Landsat time series analysis for temperate forest cover change detection in the Sierra Madre Occidental, Durango, Mexico

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

Contents lists available at ScienceDirect

Int J Appl Earth Obs Geoinformation journal homepage: www.elsevier.com/locate/jag

Landsat time series analysis for temperate forest cover change detection in the Sierra Madre Occidental, Durango, Mexico

T

Alís Novo-Fernándeza, Shannon Franksb, Christian Wehenkelc, Pablito M. López-Serranoc, ⁎ Matthieu Molinierd, Carlos A. López-Sánchezc,e, a

Department of Organisms and Systems Biology, University of Oviedo, Oviedo, Asturias, Spain NASA’s Goddard Space Flight Center, Greenbelt, United States c Instituto de Silvicultura e Industria de la Madera, Universidad Juárez del Estado de Durango, Durango, Mexico d Remote sensing team, VTT Technical Research Centre of Finland Ltd., Espoo, Finland e Department of Organisms and Systems Biology, GIS-Forest Group, University of Oviedo, 33600 Mieres, Spain b

A R T I C LE I N FO

A B S T R A C T

Keywords: Change detection Time series analysis Landsat Temperate forests Vegetation change tracker

The Sierra Madre Occidental (SMO) is the longest continuous mountain complex in Mexico and is characterised by high species diversity and a high proportion of endemism. The rate of deforestation is high in Mexico, as in other megadiverse countries, and protection of the country’s biodiversity is a top priority. Quantification of changes in vegetation cover is essential for this purpose. Temporal information is required to enable classification of vegetation cover and change processes. In this study, the disturbances that occurred in the temperate forest of the SMO in the State of Durango (Mexico) during the period 1986–2012 were quantified using Landsat Time Series Stacks (LTSS) and the Vegetation Change Tracker (VCT) algorithm. The results obtained confirmed that land cover changes were detected with high overall accuracy (97.6%). In order to analyze the forest losses corresponding to the only official data available in Mexico, we retrieved land use and vegetation mapping (USV) data from the Mexican National Institute of Statistics and Geography (INEGI). The aridity index was established and fragmentation analysis was carried out in the study area, showing that forest pests and forest fires were the principal disturbance events in the SMO of Durango, and that the climate greatly influenced the occurrence of disturbances. The LTSS-VCT analysis revealed that for the period 1986–2012, about 34% of the temperate forest cover in the SMO in Durango was lost due to different types of disturbance, representing an annual rate of loss of forest cover of 1.3% and affecting 32,840 ha of land per year. The trend analysis of USV data showed very similar changes to those indicated by the LTSS-VCT analysis in terms of loss of temperate forest. However, differences were observed in regards to the absolute values of forest cover and vegetation loss, with analysis of the USV data indicating forest losses of 28% due to disturbances and an annual disturbance rate of 1%, affecting 49,940 ha of land per year. The LTSS-VCT approach proved efficient for mapping data on forest disturbance acquired by a medium spatial resolution (Landsat) sensor in the SMO in the State of Durango, providing satisfactory results and at low cost.

1. Introduction Mexico is a megadiverse country that is home to extensive forest ecosystems. It is known as one of the 12 countries that harbour more than 10% of the total biodiversity of the planet (Sarukhán et al., 2012). Moreover, some 40% of the plant species and more than 17% of vertebrate species are endemic, leading the United Nations to consider the conservation of Mexican forests a top priority (SRNyMA-CONAFOR, 2007). Forest harvesting is an important source of income and ⁎

employment; however, forest production has decreased drastically in recent years, and the forest sector has suffered severe losses, leading to the degradation of natural forest resources (SRNyMA-CONAFOR, 2007). The deforestation rate is high (Williams-Linera and AlvarezAquino, 2010) and Mexico is one of seven countries in which the largest annual net loss of forest area occurred between 1990 and 2010 (FAOCONAFOR, 2009). As in other megadiverse countries, the reduction in natural biodiversity is affecting ecosystem services such as timber production. Timber production levels are not sufficient to meet the national consumption needs, mainly due to the direct relationship

Corresponding author at: Instituto de Silvicultura e Industria de la Madera, Universidad Juárez del Estado de Durango, Durango, Mexico. E-mail addresses: [email protected] (A. Novo-Fernández), [email protected] (S. Franks), [email protected] (C. Wehenkel), [email protected] (P.M. López-Serrano), matthieu.molinier@vtt.fi (M. Molinier), [email protected], [email protected] (C.A. López-Sánchez). https://doi.org/10.1016/j.jag.2018.06.015 Received 23 March 2018; Received in revised form 15 June 2018; Accepted 21 June 2018 0303-2434/ © 2018 Elsevier B.V. All rights reserved.

