Comparison of PlanetScope, Sentinel-2, and landsat 8 data in soybean yield estimation within-field variability with random forest regression.

Accurate timely and early-season crop yield estimation within the field variability is important for precision farming and sustainable management applications. Therefore, the ability to estimate the within-field variability of grain yield is crucial for ensuring food security worldwide, especially under climate change. Several Earth observation systems have thus been developed to monitor crops and predict yields. Despite this, new research is required to combine multiplatform data integration, advancements in satellite technologies, data processing, and the application of this discipline to agricultural practices. This study provides further developments in soybean yield estimation by comparing multisource satellite data from PlanetScope (PS), Sentinel-2 (S2), and Landsat 8 (L8) and introducing topographic and meteorological variables. Herein, a new method of combining soybean yield, global positioning systems, harvester data, climate, topographic variables, and remote sensing images has been demonstrated. Soybean yield shape points were obtained from a combine-harvester-installed GPS and yield monitoring system from seven fields over the 2021 season. The yield estimation models were trained and validated using random forest, and four vegetation indices were tested. The result showed that soybean yield can be accurately predicted at 3-, 10-, and 30-m resolutions with mean absolute error (MAE) value of 0.091 t/ha for PS, 0.118 t/ha for S2, and 0.120 t/ha for L8 data (root mean square error (RMSE) of 0.111, 0.076). The combination of the environmental data with the original bands provided further improvements and an accurate yield estimation model within the soybean yield variability with MAE of 0.082 t/ha for PS, 0.097 t/ha for S2, and 0.109 t/ha for L8 (RMSE of 0.094, 0.069, and 0.108 t/ha). The results showed that the optimal date to predict the soybean yield within the field scale was approximately 60 or 70 days before harvesting periods during the beginning bloom stage. The developed model can be applied for other crops and locations when suitable training yield data, which are critical for precision farming, are available.

Enabling country-scale land cover mapping with meter-resolution satellite imagery.

High-resolution satellite images can provide abundant, detailed spatial information for land cover classification, which is particularly important for studying the complicated built environment. However, due to the complex land cover patterns, the costly training sample collections, and the severe distribution shifts of satellite imageries caused by, e.g., geographical differences or acquisition conditions, few studies have applied high-resolution images to land cover mapping in detailed categories at large scale. To fill this gap, we present a large-scale land cover dataset, Five-Billion-Pixels. It contains more than 5 billion labeled pixels of 150 high-resolution Gaofen-2 (4 m) satellite images, annotated in a 24-category system covering artificial-constructed, agricultural, and natural classes. In addition, we propose a deep-learning-based unsupervised domain adaptation approach that can transfer classification models trained on labeled dataset (referred to as the source domain) to unlabeled data (referred to as the target domain) for large-scale land cover mapping. Specifically, we introduce an end-to-end Siamese network employing dynamic pseudo-label assignment and class balancing strategy to perform adaptive domain joint learning. To validate the generalizability of our dataset and the proposed approach across different sensors and different geographical regions, we carry out land cover mapping on five megacities in China and six cities in other five Asian countries severally using: PlanetScope (3 m), Gaofen-1 (8 m), and Sentinel-2 (10 m) satellite images. Over a total study area of 60,000 km2, the experiments show promising results even though the input images are entirely unlabeled. The proposed approach, trained with the Five-Billion-Pixels dataset, enables high-quality and detailed land cover mapping across the whole country of China and some other Asian countries at meter-resolution.

Monitoring riverine traffic from space: The untapped potential of remote sensing for measuring human footprint on inland waterways.

