydration is controlled by feedback mechanisms, e.g. stomatal responses, adjustments of osmotic potential and hydraulic conductivity. Leaf water content thus is an input into related feedback-loops controlling the balance of water uptake and loss. Apoplastic alkalisation upon leaf dehydration is hypothesized to be involved together and in interaction with abscisic acid (ABA) in water stress related signaling on tissue level. However, important questions are still unresolved, e.g. the mechanisms leading to pH changes and the exact nature of its interaction with ABA. When studying these mechanisms and their intermediate signaling steps, an experimenter has only poor means to actually control the central experimental variable, leaf water content (LWC), because it is not only dependent on external variables (e.g. air humidity), which are under experimental control, but is also governed by the biological influences controlling transpiration and water uptake. Those are often unknown in their magnitude, unpredictable and fluctuating throughout an experiment and will prevent true repetitions of an experiment. The goal of the method presented here is to experimentally control and manipulate leaf water content (LWC) of attached intact leaves enclosed in a cuvette while simultaneously measuring physiological parameters like, in this case, apoplastic pH.

RESULTS

An experimental setup was developed where LWC is measured by a sensor based on IR-transmission and its signal processed to control a pump which circulates air from the cuvette through a cold trap. Hereby a feedback-loop is formed, which by adjusting vapour pressure deficit (VPD) and consequently leaf transpiration can precisely control LWC. This technique is demonstrated here in a combination with microscopic fluorescence imaging of apoplastic pH (pH_apo) as indicated by the excitation ratio of the pH sensitive dye OregonGreen. Initial results indicate that pH_apo of the adaxial epidermis of Vicia faba is linearly related to reductions in LWC.

CONCLUSIONS

Using this setup, constant LWC levels, step changes or ramps can be experimentally applied while simultaneously measuring physiological responses. The example experiments demonstrate that bringing LWC under experimental control in this way allows better controlled and more repeatable experiments to probe quantitative relationships between LWC and signaling and regulatory processes.

Comparison of new hyperspectral index and machine learning models for prediction of winter wheat leaf water content

BACKGROUND

The leaf water content estimation model is established by hyperspectral technology, which is crucial and provides technical reference for precision irrigation.

METHODS

In this study, two consecutive years of field experiments (different irrigation times and seven wheat varieties) in 2018-2020 were performed to obtain the canopy spectra reflectance and leaf water content (LWC) data. The characteristic bands related to LWC were extracted from correlation coefficient method (CA) and x-Loading weight method (x-Lw). Five modeling methods, spectral index and four other methods (Partial Least-Squares Regression (PLSR), Random Forest Regression (RFR), Extreme Random Trees (ERT), and K-Nearest Neighbor (KNN)) based characteristic bands, were employed to construct LWC estimation models.

RESULTS

The results showed that the canopy spectral reflectance increased with the increase of irrigation times, especially in the near-infrared band (750-1350 nm). The prediction accuracy of the newly developed differential spectral index DVI (R1185, R1307) was higher than that of the existing spectral index, with R² of 0.85 and R² of 0.78 for the calibration and validation, respectively. Due to a large amount of hyperspectral data, the correlation coefficient method (CA) and x-Loading weight (x-Lw) were used to select the water characteristic bands (100 and 28 characteristic bands, respectively) from the full spectrum. We found that the accuracy of the model based on the characteristic bands was not significantly lower than that of the full spectrum-based models. Among these models, the ERT- x-Lw model performed the best (R² and RMSE of 0.88 and 1.46; 0.84 and 1.62 for the calibration and validation, respectively). In addition, the accuracy of the LWC estimation model constructed by ERT-x-Lw was higher than that of DVI (R1185, R1307).

CONCLUSION

The two models based on ERT-x-Lw and DVI (R1185, R1307) can effectively predict wheat leaf water content. The results provide a technical reference and a basis for crop water monitoring and diagnosis under similar production conditions.