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

details needed to classify many of the changes due to both natural and anthropogenic disturbance (Townshend and Justice, 1988). Change detection methods based on annual Landsat time series (i.e. trajectory-based change detection methods) such as the Breaks For Additive Season and Trend (BFAST; Verbesselt et al., 2010), Continuous Change Detection and Classification (CCDC) (Zhu and Woodcock, 2014), Landsat-based detection of Trends in Disturbance and Recovery (LandTrendr by Kennedy et al., 2010) and the Vegetation Change Tracker (VCT) (Huang et al., 2010) make better use of the temporal depth of the Landsat archive to reconstruct forest disturbance histories with annual resolution and trends, such as forest regeneration and succession. The VCT has been used extensively to map forest disturbances in previous studies (Verbesselt et al., 2010; Masek et al., 2013). VCT is a change detection algorithm based on the spectral–temporal characteristics of land cover and forest change processes using LTSS (Zhao et al., 2015). The State of Durango (with a forest area of about 4,900,000 ha) is the most important state in Mexico as regards the extent and economic value of its forest resources. Timber resources in the State of Durango amount to about one-quarter of the national resources of Mexico (SRNyMA, 2002; Wehenkel et al., 2011). Despite the economic importance of the Durango forests (SEMARNAT, 2013), multitemporal analysis of forest cover change has not yet been conducted. The objective of the present study was therefore to generate reliable information about the changes in forest cover in the state, thus contributing to the above-cited objectives of Mexican forest policies. The overall aim of the study was to use LTSS and the VCT algorithm (Huang et al., 2010) to map the changes in forest cover in SMO in the State of Durango (Mexico) that occurred between 1986 and 2012, and to compare the results with those estimated by the INEGI’s land use and vegetation maps (INEGI, 2013), the only official data source available in Mexico to characterize forest loss. The relevance of this study lies in the need to provide quality information for the monitoring disturbance trends in areas of particular importance for biodiversity and ecosystem services, in order to promote the implementation of reforestation programmes and policies that contribute to mitigating the effects of climate change.

