Predicting fish species richness in estuaries: Which modelling technique to use?

Four different modelling techniques were compared and evaluated: generalized linear models (GLM), generalized additive models (GAM), classification and regression trees (CART) and boosted regression trees (BRT). Each method was used to model fish species richness variation throughout several Portuguese estuarine systems. Model comparisons were based on goodness-of-fit and predictive performance via cross-validation. The relative influence of the most important predictors according to each of the four models was also examined. Fitted BRT, CART, GAM and GLM models accounted for 70.6%, 57.0%, 34.6% and 23.7% of total model deviance, respectively. No single variable was consistently responsible for the larger amount of percentage of relative deviance explained by the models, but several variables were selected by the four models. Nevertheless, their relative importance was highly variable, according to each modelling technique. The tree-based models (CART and BRT) presented lower prediction errors after cross-validation. The limitations and usefulness of each technique are discussed.     

A framework for coupling explanation and prediction in hydroecological modelling

Causal explanation and empirical prediction are usually addressed separately when modelling ecological systems. This potentially leads to erroneous conflation of model explanatory and predictive power, to predictive models that lack ecological interpretability, or to limited feedback between predictive modelling and theory development. These are fundamental challenges to appropriate statistical and scientific use of ecological models. To help address such challenges, we propose a novel, integrated modelling framework which couples explanatory modelling for causal understanding and input variable selection with a machine learning approach for empirical prediction. Exemplar datasets from the field of freshwater ecology are used to develop and evaluate the framework, based on 267 stream and river monitoring stations across England, UK. These data describe spatial patterns in benthic macroinvertebrate community indices that are hypothesised to be driven by meso-scale physical and chemical habitat conditions. Whilst explanatory models developed using structural equation modelling performed strongly (r² for two macroinvertebrate indices = 0.64–0.70), predictive models based on extremely randomised trees demonstrated moderate performance (r² for the same indices = 0.50–0.61). However, through coupling explanatory and predictive components, our proposed framework yields ecologically-interpretable predictive models which also maintain the parsimony and accuracy of models based on solely predictive approaches. This significantly enhances the opportunity for feedback among causal theory, empirical data and prediction within environmental modelling.     

Settlement location and population density estimation in rugged terrain using information derived from Landsat ETM and SRTM data

It is useful to have a disaggregated population database at uniform grid units in disaster situations. This study presents a method for settlement location probability and population density estimations at a 90 m resolution for northern Iraq using the Shuttle Radar Topographic Mission (SRTM) digital terrain model and Landsat Enhanced Thematic Mapper satellite imagery. A spatial model each for calculating the probability of settlement location and for estimating population density is described. A randomly selected subset of field data (equivalent to 50%) is first analysed for statistical links between settlement location probability and population density; and various biophysical features which are extracted from Landsat or SRTM data. The model is calibrated using this subset. Settlement location probability is attributed to the distance from roads and water bodies and land cover. Population density can be estimated based upon land cover and topographic features. The Landsat data are processed using a segmentation and subsequent feature–based classification approach making this method robust to seasonal variations in imagery and therefore applicable to a time series of images regardless of acquisition date. The second half of the field data is used to validate the model. Results show a reasonable estimate of population numbers (r = 0.205, p<0.001) for both rural and urban settlements. Although there is a strong overall correlation between the results of this and the LandScan model (r = 0.464, p<0.001), this method performs better than the 1 km resolution LandScan grid for settlements with fewer than 1000 people, but is less accurate for estimating population numbers in urban areas (LandScan rural r = 0.181, p<0.001; LandScan urban r = 0.303, p<0.001). The correlation between true urban population numbers is superior to that of LandScan however when the 90 m grid values are summed using a filter which corresponds to the LandScan spatial resolution (r = 0.318, p<0.001).     

