Search

Abstract Crop type mapping at the field level is necessary for a variety of applications in agricultural monitoring and food security. As remote sensing imagery continues to increase in spatial and temporal resolution, it is becoming an increasingly powerful raw input from which to create crop type maps. Still, automated crop type mapping remains constrained by a lack of field-level crop labels for training supervised classification models. In this study, we explore the use of random forests transferred across geographic distance and time and unsupervised methods in conjunction with aggregate crop statistics for crop type mapping in the US Midwest, where we simulated the label-poor setting by depriving the models of labels in various states and years. We validated our methodology using available 30 m spatial resolution crop type labels from the US Department of Agriculture's Cropland Data Layer (CDL). Using Google Earth Engine, we computed Fourier transforms (or harmonic regressions) on the time series of Landsat Surface Reflectance and derived vegetation indices, and extracted the coefficients as features for machine learning models. We found that random forests trained on regions and years similar in growing degree days (GDD) transfer to the target region with accuracies consistently exceeding 80%. Accuracies decrease as differences in GDD expand. Unsupervised Gaussian mixture models (GMM) with class labels derived using county-level crop statistics classify crops less consistently but require no field-level labels for training. GMM achieves over 85% accuracy in states with low crop diversity (Illinois, Iowa, Indiana, Nebraska), but performs sometimes no better than random when high crop diversity interferes with clustering (North Dakota, South Dakota, Wisconsin, Michigan). Under the appropriate conditions, these methods offer options for field-resolution crop type mapping in regions around the world with few or no ground labels. Highlights • Crop type maps of >80% accuracy created without in-season field data in some regions. • Fourier transform coefficients of Landsat time series distinguish crop types. • Random forest classifiers transfer well across regions of similar growing degree days. • Gaussian mixture models cluster crop types accurately in low crop diversity regions. • Aggregate crop statistics allow automatic assignment of crop types to clusters. [ABSTRACT FROM AUTHOR]

Abstract The Landsat archive presents a unique source for mapping and monitoring of shrublands. Still, efficient and accurate mapping approaches are needed that provide shrub cover fraction estimates over space and time. The spectral signal of shrubs is composed of green vegetation and non-photosynthetic vegetation as well as varying fractions of soil, grass, and shadow, which makes a direct mapping of fraction cover challenging. In this study, we mapped 32 years of shrub cover fractions from annual Landsat best observation composites covering a rockrose (genus Cistus) ecosystem in the Baixo Alentejo region of southern Portugal. Fraction mapping was based on a multi-year support vector regression model trained with synthetically mixed data from a multi-annual image spectral library. Resulting fractions maps were validated using reference information derived from high-resolution satellite imagery available for 10 out of 32 years. Fraction maps reproduced the spatial-temporal patterns of shrub cover in the study region very well, with an average mean absolute error over all validation years of 13.7%. For individual validation years, mean absolute errors ranged from 7.7% to 17.1%. Our consistent modeling framework led to a reliable annual shrub cover fraction time-series, which allowed identifying areas of stable shrub cover and areas with different types and intensities of change. Such long-term shrub cover fraction monitoring is of great value for land use assessments and contributes to a thorough understanding and management of Earth's arid and semi-arid ecosystems. Highlights • A 32-year time-series of shrub fractions was retrieved for Baixo Alentejo. • A multi-year SVR model applied to annual Landsat composites produced accurate maps. • Fraction time-series supported the assessment of shrub cover change processes. [ABSTRACT FROM AUTHOR]

Forthcoming spaceborne imaging spectrometers will provide novel opportunities for mapping urban composition globally. To move from case studies for single cities towards comparative and more operational analyses, generalized models that may be transferred throughout space are desired. In this study, we investigated how single regression models can be spatially generalized for vegetation-impervious-soil (VIS) mapping across multiple cities. The combination of support vector regression (SVR) with synthetically mixed training data generated from spectral libraries was used for fraction mapping. We developed three local models based on separate spectral libraries from Berlin (Germany), Brussels (Belgium), and Santa Barbara (U.S.), and a generalized model based on a combined multi-site spectral library. To examine the performance and transferability of the generalized model compared to local models, we first applied all model variants to simulated Environmental Mapping and Analysis Program (EnMAP) data from the three cities that were represented in the models, i.e., known sites. Next, we transferred the models to two unknown sites not represented in the models, San Francisco Bay Area (U.S.) and Munich (Germany). In the first mapping constellation, results demonstrated that the generalized model was capable of accurately mapping VIS fractions across all three known sites. Average mean absolute errors (AV-MAEs) were 8.5, 12.2, and 11.0% for Berlin, Brussels, and Santa Barbara. The performance of the generalized model was very similar to the local models, with ∆AV-MAEs falling within a range of ±0.7%. A detailed assessment of fraction maps and class-wise accuracies confirmed that modeling errors related to remaining limitations of urban mapping based on optical remote sensing data rather than to the choice between a local or generalized model. For the second mapping constellation, the generalized model proved to be useful for mapping vegetation and impervious fractions in the unknown sites. MAEs for both cover types were 5.4 and 10.9% for the San Francisco Bay Area, and 6.3 and 15.4% for Munich. In contrast, the three local models were only found to have similar accuracies as the generalized model for one of the two sites or for individual VIS categories. Despite the enhanced transferability of the generalized model to the unknown sites, deficiencies remained for accurate soil mapping. MAEs were 22.4 and 12.3%, and high over - and underestimations were observed at the low and high end of the fraction range. These shortcomings indicated possible limitations of the spectral libraries to account for the spectral characteristics of soils in the unknown sites. Overall, we conclude that the combination of SVR and synthetically mixed training data generated from multi-site libraries constitutes a flexible modeling approach for generalized urban mapping across multiple cities. [ABSTRACT FROM AUTHOR]

