CHAPTER 2. MODELINGCROPPHENOLOGY IN THE U.S. CORN BELT USING SMOS VEGETATION OPTICAL THICKNESS DATA
The demands on agriculture to feed an increasing world population continue to grow. A 2009 report by the Food and Agriculture Organization (FAO) projects that by 2050 agricultural production must rise by 70% to meet expected demand (Economist, 2016). In anticipation, farmers have started to employ automated technology and data (from drones, sensors, and satellites) to transform farming into a precisely controlled scientific laboratory. A major piece of technology under development is the use of remote sensing devices to monitor crops, forecast yield, and analyze crop trends or patterns across seasons. Rather than going out into the field to gather sample data, a farmer or agronomist can remotely access information via ground level sensors or analyze satellite data collected across a region. New research shows the European Space Agency’s Soil Moisture and Ocean Salinity (SMOS) satellite can estimate the mass of water contained in vegetation tissue. Also re- ferred to as the water column density of vegetation (Jackson et al., 2012) this variable is proportional to the amount of ground vegetation, which varies over the course of a growing season. Recent work by Hornbuckle et al. (2016) confirms this new variable τ mirrors the growth and senescence of crops. Analyzing SMOS data from intensively cultivated agri- cultural regions, τ consistently peaked around the same time as corn and soybean reached their maximum water column density.
The main goal of this study is to model the spatial extreme precipitation over the Pa- cific Northwest United States using satellite-based rainfall estimates from PERSIANN-CDR product. To do so, several max-stable models were used, including the Schlather, the Brown- Resnick, and the Extremal-t models. The inputs to these model are the bias-corrected PERSIANN-CDR data. Latitude, longitude, altitude, and mean annual precipitation infor- mation were also used as covariates. In order to identify the most suitable model over the study region, we evaluated the performances in terms of capturing the spatial dependence structure of the extreme rainfall data, and models’ capability in estimating the marginal distributions of the data. Extremal coefficient and f-madogram indices were used to as eval- uation metrics. The quality of the applied statistical models was also assessed using the TIC metric. After identifying the superior max-stable model, we found the Extremal-t model demonstrated the highest skill, and was further applied to estimate the GEV parameters and develop the extreme return levels. The return level estimates from the Etremal-t model were verified in terms of pair-wise and group-wise maxima using Q-Q plots. The gener- ated return level plots from this model were compared with the empirical quantile estimates from the bias-corrected PERSIANN-CDR data. Finally, the constructed DDF curves were compared with the gauge-based estimates at multiple gauge locations across the study area.
We investigated the use of principal components analysis (PCA) to visualize dominant patterns and identify anomalies in a multi-year land surface phenologydata set (231 m × 231 m normalized diﬀerence vegetation index (NDVI) values derived from the Moderate Resolution Imaging Spectroradiometer (MODIS)) used for detecting threats to forest health in the conterminous United States (CONUS). Our goal is to ﬁnd ways that PCA can be used with this massive data set to automate the process of detecting forest disturbance and attributing it to particular agents. We brieﬂy describe the parallel computational approaches we used to make PCA feasible, and present some examples in which we have used it to visualize the seasonal vegetation phenology for the CONUS and to detect areas where anomalous NDVI traces suggest potential threats to forest health. Keywords: phenology; MODIS; NDVI; remote sensing; principal components analysis; singular value decomposition; data mining; anomaly detection; high performance computing; parallel computing
Long-term NDVI time series is required to analyze the vegetation phenology. For that purpose, data from AVHRR (Advanced Very High Resolution Radiometer) sensors which have been archived since 1981 are commonly used in vegetation analysis. Nevertheless, the datasets suffer from quality deficiencies due to instrumentation problems, changes in sensor angle, atmospheric conditions (e.g., clouds and haze), and ground conditions (e.g., snow cover), as has been reported by Bradley et al. (2007). They argued that the identification of phenology parameters via NDVI datasets is problematic. To overcome these problems, maximum value compositing (MVC), the best index slope extraction (BISE) and spatial and temporal smoothing methods have been introduced (Bradley et al. 2007). Among different datasets based on AVHRR, data from the NASA Global Inventory Monitoring and Modeling Systems (GIMMS) group at the Laboratory for Terrestrial Physics (Tucker et al. 2005) is the most known dataset (De beurs and Henebry 2010, Fensholt et al. 2009) . Moreover, due to enhanced and high quality of dataset which was achieved from applying several correction methods (Tucker et al. 2005), this dataset is widely used.
