Crop loss identification at field parcel scale using satellite remote sensing and machine learning

Identifying crop loss at field parcel scale using satellite images is challenging: first, crop loss is caused by many factors during the growing season; second, reliable reference data about crop loss are lacking; third, there are many ways to define crop loss. This study investigates the feasibility of using satellite images to train machine learning (ML) models to classify agricultural field parcels into those with and without crop loss. The reference data for this study was provided by Finnish Food Authority (FFA) containing crop loss information of approximately 1.4 million field parcels in Finland covering about 3.5 million ha from 2000 to 2015. This reference data was combined with Normalised Difference Vegetation Index (NDVI) derived from Landsat 7 images, in which more than 80% of the possible data are missing. Despite the hard problem with extremely noisy data, among the four ML models we tested, random forest (with mean imputation and missing value indicators) achieved the average AUC (area under the ROC curve) of 0.688±0.059 over all 16 years with the range [0.602, 0.795] in identifying new crop-loss fields based on reference fields of the same year. To our knowledge, this is one of the first large scale benchmark study of using machine learning for crop loss classification at field parcel scale. The classification setting and trained models have numerous potential applications, for example, allowing government agencies or insurance companies to verify crop-loss claims by farmers and realise efficient agricultural monitoring.

Download Full-text

Crop loss identification at field parcel scale using satellite remote sensing and machine learning

10.1101/2021.05.07.443072 ◽

2021 ◽

Author(s):

Santosh Hiremath ◽

Samantha Wittke ◽

Taru Palosuo ◽

Jere Kaivosoja ◽

Fulu Tao ◽

...

Keyword(s):

Machine Learning ◽

Large Scale ◽

Satellite Images ◽

Vegetation Index ◽

Reference Data ◽

Insurance Companies ◽

Crop Loss ◽

Agricultural Field ◽

Normalised Difference Vegetation Index ◽

The Hard Problem

Identifying crop loss at field parcel scale using satellite images is challenging: first, crop loss is caused by many factors during the growing season; second, reliable reference data about crop loss are lacking; third, there are many ways to define crop loss. This study investigates the feasibility of using satellite images to train machine learning (ML) models to classify agricultural field parcels into those with and without crop loss. The reference data for this study was provided by Finnish Food Authority (FFA) containing crop loss information of approximately 1.4 million field parcels in Finland covering about 3.5 million ha from 2000 to 2015. This reference data was combined with Normalised Difference Vegetation Index (NDVI) derived from Landsat 7 images, in which more than 80\% of the possible data are missing. Despite the hard problem with extremely noisy data, among the four ML models we tested, random forest (with mean imputation and missing value indicators) achieved the average AUC (area under the ROC curve) of [[EQUATION]] over all 16 years with the range [0.602, 0.795] in identifying new crop-loss fields based on reference fields of the same year. To our knowledge, this is one of the first large scale benchmark study of using machine learning for crop loss classification at field parcel scale. The classification setting and trained models have numerous potential applications, for example, allowing government agencies or insurance companies to verify crop-loss claims by farmers and realise efficient agricultural monitoring.

Download Full-text

Identifying Ecosystem Function Shifts in Africa Using Breakpoint Analysis of Long-Term NDVI and RUE Data

Remote Sensing ◽

10.3390/rs12111894 ◽

2020 ◽

Vol 12 (11) ◽

pp. 1894

Author(s):

Thomas P. Higginbottom ◽

Elias Symeonakis

Keyword(s):

Time Series ◽

Large Scale ◽

Vegetation Index ◽

Change Point Analysis ◽

Annual Rainfall ◽

Small Scale ◽

Growth Season ◽

Continental Scale ◽

Normalised Difference Vegetation Index

Time-series of vegetation greenness data, derived from Earth-observation imagery, have become a key source of information for studying large-scale environmental change. The ever increasing length of such series allows for a range of indicators to be derived and for increasingly complex analyses to be applied. This study presents an analysis of trends in vegetation productivity—measured using the Global Inventory Monitoring and Modelling System third generation (GIMMS3g) Normalised Difference Vegetation Index (NDVI) data—for African savannahs, over the 1982–2015 period. Two annual metrics were derived from the 34 year dataset: the monthly, smoothed NDVI (the aggregated growth season NDVI) and the associated Rain Use Efficiency (growth season NDVI divided by annual rainfall). These indicators were then used in a BFAST-based change-point analysis, allowing the direction of change over time to change and the detection of one major break in the time-series. We also analysed the role of land cover type and climate zone as associations of the observed changes. Both methods agree that vegetation greening was pervasive across African savannahs, although RUE displayed less significant changes than NDVI. Monotonically increasing trends were the most common trend type for both indicators. The continental scale of the greening may suggest global processes as key drivers, such as carbon fertilization. That NDVI trends were more dynamic than RUE suggests that a large component of vegetation trends is driven by precipitation variability. Areas of negative trends were conspicuous by their minimalism. However, some patterns were apparent. In the southern Sahel and West Africa, declining NDVI and RUE overlapped with intensive population and agricultural regions. Dynamic trend reversals, in RUE and NDVI, located in Angola, Zambia and Tanzania, coincide with areas where a long-term trend of forest degradation and agricultural expansion has recently given way to increases in woody biomass. Meanwhile in southern Africa, monotonic increases in RUE with varying NDVI trend types may be indicative of shrub encroachment. However, all these processes are small-scale relative to the GIMMS NDVI data, and reconciling these conflicting drivers is not a trivial task. Our study highlights the importance of considering multiple options when undertaking trend analyses, as different inputs and methods can reveal divergent patterns.

