Applications of remote sensing vis-à-vis machine learning in air quality monitoring and modelling: a review.

Environmental contamination especially air pollution is an exponentially growing menace requiring immediate attention, as it lingers on with the associated risks of health, economic and ecological crisis. The special focus of this study is on the advances in Air Quality (AQ) monitoring using modern sensors, integrated monitoring systems, remote sensing and the usage of Machine Learning (ML), Deep Learning (DL) algorithms, artificial neural networks, recent computational techniques, hybridizing techniques and different platforms available for AQ modelling. The modern world is data-driven, where critical decisions are taken based on the available and accessible data. Today's data analytics is a consequence of the information explosion we have reached. The current research also tends to re-evaluate its scope with data analytics. The emergence of artificial intelligence and machine learning in the research scenario has radically changed the methodologies and approaches of modern research. The aim of this review is to assess the impact of data analytics such as ML/DL frameworks, data integration techniques, advanced statistical modelling, cloud computing platforms and constantly improving optimization algorithms on AQ research. The usage of remote sensing in AQ monitoring along with providing enormous datasets is constantly filling the spatial gaps of ground stations, as the long-term air pollutant dynamics is best captured by the panoramic view of satellites. Remote sensing coupled with the techniques of ML/DL has the most impact in shaping the modern trends in AQ research. Current standing of research in this field, emerging trends and future scope are also discussed.

Structure and composition of field margin vegetation in the rural-urban interface of Bengaluru, India: a case study on an unexplored dimension of agroecosystems.

Field margin vegetation (FMV) refers to the plant community in the interface between agriculture and natural environments. Substantial work has been carried out on the management of field margins in European countries with the aim of conserving field-level biodiversity and enhancing agronomic benefits. India, instead, is lagging behind in the assessment of FMV and formulating subsequent management strategies for biodiversity conservation at the field boundaries. This study is a first step to better understand the structural and functional dimensions of field margin vegetation along an agricultural transformation gradient near the megacity of Bengaluru, India. Empirical field studies along with the detection of vegetation change using remote sensing and geo-informatics technique were used to record information on field margin vegetation. The phytosociological study, revealed a total of 81 species, comprising 29 species of trees, 21 shrubs and 31 herbs at the field margins of six selected villages of northern Bengaluru. Randomly selected 355 field boundaries were delineated from high-resolution Worldview 3 images for the year 2018 and from Google Earth images for the year 2004-2005. The FMV area was around to 85.40 ha in 2004-2005 but declined to 76.69 ha in 2017-2018. The survey also indicated that local farmers have in-depth ecological knowledge on the importance of FMV in ensuring a sustainable flow of resources within the agricultural landscape. The results demonstrate that rural and transition zones of the study area have higher dominance of planted tree species on the margins, whereas urban zone exhibits comparatively uniform dominance for all species. Our study also highlights the need for conservation of FMV to ensure agroecosystem health as a prerequisite for sustainable socioecological development.

Deep learning-based prediction of plant height and crown area of vegetable crops using LiDAR point cloud.

Remote sensing has been increasingly used in precision agriculture. Buoyed by the developments in the miniaturization of sensors and platforms, contemporary remote sensing offers data at resolutions finer enough to respond to within-farm variations. LiDAR point cloud, offers features amenable to modelling structural parameters of crops. Early prediction of crop growth parameters helps farmers and other stakeholders dynamically manage farming activities. The objective of this work is the development and application of a deep learning framework to predict plant-level crop height and crown area at different growth stages for vegetable crops. LiDAR point clouds were acquired using a terrestrial laser scanner on five dates during the growth cycles of tomato, eggplant and cabbage on the experimental research farms of the University of Agricultural Sciences, Bengaluru, India. We implemented a hybrid deep learning framework combining distinct features of long-term short memory (LSTM) and Gated Recurrent Unit (GRU) for the predictions of plant height and crown area. The predictions are validated with reference ground truth measurements. These predictions were validated against ground truth measurements. The findings demonstrate that plant-level structural parameters can be predicted well ahead of crop growth stages with around 80% accuracy. Notably, the LSTM and the GRU models exhibited limitations in capturing variations in structural parameters. Conversely, the hybrid model offered significantly improved predictions, particularly for crown area, with error rates for height prediction ranging from 5 to 12%, with deviations exhibiting a more balanced distribution between overestimation and underestimation This approach effectively captured the inherent temporal growth pattern of the crops, highlighting the potential of deep learning for precision agriculture applications. However, the prediction quality is relatively low at the advanced growth stage, closer to the harvest. In contrast, the prediction quality is stable across the three different crops. The results indicate the presence of a robust relationship between the features of the LiDAR point cloud and the auto-feature map of the deep learning methods adapted for plant-level crop structural characterization. This approach effectively captured the inherent temporal growth pattern of the crops, highlighting the potential of deep learning for precision agriculture applications.