Estimation of heavy-metal contamination in soil using reflectance spectroscopy and partial least-squares regression

Lead (Pb) poisoning from anthropogenic sources continues to threaten the health of urban children. Mapping Pb distribution on a large scale is imperative to identify hotspots and reduce Pb poisoning. To assess the feasibility of using reflectance spectroscopy to map soil Pb and other heavy metal abundance, the relationship between surface soil metal concentrations and hyperspectral reflectance measurements was examined via partial least-squares regression (PLSR) modelling. Soil samples were taken from four study sites. Metal concentrations were determined by inductively coupled plasma-atomic-emission spectrometry (ICP-AES) analysis, and reflectance was measured with an ASD (Analytical Spectral Devices) field spectrometer covering the spectral region of 350–2500 nm. Pb displayed an exponential decrease as a function of distance from the roadway, demonstrating the depositional patterns from leaded gas combustion which remain on the landscape 20 years after the phase-out of leaded gasoline. Calibration samples were used to derive the PLSR algorithm, and validation samples assessed the model's predictive ability. The correlation coefficients between the lab-determined abundance and the abundance predicted from PLSR calibration for all metals except copper were at or above 0.970, with the correlation coefficient for Pb the highest of all metals (0.992). Manganese, zinc and Pb had significant coefficients of determination (0.808, 0.760 and 0.746, respectively) for the validation samples. These results suggest that Pb and other heavy metal concentrations can be retrieved from spectral reflectance at high accuracy. Reflectance spectroscopy thus has potential to map the spatial distribution of Pb abundance with the aim of improving children's health in an urban environment.     

Modelling the reflectance anisotropy of Chihuahuan Desert grass–shrub transition canopy–soil complexes

The goal of the research presented here is to assess the factors controlling the remotely sensed signal returned in the solar wavelengths from Chihuahuan Desert grass–shrub transition canopy–soil complexes. The specific objectives were twofold: to evaluate the importance of the different elements (overstorey, understorey, soil) in the bidirectional reflectance distribution function (BRDF) of a Chihuahuan Desert grass–shrub transition zone; and to explore the behaviour of simple parametric and explicit scattering models with respect to observations. The first objective was approached by simulations using the Radiosity Graphics Method (RGM), with surface parameters provided by measurements of plant locations and dimensions obtained over two contrasting 25?m² plots. The second was approached through simulations of bidirectional reflectance factors (BRFs) by both the RGM and a Simplified Geometric Model (SGM) developed for inversion purposes. The modelled BRFs were assessed against multi-angle observations (MAO) – samples of the BRDF at a wavelength of 650?nm acquired from the air at up to six view zenith angles and three solar zenith angles. The results show that the understorey of small forbs and sub-shrubs plays an important role in determining the brightness and reflectance anisotropy of grass–shrub transition landscapes in relation to that of larger shrubs such as mesquite and ephedra. This is owing to the potentially high density of these plants and to the fact that there is also a varying proportion of black grama grass and prone grass litter associated with snakeweed abundance. Both of these components darken the scene. The SGM performed well measured against both the RGM and the MAO at the MAO acquisition angles (R² of 0.98 and 0.92, respectively) and good correlations were obtained between RGM and SGM when modelling was performed at a wider range of angular configurations (R²≈0.90). The SGM was shown to be highly sensitive to its adjustable parameters. Both models underestimated BRF magnitude with respect to the MAO by a small amount (<6%), showing increasing divergence from the backscattering into the forward-scattering direction. A remaining problem for operational model inversions using MAO is the a priori estimation of understorey and grass abundance.     

Modelling yearly flooding extent of the Waza-Logone floodplain in northern Cameroon based on MODIS and rainfall data

The Sahelian floodplains are of high ecological and economical importance, providing water and fresh pasture in the dry season. A spatial model is presented to predict the yearly flooding extent of the Waza-Logone floodplain based on cumulative runoff in the catchment area and estimations of the soil moisture prior to the flooding. Observations of flooding extent were based on thresholding 16-day composite Moderate Resolution Imaging Spectroradiometer (MODIS) shortwave infrared (SWIR) images. The Soil Conservation Service Curve Number (SCS-CN) method was used to calculate cumulative runoff within the Logone catchment area based on rainfall estimates (RFEs) for Africa. MODIS SWIR images acquired prior to the flooding were used as indicators for soil moisture. The mean observed flooding extent of the Waza-Logone floodplain during the period 2000–2005 was 6747 km² with a standard deviation of 1838 km². Multiple regression analysis was performed to create a predictive model forecasting flooding extent 1.5 months in advance with a coefficient of determination (R ²) equal to 0.957. Multiple regression modelling was also performed for three subregions separately. The 1.5-month forecast model for the Waza subregion resulted in the highest accuracy (R ²?=?0.950). A floodwater distribution map was created for this subregion model, allowing determination where the flooding occurs for an estimated flood size. The average additional error caused by the mapping procedure was 138 km², which is relatively small compared to an average flooded area of 3211 km² for the Waza subregion. As the flooding extent in the Waza-Logone floodplain is highly correlated to the amount of natural resources available in the dry season, the model may be a valuable tool for sustainable management of these resources.     