Mass urbanisation and intensive agricultural development across river deltas have driven ecosystem degradation, impacting deltaic socio-ecological systems and reducing their resilience to climate change. Assessments of the drivers of these changes have so far been focused on human activity on the subaerial delta plains. However, the fragile nature of deltaic ecosystems and the need for biodiversity conservation on a global scale require more accurate quantification of the footprint of anthropogenic activity across delta waterways. To address this need, we investigated the potential of deep learning and high spatiotemporal resolution satellite imagery to identify river vessels, using the Vietnamese Mekong Delta (VMD) as a focus area. We trained the Faster R-CNN Resnet101 model to detect two classes of objects: (i) vessels and (ii) clusters of vessels, and achieved high detection accuracies for both classes (f-score = 0.84-0.85). The model was subsequently applied to available PlanetScope imagery across 2018-2021; the resultant detections were used to generate monthly, seasonal and annual products mapping the riverine activity, termed here the Human Waterway Footprint (HWF), with which we showed how waterborne activity has increased in the VMD (from approx. 1650 active vessels in 2018 to 2070 in 2021 - a 25 % increase). Whilst HWF values correlated well with population density estimates (R2 = 0.59-0.61, p < 0.001), many riverine activity hotspots were located away from population centres and varied spatially across the investigated period, highlighting that more detailed information is needed to fully evaluate the extent, and type, of human footprint on waterways. High spatiotemporal resolution satellite imagery in combination with deep learning methods offers great promise for such monitoring, which can subsequently enable local and regional assessment of environmental impacts of anthropogenic activities on delta ecosystems around the globe.

Detecting gold mining impacts on insect biodiversity in a tropical mining frontier with SmallSat imagery.

Gold mining is a major driver of Amazonian forest loss and degradation. As mining activity encroaches on primary forest in remote and inaccessible areas, satellite imagery provides crucial data for monitoring mining-related deforestation. High-resolution imagery, in particular, has shown promise for detecting artisanal gold mining at the forest frontier. An important next step will be to establish relationships between satellite-derived land cover change and biodiversity impacts of gold mining. In this study, we set out to detect artisanal gold mining using high-resolution imagery and relate mining land cover to insects, a taxonomic group that accounts for the majority of faunal biodiversity in tropical forests. We applied an object-based image analysis (OBIA) to classify mined areas in an Indigenous territory in Guyana, using PlanetScope imagery with ~3.7 m resolution. We complemented our OBIA with field surveys of insect family presence or absence in field plots (n = 105) that captured a wide range of mining disturbances. Our OBIA was able to identify mined objects with high accuracy (>90% balanced accuracy). Field plots with a higher proportion of OBIA-derived mine cover had significantly lower insect family richness. The effects of mine cover on individual insect taxa were highly variable. Insect groups that respond strongly to mining disturbance could potentially serve as bioindicators for monitoring ecosystem health during and after gold mining. With the advent of global partnerships that provide universal access to PlanetScope imagery for tropical forest monitoring, our approach represents a low-cost and rapid way to assess the biodiversity impacts of gold mining in remote landscapes.

Remote sensing to detect harmful algal blooms in inland waterbodies.

Harmful algal blooms (HABs) are an issue of concern for water management worldwide. As such, effective monitoring strategies of HAB spatio-temporal variability in waterbodies are needed. Remote sensing has become an increasingly important tool for HAB detection and monitoring in large lakes. However, accurate HAB detection in small-medium waterbodies via satellite data remains a challenge. Current barriers include the waterbody size, the limited freely available high resolution satellite data, and the lack of field calibration data. To test the applicability of remote sensing for detecting HABs in small-medium waterbodies, three satellites (Planetscope, Sentinel-2 and Landsat-8) were used to understand how spatial resolution, the availability of spectral bands, and the waterbody size itself effect HAB detection skill. Different algorithms and a non-parametric method, Self-Organizing Map (SOM), were tested. Curvature Around Red and NIR minus Red had the best HAB detection skill of the 20 existing algorithms that were tested. Landsat 8 and Sentinel 2 were the best satellites for HAB detection in small to medium waterbodies. The most critical attribute for detecting HABs were the available satellite bands, which determine the detection algorithms that can be used. Importantly, algorithm performance was mostly unrelated to waterbody size. However, there remain some barriers in utilizing satellite data for HAB detection, including algae dynamics, macrophyte cover within the waterbody, weather effects, and the correction models for satellite data. Moreover, it is important to consider the match time between satellite overpass and sampling activities for calibration. Given these challenges, integrating regular sampling activities and remote sensing is recommended for monitoring and managing small-medium waterbodies.

Landscape and vegetation traits of urban green space can predict local surface temperature.