Download Full-text

Estimating causal networks in biosphere–atmosphere interaction with the PCMCI approach

Biogeosciences ◽

10.5194/bg-17-1033-2020 ◽

2020 ◽

Vol 17 (4) ◽

pp. 1033-1061

Author(s):

Christopher Krich ◽

Jakob Runge ◽

Diego G. Miralles ◽

Mirco Migliavacca ◽

Oscar Perez-Priego ◽

...

Keyword(s):

Time Series ◽

Large Scale ◽

Vegetation Index ◽

Terrestrial Ecosystems ◽

Causal Discovery ◽

Mediterranean Ecosystem ◽

Climate Data ◽

Normalised Difference Vegetation Index ◽

Causal Networks ◽

Causal Graphs

Abstract. The dynamics of biochemical processes in terrestrial ecosystems are tightly coupled to local meteorological conditions. Understanding these interactions is an essential prerequisite for predicting, e.g. the response of the terrestrial carbon cycle to climate change. However, many empirical studies in this field rely on correlative approaches and only very few studies apply causal discovery methods. Here we explore the potential for a recently proposed causal graph discovery algorithm to reconstruct the causal dependency structure underlying biosphere–atmosphere interactions. Using artificial time series with known dependencies that mimic real-world biosphere–atmosphere interactions we address the influence of non-stationarities, i.e. periodicity and heteroscedasticity, on the estimation of causal networks. We then investigate the interpretability of the method in two case studies. Firstly, we analyse three replicated eddy covariance datasets from a Mediterranean ecosystem. Secondly, we explore global Normalised Difference Vegetation Index time series (GIMMS 3g), along with gridded climate data to study large-scale climatic drivers of vegetation greenness. We compare the retrieved causal graphs to simple cross-correlation-based approaches to test whether causal graphs are considerably more informative. Overall, the results confirm the capacity of the causal discovery method to extract time-lagged linear dependencies under realistic settings. For example, we find a complete decoupling of the net ecosystem exchange from meteorological variability during summer in the Mediterranean ecosystem. However, cautious interpretations are needed, as the violation of the method's assumptions due to non-stationarities increases the likelihood to detect false links. Overall, estimating directed biosphere–atmosphere networks helps unravel complex multidirectional process interactions. Other than classical correlative approaches, our findings are constrained to a few meaningful sets of relations, which can be powerful insights for the evaluation of terrestrial ecosystem models.

Download Full-text

Large-scale vegetation responses to terrestrial moisture storage changes

10.5194/hess-2016-545 ◽

2016 ◽

Author(s):

Robert L. Andrew ◽

Huade Guan ◽

Okke Batelaan

Keyword(s):

Land Use ◽

Time Series ◽

Large Scale ◽

Vegetation Index ◽

Water Storage ◽

Temporal Variations ◽

Discrete Wavelet ◽

Large Spatial Scale ◽

Normalised Difference Vegetation Index ◽

Vegetation Activity

Abstract. The Normalised Difference Vegetation Index (NDVI) is a useful tool for studying vegetation activity and ecosystem performance at a large spatial scale. In this study we use the Gravity Recovery and Climate Experiment (GRACE) total water storage (TWS) estimates to examine temporal variability of NDVI across Australia. We aim to demonstrate a new method that reveals the moisture dependence of vegetation cover at different temporal resolutions. Time series of monthly GRACE TWS anomalies are decomposed into different temporal frequencies using a discrete wavelet transform and analysed against time series of NDVI anomalies in a stepwise regression. Results show that combinations of different frequencies of decomposed GRACE TWS data explain NDVI temporal variations better than raw GRACE TWS alone. Generally, NDVI appears to be more sensitive to inter-annual changes in water storage than shorter changes, though grassland-dominated areas are sensitive to higher frequencies of water storage changes. Different types of vegetation, defined by areas of land use type show distinct differences in how they respond to the changes in water storage which is generally consistent with our physical understanding. This unique method provides useful insight into how NDVI is affected by changes in water storage at different temporal scales across land use types.