Habitat prediction and knowledge extraction for spawning European grayling (Thymallus thymallus L.) using a broad range of species distribution models

This study aims to apply seven data-driven methods (i.e. artificial neural networks [ANNs], classification and regression trees [CARTs], fuzzy habitat suitability models [FHSMs], generalized additive models [GAMs], generalized linear models [GLMs], random forests [RF] and support vector machines [SVMs]) to develop data-driven species distribution models (SDMs) for spawning European grayling (Thymallus thymallus), and to compare the predictive performance and the ecological relevance, quantified by the habitat information retrieved from these SDMs (i.e. variable importance and habitat suitability curves [HSCs]). The results suggest RF to yield the most accurate SDM, followed by SVM, CART, ANN, GAM, FHSM and GLM. However, inconsistencies between different performance measures were observed, indicating that different models may obtain a high score on a particular aspect and perform worse on other aspects. Despite their lower predictive ability, GAM, GLM and FHSM proved to be useful, since HSCs could be obtained and thus these techniques allow testing of ecological relevance and habitat suitability. Water depth and flow velocity appeared to be important variables for spawning grayling. The HSCs clearly indicate higher habitat suitability at a lower water depth, a low to medium flow velocity and a higher percentage of medium-sized gravel, whereas the models disagreed on the habitat suitability for the percentage of small-sized gravel. These findings demonstrate the applicability of data-driven SDMs for both habitat prediction and ecological knowledge extraction that are useful for management of a target species.     

Mapping population density distribution at multiple scales in Zhejiang Province using Landsat Thematic Mapper and census data

Population density is usually calculated from the census data, but it is dynamic over time and updating population data is often challenging because it is time-consuming and costly. Another problem is that population data for public use are often too coarse, such as at the county scale in China. Previous research on population estimation mainly focused on megacities due to their importance in socio-economic conditions, but has not paid much attention to the township or village scale because of the sparse population density and less importance in economic conditions. In reality, population density in townships and villages plays an important role in land-use/cover change and environmental conditions. It is an urgent task to timely update population density at the township and cell-size scales. Therefore, this article aims to develop an approach to estimate population density at the township scale and at a cell size of 1 km by 1 km through downscaling the population density from county to township and then to cell size. We estimated population density using Landsat Thematic Mapper (TM) and census data in Zhejiang Province, China. Landsat TM images in 2010 were used to map impervious surface area (ISA) distribution using a hybrid approach, in which a decision tree classifier was used to extract ISA data and cluster analysis was used to further modify the ISA results. A population density estimation model was developed at the county scale, and this model was then transferred to the township scale. The population density was finally redistributed to cell-size scale based on the assumption that population only occupied the sites having ISA. This research indicates that most townships have residuals within ±50 persons/km² with a root mean squared error (RMSE) of 71.56 persons/km², and a relative RMSE of 27.6%. The spatial patterns of population density distribution at the 1 km² cell size are much improved compared to the township and county scales. This research indicates the importance of using the ISA for population density estimation, where ISA can be accurately extracted from remotely sensed data.     

Quantitative analysis of salt-affected soil reflectance spectra: A comparison of two adaptive methods (PLSR and ANN)