Military operations are highly dependent on accurate knowledge of the terrain. Today, in modern warfare, there are many situations wherein armed forces are expected to travel “off-road”. When these situations occur, decision makers and strategic planners rely on estimates of the geophysical properties of the terrain to evaluate its capacity to support military vehicles and personnel. Agri- cultural outputs are also highly dependent on accurate knowledge of the terrain. This is because soil compaction reduces crop yields. Subsoiling is a practice that breaks up the terrain to allow for increased water movement, aeration, and access to nutrients and minerals . But, subsoiling is also a practice that is time consuming and costly. Therefore, at the start of each growing season, farmers rely on estimates of the geophysical properties of the terrain to identify the areas where soil compaction exists, and where subsoiling is needed.
The spatial and physical characteristics of urban features, urban patterns and its forms may be quantified using spatial metrics . These indices can be obtained directly from thematic maps derived from remote sensing data . The availability of remotelysenseddata from multiple dates enables us to carry out studies on urban modeling , urban landscape pattern analysis , and urban growth studies [13,16]. Globally, different studies on urban growth and model analysis have been carried out [8,9,12,15,17]. However, with a few exceptions, such studies are scarce for India [16,18–20]. The city of Chennai, India has been one of the fastest growing urban areas in the country in the last three decades. This has resulted in traffic congestion, air and water pollution, uncontrolled increase of population, encroachment, water and land scarcity, the growth of slums, and the degradation of vegetation within and in the peripheral areas of the city . Thus, such a study would benefit urban planners that need to understand the spatiotemporal changes of urban areas to better address these environmental problems and, at the same time, to ensure the provision of basic infrastructures and facilities without disturbing ecosystems. This study (1) studies land- use and land-cover changes from 1991 to 2016; (2) examines the spatiotemporal urban growth pattern using entropy and spatial metrics; and (3) predicts the urban growth and the urban sprawl for the year 2027.
Third, cloud computing provides a way for QuakeSim to work with its collaborators to more efficiently share data sets (as in Figure 2). As described previously, QuakeSim does not process raw observational data but is instead downstream consumer of data. Maintaining consistent copies with our upstream data providers is a challenge. QuakeSim’s QuakeTables database houses some processed InSAR data products and also the complete set of processed repeat pass interferometry products from the airborne UAVSAR InSAR project. These data provide essential information for modeling earthquake processes and particularly for developing accurate fault models. We are collaborating with data product providers to ensure standard interfaces formats as well as jointly used cloud infrastructure where appropriate. The infrastructure must be flexible enough to support other data sets and use cases. Under the present cloud models, storage at existing data center appears more cost effective than storage on the cloud where recurring costs are at present prohibitive. However, this may change in the future. Microsoft Azure’s Blob storage service, Amazon’s S3, and the Lustre file system- based Whamcloud are examples of unstructured storage, and BigTable, HBase, and the Azure Table Service are examples of structured data storage. We will evaluate these for the storage and access of large collections of individually large data sets. A key observation from our
calculated hydraulic slopes, understanding the variables that control the error could provide insight into the applicability and constraints of using the DEM slope for hydraulic modeling.
The slope residual and the residuals for each of the three models are normally distributed as seen on Figure 5.3. This indicates that inferential statistics regarding probable accuracy can be made when using the DEM derived slope and the hydraulic models. The ranked distributions also indicate that the DEM slope is unbiased relative to the hydraulic slope. This is indicated by the coincident residual distribution relative to a normal distribution with a mean of zero and the same standard deviation. However, the residuals from Models 1, 2 and 3 appear to be biased, as shown by the fact that tlie residuals plot either above (in the case of Model 1) or below (in the case of Model 2 and Model 3) the normal distribution. This suggests that these models could be linearly corrected by adjusting the magnitude of tlie coefficient of each model. However, because the models are used to predict only the mean annual flow', adjusting the model coefficients could result in greater errors for higher and low'er flows if the models were used to predict a wider range of discharge at each station.
