Jonas Franke

Multi-temporal wheat disease detection by multi-spectral remote sensing

For the implementation of site-specific fungicide applications, the spatio-temporal dynamics of crop diseases must be well known. Remote sensing can be a useful tool to monitor the heterogeneity of crop vitality within agricultural sites. However, the identification of fungal infections at an early growth stage is essential. This study examines the potential of multi-spectral remote sensing for a multi-temporal analysis of crop diseases. Within an experimental field, a 6 ha plot of winter wheat was grown, containing all possible infective stages of the powdery mildew (Blumeria graminis) and leaf rust (Puccinia recondita) pathogens. Three high-resolution remote sensing images were used to execute a spatio-temporal analysis of the infection dynamics. A decision tree, using mixture tuned matched filtering (MTMF) results and the Normalized Difference Vegetation Index (NDVI), was applied to classify the data into areas showing different levels of disease severity. Classification results were compared to ground truth data. The classification accuracy of the first scene was only 56.8%, whereas the scenes from May 28th and June 20th achieved considerably higher accuracies of 65.9% and 88.6% respectively. The results showed that high-resolution multi-spectral data are generally suitable to detect in-field heterogeneities of crop vigour but are only moderately suitable for early detection of crop infections.

Soil heterogeneity at the field scale: a challenge for precision crop protection

Crop protection seldom takes into account soil heterogeneity at the field scale. Yet, variable site characteristics affect the incidence of pests as well as the efficacy and fate of pesticides in soil. This article reviews crucial starting points for incorporating soil information into precision crop protection (PCP). At present, the lack of adequate field maps is a major drawback. Conventional soil analyses are too expensive to capture soil heterogeneity at the field scale with the required spatial resolution. Therefore, we discuss alternative procedures exemplified by our own results concerning (i) minimally and non-invasive sensor techniques for the estimation of soil properties, (ii) the evidence of soil heterogeneity with respect to PCP, and (iii) current possibilities for incorporation of high resolution soil information into crop protection decisions. Soil organic carbon (SOC) and soil texture are extremely interesting for PCP. Their determination with minimally invasive techniques requires the sampling of soils, because the sensors must be used in the laboratory. However, this technique delivers precise information at low cost. We accurately determined SOC in the near-infrared. In the mid-infrared, texture and lime content were also exactly quantified. Non-invasive sensors require less effort. The airborne HyMap sensor was suitable for the detection of variability in SOC at high resolution, thus promising further progress regarding SOC data acquisition from bare soil. The apparent electrical conductivity as measured by an EM38 sensor was shown to be a suitable proxy for soil texture and layering. A survey of arable fields near Bonn (Germany) revealed widespread within-field heterogeneity of texture-related ECa, SOC and other characteristics. Maps of herbicide sorption and application rate were derived from sensor data, showing that optimal herbicide dosage is strongly governed by soil variability. A phytoassay with isoproturon confirmed the reliability of spatially varied herbicide application rates. Mapping areas with an enhanced leaching risk within fields allows them to be kept free of pesticides with related regulatory restrictions. We conclude that the use of information on soil heterogeneity within the concept of PCP is beneficial, both economically and ecologically.

Spectral requirements on airborne hyperspectral remote sensing data for wheat disease detection

Remote sensing approaches are of increasing importance for agricultural applications, particularly for the support of selective agricultural measures that increase the productivity of crop stands. In contrast to multi-spectral image data, hyperspectral data has been shown to be highly suitable for the detection of crop growth anomalies, since they allow a detailed examination of stress-dependent changes in certain spectral ranges. However, the entire spectrum covered by hyperspectral data is probably not needed for discrimination between healthy and stressed plants. To define an optimal sensor-based system or a data product designed for crop stress detection, it is necessary to know which spectral wavelengths are significantly affected by stress factors and which spectral resolution is needed. In this study, a single airborne hyperspectral HyMap dataset was analyzed for its potential to detect plant stress symptoms in wheat stands induced by a pathogen infection. The Bhattacharyya distance (BD) with a forward feature search strategy was used to select relevant bands for the differentiation between healthy and fungal infected stands. Two classification algorithms, i.e. spectral angle mapper (SAM) and support vector machines (SVM) were used to classify the data covering an experimental field. Thus, the original dataset as well as datasets reduced to several band combinations as selected by the feature selection approach were classified. To analyze the influence of the spectral resolution on the detection accuracy, the original dataset was additionally stepwise spectrally resampled and a feature selection was carried out on each step. It is demonstrated that just a few phenomenon-specific spectral features are sufficient to detect wheat stands infected with powdery mildew. With original spectral resolution of HyMap, the highest classification accuracy could be obtained by using only 13 spectral bands with a Kappa coefficient of 0.59 in comparison to Kappa 0.57 using all spectral bands of the HyMap sensor. The results demonstrate that even a few hyperspectral bands as well as bands with lower spectral resolution still allow an adequate detection of fungal infections in wheat. By focusing on a few relevant bands, the detection accuracy could be enhanced and thus more reliable information could be extracted which may be helpful in agricultural practice.