In this paper the possibility of predicting salt concentrations in soils from measured reflectance spectra is studied using partial least squares regression (PLSR) and artificial neural network (ANN). Performance of these two adaptive methods has been compared in order to examine linear and non-linear relationship between soil reflectance and salt concentration.Experiment-, field- and image-scale data sets were prepared consisting of soil EC measurements (dependent variable) and their corresponding reflectance spectra (independent variables). For each data set, PLSR and ANN predictive models of soil salinity were developed based on soil reflectance data. The predictive accuracies of PLSR and ANN models were assessed against independent validation data sets not included in the calibration or training phase.The results of PLSR analyses suggest that an accurate to good prediction of EC can be made based on models developed from experiment-scale data (R² > 0.81 and RPD (ratio of prediction to deviation) > 2.1) for soil samples salinized by bischofite and epsomite minerals. For field-scale data sets, the PLSR predictive models provided approximate quantitative EC estimations (R² = 0.8 and RPD = 2.2) for grids 1 and 6 and poor estimations for grids 2, 3, 4 and 5. The salinity predictions from image-scale data sets by PLSR models were very reliable to good (R² between 0.86 and 0.94 and RPD values between 2.6 and 4.1) except for sub-image 2 (R² = 0.61 and RPD = 1.2).The ANN models from experiment-scale data set revealed similar network performances for training, validation and test data sets indicating a good network generalization for samples salinized by bischofite and epsomite minerals. The RPD and the R² between reference measurements and ANN outputs of theses models suggest an accurate to good prediction of soil salinity (R² > 0.92 and RPD > 2.3). For the field-scale data set, prediction accuracy is relatively poor (0.69 > R² > 0.42). The ANN predictive models estimating soil salinity from image-scale data sets indicate a good prediction (R² > 0.86 and RPD > 2.5) except for sub-image 2 (R² = 0.6 and RPD = 1.2).The results of this study show that both methods have a great potential for estimating and mapping soil salinity. Performance indexes from both methods suggest large similarity between the two approaches with PLSR advantages. This indicates that the relation between soil salinity and soil reflectance can be approximated by a linear function.     

Mapping of rice-cropping pattern and cultural type using remote-sensing and ancillary data: a case study for South and Southeast Asian countries

A geospatial database on the spatial distribution of rice areas and rice cultural types of major rice-producing countries of South and Southeast Asia has been developed in this study using remote-sensing and ancillary data sets. Multitemporal SPOT VGT normalized difference vegetation index (NDVI) data for the period 2009–2010 were used for the analysis. The classification was performed adopting ISODATA clustering to build a non-agricultural area mask followed by rice area mapping. The derived rice area was stratified by logical modelling of ancillary data sets into five rice cultural types: irrigated wet, upland, flood-prone, drought-prone, and deep-water. The uniqueness of this study is a synergistic approach based solely on single-source, high-temporal remote-sensing data coupled with ancillary data, which demonstrate the application of SPOT VGT NDVI data in building a geospatial database for rice crops over a wide spatial extent. This approach was adopted for cost effectivity as the study extent was vast and thus lacking ground truth information. Comparison of the derived rice area against the reported literature values for validation yielded a good correlation (linear coefficient of determination, R² = 0.95–0.99). The high-temporal resolution NDVI data enabled effective characterization of vegetation phenology. The derived spatial outputs can be used in various studies associated with the assessment of greenhouse gas emissions from paddy fields, change detection, and inputs to crop simulation models, which are significantly related to different rice cultural types.     

Integrating AVHRR satellite data and NOAA ground observations to predict surface air temperature: a statistical approach

Ground station temperature data are not commonly used simultaneously with the Advanced Very High Resolution Radiometer (AVHRR) to model and predict air temperature or land surface temperature. Technology was developed to acquire near-synchronous datasets over a 1?000?000?km² region with the goal of improving the measurement of air temperature at the surface. This study compares several statistical approaches that combine a simple AVHRR split window algorithm with ground meterological station observations in the prediction of air temperature. Three spatially dependent (kriging) models were examined, along with their non-spatial counterparts (multiple linear regressions). Cross-validation showed that the kriging models predicted temperature better (an average of 0.9°C error) than the multiple regression models (an average of 1.4°C error). The three different kriging strategies performed similarly when compared to each other. Errors from kriging models were unbiased while regression models tended to give biased predicted values. Modest improvements seen after combining the data sources suggest that, in addition to air temperature modelling, the approach may be useful in land surface temperature modelling.     

Mapping urban areas on a global scale: which of the eight maps now available is more accurate?

Eight groups from government and academia have created 10 global maps that offer a ca 2000 portrait of land in urban use. Our initial investigation found that their estimates of the total amount of urban land differ by as much as an order of magnitude (0.27–3.52 ×10⁶ km²). Since it is not possible for these heterogeneous maps to all represent urban areas accurately, we undertake the first global accuracy assessment of these maps using a two-tiered approach that draws on a stratified random sample of 10 000 high-resolution Google Earth validation sites and 140 medium-resolution Landsat-based city maps. Employing a wide range of accuracy measures at different spatial scales, we conclude that the new MODIS 500 m resolution global urban map has the highest accuracy, followed by a thresholded version of the Global Impervious Surface Area map based on the Night-time Lights and LandScan datasets.     

