Domain Adaptation in remote sensing: increasing the portability of land-cover classifiers

Among the types of remote sensing acquisitions, optical images are certainly one of the most widely relied upon data sources for Earth observation. They provide detailed measurements of the electromagnetic radiation reflected or emitted by each pixel in the scene. Through a process termed supervised land-cover classification, this allows to automatically yet accurately distinguish objects at the surface of our planet. In this respect, when producing a land-cover map of the surveyed area, the availability of training examples representative of each thematic class is crucial for the success of the classification procedure. However, in real applications, due to several constraints on the sample collection process, labeled pixels are usually scarce. When analyzing an image for which those key samples are unavailable, a viable solution consists in resorting to the ground truth data of other previously acquired images. This option is attractive but several factors such as atmospheric, ground and acquisition conditions can cause radiometric differences between the images, hindering therefore the transfer of knowledge from one image to another. The goal of this Thesis is to supply remote sensing image analysts with suitable processing techniques to ensure a robust portability of the classification models across different images. The ultimate purpose is to map the land-cover classes over large spatial and temporal extents with minimal ground information. To overcome, or simply quantify, the observed shifts in the statistical distribution of the spectra of the materials, we study four approaches issued from the field of machine learning. First, we propose a strategy to intelligently sample the image of interest to collect the labels only in correspondence of the most useful pixels. This iterative routine is based on a constant evaluation of the pertinence to the new image of the initial training data actually belonging to a different image. Second, an approach to reduce the radiometric differences among the images by projecting the respective pixels in a common new data space is presented. We analyze a kernel-based feature extraction framework suited for such problems, showing that, after this relative normalization, the cross-image generalization abilities of a classifier are highly increased. Third, we test a new data-driven measure of distance between probability distributions to assess the distortions caused by differences in the acquisition geometry affecting series of multi-angle images. Also, we gauge the portability of classification models through the sequences. In both exercises, the efficacy of classic physically- and statistically-based normalization methods is discussed. Finally, we explore a new family of approaches based on sparse representations of the samples to reciprocally convert the data space of two images. The projection function bridging the images allows a synthesis of new pixels with more similar characteristics ultimately facilitating the land-cover mapping across images.

Soil water dynamics related to the degree of compaction of two brazilian oxisols under no-tillage

Soil water properties are related to crop growth and environmental aspects and are influenced by the degree of soil compaction. The objective of this study was to determine the water infiltration and hydraulic conductivity of saturated soil under field conditions in terms of the compaction degree of two Oxisols under a no-tillage (NT). Two commercial fields were studied in the state of Rio Grande do Sul, Brazil: one a Haplortox after 14 years under NT; the other a Hapludox after seven years under NT. Maps (50 x 30 m) of the levels of mechanical penetration resistance (PR) were drawn based on the kriging method, differentiating three compaction degrees (CD): high, intermediate and low. In each CD area, the infiltration rate (initial and steady-state) and cumulative water infiltration were measured using concentric rings, with six replications, and the saturated hydraulic conductivity (K(θs)) was determined using the Guelph permeameter. Statistical evaluation was performed based on a randomized design, using the least significant difference (LSD) test and regression analysis. The steady-state infiltration rate was not influenced by the compaction degree, with mean values of 3 and 0.39 cm h-1 in the Haplortox and the Hapludox, respectively. In the Haplortox, saturated soil hydraulic conductivity was 26.76 cm h-1 at a low CD and 9.18 cm h-1 at a high CD, whereas in the Hapludox, this value was 5.16 cm h-1 and 1.19 cm h-1 for the low and high CD, respectively. The compaction degree did not affect the initial and steady-state water infiltration rate, nor the cumulative water infiltration for either soil type, although the values were higher for the Haplortox than the Hapludox.