Reliable fire severity mapping is a vital resource for fire scientists and land management agencies globally. Satellite derived pre- and post-fire differenced severity indices (∆FSI), such as the differenced Normalised Burn Ratio (∆NBR), are widely used to map the severity of large wildfires. Fire severity classification is commonly undertaken through the identification of severity class thresholds in ∆FSI. However, several shortcomings have been identified with severity classifications using ∆FSI, including poor delineation of low fire severity classes, and unsatisfactory performance when ∆FSI classification thresholds are applied to new landscapes. Our study assesses the performance of the Random Forest classifier (RF) for improving the accuracy of satellite based wildfire severity mapping across heterogeneous landscapes using Landsat imagery. We collected point based fire severity training data (n = 10,855) from sixteen large wildfires occurring across south-eastern Australia between 2006 and 2016. The predictive accuracy of fire severity classification using ∆NBR and the RF incorporating numerous spectral indices, was assessed using bootstrapping and cross validation. Image acquisition and index calculation for each fire was undertaken in Google Earth Engine (GEE). Results of the bootstrapping validation show that the RF classifier had very high classification accuracy (>95%) for unburnt (UB), crown scorch (CS) and crown consumption (CC) severity classes, and high classification accuracy (>74%) for low severity classes (crown unburnt, CU; partial crown scorch, PCS). The RF classification outperformed the ∆NBR classification for all severity classes, increasing classification accuracy by between 6%–21%. Cross validation using independent fires produced similar median classification accuracy as the bootstrapping validation, though the RF classification of CU was substantially reduced to ~55%. ∆NBR, ∆NDWI and ∆NDVI were the three most important variables in the RF model. The Landsat satellite platform used to derive the indices had little effect on classification accuracy. Maps produced using the RF classifier in GEE had similar spatial patterns in fire severity classes as maps produced using time-consuming hand digitisation of aerial images. GEE was found to be a highly efficient platform for image acquisition, processing and production of severity maps. Our study shows that fire severity mapping using RF classifiers provides a robust method for broad scale mapping of fire severity across heterogeneous landscapes. Furthermore, the GEE-based classification framework supports the operational application of this approach in a land management agency context for the rapid production of fire severity maps. [ABSTRACT FROM AUTHOR]

Monitoring biodiversity and how anthropogenic pressures impact this is critical, especially as anthropogenically driven climate change continues to affect all ecosystems. Intertidal areas are exposed to particularly high levels of pressures owing to increased population density in coastal areas. Traditional methods of monitoring intertidal areas do not provide datasets with full coverage in a cost-effective or timely manner, and so the use of remote sensing to monitor these areas is becoming more common. Monitoring of ecologically important monospecific habitats, such as seagrass beds, using remote sensing techniques is well documented. However, the ability for multispectral data to distinguish efficiently and accurately between classes of vegetation with similar pigment composition, such as seagrass and green algae, has proved difficult, often requiring hyperspectral data. A machine learning approach was used to differentiate between soft-bottom intertidal vegetation classes when exposed at low tide, comparing 6 different multi- and hyperspectral remote and in situ sensors. For the library of 366 spectra, collected across Northern Europe, high accuracy (>80%) was found across all spectral resolutions. While a higher spectral resolution resulted in higher accuracy, there was no discernible increase in accuracy above 10 spectral bands (95%: Sentinel-2 MSI sensor with a spatial resolution of 20 m). This work highlights the ability of multispectral sensors to discriminate intertidal vegetation types, while also showing the most important wavelengths for this discrimination (∼530 and ∼ 730 nm), giving recommendations for spectral ranges of future satellite missions. The ability for multispectral sensors to aid in accurate and rapid intertidal vegetation classification at the taxonomic resolution of classes, could be a significant contribution for future sustainable and effective ecosystem management. • Multispectral data can distinguish green macrophytes with high accuracy. • This accuracy is achievable at satellite spectral resolutions currently available. • This opens new avenues for biodiversity studies of intertidal ecosystems. • Minimal gain in accuracy is achieved by increased spectral resolution. [ABSTRACT FROM AUTHOR]

Smallholder farms dominate in many parts of the world, particularly Sub-Saharan Africa. These systems are characterized by small, heterogeneous, and often indistinct field patterns, requiring a specialized methodology to map agricultural land cover. Using a variety of sites in South Africa, we present a new approach to mapping agricultural fields, based on efficient extraction of a vast set of simple, highly correlated, and interdependent features, followed by a random forest classifier. We achieved similar high performance across agricultural types, including the spectrally indistinct smallholder fields as well as the more easily distinguishable commercial fields, and demonstrated the ability to generalize performance across large geographic areas. In sensitivity analyses, we determined multi-temporal information provided greater gains in performance than the addition of multi-spectral bands available in DigitalGlobe Worldview-2 imagery. [ABSTRACT FROM AUTHOR]