Evaluation of different water absorption bands, indices and multivariate models for water-deficit stress monitoring in rice using visible-near infrared spectroscopy

Accurate estimation of plant water status is a major factor in the decision-making process regarding general land use, crop water management and drought assessment. Visible-near infrared (VNIR) spectroscopy can provide an effective means for real-time and non-invasive monitoring of leaf water content (LWC) in crop plants. The current study aims to identify water absorption bands, indices and multivariate models for development of non-destructive water-deficit stress phenotyping protocols using VNIR spectroscopy and LWC estimated from 10 different rice genotypes. Existing spectral indices and band depths at water absorption regions were evaluated for LWC estimation. The developed models were found efficient in predicting LWC of the samples kept in the same environment with the ratio of performance to deviation (RPD) values varying from 1.49 to 3.05 and 1.66 to 2.63 for indices and band depths, respectively during validation. For identification of novel indices, ratio spectral indices (RSI) and normalised difference spectral indices (NDSI) were calculated in every possible band combination and correlated with LWC. The best spectral indices for estimating LWC of rice were RSI (R₁₈₃₀, R₁₈₃₄) and NDSI (R₁₈₃₀, R₁₈₃₄) with R² greater than 0.90 during training and validation, respectively. Among the multivariate models, partial least squares regression (PLSR) provided the best results for prediction of LWC (RPD = 6.33 and 4.06 for training and validation, respectively). The approach developed in this study will also be helpful for high-throughput water-deficit stress phenotyping of other crops.

Indirect effects of water availability in driving and predicting productivity in the Gobi desert

Climate is the fundamental determinant of plant metabolism and net primary productivity (NPP). However, whether climate drives NPP directly or indirectly is not well understand. The Gobi desert across a precipitation gradient in the arid zone provides an ideal naturally-controlled platform for studying the precipitation-productivity relationships. We conducted 3-year experiments in four Gobi desert shrublands across an aridity gradient in Gansu Province of China to test the relationship between water availability and shrub productivity as well as the relative importance of the possible factors driving productivity (using piecewise structural equation modeling) and to explore the appropriate variables for predicting productivity (using three spatial models). The results showed that water availability indirectly affected the NPP via stand biomass, while stand biomass had a significant direct effect on NPP regardless of whether the leaf water content and stand height were considered. The model based on stand size (71.6%) and the model that contained both stand size and water availability (72.3%) explained more of the variation in the water-NPP relationships than the model based on water availability (37.3%). Our findings suggest that even in extremely water-limited areas, the effects of water availability on plant growth and the kinetics of plant metabolism could be indirect via plant size, demonstrating the importance of plant size as an indicator of shrub productivity. This study explains the mechanisms underlying the NPP driving pattern and proposes a practical NPP model for arid ecosystems.

Evaluating the utility of solar-induced chlorophyll fluorescence for drought monitoring by comparison with NDVI derived from wheat canopy

Normalized Difference Vegetation Index (NDVI) has been extensively used in continuous and long-term drought monitoring over large-scale, but with late response to drought-related changes of photosynthesis. Instead, solar-induced chlorophyll fluorescence (SIF) is more closely related to photosynthesis and thus is proposed to track the impacts of drought on vegetation growth. However, the detailed difference between SIF and NDVI in responding to drought has not been thoroughly explored. Here we present continuous ground measurements of NDVI and SIF at 760nm over four plots of wheat with different intensities of drought (well-watered treatment, moderate drought, severe drought and extreme drought). The average values of seasonal SIF were significantly lower under severe drought and extreme drought, while NDVI means only showed significant reduction in extreme drought. In the seasonal patterns, daily SIF could clearly separate the difference of drought gradient, while the difference of daily NDVI was clearer in the end of the field campaign. Daily SIF also significantly and positively correlated with soil moisture, indicating that SIF could be considered as an estimator of soil moisture to detect the information about agricultural drought. Furthermore, in extreme drought plot, the correlation of SIF and soil moisture was higher than that of NDVI and soil moisture in a shorter time lag (<15-day) but weaker in a longer time lag (longer than 30-day). The relationships of growth parameters with SIF and NDVI were further analyzed, showing a saturation of NDVI and unsaturation of SIF at high values of leaf area index and relative water content. These results suggested that SIF is better fit in early drought monitoring, especially over closure canopy, while NDVI is more feasible when drought lasted over a long time scale. Our findings in the study might provide deep insight into the utility of SIF in drought monitoring.