Microbial alterations of the soil influenced by induced compaction

Compaction is one of the most destructive factors of soil quality, however the effects on the microbial community and enzyme activity have not been investigated in detail so far. The objective of this study was to evaluate the effects of soil compaction caused by the traffic of agricultural machines on the soil microbial community and its enzyme activity. Six compaction levels were induced by tractors with different weights driving over a Eutrustox soil and the final density was measured. Soil samples were collected after corn from the layers 0-0.10 and 0.10-0.20 m. The compaction effect on all studied properties was evident. Total bacteria counts were reduced significantly (by 22-30 %) and by 38-41 % of nitrifying bacteria in the soil with highest bulk density compared to the control. On the other hand, fungi populations increased 55-86 % and denitrifying bacteria 49-53 %. Dehydrogenase activity decreased 20-34 %, urease 44-46 % and phosphatase 26-28 %. The organic matter content and soil pH decreased more in the 0-0.10 than in the 0.10-0.20 m layer and possibly influenced the reduction of the microbial counts, except denitrifying bacteria, and all enzyme activities, except urease. Results indicated that soil compaction influences the community of aerobic microorganisms and their activity. This effect can alter nutrient cycling and reduce crop yields.

Spatial variability of soil chemical properties after coffee tree removal

Assessing the spatial variability of soil chemical properties has become an important aspect of soil management strategies with a view to higher crop yields with minimal environmental degradation. This study was carried out at the Centro Experimental of the Instituto Agronomico, in Campinas, São Paulo, Brazil. The aim was to characterize the spatial variability of chemical properties of a Rhodic Hapludox on a recently bulldozer-cleaned area after over 30 years of coffee cultivation. Soil samples were collected in a 20 x 20 m grid with 36 sampling points across a 1 ha area in the layers 0.0-0.2 and 0.2-0.4 m to measure the following chemical properties: pH, organic matter, K+, P, Ca2+, Mg2+, potential acidity, NH4-N, and NO3-N. Descriptive statistics were applied to assess the central tendency and dispersion moments. Geostatistical methods were applied to evaluate and to model the spatial variability of variables by calculating semivariograms and kriging interpolation. Spatial dependence patterns defined by spherical model adjusted semivariograms were made for all cited soil properties. Moderate to strong degrees of spatial dependence were found between 31 and 60 m. It was still possible to map soil spatial variability properties in the layers 0-20 cm and 20-40 cm after plant removal with bulldozers.

Compressibility of Oxisol aggregates under no-till in response to soil water potential

The system of no-till sowing stands out as being a technology that suits the objectives of more rational use of the soil and greater protection against the erosion. However, through till, any of it, occurs modifications of the soil's structure. This current work aims to study the influence of the energy state of the water and of the organic matter on the mechanism of compaction of Red Oxisol under no-till management system. Humid and non-deformed sample were collected in horizon AP of two agricultural areas under no-till, with and without rotation of cultures. In the laboratory, these samples were broken into fragments and sifted to obtain aggregates of 4 to 5 mm sized, which were placed in equilibrium under four matrix potentials. Thereafter, they were exposed to uni-dimensional compression with pressures varying from 32 to 1,000 kPa. The results in such a way show that the highest compressibility of aggregates both for the tilling with rotation of cultures as for the tilling without rotation of cultures, occurred for matrix potential -32 kPa (humidity of 0.29-0.32 kg kg-1, respectively), while the minor occurred for the potentials of -1 and -1,000 kPa (humidity of 0.35 and 0.27 kg kg-1, respectively), indicating that this soil should not be worked with humidity ranging around 0.29 to 0.32 kg kg-1 and the highest reduction of volume of aggregates was obtained for the mechanical pressures lower than 600 inferior kPa, indicating that these soils showed to be very influenced by compression, when exposed to mechanical work. Also, the aggregates of soil under no-till and rotation of crops presented higher sensitivity to the compression than the aggregates of soil under no-till and without rotation of crops, possibly for having better structural conditions given to a higher content of organic matter.