Comparison of multi- and hyperspectral imaging data of leaf rust infected wheat plants

In the context of precision agriculture, several recent studies have focused on detecting crop stress caused by pathogenic fungi. For this purpose, several sensor systems have been used to develop in-field-detection systems or to test possible applications of remote sensing. The objective of this research was to evaluate the potential of different sensor systems for multitemporal monitoring of leaf rust (puccinia recondita) infected wheat crops, with the aim of early detection of infected stands. A comparison between a hyperspectral (120 spectral bands) and a multispectral (3 spectral bands) imaging system shows the benefits and limitations of each approach. Reflectance data of leaf rust infected and fungicide treated control wheat stand boxes (1sqm each) were collected before and until 17 days after inoculation. Plants were grown under controlled conditions in the greenhouse and measurements were taken under consistent illumination conditions. The results of mixture tuned matched filtering analysis showed the suitability of hyperspectral data for early discrimination of leaf rust infected wheat crops due to their higher spectral sensitivity. Five days after inoculation leaf rust infected leaves were detected, although only slight visual symptoms appeared. A clear discrimination between infected and control stands was possible. Multispectral data showed a higher sensitivity to external factors like illumination conditions, causing poor classification accuracy. Nevertheless, if these factors could get under control, even multispectral data may serve a good indicator for infection severity.

Monitoring Fire and Selective Logging Activities in Tropical Peat Swamp Forests

Climate change mitigation schemes, such as REDD and biodiversity conservation in tropical rainforests, necessitate remote sensing based forest monitoring capabilities with high spatial resolution and temporal coverage. Regular monitoring has to be capable of detecting rapid changes in forest extent, i.e. deforestation, and subtle changes to the forest cover caused by logging and/or fire, described as forest degradation. Particularly the early detection of illegal logging activities is important for the conservation of tropical forests. In the present study, a forest disturbance monitoring approach was developed and tested, which makes use of high resolution satellite imagery. A time series consisting of three images, acquired between May 2009 and June 2010, was analyzed covering a remote area of tropical peat swamp forest in Central Kalimantan, Indonesia. The forest area was assessed by an object-oriented classification. Logging activities and the impact of fire were detected by a pixel-based spectral mixture analysis. Forest, non-forest and logging trails could be differentiated with an overall classification accuracy of 91.5% (Kappa of 0.87). A high forest disturbance rate of 8.7% was found in the study area. Low impact logging could be detected reliably and the progress was tracked over time. The results show that the timely detection of forest disturbances is necessary because of the fast regrowth of vegetation. The study emphasises the importance of high resolution satellite imagery for tropical forest monitoring and for timely updating forest status assessments, which is important for the implementation of REDD.

Assessment of NDVI- Differences Caused by Sensor Specific Relative Spectral Response Functions

The normalized difference vegetation index (NDVI) is the most often used remote sensing-based indicator to monitor dynamics of land surfaces and environmental changes. Due to different sensor characteristics, the NDVI values vary according to the recording system. This study focuses on the factor of spectral sensor characteristics, which can complicate the interpretation of multisensoral NDVI data. Therefore, multispectral bands of Landsat 5TM, QuickBird and SPOT5 were simulated from hyperspectral data. These simulated data sets show identical characteristics (except spectrally) like sensor geometry, atmospheric conditions, topography and spatial resolution. This allows a direct comparison of NDVI differences caused by the factor of different spectral characteristics. The results show substantial NDVI differences with a systematic nonlinear offset between the sensor systems, solely caused by different relative spectral response functions of the sensor bands. Thus, a step-by-step sensor intercalibration is desirable, which first takes the spectral characteristics into account, because this NDVI-differences causing factor is clearly determinable against others.

Geostatistical analysis of the spatiotemporal dynamics of powdery mildew and leaf rust in wheat.

Plant diseases are dynamic systems that progress or regress in spatial and temporal dimensions. Site-specific or temporally optimized disease control requires profound knowledge about the development of each stressor. The spatiotemporal dynamics of leaf rust (Puccinia recondite f. sp. tritici) and powdery mildew (Blumeria graminis f. sp. tritici) in wheat was analyzed in order to evaluate typical species-dependent characteristics of disease spread. During two growing seasons, severity data and other relevant plant growth parameters were collected in wheat fields. Spatial characteristics of both diseases were assessed by cluster analyses using spatial analysis by distance indices, whereas the temporal epidemic trends were assessed using statistical parameters. Multivariate statistics were used to identify parameters suitable for characterizing disease trends into four classes of temporal dynamics. The results of the spatial analysis showed that both diseases generally occurred in patches but a differentiation between the diseases by their spatial patterns and spread was not possible. In contrast, temporal characteristics allowed for a differentiation of the diseases, due to the fact that a typical trend was found for leaf rust which differed from the trend of powdery mildew. Therefore, these trends suggested a high potential for temporally optimized disease control. Precise powdery mildew control would be more complicated due to the observed high variability in spatial and temporal dynamics. The general results suggest that, in spite of the high variability in spatiotemporal dynamics, disease control that is optimized in space and time is generally possible but requires consideration of disease- and case-dependent characteristics.