Download Full-text

Segmentation of Agricultural Parcels in Satellite Images Based on Historical Vegetation Index Data

10.20944/preprints202112.0243.v1 ◽

2021 ◽

Author(s):

Maria Pavlova ◽

Valerii Timofeev ◽

Dmitry Bocharov ◽

Irina Kunina ◽

Anna Smagina ◽

...

Keyword(s):

Satellite Images ◽

Vegetation Index ◽

Connected Components ◽

Agricultural Field ◽

Index Data ◽

Boundary Recognition ◽

History Of Vegetation ◽

Index Approach ◽

History Of

This paper considered the issue of agricultural fields boundary recognition in satellite images. A novel algorithm based on the aggregated history of vegetation index data obtained via open satellite data, Sentinel-2, was proposed. The proposed algorithm included several basic steps, namely the detection of parcel regions on aggregated index data; the calculation of aggregated edge maps; the segmentation of parcel regions using the edges obtained; the computation of connected components and their contour extraction. In this paper, we showed that the use of aggregated vegetation index data and boundary maps allow for much more accurate agricultural field segmentation compared to the instant vegetation index approach. The quality of segmentation within regions of Russia and the Ukraine was estimated. The dataset that was used and Python implementation of the proposed algorithm were provided.

Download Full-text

Application of Machine Learning Techniques to Predict the Impact of Health Insurance on the Wellbeing of an Individual

International Journal of Innovative Technology and Exploring Engineering - Special Issue ◽

10.35940/ijitee.b7247.129219 ◽

2019 ◽

Vol 9 (2) ◽

pp. 3065-3070

Keyword(s):

Machine Learning ◽

Health Insurance ◽

Large Scale ◽

Insurance Industry ◽

Financial Burden ◽

Machine Learning Algorithms ◽

Machine Learning Techniques ◽

Insurance Companies ◽

The Government ◽

The Impact

The healthcare domain in India has suffered considerably despite the advancement in technology. Several financing schemes are endorsed by the insurance companies to lessen the financial burden faced by the government and people. Nonetheless, Health Insurance segment in India remains underdeveloped due to various complexities that it faces. This paper exploits a heuristic sampling approach combined with the ensemble Machine Learning algorithms on the large-scale insurance business data to realize the current shape of the Health Insurance industry in India. Through the courtesy of Data Mining and Data Analytics, it is plausible to furnish insights that assist the common people in acquiring closure that helps in the process of decision making.

Download Full-text

Spectral analysis of charcoal on soils: implicationsfor wildland fire severity mapping methods

International Journal of Wildland Fire ◽

10.1071/wf09057 ◽

2010 ◽

Vol 19 (7) ◽

pp. 976 ◽

Cited By ~ 36

Author(s):

Alistair M. S. Smith ◽

Jan U. H. Eitel ◽

Andrew T. Hudak

Keyword(s):

Spectral Analysis ◽

Soil Type ◽

Large Scale ◽

Vegetation Index ◽

Wildland Fire ◽

Fire Severity ◽

Mapping Method ◽

Normalised Difference Vegetation Index ◽

National Programs

Recent studies in the Western United States have supported climate scenarios that predict a higher occurrence of large and severe wildfires. Knowledge of the severity is important to infer long-term biogeochemical, ecological, and societal impacts, but understanding the sensitivity of any severity mapping method to variations in soil type and increasing charcoal (char) cover is essential before widespread adoption. Through repeated spectral analysis of increasing charcoal quantities on six representative soils, we found that addition of charcoal to each soil resulted in linear spectral mixing. We found that performance of the Normalised Burn Ratio was highly sensitive to soil type, whereas the Normalised Difference Vegetation Index was relatively insensitive. Our conclusions have potential implications for national programs that seek to monitor long-term trends in wildfire severity and underscore the need to collect accurate soils information when evaluating large-scale wildland fires.