Modelling the impact of land-cover change on potential soil loss in the Taita Hills,Kenya, between 1987 and 2003 using remote-sensing and geospatial data

In sub-Saharan Africa, natural vegetation is being transformed into agricultural lands at a fast rate, endangering ecosystem services and increasing soil-loss potential, which may trigger land degradation. For the Taita Hills study area in Kenya, multi-temporal land-cover models of 1987, 1999 and 2003, derived from Satellite Pour l'Observation de la Terre (SPOT) imagery using a multi-scale segmentation/object relationship modelling (MSS/ORM) methodology and a rainfall layer, a digital elevation model (DEM) and a digital soil map were applied to model potential soil loss. Population growth in the area has led to a shortage of agricultural land and movement of people to the lowlands, evidenced by a 39% (9.3 km²) increase in croplands from 30% to 41% of the study area during the research time frame. Expansion took place mostly in surrounding foothills and lowlands, at the expense of natural shrubland and grassland, but also occurred in the hills. Universal soil-loss equation (USLE) model results showed a 60% (4 km²) increase in the area of very high potential soil loss, from 7% of the study area in 1987 to 12% in 2003, due mainly to very high soil-loss potential in croplands. Whilst the area of croplands as a whole increased, the relative proportion of very high soil-loss potential in croplands remained 20%, both in 1987 and in 2003, indicating that newly cleared agricultural lands with vulnerable soils are the most at-risk areas.     

Two-level and hybrid ensembles of decision trees for high performance concrete compressive strength prediction

Accurate prediction of high performance concrete (HPC) compressive strength is very important issue. In the last decade, a variety of modeling approaches have been developed and applied to predict HPC compressive strength from a wide range of variables, with varying success. The selection, application and comparison of decent modeling methods remain therefore a crucial task, subject to ongoing researches and debates. This study proposes three different ensemble approaches: (i) single ensembles of decision trees (DT) (ii) two-level ensemble approach which employs same ensemble learning method twice in building ensemble models (iii) hybrid ensemble approach which is an integration of attribute-base ensemble method (random sub-spaces RS) and instance-base ensemble methods (bagging Bag, stochastic gradient boosting GB). A decision tree is used as the base learner of ensembles and its results are benchmarked to proposed ensemble models. The obtained results show that the proposed ensemble models could noticeably advance the prediction accuracy of the single DT model and for determining average determination of correlation, the best models for HPC compressive strength forecasting are GB–RS DT, RS–GB DT and GB–GB DT among the eleven proposed predictive models, respectively. The obtained results show that the proposed ensemble models could noticeably advance the prediction accuracy of the single DT model and for determining determination of correlation (R²_max), the best models for HPC compressive strength forecasting are GB–RS DT (R²=0.9520), GB–GB DT (R²=0.9456) and Bag–Bag DT (R²=0.9368) among the eleven proposed predictive models, respectively.     

Simulating slum growth in Lagos: An integration of rule based and empirical based model

Demographic forecasts put Lagos as one of the cities with the highest population growth. Past trends show correlations between urban growth and slum growth, thereby creating a major challenge for sustainable city planning. This study explores the drivers of slum development in Lagos, and simulates scenarios for slum growth through coupling logistic regression with the cellular automata-based SLEUTH model. RapidEye (2009 and 2015) and Sentinel-2 (2015) imagery were used to create slum extents maps for each time point, and then used for the calibration and prediction, respectively, of the model. The driving forces of slum development in Lagos were analyzed, and the correlated spatial drivers compiled to create a probability map of slum development using the logistic regression model. The probability map was incorporated with the exclusion layer of the modified SLEUTH to simulate scenarios of slum growth in Lagos by 2035. Three scenarios were designed based on the modification of the exclusion layer and the transition rules. The Scenario 1 ‘business as usual’, depicts slum development following the present trend; the scenario 2 ‘excessive growth’, considers the demographic projection for the city; while the scenario 3 ‘limited government influence’, asserts limited interference by the government in slum management/control. Factors including distance to markets, distance to shoreline, distance to local government administrative buildings, land prices, etc. were predictors of slum development in Lagos. The prediction model, based on the logistic regression, reached an overall accuracy of 79.17% and a relative operation characteristics value of 0.85. The three scenarios show further densification of the existing slums, and increase in their area by 1.18 km² (scenario 1), 4.02 km² (scenario 2), and 1.28 km² (scenario 3). New slums are predicted at the fringe of the south-eastern part of the city. The foreseen limited spatial growth of the slums is due to the high density of the city, thus new slums may likely develop in the neighboring zones to Lagos when land in the city is no longer available.     

