Search

Machine learning is frequently combined with imagery acquired from uncrewed aircraft systems (UASs) to detect invasive plants. Having prior knowledge of which machine learning algorithm will produce the most accurate results is difficult. This study examines the efficacy of a voting-based ensemble classifier to identify invasive Phragmites australis from three-band (red, green, blue; RGB) and five-band (red, green, blue, red edge, near-infrared; multispectral; MS) UAS imagery acquired over multiple Minnesota wetlands. A Random Forest, histogram-based gradient-boosting classification tree, and two artificial neural networks were used within the voting-based ensemble classifier. Classifications from the RGB and multispectral imagery were compared across validation sites both with and without post-processing from an object-based image analysis (OBIA) workflow (post-machine learning OBIA rule set; post-ML OBIA rule set). Results from this study suggest that a voting-based ensemble classifier can accurately identify invasive Phragmites australis from RGB and multispectral imagery. Accuracies greater than 80% were attained by the voting-based ensemble classifier for both the RGB and multispectral imagery. The highest accuracy, 91%, was achieved when using the multispectral imagery, a canopy height model, and a post-ML OBIA rule set. The study emphasizes the need for further research regarding the accurate identification of Phragmites australis at low stem densities.

Uncrewed aircraft systems (UASs) are a popular tool when surveilling for invasive alien plants due to their high spatial and temporal resolution. This study investigated the efficacy of a UAS equipped with a three-band (i.e., red, green, blue; RGB) sensor to identify invasive Phragmites australis in multiple Minnesota wetlands using object-based image analysis (OBIA) and machine learning (ML) algorithms: artificial neural network (ANN), random forest (RF), and support vector machine (SVM). The addition of a post-ML classification OBIA workflow was tested to determine if ML classifications can be improved using OBIA techniques. Results from each ML algorithm were compared across study sites both with and without the post-ML OBIA workflow. ANN was identified as the best classifier when not incorporating a post-ML OBIA workflow with a classification accuracy of 88%. Each of the three ML algorithms achieved a classification accuracy of 91% when incorporating the post-ML OBIA workflow. Results from this study suggest that a post-ML OBIA workflow can increase the ability of ML algorithms to accurately identify invasive Phragmites australis and should be used when possible. Additionally, the decision of which ML algorithm to use for Phragmites mapping becomes less critical with the addition of a post-ML OBIA workflow.

The means to accurately monitor wetland change over time are crucial to wetland management. This paper explores the applicability of LandTrendr, a temporal segmentation algorithm designed to identify significant interannual trends, to monitor wetlands by modeling surface water presence in Minnesota from 1984 to 2019. A time series of harmonized Landsat and Sentinel-2 data in the spring is developed in Google Earth Engine, and calculated to sub-pixel water fraction. The optimal parameters for modeling this time series with LandTrendr are identified by minimizing omission of known surface water locations, and the result of this optimal model of sub-pixel water fraction is evaluated against reference images and qualitatively. Accuracy of this method is high: overall accuracy is 98% and producer’s and user’s accuracies for inundation are 82% and 88% respectively. Maps summarizing the trendlines of multiple pixels, such as frequency of inundation over the past 35 years, also show LandTrendr as applied here can accurately model long-term trends in surface water presence across wetland types. However, the tendency of omission for more variable prairie pothole wetlands and the under-prediction of inundation for small or emergent wetlands suggests the algorithm will require careful development of the segmented time series to capture inundated conditions more accurately.

Invasive plant species are an increasing worldwide threat both ecologically and financially. Knowing the location of these invasive plant infestations is the first step in their control. Surveying for invasive Phragmites australis is particularly challenging due to limited accessibility in wetland environments. Unoccupied aircraft systems (UAS) are a popular choice for invasive species management due to their ability to survey challenging environments and their high spatial and temporal resolution. This study tested the utility of three-band (i.e., red, green, and blue; RGB) UAS imagery for mapping Phragmites in the St. Louis River Estuary in Minnesota, U.S.A. and Saginaw Bay in Michigan, U.S.A. Iterative object-based image analysis techniques were used to identify two classes, Phragmites and Not Phragmites. Additionally, the effectiveness of canopy height models (CHMs) created from two data types, UAS imagery and commercial satellite stereo retrievals, and the RADARSAT-2 horizontal-horizontal (HH) polarization were tested for Phragmites identification. The highest overall classification accuracy of 90% was achieved when pairing the UAS imagery with a UAS-derived CHM. Producer’s accuracy for the Phragmites class ranged from 3 to 76%, and the user’s accuracies were above 90%. The Not Phragmites class had user’s and producer’s accuracies above 88%. Inclusion of the RADARSAT-2 HH polarization caused a slight reduction in classification accuracy. Commercial satellite stereo retrievals increased commission errors due to decreased spatial resolution and vertical accuracy. The lowest classification accuracy was seen when using only the RGB UAS imagery. UAS are promising for Phragmites identification, but the imagery should be used in conjunction with a CHM.