Download Full-text

Geospatial Analysis of Changes in Vegetation Cover over Nigeria

Bulletin of Geography Physical Geography Series ◽

10.1515/bgeo-2017-0010 ◽

2017 ◽

Vol 13 (1) ◽

pp. 17-27 ◽

Cited By ~ 7

Author(s):

Olutoyin Fashae ◽

Adeyemi Olusola ◽

Oluwatola Adedeji

Keyword(s):

Information Technology ◽

Markov Chain ◽

Environmental Sustainability ◽

Vegetation Cover ◽

Satellite Images ◽

Vegetation Index ◽

Normalised Difference Vegetation Index ◽

Ecological Hazards ◽

Spatio Temporal

AbstractVegetation cover over Nigeria has been on the decrease recently, hence the need for adequate monitoring using geo-information technology. This study examined the spatio-temporal variation of vegetation cover over Nigeria for thirty years with a view to developing a strategy for enhancing environmental sustainability. In order to predict the spatial extent of vegetation cover in 2030, the study utilised satellite images from between 1981 and 2010 using the Normalised Difference Vegetation Index (NDVI) coupled with cellular automata and Markov chain techniques in ArcGIS 10.3. The results showed that dense vegetal areas decreased in area from 358,534.2 km2in 1981 to 207,812 km2in 2010, while non-vegetal areas increased from 312,640.8 km2in 1981 to 474,436.4 km2in 2010 with a predicted increase to 501,504.9 km2by 2030, i.e. an increase of about 27,068.4 km2between 2010 and 2030. The study concluded that geoinformation techniques are effective in monitoring long-term intra- and inter-annual variability of vegetation and also useful in developing sustainable strategies for combating ecological hazards.

Download Full-text

Automated spaceborne detection of degraded vegetation around Monchegorsk, Kola Peninsula, Russia

Polar Record ◽

10.1017/s0032247411000337 ◽

2011 ◽

Vol 48 (1) ◽

pp. 107-112 ◽

Cited By ~ 3

Author(s):

W. G. Rees

Keyword(s):

Time Series ◽

Kola Peninsula ◽

Industrial Pollution ◽

Satellite Images ◽

Vegetation Index ◽

Simple Method ◽

Data Analyst ◽

Normalised Difference Vegetation Index ◽

The Town ◽

Analysis Of Time Series

ABSTRACTThis paper develops a simple method for the detection of ‘vegetation anomalies’, locations where the amount of vegetation, estimated through the use of the normalised difference vegetation index (NDVI), is significantly lower than expected on the basis of topographic factors alone. The method is developed and tested using satellite imagery from the area around the town of Monchegorsk on the Kola Peninsula, Russia. This area has been subject to heavy levels of airborne industrial pollution for many years and the intended purpose of the method is to allow the extent of pollution damaged vegetation to be estimated with as few operational decisions as possible by the data analyst, thus suiting it for automation and for the analysis of time-series of satellite images. While the work described in this paper is to some extent preliminary, it does establish that spatial variations in the NDVI of undisturbed vegetation can, at least in the study area, be modelled satisfactorily using topographic variables, and that negative departures from this modelled variation are very strongly associated with industrially mediated damage.

Download Full-text

The Impact of Pan-Sharpening and Spectral Resolution on Vineyard Segmentation through Machine Learning

Remote Sensing ◽

10.3390/rs12060934 ◽

2020 ◽

Vol 12 (6) ◽

pp. 934 ◽

Cited By ~ 3

Author(s):

Eriita G. Jones ◽

Sebastien Wong ◽

Anthony Milton ◽

Joseph Sclauzero ◽

Holly Whittenbury ◽

...

Keyword(s):

Machine Learning ◽

Remote Sensing ◽

Vegetation Index ◽

A Priori ◽

Elevation Angle ◽

Model Performance ◽

Machine Learning Techniques ◽

Normalised Difference Vegetation Index ◽

Model Training ◽

The Impact

Precision viticulture benefits from the accurate detection of vineyard vegetation from remote sensing, without a priori knowledge of vine locations. Vineyard detection enables efficient, and potentially automated, derivation of spatial measures such as length and area of crop, and hence required volumes of water, fertilizer, and other resources. Machine learning techniques have provided significant advancements in recent years in the areas of image segmentation, classification, and object detection, with neural networks shown to perform well in the detection of vineyards and other crops. However, what has not been extensively quantitatively examined is the extent to which the initial choice of input imagery impacts detection/segmentation accuracy. Here, we use a standard deep convolutional neural network (CNN) to detect and segment vineyards across Australia using DigitalGlobe Worldview-2 images at ∼50 cm (panchromatic) and ∼2 m (multispectral) spatial resolution. A quantitative assessment of the variation in model performance with input parameters during model training is presented from a remote sensing perspective, with combinations of panchromatic, multispectral, pan-sharpened multispectral, and the spectral Normalised Difference Vegetation Index (NDVI) considered. The impact of image acquisition parameters—namely, the off-nadir angle and solar elevation angle—on the quality of pan-sharpening is also assessed. The results are synthesised into a ‘recipe’ for optimising the accuracy of vineyard segmentation, which can provide a guide to others aiming to implement or improve automated crop detection and classification.

Download Full-text