Remote sensing of woody shrub cover in desert grasslands using MISR with a geometric-optical canopy reflectance model

A new method is described for the retrieval of fractional cover of large woody plants (shrubs) at the landscape scale using moderate resolution multi-angle remote sensing data from the Multiangle Imaging SpectroRadiometer (MISR) and a hybrid geometric-optical (GO) canopy reflectance model. Remote sensing from space is the only feasible method for regularly mapping woody shrub cover over large areas, an important application because extensive woody shrub encroachment into former grasslands has been seen in arid and semi-arid grasslands around the world during the last 150 years. The major difficulty in applying GO models in desert grasslands is the spatially dynamic nature of the combined soil and understory background reflectance: the background is important and cannot be modeled as either a Lambertian scatterer or by using a fixed bidirectional reflectance distribution function (BRDF). Candidate predictors of the background BRDF at the Sun-target-MISR angular sampling configurations included the volume scattering kernel weight from a Li-Ross BRDF model; diffuse brightness (ρ0) from the Modified Rahman-Pinty-Verstraete (MRPV) BRDF model; other Li-Ross kernel weights (isotropic, geometric); and MISR near-nadir bidirectional reflectance factors (BRFs) in the blue, green, and near infra-red bands. The best method was multiple regression on the weights of a kernel-driven model and MISR nadir camera blue, green, and near infra-red bidirectional reflectance factors. The results of forward modeling BRFs for a 5.25 km² area in the USDA, ARS Jornada Experimental Range using the Simple Geometric Model (SGM) with this background showed good agreement with the MISR data in both shape and magnitude, with only minor spatial discrepancies. The simulations were shown to be accurate in terms of both absolute value and reflectance anisotropy over all 9 MISR views and for a wide range of canopy configurations (r² = 0.78, RMSE = 0.013, N = 3969). Inversion of the SGM allowed estimation of fractional shrub cover with a root mean square error (RMSE) of 0.03 but a relatively weak correlation (r² = 0.19) with the reference data (shrub cover estimated from high resolution IKONOS panchromatic imagery). The map of retrieved fractional shrub cover was an approximate spatial match to the reference map. Deviations reflect the first-order approximation of the understory BRDF in the MISR viewing plane; errors in the shrub statistics; and the 12 month lag between the two data sets.     

Enhanced predictive modelling process of broadband services adoption based on time series data

In this paper, the importance of the predictive modelling process of broadband services adoption is described. A detailed overview of different analytical models used for prediction, i.e., fitting and forecasting processes of broadband services adoption are presented. Furthermore, a comparison of several analytical models commonly used for prediction of broadband adoption is conducted. In order to more accurately fit to the existing broadband adoption time series data, and to forecast the future broadband services adoption paths, the features of the most accurate common predictive models have been identified for different phases of broadband services adoption. Considering the given results, usage of additional models in the predictive modelling process is analyzed. The objective of these analyses is set to improve the accuracy of the existing predictive modelling process. The accuracy of the predictive modelling process using additional models is tested and compared in different phases of broadband adoption. The model which gives the most accurate results is identified. Finally, in order to enable the usage of this model within a whole broadband service life cycle, as well as to include a greater number of explanatory parameters in predictive modelling process, an enhanced predictive modelling process is proposed.     

Technical Note: Mapping and forecasting of North Indian winter fog: an application of spatial technologies

In India, the Indo‐Gangetic plain (part of Northern India) is invariably affected by dense fog in the winter months every year due to typical meteorological, environmental and prevailing terrain conditions. Pollution also plays an important role in the formation of fog (smoke+fog = smog) in India. Using National Oceanic and Space Administration‐advanced very high resolution radiometer data the fog‐affected regions in Northern India were delineated and the spatial extent of fog for the winter months of the years 2002–03, 2003–04 and 2004–05 (December–February) were studied and mapped. Forecast for future fog based on the analysis of satellite and meteorological (air temperature, relative humidity and wind speed) data was also done. The fog‐affected areas were classified into maximum‐fog‐affected area, moderately fog‐affected area and least fog‐affected area. It has been found that in the winter months of the years 2002–03, 2003–04 and 2004–05, the fog‐affected area in Northern India was about 867 000 km², 625 000 km² and 706 800 km² respectively. The maximum fog‐affected area was found to be 606 400 km², the moderately fog‐affected area was found to be 230 400 km² and the least fog‐affected area was found to be 404 500 km². Further, based on meteorological parameters, such as temperature, humidity and wind speed along with elevation data was used to derive an approach for future fog prediction in this region.     