between forest degradation and rural poverty in this sector and to the pressure due to gradual overexploitation, which has led to deforestation and soil degradation (SEMARNAT-CONAFOR, 2001a, b; SEMARNAT, 2013). Forest fires, changes in land use, intensive grazing and clandestine felling have also led to deterioration of the forests (SARH, 1994). Deforestation and degradation of forest lead to carbon release, thus minimizing the potential of Mexican ecosystems for carbon capture and hampering conservation of the resources and the search for other forms of income for the inhabitants of forest areas (FAO-CONAFOR, 2009). Since 2000, Mexico has created diverse programs for the sustainable development of the forest sector in relation to climate change, deforestation and biodiversity, including The Climate Change Law (Ley General de Cambio Climático, 2012) and the Strategic Forest Plan for Mexico 2025 (Programa Estratégico Forestal para México 2025), which recognises the difficulty in quantifying the extent of deforestation, mainly due to the lack of reliable methods of estimating deforestation rates. The program also highlights the need to strengthen national forestry information systems, to make them reliable, transparent and helpful for decision-making in the sector (Lal, 2001). Studies of forest disturbance are therefore important and necessary in Mexico. Forest disturbance (caused by e.g. insects, fire, cultivation, extraction of resources, and anthropogenic activities associated with settlement) and post-disturbance recovery are key processes in the development of forest ecosystems. These processes influence biomass level, biogeochemical cycling, productivity and resource availability across a broad range of spatial and temporal dimensions (Peterken, 2001; Hilker et al., 2009). Understanding the temporal and spatial aspects of these processes is crucial for modelling ecosystem characteristics, as well as for detecting changes in the terrestrial carbon cycle and mapping the quality and abundance of wildlife habitats (Hirsch et al., 2004; Law et al., 2004). Ground-based measurements can provide accurate information on forest and biomass (Lopez-Serrano et al., 2016a, 2016b, 2016c, 2016d; Molinier et al., 2016; Lopez-Sanchez et al., 2017; Vargas-Larreta et al., 2017), but lack the spatial and temporal coverage capacity of remote sensing imagery. Remote sensing is well suited to monitoring land cover change, and a myriad of approaches have been developed for this purpose (Coppin et al., 2004; Hussain et al., 2013; Lu et al., 2004; SINGH, 1989; Tewkesbury et al., 2015). Temporal information is essential for classifying vegetation cover and change processes (Stibig et al., 2014; Zhu and Woodcock, 2014; White et al., 2017). Time series of medium spatial resolution optical data have been shown to be capable of characterizing environmental phenomena, describing trends as well as discrete change events (Gomez et al., 2016). This type of data has been used to map forest disturbance (Kennedy et al., 2010) and surface water bodies (Tulbure and Broich, 2013), to characterize land cover change (Zhu and Woodcock, 2014), to identify the nature of land cover changes (Olthof and Fraser, 2014), and to model and estimate ecosystem structural variables, aboveground biomass (Gomez et al., 2014; Main-Knorn et al., 2013; Deo et al., 2017a,b), forest carbon sinks (Gomez et al., 2012), forest degradation (Shimabukuro et al., 2014) and forest disturbance (Schroeder et al., 2014). Strategies have been developed to deal with irregular and sparse time series of data (Gomez et al., 2011), and annual-time series have been found to be most appropriate for extracting information from vegetated ecosystems (Gomez et al., 2016; Zhu, 2017). Many approaches to detecting change have been proposed in recent years, with the aim of improving the detection of forest disturbances from Landsat images (Hilker et al., 2009; Zhu et al., 2012; Verbesselt et al., 2012; Zhu et al., 2012; Xin et al., 2013; Reiche et al., 2015; Hamunyela et al., 2016; Hansen et al., 2016). Landsat Time Series Stacks (LTSS) provide a unique source of information for reconstructing forest disturbance history in many areas of the world. These collections can be analysed in order to document forest change, while the spatial resolution of the images from Landsat sensors provides the spatial

2. Materials and Methods 2.1. Study area The study area is part of the Sierra Madre Occidental (SMO) mountain range that runs through the State of Durango (Fig. 1), one of the main timber supplying regions in Mexico. The SMO is the longest continuous mountain complex in Mexico, extending from close to the US border to the north of Jalisco. It is of great economic and environmental value, partly because of the high diversity of species and the high proportion of endemism that it harbours. It is also an important biological corridor for boreal and tropical mountain species, as it is linked to the Colorado Plains and the Rocky Mountains to the north and connects to the south with the Trans-Mexican Volcanic Belt (GonzálezElizondo et al., 2012). Pine-oak forests harbour the highest level of floristic diversity in Mexico (Rzedowski, 1978), and this region is home to the greatest diversity of pines, oaks and strawberry trees worldwide (González-Elizondo et al., 2012). The State of Durango occupies an area of 12.3 million hectares (6.3% of the total surface area of Mexico), and the forested area covers around 9.1 million hectares totalling 74.35% of the surface area of the state (Diario Oficial de la Federación, 2017), occupying 4th position as regards the national forest area. Some 44.67% of the area is occupied by temperate forests and rainforests, covering an area of 5.4 million hectares. The total afforested area of 4.9 million hectares (40.64% of the surface area of the state corresponds to temperate-cold forest and 495,020 ha (4.03% of the surface area of the state) to tropical rainforest (SRNyMA-CONAFOR, 2007), with arid zones, hydrophilous and 231

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Fig. 1. Location of the study area.