Carbon content in Amazonian Oxisols after forest conversion to pasture

Soil plays an important role in the C cycle, and substitution of tropical forest by cultivated land affects C dynamic and stock. This study was developed in an area of expansion of human settlement in the Eastern Amazon, in Itupiranga, State of Pará, to evaluate the effects of native forest conversion to Brachiaria brizantha pasture on C contents of a dystrophic Oxisol. Soil samples were collected in areas of native forest (NF), of 8 to 10 year old secondary forest (SF), 1 to 2 year old SF (P1-2), 5 to 7 year old SF (P5-7), and of 10 to 12 year old SF (P10-12), and from under pastures, in the layers 0-2, 2-5 and 5-10 cm, to evaluate C levels and stocks and carry out separation of OM based on particle size. After deforestation, soil density increased to a depth of 5 cm, with greater increase in older pastures. Variation in C levels was greatest in the top soil layer; C contents increased with increasing pasture age. In the layers 2-5 and 5-10 cm, C content proved to be stable for the types of plant cover evaluated. Highest C concentrations were found in the silt fraction; however, C contents were highest in the clay fraction, independent of the plant cover. An increase in C associated with the sand fraction in the form of little decomposed organic residues was observed in pastures, confirming greater sensitivity of this fraction to change in soil use.

The combined use of reflectance, emissivity and elevation Aster/Terra data for tropical soil studies

Reflectance, emissivity and elevation data of the sensor ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer)/Terra were used to characterize soil composition variations according to the toposequence position. Normalized data of SWIR (shortwave infrared) reflectance and TIR (thermal infrared) emissivity, coupled to a soil-fraction image from a spectral mixture model, were evaluated to separate bare soils from nonphotosynthetic vegetation. Regression relationships of some soil properties with reflectance and emissivity data were then applied on the exposed soil pixels. The resulting estimated values were plotted on the ASTER-derived digital elevation model. Results showed that the SWIR bands 5 and 6 and the TIR bands 10 and 14 measured the clay mineral absorption band and the quartz emissivity feature, respectively. These bands improved also the discrimination between nonphotosynthetic vegetation and soils. Despite the differences in pixel size and field sampling size, some soil properties were correlated with reflectance (R² of 0.65 for Al2O3 in band 6; 0.61 for Fe2O3 in band 3) and emissivity (R² of 0.65 for total sand fraction in the 10/14 band ratio). The combined use of reflectance, emissivity and elevation data revealed variations in soil composition with topography in specific parts of the landscape. From higher to lower slope positions, a general decrease in Al2O3 and increase in total sand fraction was observed, due to the prevalence of Rhodic Acrustox at the top and its gradual transition to Typic Acrustox at the bottom.

Characterization of soil structure and porosity under long-term conventional tillage and no-tillage systems

Different management systems tend to modify soil structure and porosity over the years. The aim of this study was to study modifications in the morphostructure and porosity of dystroferric Red Latosol (Oxisol) under conventional tillage and no-tillage over a 31- year period. The study began with the description of soil profiles based on the cropping profile method, to identify the most compact structures, define sample collection points for physical and chemical analysis, and determine the water retention curve. A forest soil profile was described and used as reference. The results showed that, under conventional tillage, the microaggregate structure of the Oxisol was fragmented between 0 and 0.20 m, and compact (bulk density = 1.52 Mg m-3) in the sub-surface layer between 0.20 and 0.50 m. Under no-tillage, the structure became compacted (bulk density = 1.40 Mg m-3) between 0 and 0.60 m, but contained fissures and biopores. The volume of the class with a pore diameter of > 100 µm under no-tillage was limited, but practically non-existent in the conventional management system. On the other hand, the classes with a pore diameter of < 100 µm were not affected by the type of soil management system.

Nutrient supply by mass flow and diffusion to maize plants in response to soil aggregate size and water potential

Nutrients are basically transported to the roots by mass flow and diffusion. The aim of this study was to quantify the contribution of these two mechanisms to the acquisition of macronutrients (N, P, K, Ca, Mg, and S) and cationic micronutrients (Fe, Mn, Zn, and Cu) by maize plants as well as xylem exudate volume and composition in response to soil aggregate size and water availability. The experiment was conducted in a greenhouse with samples of an Oxisol, from under two management systems: a region of natural savanna-like vegetation (Cerradão, CER) and continuous maize under conventional management for over 30 years (CCM). The treatments were arranged in a factorial [2 x (1 + 2) x 2] design, with two management systems (CER and CCM), (1 + 2) soil sifted through a 4 mm sieve and two aggregate classes (< 0.5 mm and 0.5 - 4.0 mm) and two soil matric potentials (-40 and -10 kPa). These were evaluated in a randomized block design with four replications. The experiment was conducted for 70 days after sowing. The influence of soil aggregate size and water potential on the nutrient transport mechanisms was highest in soil samples with higher nutrient concentrations in solution, in the CER system; diffusion became more relevant when water availability was higher and in aggregates < 0.5 mm. The volume of xylem exudate collected from maize plants increased with the decrease in aggregate size and the increased availability of soil water in the CER system. The highest Ca and Mg concentrations in the xylem exudate of plants grown on samples from the CER system were related to the high concentrations of these nutrients in the soil solution of this management system.