A feasible method for fractional snow cover mapping in boreal zone based on a reflectance model

A feasible method for mapping the fraction of Snow Covered Area (SCA) in the boreal zone is presented. The method (SCAmod) is based on a semi-empirical model, where three reflectance contributors (wet snow, snow-free ground and forest canopy), interconnected by an effective canopy transmissivity and SCA, constitute the observed reflectance from the target area. Given the reflectance observation, SCA is solved from the model. The predetermined values for the reflectance contributors can be adjusted to an optional wavelength region, which makes SCAmod adaptable to various optical sensors. The effective forest canopy transmissivity specifies the effect of forests on the local reflectance observation; it is estimated using Earth observation data similar to that employed in the actual SCA estimation. This approach enables operational snow mapping for extensive areas, as auxiliary forest data are not needed.Our study area covers 404 000 km², comprising all drainage basins of Finland (with an average area of 60 km²) and some transboundary drainage basins common with Russia, Norway and Sweden. Applying SCAmod to NOAA/AVHRR cloud-free data acquired during melting periods 2001-2003, we estimated the areal fraction of snow cover for all the 5845 basins. The validation against in situ SCA from the Finnish snow course network indicates that with SCAmod, 15% (absolute SCA-units) accuracy for SCA is gained. Good results were also obtained from the validation against snow cover information provided by the Finnish weather station network, for example, 94% of snow-free and fully snow-covered basins were recognized. A general formula for deriving the statistical accuracy of SCA estimates provided by SCAmod is presented, complemented by the results when the AVHRR data are employed.Snow melting in Finland has been operatively monitored with SCAmod in Finnish Environment Institute (SYKE) since year 2001. The SCA estimates have been assimilated to the Finnish national hydrological modelling and forecasting system since 2003, showing a substantial improvement in forecasts.     

Benchmarking classifiers to optimally integrate terrain analysis and multispectral remote sensing in automatic rock glacier detection

The performance improvements that can be achieved by classifier selection and by integrating terrain attributes into land cover classification are investigated in the context of rock glacier detection. While exposed glacier ice can easily be mapped from multispectral remote-sensing data, the detection of rock glaciers and debris-covered glaciers is a challenge for multispectral remote sensing. Motivated by the successful use of digital terrain analysis in rock glacier distribution models, the predictive performance of a combination of terrain attributes derived from SRTM (Shuttle Radar Topography Mission) digital elevation models and Landsat ETM+ data for detecting rock glaciers in the San Juan Mountains, Colorado, USA, is assessed. Eleven statistical and machine-learning techniques are compared in a benchmarking exercise, including logistic regression, generalized additive models (GAM), linear discriminant techniques, the support vector machine, and bootstrap-aggregated tree-based classifiers such as random forests. Penalized linear discriminant analysis (PLDA) yields mapping results that are significantly better than all other classifiers, achieving a median false-positive rate (mFPR, estimated by cross-validation) of 8.2% at a sensitivity of 70%, i.e. when 70% of all true rock glacier points are detected. The GAM and standard linear discriminant analysis were second best (mFPR: 8.8%), followed by polyclass. For comparison, the predictive performance of the best three techniques is also evaluated using (1) only terrain attributes as predictors (mFPR: 13.1-14.5% for best three techniques), and (2), only Landsat ETM+ data (mFPR: 19.4-22.7%), yielding significantly higher mFPR estimates at a 70% sensitivity. The mFPR of the worst three classifiers was by about one-quarter higher compared to the best three classifiers, and the combination of terrain attributes and multispectral data reduced the mFPR by more than one-half compared to remote sensing only. These results highlight the importance of combining remote-sensing and terrain data for mapping rock glaciers and other debris-covered ice and choosing the optimal classifier based on unbiased error estimators. The proposed benchmarking methodology is more generally suitable for comparing the utility of remote-sensing algorithms and sensors.