2.2. Data

halophytic vegetation and disturbed areas also present (SARH, 1994). The forest land is managed by 13 Regional Forest Management Units (UMAFOR), defined as “land whose physical, environmental, social and economic conditions are similar for ordination purposes, sustainable forest management and conservation of resources”, and which were delimited throughout the country by The National Forest Commission (CONAFOR), in coordination with the federal entities and within the framework of the General Law of Sustainable Forest Development (Diario Oficial de la Federación, 2017). Two-thirds of the total area (latitude: 107° 11′ W–104° 20′W; longitude: 26°50′N–22°20′N) is covered by mixed and uneven-aged pine-oak forests (Pinus spp. and Quercus spp.) with canopy cover ranging from 32% to 100% (Wehenkel et al., 2011). These forests include at least 27 types of coniferous species (of which 23 are Pinus species), and 43 Quercus species. The dominant forest includes pines and evergreen oaks, usually mixed with Arbutus species and Juniperus species, amongst others (Zhao et al., 2014). These unique forests are mostly heterogeneous and have been subjected to selective harvesting for almost a century, providing a combination of ecosystem services to the local communities. Less than 0.3% of the productive forest area is afforested land (SEMARNAT-SNIARN, 2017). The heterogeneity refers to the spatial distribution of the trees (vertical and horizontal irregularities), tree volume variations and age structure. The structure results from the management history and depends on land ownership, economic and social changes that have taken place (Wehenkel et al., 2011). The climate is classified as humid temperate, with some rainfall in summer (relative humidity, 50.1%) and winter frost due to low temperatures and humid winds from the Pacific Ocean (SRNyMACONAFOR, 2007). Precipitation ranges from 443 to 1452 mm, with an annual average of 917 mm, whereas the mean annual temperature varies from 8.2 to 26.2 °C, with an annual average of 13.3 °C. The elevation above sea level varies between 363 and 3200 m (average 2264 m). This study considers eleven UMAFORs (numbers 1001–1011), corresponding to the temperate forest of the SMO in the State of Durango (Fig. 1). Some of the characteristics of the 11 UMAFORs under study are summarised in Tables 1 and 2.

Landsat Time Series Stacks (LTSS) for the period 1986–2015 were selected for study. The LTSS consisted of one image per year for those years that were free or almost free of cloud (less than 5% cover) acquired in a specific period (Huang et al., 2015). We attempted to select, as far as possible, dates between March and June, as this is when rainfall is lowest and cloud cover is therefore lowest over the State of Durango. To cover the entire study area, which comprises 11 UMAFORs that manage the temperate forest areas of the SMO, eight Landsat WRS-2 path/rows were acquired (Fig. 2). For each of these scenes, data from 26 years (1986–2015) were analysed. A total of 242 images acquired by Thematic Mapper (Landsat4 and Landsat-5 satellites) and Enhanced Thematic Mapper Plus (Landsat-7 satellite) sensors were required (Fig. 2 and Table 3) and used to assemble the LTSS. In the study area, there are no applicable images of the required quality (defined as having minimal contamination from cloud and shadow and minimal instrument or processing related errors) for the period 1987–1989 and 1991, so that data from these years were not included in the study. Table 3 shows the Landsat satellite sensors (specifications in: (NASA, 2017)) used for each WRS-2 path/row, as well as the acquisition date for all images used. The images were selected so that the cloud cover was in the range 0–3 %. The images used are freely available from the National Landsat Archive Processing System (NLAPS). The Landsat 4–5 TM and Landsat 7 ETM + product level 1 of surface reflectance (radiometrically and atmospherically corrected) was processed using the Standard Landsat Product Generation System (LPGS) via the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) algorithm available at (USGS, 2017), which uses a 6S radiative transfer model to correct for atmospheric effects on a given date (Kotchenova et al., 2006). 2.3. VCT algorithm The VCT algorithm was used to build the forest disturbance history from LTSS. This algorithm is based on the relationship between forest structure and spectral properties over time, disturbance and the posterior forest recovery process (Huang et al., 2010). The original spatial 232

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Table 1 Land use and tenancy of the UMAFORs. Source: Regional Forest Studies in the different UMAFORs (SRNyMA-CONAFOR, 2017). UMAFOR

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011

Forest Area Under Management (%)

37.1 63.0 71.7 68.9 49.0 72.5 32.2 64.0 72.9 71.8 39.2

Type of forest property (%) Cooperative

Communal

Private

Other

59.4 8.7 49.9 16.2 50.8 65.4 67.0 88.0 64.9 12.0 2.5

6.4 81.2 25.6 59.2 27.7 13.2 0.0 7.0 0.0 76.8 88.7

34.2 2.5 2.2 24.5 19.5 20.3 30.0 5.0 34.7 11.2 8.8

0.0 7.5 22.2 0.0 2.1 1.2 3.0 0.0 0.4 0.0 0.0

Roads (%)