Information of cropphenology is essential for evaluating crop productivity and crop management. Therefore we developed a new method for remotely determining phenological stages of paddy rice. The method consists of three procedures: (i) prescription of multi- temporal MODIS/Terra data; (ii) filtering time-series Enhanced Vegetation Index (EVI) data by time-frequency analysis; and (iii) specifying the phenological stages by detecting the maximum point, minimal point and inflection point from the smoothed EVI time profile. Applying this method to MODIS data, we determined the planting date, heading date, harvesting date, and growing period in 2002. And we validated the performance of the method against statistical data in 30 paddy fields. As for the filtering, we adopted wavelet and Fourier transforms. Three types of mother wavelet (Daubechies, Symlet and Coiflet) were used in Wavelet transform. As the results of validation, the wavelet transform performed better than the Fourier transform. Specifically, the case using Coiflet (order = 4) gave remarkably good results in determining phenological stages and growing periods. The root mean square errors of the estimated phenological dates against the statistical data were: 12.1 days for planting date, 9.0 days for heading date, 10.6 days for harvesting date, and 11.0 days for growing period. The method using wavelet transform with Coiflet (order = 4) allows the determination of regional characteristics of rice phenology. We proposed this new method using the wavelet transform; Wavelet based Filter for determining CropPhenology (WFCP).
months, at different spatial resolutions, and available on line for free. According to these intrinsic characteristics, RS have shown its capacity to detect LCC and assemble reliable time series [5-7]. Despite the achievement of regular data, characterizing LCC using RS data is still complex because of the combination of several processes occurring at the same time: seasonal changes, abrupt changes, climate alterations and acquisition errors . The Normalized Difference Vegetation Index (NDVI), calculated from near infrared and red reflectances, was described firstly by Rouse et al.  and is widely the most used vegetation index. This index is generally rec- ognized as a good indicator of vegetation activity [6, 10-12] and has been related with several biophysical variables such as: fraction of absorbed photosynthetically active radiation [13,14], leaf area index [14,15], primary production [16,17], among others. Also, NDVI is espe- cially useful in multi temporal datasets because they permit to describe vegetation phenology [5,12,18,19].
Each LUC class can be defined using different dataset and rules according to characteristics of LUC. In this chapter, object and pixel based classifications were evaluated in a Mediterranean agricultural land called Lower Seyhan Plane (LSP) (figure 14).
Especially in agricultural land, object based classification is the most suitable technique. Most of the agricultural fields has regular shape and contains one dominant crop in a field in one time. In winter time dominant crop is wheat in the study area, summer period includes corn, soybean and cotton. Mapping the farmlands may be inappropriateusing only one optical image. Multitemporal object based classification approach was used to map LUC in LSP. Two Landsat TM images from March and April were classified together, and June, August images and some of physical variables like distance from cost line and distance from built up areas were added to create rules for each LUC. In this chapter only winter crop pattern discussed using LDA classifier, and rule dependent object based classification were compared each other to see accuracy difference in each LUC (figure 15).
In our study, we chose to use MODIS because it provides data products at spatial and temporal scales suitable for comparison with eddy covariance data. The advent of MODIS collection 6 now provides a standardized surface reflectance product, including land and ocean bands. Consequently, a global CCI is now becoming widely available, and offers a practical means of assessing pigment dynamics associated with seasonal changes in photosynthetic activity in evergreens, where established greenness indices (e.g., NDVI, fPAR, leaf area index) cannot properly capture this seasonal photosynthetic activity. Of particular significance is the similar behavior of the CCI across three evergreen conifer stands and several spatial scales, including ground sampling at leaf and stand scales and whole-ecosystem satellite measurements. This scale independence suggests that the CCI can provide a potent metric of evergreen photosynthetic phenology from a variety of remote sensing platforms, and can be supported by ground sampling that assesses pigment levels or foliage optical properties.
Linear Discriminate Analysis (LDA) is a method to dis- criminate between two or more groups of samples. The groups to be discriminated can be defined either naturally by the problem under investigation, or by some preced- ing analysis, such as a cluster analysis. The number of groups is not restricted to two, although the discrimina- tion between two groups is the most common approach. Linear Discrimination Analysis (LDA) is a commonly used technique for data classification. LDA approach is explained by . It easily handles the case where the within-class frequencies are unequal and their perform- ance has been examined on randomly generated test data. This method maximizes the ratio of between-class vari- ance to the within-class variance in any particular data set thereby guaranteeing maximal separability. LDA doesn’t change the location but only tries to provide more class separability and draw a decision region be- tween the given classes. This method also helps to better understand the distribution of the feature data. In the current study, Class-independent transformation type of LDA was performed. This approach involves maximiz- ing the ratio of overall variance to within class variance. It uses only one optimizing criterion to transform the data sets and hence all data points irrespective of their class identity are transformed using this transform. In this type of LDA, each class is considered as a separate class against other classes. In LDA, within-class and between- class scatter are used to formulate criteria for class sepa- rability. Within-class scatter is the expected covariance of each of the classes. The scatter measures are computed using Equations (3) and (4).