Isolation and characterization of two plant growth-promoting bacteria from the rhizoplane of a legume (Lupinus albescens) in sandy soil

Two bacterial strains that amplified part of the nifH gene, RP1p and RP2p, belonging to the genus Enterobacter and Serratia, were isolated from the rhizoplane of Lupinus albescens. These bacteria are Gram-negative, rod-shaped, motile, facultative anaerobic, and fast-growing; the colonies reach diameters of 3-4 mm within 24 h of incubation at 28 ºC. The bacteria were also able to grow at temperatures as high as 40 ºC, in the presence of high (2-3 % w/v) NaCl concentrations and pH 4 -10. Strain RP1p was able to utilize 10 of 14 C sources, while RP2p utilized nine. The isolates produced siderophores and indolic compounds, but none of them was able to solubilize phosphate. Inoculation of L. albescens with RP1p and RP2p strains resulted in a significant increase in plant dry matter, indicating the plant-growth-promoting abilities of these bacteria.

Nitrate and ammonium in soil solution in tobacco management systems

Tobacco farmers of southern Brazil use high levels of fertilizers, without considering soil and environmental attributes, posing great risk to water resources degradation. The objective of this study was to monitor nitrate and ammonium concentrations in the soil solution of an Entisol in and below the root zone of tobacco under conventional tillage (CT), minimum tillage (MT) and no-tillage (NT). The study was conducted in the small-watershed Arroio Lino, in Agudo, State of Rio Grande do Sul, Brazil. A base fertilization of 850 kg ha-1 of 10-18-24 and topdressing of 400 kg ha-1 of 14-0-14 NPK fertilizer were applied. The soil solution was sampled during the crop cycle with a tension lysimeter equipped with a porous ceramic cup. Ammonium and nitrate concentrations were analyzed by the distillation and titration method. Nitrate concentrations, ranging from 8 to 226 mg L-1, were highest after initial fertilization and decreased during the crop cycle. The average nitrate (N-NO3-) concentration in the root zone was 75 in NT, 95 in MT, and 49 mg L-1 in CT. Below the root zone, the average nitrate concentration was 58 under NT, 108 under MT and 36 mg L-1 under CT. The nitrate and ammonium concentrations did not differ significantly in the management systems. However, the nitrate concentrations measured represent a contamination risk to groundwater of the watershed. The ammonium concentration (N-NH4+) decreased over time in all management systems, possibly as a result of the nitrification process and root uptake of part of the ammonium by the growing plants.

Properties of top soil and the relationship between soil and trees in a boreal Scots pine stand

Summary

Soil-applied silicon decreases severity of wheat spot blotch on silicon-deficient soils

Spot bloth caused by Bipolaris sorokiniana is an important wheat desease mainly in hot and humid regions. The aim of this study was to evaluate the response of wheat to different sources and modes of Si application, as related to the severity of wheat spot blotch and plant growth, in two Si-deficient Latosols (Oxisols). An greenhouse experiment was arranged in a 2 x 5 factorial completely randomized design, with eight replications. The treatments consisted of two soils (Yellow Latosol and Red Latosol) and five Si supply modes (no Si application; Si applied as calcium silicate and monosilicic acid to the soil; and Si applied as potassium silicate or monosilicic acid to wheat leaves). No significant differences were observed between the two soils. When Si was applied to the soil, regardless the Si source, the disease incubation period, the shoot dry matter yield and the Si content in leaves were greater. Additionally, the final spot blotch severity was lower and the area under the spot blotch disease progress curve and the leaf insertion angle in the plant were smaller. Results of Si foliar application were similar to those observed in the control plants.