Natural Protected Areas (%)

Urbanized land (%)

0.57 0.65 0.53 0.43 0.54 0.53 0.68 0.77 0.84 0.63 0.49

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 4.2 60.9 39.5

0.18 0.14 0.05 0.14 0.38 0.11 0.60 0.55 0.64 0.15 0.04

resolution of the Landsat imagery (30 m) was maintained and used throughout the VCT process, which consists of two steps: (i) creation of masks and normalization (radiometric) of each image; and (ii) time series analysis to detect disturbance. The original VCT algorithm developed by Huang et al. (Huang et al., 2010) is not publicly available and therefore was not used in this study. We used a version written in Interactive Data Language (IDL) (Harris Geospatial Solutions) and provided by NASA’s Goddard Space Flight Center Biospheric Sciences Laboratory (https://science.gsfc.nasa.gov/ earth/biosphere/). Other researchers have found the IDL version of the VCT comparable for their needs (Pickell et al., 2014; Mascorro et al., 2015). Hereafter, when we refer to the VCT, we are referring to the IDL version rather than the original version. The version of the VCT used is not as fully automated as the original and needs several inputs for initiation. Moreover, while both algorithms are used to detect deviations from time series trends, the IDL version does this by searching for anomalies in the tasseled-cap based Disturbance Index (DI) (Healey et al., 2005; Crist and Cicone, 1984) rather than in the Forestness Index (integrated forest z-score - IFZ) as in the original applications, albeit these indices are the inverse of each other and both use the short-wave infrared (SWIR) band for disturbance detection. The roots of the algorithms, however, are the same in that they select forest training samples in each of the Landsat images and calculate the distance in spectral space from the calculated means from each of the “pure” forest pixels. The distance from the centroids to each corresponding pixel is calculated in relation to the centroid of the forest population. This trend analysis is conducted over all the pixels and classifications are derived on this basis. Another important difference is that the IDL version of the VCT used in this study did not measure the magnitude of disturbance.

create initial masks and to normalize the image using existing forest samples. This step includes the following: (i) creation of cloud masks and cloud shadow projections; (ii) identification of forest samples (by thresholding normalized difference vegetation index (NDVI) and MODIS Vegetation Continuous Fields (VCF) product (Hansen et al., 2002) to identify “pure” forest pixels); (iii) calculation of forest indices of Disturbance Index; and (iv) creation of a water mask.

2.3.1. Image mask creation and normalization The purpose of the first step is to analyze each image individually to

2.3.3. Data set integration A mosaic of the entire study area was created from the annual

2.3.2. Time series analysis Once the above step was completed, a pixel-level time interpolation was performed on all LTSS images to produce values labelled as cloud, shadow and other anomalies. The resulting masks and indices were then used to determine the disturbance, or lack thereof, and to obtain a set of attributes that characterize such mapped disturbance, to ultimately use the spectral trajectory of each detected disturbance and be able to track the recovery process after the disturbance had occurred (Huang et al., 2010; Vogelmann et al., 2011). In this study, physical interpretation of the Disturbance Index (DI) is required for LTSS analysis in order to determine the disturbance, or lack thereof. The DI measures the probability that a given pixel belongs to the type of forest, as its value changes as the forest does (Huang et al., 2010; Healey et al., 2005; Masek et al., 2008). The parameters used to calculate DI and initiate the VCT algorithm are shown in Table 4. Although the time series for the period 1986–2015 was considered, disturbance years were generated until 2012. This is because the VCT algorithm needs data from three consecutive years in order to categorize a pixel (in this case) as disturbance. This means that data from 2013 to 2015 are necessary to calculate disturbance values in 2012, but do not have values for that time period per se.

