Amanda Ramcharan, Kelsee Baranowski, Peter McCloskey, Babuali Ahmed, James Legg, and David P. Hughes
Cassava is the third largest source of carbohydrates for human food in the world but is vulnerable to virus diseases, which threaten to destabilize food security in sub-Saharan Africa. Image recognition offers both a cost effective and scalable technology for disease detection. New deep learning models offer an avenue for this technology to be easily deployed on mobile devices. Using a dataset of cassava disease images taken in the field in Tanzania, we applied transfer learning to train a deep convolutional neural network to identify three diseases and two types of pest damage (or lack thereof). The best trained model accuracies were 98% for brown leaf spot (BLS), 96% for red mite damage (RMD), 95% for green mite damage (GMD), 98% for cassava brown streak disease (CBSD), and 96% for cassava mosaic disease (CMD). The best model achieved an overall accuracy of 93% for data not used in the training process. Our results show that the transfer learning approach for image recognition of field images offers a fast, affordable, and easily deployable strategy for digital plant disease detection.
SoilGrids100m: Soil Property and Class Maps of the Conterminous US at 100 meter Spatial Resolution based on a Compilation of National Soil Point Observations and Machine Learning
Amanda Ramcharan, Travis Nauman, Tomislav Hengl, Sharon Waltman , Colby Brungard, Skye Wills, James Thompson
With growing concern for the depletion of soil resources, conventional soil maps need to be updated and provided at finer and finer resolutions to be able to support spatially explicit human-landscape models. Three US soil point datasets—the National Cooperative Soil Survey Characterization Database, the National Soil Information System, and the Rapid Carbon Assessment dataset—were combined with a stack of over 200 environmental datasets to generate complete coverage gridded predictions at 100-m spatial resolution of soil properties (percent organic C, total N, bulk density, pH, and percent sand and clay) and US soil taxonomic classes (291 great groups [GG] and 78 modified particle size classes [mPSCs]) for the conterminous United States. Models were built using an ensemble of parallelized random forest and gradient boosting algorithms as implemented in the ranger and xgboost packages for R. Soil property predictions were generated at seven standard soil depths (0, 5, 15, 30, 60, 100, and 200 cm). Prediction probability maps for US soil taxonomic classifications were also generated. Model validation results indicate an out-of-bag classification accuracy of 60% for GGs and 66% for mPSCs; for soil properties RMSE for leave-location-out cross-validation was 3.63 wt% (R2 =0.41), 17.8 wt% (R2=0.57), 12 wt% (R2=0.46), 0.2 g cm-3 (R2=0.42), 0.74 (R2=0.68), and 0.27 wt% (R2=0.39) for weight percent organic C, percent sand and clay, bulk density, pH, and weight percent total N respectively. Nine independent validation datasets were used to assess prediction accuracies for soil class models, and results ranged between 24-58% and 24-93% for great group and mPSC prediction accuracies, respectively. Although mapping accuracies were variable and likely lower than gSSURGO in some areas, this modeling approach can enable easier integration of soil information with spatially explicit models, compared to multicomponent map units.
A soil bulk density pedotransfer function based on machine learning: A case study with the NCSS characterization database
Amanda Ramcharan, Tomislav Hengl, Dylan Beaudette, Skye Wills
Thus far pedotranfer functions (PTFs) to calculate bulk density have been developed with limited soil data resulting in much uncertainty in applying PTFs to other soil climates. Without reliable pedotransfer functions for bulk density, we cannot calculate important weight-to-volume conversions of soil properties such as soil organic carbon, an important metric for managing agricultural landscapes. There exists a wealth of data in the NCSS Characterization Database that can be used with machine learning methods to build a bulk density pedotransfer function applicable in a wide range of soil climates.With a bulk density pedotransfer function built from data collected at a national scale, the function can be effectively applied to a wide range of soil climates and can be used to gap-fill missing data in any location with known uncertainty values.
Carbon and nitrogen environmental trade-offs of winter rye cellulosic biomass in the Chesapeake Watershed
Amanda Ramcharan, Tom Richard
Cellulosic biomass from winter crops such as winter rye (Secale cereale L.) can complement corn stover harvested from corn (Zea mays L.) – soybean (Glycine max L.) rotations. In this study, we assessed on-field environmental impacts related to carbon (C) and nitrogen (N) of harvesting biomass in such double cropping systems by modeling representative agro-ecological conditions prevalent in the mid-Atlantic region of the United States. While difficult to simultaneously maximize nitrogen remediation and renewable energy production, these results offer a guide to help farm operators and policy-makers manage these trade-offs, while tapping into a potentially important low CO2eq emission energy source.