In an end-to-end computational infrastructure users should be able to evaluate data, develop science models, produce improved earthquake forecasts, and respond to disasters in intuitive map-based interfaces. Fault models can be constrained and improved not just by geology, but also by feature identification from InSAR (UAVSAR) and inversions of both GPS and InSAR crustal deformation data [2-3]. Forecasting is improved by development of better interacting fault models, pattern analysis, and fusion of both seismicity and crustal deformation data. Increasing the accessibility and utility of GPS, InSAR, and geologic data, addresses science challenges such as earthquake forecasting or fluid migration. Intuitive computational infrastructure also can enable new observations by providing tools to conduct simulation experiments and new information products for use in a wide variety of fields ranging from earthquake research to earthquake response. Timely and affordable delivery of information to users in the form of high-level products is necessary for earthquake forecasting and emergency response, but it also necessary for exploiting crustal deformation to enable new discoveries and uses. There are numerous practical issues to establishing an effective computational infrastructure. Of chief importance is that tools be intuitive and easily accessible. Some QuakeSim tools are public and reside outside of any required login. This mode of operation is often preferred by users as it avoids the need to remember another login and password combination and allows for greater privacy. However, there are also limitations. Chief of these is that
MONTHLY AGRICULTURAL YIELD SURVEY DATA
Survey data is also being mapped to provide a supplemental visual tool along with AVHRR and crop progress and condition. Agricultural Yield Surveys are conducted monthly during May through November. All States, except Alaska and Hawaii, participate in the Survey. The months for which individual States participate will depend on the estimating program needs for that State. The survey provides the primary indications for the monthly Crop Production report that publishes forecasts of production during the growing season. The crop acreage and yield data are collected by mail and telephone. The sample consists of a sub-sample of operators who reported the crop of interest during the March and June Agricultural Surveys. For the visualization, we take the SAS data set of survey data created by Headquarters and generate summary statistics in Arc/Info to produce county level yield responses for corn and soybeans.
Figure 1. Many variables will contribute to the formation of an algal bloom, but there are still so many unknown variables that cause a particular aquatic environment to produce an algal or cyanobacterial bloom…………………………………………………………………………...…3 Figure 2. Landsat 8 images (USGS) cover a large spatial region, 170 km x 183 km, resulting in ~325 km of Ohio River captured in one satellite image. Polygons in the image on the right reveal the area where samples were taken on the Ohio River, near Huntington, West Virginia…….…...8 Figure 3. One scene of Landsat 8 OLI, as shown in the green image insert, covers Boat Launch, Lock 27, and Harris Park are located within the Greenup Pool of the Ohio River……………...10 Figure 4. A 500 mL Nalgene bottle was attached to the end of a fishing pole and cast into the Ohio River to obtain samples away from shore not using a boat………………..……………....11 Figure 5. Landsat 8 band coverage showing reflectance area. The Aerosal band (1) and the Blue band (2) are next to each other in the spectrum………………………………………………….12 Figure 6. Buffer polygons (200 m x 800 m) were created to extract pixels on the Ohio River from the Landsat 8 images. Polygons were sized to capture the center 50% of the river near the sampling point……….…………………………………………………………………………...14 Figure 7. Actual chlorophyll a compared to predicted chlorophyll a. Chlorophyll a was predicted using band 2 (Blue) and band 5 (NIR) of Landsat 8……………………………………………..17 Figure 8. Actual chlorophyll a and b compared to predicted chlorophyll a and b. Chlorophyll a and b were predicted using band 2 (Blue), band 5 (NIR), and band 6 (SWIR 1) of Landsat
precise accuracy. Variations o f spectral responses may also lead to feature heterogeneity. Therefore, there are challenges for estimating the forest species composition in a complex forest ecosystem characterized by its heterogeneous and high density multilayer canopies. On the other hand, the spectral resolution of remotelysenseddata also plays an important role in order to achieve high classification accuracy. High spectral resolution with high spatial resolution results in well-define signature vectors for target o f interest, hence, able to distinguish similar features.