Table 2 Land use and type of vegetation cover in the UMAFORs. Source: USV V (INEGI, 2011). Type of vegetation

Tree Vegetation Secondary Vegetation Scrubland Rainforest Pasture Agriculture Human settlements Water bodies No vegetation

UMAFOR 1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

37.95 38.62 0.62 0.00 15.55 6.94 0.08 0.13 0.11

63.06 27.88 0.24 1.88 1.91 5.00 0.03 0.00 0.00

71.70 4.78 0.00 16.76 0.88 5.87 0.00 0.00 0.00

68.89 6.54 0.00 17.10 1.72 5.74 0.00 0.00 0.00

49.00 27.47 0.08 10.13 4.85 8.39 0.09 0.00 0.00

72.55 7.76 0.00 15.17 3.15 1.27 0.00 0.07 0.03

32.17 33.97 1.62 0.00 11.88 17.66 0.22 2.47 0.00

63.99 25.25 0.00 6.14 3.05 1.40 0.16 0.00 0.00

72.95 13.64 0.99 0.28 6.11 5.60 0.25 0.18 0.00

71.84 10.86 0.00 11.25 4.87 1.11 0.00 0.06 0.00

39.34 50.43 0.00 6.63 2.19 1.39 0.01 0.00 0.00

233

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Fig. 2. Spatial distribution of Landsat WRS-2 path/rows for the study area.

Table 3 List of images with acquisition date and Landsat sensor distribution.

234

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

occurrence of disturbance in the different UMAFORs. The aridity index is a powerful climatic variable for describing the patterns of tree species distribution and diversity in the study area (Silva-Flores et al., 2014). The average value of the aridity index (UNEP, 1997) for UMAFORs was estimated from the Global Aridity Index datasets (available at the CGIAR-CSI GeoPortal, http://www.csi.cgiar.org) by using the Zonal Raster Statistics tool of SAGA® software (Conrad et al., 2015). We used quantitative landscape parameters to characterize forest disturbance. Thus, we used FRAGSTATS 4.2 (McGarigal et al., 2002) to quantify the total area (TA), mean patch area (MPA), and we used the largest patch index (LPI; the percentage of the landscape encompassed by the largest patch) of the forest disturbances patches in the study area to estimate the spatial and temporal patterns of the forest loss (deforestation) for both the persistent forest and for the occurrence of disturbances. We also quantified fragmentation by using the aggregation index (AI), which equals 0 when forest disturbances patches are maximally disaggregated into single grid cell patches disconnected from all other patches and increases to 1 as forest disturbances patches is increasingly aggregated into a single, compact patch. Finally, we calculated descriptive statistics and conducted Spearman’s Correlation Analysis (α = 0.01), using R statistical environment (R Core team, 2015).

Table 4 Sample parameters used in DI creation and VCT time series analysis. Parameter

Purpose

Value

NDVI_thresh

Minimum NDVI for obtaining forest population during Tasseled-Cap normalization Minimum treecover percentage for obtaining forest population during Tassled-Cap normalization max number of total null dates max number of consecutive null dates max number of total cloudy dates max number of consecutive cloudy dates consecutive high DI values (must be > = for disturb) consecutive low DI values (must be > = for forest)

0.6

VCF_thresh Null_thresh c_Null_thresh Clouds_thresh c_Clouds_thresh CHV_thresh CLV_thresh

40% 6 3 6 3 3 4

disturbance map resulting from each scene. This mosaic was overlaid with the UMAFORs polygons of the SMO, which is a representation of the changes in the temperate forest cover during the study period (1986–2015) for each of the UMAFORs analyzed. This geoprocessing was performed using the Zonal Raster Statistics tool of SAGA® software (Conrad et al., 2015). The results of application of the VCT algorithm are presented as several categories of classes. The first categories comprise static classes, which include the following: (i) persisting forest, i.e. pixels that included forest vegetation during the entire time series; (ii) persisting non-forest areas, i.e. pixels that never included forest cover during the entire observation period; (iii) water, i.e. pixels that during the entire time series correspond to water bodies; (iv) unclassified: pixels that were not able to be classified due to too many null values; (v) too cloudy: pixels that do not enable classification due to persistent cloud or shadow cover. The second categories are pixels that are not classified as any type of persisting cover, in which disturbance has occurred during the observation period, and the pixel label is assigned to the year in which the disturbance occurred. Table 4 specifies how each of these classifications are determined. For example, if a pixel that has displayed “forest” attributes (CLV_thresh) reaches a value above a certain DI threshold (not given in table, but an arbitrary value determined by an analyst familiar with the region of interest) for a number of observations (CHV_thresh), it will be flagged as a disturbance and the pixel will be classified with the year of that disturbance. For details of the VCT algorithm, see Huang et al. (2010).