Societal and technological advances have triggered demands to improve urban environmental quality. Urban green space (UGS) can provide effective cooling service and thermal comfort to alleviate warming impacts. We investigated the relative influence of a comprehensive spectrum of UGS landscape and vegetation factors on surface temperature in arid Urumqi city in northwest China. Built-up area range was extracted from Luojia 1-01 (LJ1-01) satellite data, and within this range, the landscape metric information and vegetation index information of UGS were obtained based on PlanetScope data, and a total of 439 sampling grids (1 km × 1 km) were generated. The urban surface temperature of built-up areas was extracted from Landsat8-TIRS images. The 12 landscape metrics and 14 vegetation indexes were assigned as independent variables, and surface temperature the dependent variable. Support Vector Machine (SVM), Gradient Boost Regression Tree (GBRT) and Random Forest (RF) were enlisted to establish numerical models to predict surface temperature. The results showed that: (1) It was feasible to predict local surface temperature using a combination of landscape metrics and vegetation indexes. Among the three models, RF demonstrated the best accuracy. (2) Collectively, all the factors play a role in the surface-temperature prediction. The most influential factor was Difference Vegetation Index (DVI), followed by Green Normalized Difference Vegetation Index (GNDVI), Class Area (CA) and AREA. This study developed remote sensing techniques to extract a basket of UGS factors to predict the surface temperature at local urban sites. The methods could be applied to other cities to evaluate the cooling impacts of green infrastructures. The findings could provide a scientific basis for ecological spatial planning of UGS to optimize cooling benefits in the arid region.

High Spatial-Resolution Red Tide Detection in the Southern Coast of Korea Using U-Net from PlanetScope Imagery.

Red tides caused by Margalefidinium polykrikoides occur continuously along the southern coast of Korea, where there are many aquaculture cages, and therefore, prompt monitoring of bloom water is required to prevent considerable damage. Satellite-based ocean-color sensors are widely used for detecting red tide blooms, but their low spatial resolution restricts coastal observations. Contrarily, terrestrial sensors with a high spatial resolution are good candidate sensors, despite the lack of spectral resolution and bands for red tide detection. In this study, we developed a U-Net deep learning model for detecting M. polykrikoides blooms along the southern coast of Korea from PlanetScope imagery with a high spatial resolution of 3 m. The U-Net model was trained with four different datasets that were constructed with randomly or non-randomly chosen patches consisting of different ratios of red tide and non-red tide pixels. The qualitative and quantitative assessments of the conventional red tide index (RTI) and four U-Net models suggest that the U-Net model, which was trained with a dataset of non-randomly chosen patches including non-red tide patches, outperformed RTI in terms of sensitivity, precision, and F-measure level, accounting for an increase of 19.84%, 44.84%, and 28.52%, respectively. The M. polykrikoides map derived from U-Net provides the most reasonable red tide patterns in all water areas. Combining high spatial resolution images and deep learning approaches represents a good solution for the monitoring of red tides over coastal regions.

High Resolution, Annual Maps of Field Boundaries for Smallholder-Dominated Croplands at National Scales.

Mapping the characteristics of Africa's smallholder-dominated croplands, including the sizes and numbers of fields, can provide critical insights into food security and a range of other socioeconomic and environmental concerns. However, accurately mapping these systems is difficult because there is 1) a spatial and temporal mismatch between satellite sensors and smallholder fields, and 2) a lack of high-quality labels needed to train and assess machine learning classifiers. We developed an approach designed to address these two problems, and used it to map Ghana's croplands. To overcome the spatio-temporal mismatch, we converted daily, high resolution imagery into two cloud-free composites (the primary growing season and subsequent dry season) covering the 2018 agricultural year, providing a seasonal contrast that helps to improve classification accuracy. To address the problem of label availability, we created a platform that rigorously assesses and minimizes label error, and used it to iteratively train a Random Forests classifier with active learning, which identifies the most informative training sample based on prediction uncertainty. Minimizing label errors improved model F1 scores by up to 25%. Active learning increased F1 scores by an average of 9.1% between first and last training iterations, and 2.3% more than models trained with randomly selected labels. We used the resulting 3.7 m map of cropland probabilities within a segmentation algorithm to delineate crop field boundaries. Using an independent map reference sample (n = 1,207), we found that the cropland probability and field boundary maps had respective overall accuracies of 88 and 86.7%, user's accuracies for the cropland class of 61.2 and 78.9%, and producer's accuracies of 67.3 and 58.2%. An unbiased area estimate calculated from the map reference sample indicates that cropland covers 17.1% (15.4-18.9%) of Ghana. Using the most accurate validation labels to correct for biases in the segmented field boundaries map, we estimated that the average size and total number of field in Ghana are 1.73 ha and 1,662,281, respectively. Our results demonstrate an adaptable and transferable approach for developing annual, country-scale maps of crop field boundaries, with several features that effectively mitigate the errors inherent in remote sensing of smallholder-dominated agriculture.