Ultra-high-resolution hyperspectral imagery datasets for precision agriculture applications.

Technology infusion in agriculture has been progressing steadily, touching upon various spheres of agriculture such as crop identification, soil classification, yield prediction, disease detection, and weed-crop discrimination. On-demand crop type detection, often realized as crop mapping, is a primary requirement in agriculture. Alongside the topographic LiDAR and thermal imaging, hyperspectral remote sensing is a versatile technique for mapping and predicting various parameters of interest in agriculture. The ongoing developments in the methods and algorithms of remote sensing data analyses for crop mapping require the availability of curated, high-resolution hyperspectral datasets, varied by crop type, nutrient supply (nitrogen level), and ground truth data. Aimed at enabling the development and validation of approaches for crop mapping at the plant level, we present a high-resolution ground-based hyperspectral imaging dataset acquired over fields of two vegetable crops (cabbage, eggplant). These crops were grown on experimental plots of the University of Agricultural Sciences, Bengaluru, India, maintaining three different nitrogen levels (high, medium, and low). The datasets contain hyperspectral imagery of the vegetable crops grown under two configurations: (i) imagery, which contains only a single crop type in a scene, and (ii) imagery, which contains both crops in a single scene. In both configurations, each crop has plots representing three different nitrogen levels. Ultra-high spatial resolution hyperspectral imaging data were acquired in 400 to 900 nm with an effective spectral resolution of 3 nm and spatial resolution of 3 mm using a ground-based push-broom hyperspectral imaging system (Headwall Photonics, USA). Ground truth data were also presented. The datasets are valuable for developing and validating various methods and algorithms for precision agriculture applications, such as machine learning methods for crop mapping at plants and estimating crop growth responses to different nitrogen levels.

Precision crop mapping: within plant canopy discrimination of crop and soil using multi-sensor hyperspectral imagery.

Leveraging diverse optomechanical and imaging technologies for precision agriculture applications is gaining attention in emerging economies. The precise spatial detection of plant objects in farms is crucial for optimizing plant-level nutrition and managing pests and diseases. High-resolution remote sensors mounted on drones have been increasingly deployed for large-scale crop mapping and field variability characterization. While field-level crop identification and crop-soil discrimination have been studied extensively, within-plant canopy discrimination of crop and soil has not been explored in real agricultural farms. The objectives of this study are: (i) adoption and assessment of spectral unmixing for discriminating crop and soil at within-plant canopy level, and (ii) generation of benchmark terrestrial and drone-based hyperspectral datasets for plant or sub-plant level discrimination using various spectral mixture modelling approaches and sources of endmembers. We acquired hyperspectral imagery of vegetable crops using a frame-based sensor mounted on a drone flying at different heights. Further, several linear, non-linear, and sparse-based spectral unmixing methods were used to discriminate plant and soil based on spectral signatures (endmembers) extracted from different spectral libraries prepared using in situ or field, ground-based, and drone-based hyperspectral imagery. The results, validated against pixel-to-pixel ground truth data, indicate an overall crop-soil discrimination accuracy of 99-100%, subject to a combination of endmember source and flying height. The influences of different endmember sources, spatial resolution as indicated by flying height, and inversion algorithms on the quality of estimated abundances are assessed from a verifiable and functionally relevant perspective. The generated hyperspectral datasets and ground truth data can be used for developing and testing new methods for sub-canopy level soil-crop discrimination in various agricultural applications of remote sensing.

LiDAR-based reference aboveground biomass maps for tropical forests of South Asia and Central Africa.

Accurate mapping and monitoring of tropical forests aboveground biomass (AGB) is crucial to design effective carbon emission reduction strategies and improving our understanding of Earth's carbon cycle. However, existing large-scale maps of tropical forest AGB generated through combinations of Earth Observation (EO) and forest inventory data show markedly divergent estimates, even after accounting for reported uncertainties. To address this, a network of high-quality reference data is needed to calibrate and validate mapping algorithms. This study aims to generate reference AGB datasets using field inventory plots and airborne LiDAR data for eight sites in Central Africa and five sites in South Asia, two regions largely underrepresented in global reference AGB datasets. The study provides access to these reference AGB maps, including uncertainty maps, at 100 m and 40 m spatial resolutions covering a total LiDAR footprint of 1,11,650 ha [ranging from 150 to 40,000 ha at site level]. These maps serve as calibration/validation datasets to improve the accuracy and reliability of AGB mapping for current and upcoming EO missions (viz., GEDI, BIOMASS, and NISAR).