The ability to predict spatially explicit nitrogen uptake (NUP) in maize (Zea mays L.) during the early development stages provides clear value for making in-season nitrogen fertilizer applications that can improve NUP efficiency and reduce the risk of nitrogen loss to the environment. Aerial hyperspectral imaging is an attractive agronomic research tool for its ability to capture spectral data over relatively large areas, enabling its use for predicting NUP at the field scale. The overarching goal of this work was to use supervised learning regression algorithms—Lasso, support vector regression (SVR), random forest, and partial least squares regression (PLSR)—to predict early season (i.e., V6–V14) maize NUP at three experimental sites in Minnesota using high-resolution hyperspectral imagery. In addition to the spectral features offered by hyperspectral imaging, the 10th percentile Modified Chlorophyll Absorption Ratio Index Improved (MCARI2) was made available to the learning models as an auxiliary feature to assess its ability to improve NUP prediction accuracy. The trained models demonstrated robustness by maintaining satisfactory prediction accuracy across locations, pixel sizes, development stages, and a broad range of NUP values (4.8 to 182 kg ha−1). Using the four most informative spectral features in addition to the auxiliary feature, the mean absolute error (MAE) of Lasso, SVR, and PLSR models (9.4, 9.7, and 9.5 kg ha−1, respectively) was lower than that of random forest (11.2 kg ha−1). The relative MAE for the Lasso, SVR, PLSR, and random forest models was 16.5%, 17.0%, 16.6%, and 19.6%, respectively. The inclusion of the auxiliary feature not only improved overall prediction accuracy by 1.6 kg ha−1 (14%) across all models, but it also reduced the number of input features required to reach optimal performance. The variance of predicted NUP increased as the measured NUP increased (MAE of the Lasso model increased from 4.0 to 12.1 kg ha−1 for measured NUP less than 25 kg ha−1 and greater than 100 kg ha−1, respectively). The most influential spectral features were oftentimes adjacent to each other (i.e., within approximately 6 nm), indicating the importance of both spectral precision and derivative spectra around key wavelengths for explaining NUP. Finally, several challenges and opportunities are discussed regarding the use of these results in the context of improving nitrogen fertilizer management.

Ash trees (Fraxinus spp.) are a prominent species in Minnesota forests, with an estimated 1.1 billion trees in the state, totaling approximately 8% of all trees. Ash trees are threatened by the invasive emerald ash borer (Agrilus planipennis Fairmaire), which typically results in close to 100% tree mortality within one to five years of infestation. A detailed, wall-to-wall map of ash presence is highly desirable for forest management and monitoring applications. We used Google Earth Engine to compile Landsat time series analysis, which provided unique information on phenologic patterns across the landscape to identify ash species. Topographic position information derived from lidar was added to improve spatial maps of ash abundance. These input data were combined to produce a classification map and identify the abundance of ash forests that exist in the state of Minnesota. Overall, 12,524 km2 of forestland was predicted to have greater than 10% probability of ash species present. The overall accuracy of the composite ash presence/absence map was 64% for all ash species and 72% for black ash, and classification accuracy increased with the length of the time series. Average height derived from lidar was the best model predictor for ash basal area (R2 = 0.40), which, on average, was estimated as 16.1 m2 ha−1. Information produced from this map will be useful for natural resource managers and planners in developing forest management strategies which account for the spatial distribution of ash on the landscape. The approach used in this analysis is easily transferable and broadly scalable to other regions threatened with forest health problems such as invasive insects.

Wetland mapping at the landscape scale using remotely sensed data requires both affordable data and an efficient accurate classification method. Random forest classification offers several advantages over traditional land cover classification techniques, including a bootstrapping technique to generate robust estimations of outliers in the training data, as well as the capability of measuring classification confidence. Though the random forest classifier can generate complex decision trees with a multitude of input data and still not run a high risk of over fitting, there is a great need to reduce computational and operational costs by including only key input data sets without sacrificing a significant level of accuracy. Our main questions for this study site in Northern Minnesota were: (1) how does classification accuracy and confidence of mapping wetlands compare using different remote sensing platforms and sets of input data; (2) what are the key input variables for accurate differentiation of upland, water, and wetlands, including wetland type; and (3) which datasets and seasonal imagery yield the best accuracy for wetland classification. Our results show the key input variables include terrain (elevation and curvature) and soils descriptors (hydric), along with an assortment of remotely sensed data collected in the spring (satellite visible, near infrared, and thermal bands; satellite normalized vegetation index and Tasseled Cap greenness and wetness; and horizontal-horizontal (HH) and horizontal-vertical (HV) polarization using L-band satellite radar). We undertook this exploratory analysis to inform decisions by natural resource managers charged with monitoring wetland ecosystems and to aid in designing a system for consistent operational mapping of wetlands across landscapes similar to those found in Northern Minnesota.

Monitoring of water clarity trends is necessary for water resource managers. Remote sensing based methods are well suited for monitoring clarity in water bodies such as the inland lakes in Minnesota, United States. This study evaluated the potential of using imagery from NASA’s MODIS sensor to study intra-annual variations in lake clarity. MODIS reflectance images from six dates throughout the 2006 growing season were used with field collected Secchi disk transparency data to estimate water clarity in large lakes throughout Minnesota. The results of this research indicate the following: water clarity estimates derived from MODIS imagery are largely similar to those derived from lower temporal resolution sensors such as Landsat, robust water clarity estimates can be derived using MODIS for many dates throughout a growing season (R2 values between 0.32 and 0.71), and the relatively low spatial resolution of MODIS restricts its applicability to a subset of the largest inland lakes (>160 ha, or 400 acres). This study suggests that water clarity maps developed with MODIS imagery and bathymetry data may be useful tools for resource managers concerned with intra- and inter-annual variations in large inland lakes.