Aerial imagery for monitoring land use in East African wetland ecosystems

Anthropogenic pressure and environmental change processes are key drivers of the recent intensification in the agricultural use of East African wetlands. Land shortage and degradation of upland areas as well as climate change effects turn wetland ecosystems into focal points of production by commercial and traditional users, entailing rapid wetland use changes and, in some instances, severe wetland degradation. An ecosystem inventory by mapping land cover and monitoring land use changes with remote sensing improves our understanding of change processes in wetlands and will contribute to the provision of decision support for sustainable use of wetland ecosystems. However, the spatial resolution of satellite systems is often too coarse to derive land use information at the plot level. In particular, small wetlands often exhibit abrupt transitions into different types of land use and landscape elements. Hence, monitoring of small wetlands requires spatially high-resolution remote sensing data, accounting for the prevailing small-scale diversity in land use. High-resolution aerial imagery, which is not available for most parts of East Africa, may provide information of wetland use/change at the required plot-level scale. Therefore, image acquisition campaigns over Kenyan and Tanzanian wetlands were realized with a common Nikon D-200 in September 2008 and February 2009, respectively. A comprehensive geo-referenced image data set that displays land use units at the plot level was obtained, used to discriminate various land cover types. Land cover/-land use maps can be derived that reveal land use trends fundamental for providing decision support for a sustainable wetland use.

Earth observation in support of malaria control and epidemiology: MALAREO monitoring approaches

Malaria affects about half of the worlds population, with the vast majority of cases occuring in Africa. National malaria control programmes aim to reduce the burden of malaria and its negative, socioeconomic effects by using various control strategies (e.g. vector control, environmental management and case tracking). Vector control is the most effective transmission prevention strategy, while environmental factors are the key parameters affecting transmission. Geographic information systems (GIS), earth observation (EO) and spatial modelling are increasingly being recognised as valuable tools for effective management and malaria vector control. Issues previously inhibiting the use of EO in epidemiology and malaria control such as poor satellite sensor performance, high costs and long turnaround times, have since been resolved through modern technology. The core goal of this study was to develop and implement the capabilities of EO data for national malaria control programmes in South Africa, Swaziland and Mozambique. High- and very high resolution (HR and VHR) land cover and wetland maps were generated for the identification of potential vector habitats and human activities, as well as geoinformation on distance to wetlands for malaria risk modelling, population density maps, habitat foci maps and VHR household maps. These products were further used for modelling malaria incidence and the analysis of environmental factors that favour vector breeding. Geoproducts were also transferred to the staff of national malaria control programmes in seven African countries to demonstrate how EO data and GIS can support vector control strategy planning and monitoring. The transferred EO products support better epidemiological understanding of environmental factors related to malaria transmission, and allow for spatio-temporal targeting of malaria control interventions, thereby improving the cost-effectiveness of interventions.

Multitemporal spectroscopy for crop stress detection using band selection methods

A fast and precise sensor-based identification of pathogen infestations in wheat stands is essential for the implementation of site-specific fungicide applications. Several works have shown possibilities and limitations for the detection of plant stress using spectral sensor data. Hyperspectral data provide the opportunity to collect spectral reflectance in contiguous bands over a broad range of the electromagnetic spectrum. Individual phenomena like the light absorption of leaf pigments can be examined in detail. The precise knowledge of stress-dependent shifting in certain spectral wavelengths provides great advantages in detecting fungal infections. This study focuses on band selection techniques for hyperspectral data to identify relevant and redundant information in spectra regarding a detection of plant stress caused by pathogens. In a laboratory experiment, five 1 sqm boxes with wheat were multitemporarily measured by a ASD Fieldspec® 3 FR spectroradiometer. Two stands were inoculated with Blumeria graminis - the pathogen causing powdery mildew - and one stand was used to simulate the effect of water deficiency. Two stands were kept healthy as control stands. Daily measurements of the spectral reflectance were taken over a 14-day period. Three ASD Pro Lamps were used to illuminate the plots with constant light. By applying band selection techniques, the three types of different wheat vitality could be accurately differentiated at certain stages. Hyperspectral data can provide precise information about pathogen infestations. The reduction of the spectral dimension of sensor data by means of band selection procedures is an appropriate method to speed up the data supply for precision agriculture.