Assessment of MODIS-derived indices (2001-2013) to drought across Taiwan's forests

Tropical and subtropical ecosystems, the largest terrestrial carbon pools, are very susceptible to the variability of seasonal precipitation. However, the assessment of drought conditions in these regions is often overlooked due to the preconceived notion of the presence of high humidity. Drought indices derived from remotely sensed imagery have been commonly used for large-scale monitoring, but feasibility of drought assessment may vary across regions due to climate regimes and local biophysical conditions. Therefore, this study aims to evaluate the feasibility of 11 commonly used MODIS-derived vegetation/drought index in the forest regions of Taiwan through comparison with the station-based standardized precipitation index with a 3-month time scale (SPI3). The drought indices were further transformed (standardized anomaly, SA) to make them better delineate the spatiotemporal variations of drought conditions. The results showed that the Normalized Difference Infrared Index utilizing the near-infrared and shortwave infrared bands (NDII6) may be more superior to other indices in delineating drought patterns. Overall, the NDII6 SA-SPI3 pair yielded the highest correlation (mean r ± standard deviation = 0.31 ± 0.13) and was most significant in central and south Taiwan (r = 0.50-0.90) during the cold, dry season (January and April). This study illustrated that the NDII6 is suitable to delineate drought conditions in a relatively humid region. The results suggested the better performance of the NDII6 SA-SPI3 across the high climate gradient, especially in the regions with dramatic interannual amplifications of rainfall. This synthesis was conducted across a wide bioclimatic gradient, and the findings could be further generalized to a much broader geographical extent.

Gaussian processes retrieval of leaf parameters from a multi-species reflectance, absorbance and fluorescence dataset

Biochemical and structural leaf properties such as chlorophyll content (Chl), nitrogen content (N), leaf water content (LWC), and specific leaf area (SLA) have the benefit to be estimated through nondestructive spectral measurements. Current practices, however, mainly focus on a limited amount of wavelength bands while more information could be extracted from other wavelengths in the full range (400-2500nm) spectrum. In this research, leaf characteristics were estimated from a field-based multi-species dataset, covering a wide range in leaf structures and Chl concentrations. The dataset contains leaves with extremely high Chl concentrations (>100μgcm(-2)), which are seldom estimated. Parameter retrieval was conducted with the machine learning regression algorithm Gaussian Processes (GP), which is able to perform adaptive, nonlinear data fitting for complex datasets. Moreover, insight in relevant bands is provided during the development of a regression model. Consequently, the physical meaning of the model can be explored. Best estimates of SLA, LWC and Chl yielded a best obtained normalized root mean square error of 6.0%, 7.7%, 9.1%, respectively. Several distinct wavebands were chosen across the whole spectrum. A band in the red edge (710nm) appeared to be most important for the estimation of Chl. Interestingly, spectral features related to biochemicals with a structural or carbon storage function (e.g. 1090, 1550, 1670, 1730nm) were found important not only for estimation of SLA, but also for LWC, Chl or N estimation. Similar, Chl estimation was also helped by some wavebands related to water content (950, 1430nm) due to correlation between the parameters. It is shown that leaf parameter retrieval by GP regression is successful, and able to cope with large structural differences between leaves.

Influence of leaf traits on the spatial distribution of insect herbivores associated with an overstorey rainforest tree

The spatial distribution of insect herbivores associated with the Australian rainforest treeArgyrodendron actinophyllum (Sterculiaceae) was investigated by restricted canopy fogging. The foliage of this species was low in nitrogen and water content, and high in fibre content. Herbivore abundance was positively correlated with the amount of young foliage present within the samples and in adjacent samples, and with the nitrogen content of young leaves. In particular, the occurrence of phloem-feeders was correlated with the magnitude of translocation within the samples. The influence of leaf water content upon herbivore distribution was marginal, presumably because this factor is not limiting in rain-forest environments during the wet season, which usually coincides with the season of leaf-flush. Specific leaf weight, leaf size and foliage compactness had little or no apparent effect on herbivore distribution. Since the magnitude of leaf turnover affected both the quantity and the quality, as exemplified by translocation effects, of young foliage available, this factor may be critical to herbivores associated with evergreen rainforest trees which are particularly low in foliar nutrients, such asA. actinophyllum.