[Remote Sensing Classification of Urban Black-odor Water Based on Decision Tree].

Remote sensing monitoring of black-odor water is an important method for understanding the current status of urban water quality, and comprehensively evaluating the effect of urban water environment treatment. A total of 171 samples were collected in Nanjing, Changzhou, Wuxi, and Yangzhou cities and water quality parameters and optical parameters were measured simultaneously. Based on the analysis of the water color and optical characteristics of the black-odor water and non-black-odor water (denoted as general water), a decision tree was constructed to identify the severe, mild black-odor water, and general water as green and yellow water. The results found that:①According to the water color, the water bodies can be divided into six types. Among them, type 1 to 4 water bodies are black-odor water, which are gray black, dark gray, gray, and light gray water, respectively, and type 5 and 6 water bodies are general water, which are green and yellow water, respectively; ②Type 1 water body contains high contents of non-pigmented particulate matter and colored dissolved organic matter(CDOM), however, the absorption of pigmented particulate matter is not dominant. Type 2 and 5 water bodies are dominated by pigmented particulate matter. Type 3, 4, and 6 water bodies are dominated by non-pigmented particulate matter; ③After water color classification, and according to the differences of the reflection spectrums of the six types of water bodies, the difference of black-odorous water index (DBWI), green-red-nir area water index (G-R-NIR AWI), the green band reflectance and the normalized difference black-odorous water index (NDBWI) were used to construct a decision tree to identify the severe, mild black-odor water, and general water; ④The decision tree was applied to the PlanetScope satellite image of Yangzhou City on April 9, 2019, and 10 synchronous sampling points were used for verification. The overall recognition accuracy reached 80.00%, and the K value reached 0.67. The urban water classification model, after water color classification, can be applied to other similar water bodies, and provides a technical method for the supervision of black-odor water bodies.

Partial Least Square Discriminant Analysis Based on Normalized Two-Stage Vegetation Indices for Mapping Damage from Rice Diseases Using PlanetScope Datasets.

In recent decades, rice disease co-epidemics have caused tremendous damage to crop production in both China and Southeast Asia. A variety of remote sensing based approaches have been developed and applied to map diseases distribution using coarse- to moderate-resolution imagery. However, the detection and discrimination of various disease species infecting rice were seldom assessed using high spatial resolution data. The aims of this study were (1) to develop a set of normalized two-stage vegetation indices (VIs) for characterizing the progressive development of different diseases with rice; (2) to explore the performance of combined normalized two-stage VIs in partial least square discriminant analysis (PLS-DA); and (3) to map and evaluate the damage caused by rice diseases at fine spatial scales, for the first time using bi-temporal, high spatial resolution imagery from PlanetScope datasets at a 3 m spatial resolution. Our findings suggest that the primary biophysical parameters caused by different disease (e.g., changes in leaf area, pigment contents, or canopy morphology) can be captured using combined normalized two-stage VIs. PLS-DA was able to classify rice diseases at a sub-field scale, with an overall accuracy of 75.62% and a Kappa value of 0.47. The approach was successfully applied during a typical co-epidemic outbreak of rice dwarf (Rice dwarf virus, RDV), rice blast (Magnaporthe oryzae), and glume blight (Phyllosticta glumarum) in Guangxi Province, China. Furthermore, our approach highlighted the feasibility of the method in capturing heterogeneous disease patterns at fine spatial scales over the large spatial extents.