In-situ and airborne hyperspectral data for detecting agricultural activities in a dense forest landscape.

Maintaining rich biodiversity and being a habitat and resource for humans, tropical forests are one of the most important global biomes. These forest ecosystems have been experiencing a host of unregulated anthropogenic activities including illegal tourism, and shifting cultivation. The presence of human-habitats in the restricted zones of forest ecosystems is a direct indicator of the human activities that may accelerate deterioration of forest quality by area and tree species composition. Remote sensing data have been extensively used for mapping forest types, and biophysical characterization at various spatial scales. Several remote sensing datasets from multispectral, hyperspectral and LIDAR sensors are available for developing and validating a host of methodologies for remote sensing application in forestry. However, quantifying the quality of forest stands and detecting potential threats from the sporadic and small-scale human activities requires sub-pixel level remote sensing data analysis methods such as, spectral mixture modelling. Generally, most of the studies employ pixel-level supervised learning-based analysis techniques to detect infrastructure and settlements. However, if the settlements are smaller than the ground sampling distance and are under the canopy, pixel-based techniques are not suitable. Reinvigorated with progressive availability of hyperspectral imagery, spectral mixture modelling based sub-pixel image analysis is gaining prominence in the contemporary remote sensing application development. However, there is a paucity of high-resolution hyperspectral imagery and associated ground truth spectral measurements for assessing various methodological approaches on studies related to anthropogenic activities and forest disturbance. Most of the studies have relied upon simulating and synthesising the hyperspectral imagery and its associated ground truth spectra for implementation of methods and algorithms. This article presents a distinct dataset of high-resolution hyperspectral imagery and associated ground truth spectra of various vegetable crops acquired over a tropical forest ecosystem. The dataset is valuable for research on developing new discrimination models of forest and cultivated vegetation, classification methods, spectral matching analysis techniques, and sub-pixel target detection methods.

Multi-resolution terrestrial hyperspectral dataset for spectral unmixing problems.

Recent developments in the miniaturization of hyperspectral imaging sensors have given rise to the increased use of hyperspectral imagery as the primary data for evaluating spectral unmixing algorithms in applications such as industrial quality control, agriculture, mineral mapping, military, etc. This article presents an ultra-high-resolution hyperspectral imagery dataset for undertaking benchmark studies on spectral unmixing. A terrestrial hyperspectral imager (THI) is used for imaging the target scene with the camera sensor pointing horizontally towards the target scene. The datasets are acquired at various spatial resolutions ranging from 1 mm to 2 cm. The targeted scene contains several paper-based panels, each size of 2 cm x 2 cm and filled with different colours and proportions, glued to a black background board that maintains a distinguishable distance between one another. In addition to the hyperspectral imagery data acquisitions, reference spectral signatures of the candidate mixture materials are obtained by in-situ hyperspectral reflectance measurements using a spectroradiometer. The hyperspectral image acquisition and the in-situ spectral signatures of the target scene are collected under natural illumination conditions. The proposed datasets are designed for undertaking proof-of-the-concept (PoC) studies in spectral unmixing. The datasets are also valuable for evaluating the performance of different statistical and machine learning algorithms for target detection, classification, and sub-pixel classification algorithms.

Multi-platform optical remote sensing dataset for target detection.

Target detection in remote sensing has vital applications in mineral mapping, law enforcement, precision agriculture, strategic surveillance, etc. We present the acquisition of a first-of-its-kind high-resolution multi-platform (ground, airborne, and space-borne) remote sensing-based benchmark dataset for target detection studies. The dataset includes imagery acquired from terrestrial hyperspectral imager (THI), airborne hyperspectral sensor (AVIRIS-NG), and space-borne multi-spectral (Sentinel-2) sensor on 20th March 2018. Five engineered targets of different materials and colours were placed on different surface backgrounds. Besides, in-situ reflectance spectra of the targets were also acquired using a spectroradiometer for serving as a spectral reference source. The airborne and space-borne imagery were processed to remove un-calibrated/noisy bands and were atmospherically corrected using a radiative transfer method based Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) model. The in-situ target reflectance spectra were resampled to spectrally match with airborne and space-borne imagery. Further, a target region of interest (ROI) was designated for each of the targets in both airborne and space-borne imagery using the known ground position of targets using a GPS device. This article provides a ground to space integrated target detection dataset, including ground positions ROI of the targets, point, and pixel-based in-situ target reference spectra, and the processed airborne and space-borne imagery to make the dataset ready for use. The data acquired in this experiment is an attempt to assess the potential of engineered material target detection in a multi-scale multi-platform view setup. The dataset is a valuable resource for testing and validation of target detection algorithms from various strategic and civilian application perspectives of remote sensing.