2.4. Analysis of other sources of vegetation cover geospatial databases In order to analyze the forest loss, the extent of disturbance estimated using the VCT algorithm was compared with the INEGI’s land use and vegetation maps (INEGI, 2013), the only official data source available in Mexico. The INEGI series I–V provides users with an excellent overview of the main types of vegetation in Mexico at 5 to 10 year intervals. While the INEGI series provide valid representations of the vegetation and land use in Mexico, they show serious limitations due to frequent switches in production methods and classification methods. In addition, the data used (from aerial photos to various satellite imagery sources) are not suitable for modern planning, decision making or for reporting greenhouse gas emissions. Moreover, the INEGI series have a minimum mapping unit (MMU) of 25 ha for land use and of 50 ha for vegetation (Gebhardt et al., 2015). This vectorial cartography was developed in five editions (Table 5) elaborated in different stages for the period 1986–2012 at a scale of 1:250,000 and with about 70 classes (INEGI, 1989, 1996, 2005, 2010, 2015). The cartography was obtained by comparison of visual examination of satellite images of the study area with previous information and confirmed with additional information obtained from field visits and laboratory analysis of botanical samples from the verification points (INEGI, 2014). For comparison of the databases, we considered vegetation cover from the INEGI series corresponding to temperate forest (pine forest, mixed-conifer forest, pine-oak forest, oak forest and temperate mesophytic forest, as well as communities of primary and secondary chaparral and montane meadow vegetation) and we also compared the forest area obtained with the VCT algorithm for Persisting Forest classification pixels and all of the forest cover with disturbance occurring after the year of study. The results obtained with the VCT algorithm were binned into the five periods established by INEGI series (Table 5):

2.3.4. Model assessment and analysis Because the current version of the VCT algorithm is unable to identify specific types of disturbance, we had to develop validation sites in order to detect changes. A stratified random sampling strategy was used to select validation sites (Olofsson et al., 2013) for assessing the accuracy of the VCT algorithm (Goodchild, 1994). A total of 1000 random point samples (with at least 20 samples per year in the areas where change occurred) were manually interpreted by the same trained person using the LTSS images; where necessary and availability allowed, the high-resolution imagery available from Google Earth™ was also checked (Hermosilla et al., 2015). The validation was aimed at major disturbance patches consisting of at least 3 × 3 continuous pixels. This was because patches smaller than 3 × 3 continuous pixels are difficult to process by visual interpretation, partly due to residual misregistration errors and to known impacts of the point spread function of the sensor (Huang et al., 2002). These patches were flagged as either no change or change, and in the case of change, the year of occurrence was also recorded. We estimated the uncertainty associated with the spatial detection of changes through estimated error matrix in terms of the unbiased estimator of the proportions of no change or change cover area of the error matrix, using the equations proposed by Olofsson et al. (2013). We used the aridity index (the ratio of mean annual precipitation and mean annual potential evapo-transpiration) to compare the

Table 5 Periods established by INEGI cartography.

235

Series

Date

References

USV USV USV USV USV

1986 1993 2002 2007 2011

(INEGI, (INEGI, (INEGI, (INEGI, (INEGI,

I II III IV V

1989) 1996) 2005) 2010) 2015)

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

than commission error (8.6%), whereas the no change category was more affected by commission (1.4%) than omission errors (1.3%).

Table 6 Estimated error matrix based of sample counts from the accuracy assessment sample. Classes

Reference

Change No change Total

Change 457 7 464

No change 43 493 536

Total

Map area (km2)

Proportion area

500 500 1,000

8,866.79 56598.74 65465.73

0.1354 0.8646 1.0000

3.1.2. Change characterization The product of the VCT algorithm is an annual disturbance map of the temperate forest of the SMO in the State of Durango. The map includes all of the land cover during the observation period (1986–2012). Fig. 3 shows a part of the map. Table 8 shows the results of the temperate forest area of each UMAFOR using LTSS-derived detection of disturbance for the period 1986–2012. The VCT analysis indicated that the total forest area was 2,593,890 ha in 1986 and that this had decreased to 1,707,211 ha by 2012. Thus, a total of 886,679 ha of temperate forest was affected by disturbance in the 11 UMAFORs in the SMO in Durango state. This represents loss of forest area at a rate of 34.2% during the period 1986–2012, which equates to an annual rate of 1.3%, or loss of 32,840 ha per year. The results for each UMAFOR (Table 8) show that the lowest absolute rate of forest loss (1099 ha/year) occurred in UMAFOR 1003, while the highest rate of forest loss (4958 ha/year) occurred in UMAFOR 1001. Considering the percentage loss, the lowest relative rate of forest loss (a total of 25.8%, 1% annually) occurred in UMAFOR 1006, while the highest rate of loss (a total of 52.6%, 2% annually) occurred in UMAFOR 1007. The percentage of pixels contaminated by cloud or shadow within each image was below 1%. Figs. 4 and 5 and S1 in Supplementary Material show the cumulative area of temperate forest disturbance per year respectively for each UMAFOR, for the UMAFORs superimposed on each other and for all UMAFORs considered as a single set. Fig. S2 in Supplementary Material shows the mosaic created from the annual disturbance map resulting from each scene and overlaid with the UMAFOR polygons. The analysis showed four periods during which loss of forest cover increased: 1986, 1992, 1997–1999, and 2010–2011.

Table 7 Error matrix of change detection process. Error matrix of change detection with cell entries based on Table 6 and expressed in terms of proportion of area. Classes

Change No change Total

Reference Change 0.1238 0.0121 0.1359

No change 0.0116 0.8525 0.8614

Accuracy User´s 0.9140 0.9860

Producer´s 0.9109 0.9865

Overall 0.9762

3. Results 3.1. VCT algorithm 3.1.1. Change detection assessment We estimated the area in which changes occurred with a margin of error (at approximate 95% confidence interval) of 8896.63 ± 635.62 km2 (standard error) in the temperate forest of the SMO in the State of Durango. The confusion matrix of the accuracy assessment is listed in Table 6, which includes user and producer accuracy for each class. The accuracy of identification of spatial detection changes was 97.62% (Table 7). Changes were detected with a higher omission error (8.9%)

Fig. 3. Forest disturbance patterns mapped by the VCT algorithm during the period 1986–2012 for the temperate forest of the Sierra Madre Occidental of Durango State. The legend details the map classification system. The first four map categories are static classes which are consistent throughout the time series. Forest change pixels are classified according to the year in which the disturbance occurred. 236

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Table 8 Forest disturbance detected using the VCT algorithm for the forest area in each UMAFOR. UMAFOR

Forest area in 1986 (ha)

Forest area in 2012 (ha)

Forest area loss (ha)

Forest area loss (%)

Annual forest area loss (%)

Forest area loss (ha/year)

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 TOTAL

345,136 295,156 89,956 193,077 299,906 339,314 120,458 332,784 107,746 239,451 230,905 2,593,890

211,274 185,562 60,295 128,178 191,580 251,930 57,133 238,523 74,057 170,698 137,980 1,707,211

133,862 109,594 29,661 64,899 108,326 87,384 63,325 94,261 33,689 68,754 92,925 886,679

38.79 37.13 32.97 33.61 36.12 25.75 52.57 28.33 31.27 28.71 40.24 34.18

1.44 1.38 1.22 1.24 1.34 0.95 1.95 1.05 1.16 1.06 1.49 1.27

4958 4,059 1099 2,404 4,012 3,236 2,345 3,491 1,248 2,546 3,442 32,840

Fig. 4. Disturbance statistics per year, with the UMAFORs superimposed.

highest occurrence of disturbance correspond to high values of fragmentation indices.

Moreover, analysis of the aridity index revealed that loss of forest area (%) was strongly affected by the degree of aridity (Spearman’s coefficient rs = −0.82) in the UMAFORs (Fig. 6). Thus, the aridity analysis indicated that the UMAFORs characterized by the highest degree of aridity (UMAFORs 1001, 1007 and 1011) correspond to those in which forest loss was highest, and vice versa (i.e. UMAFORs 1006 and 1008). Fig. 7 shows the aridity map overlaid with the UMAFOR polygons. Finally, we analyzed the spatial and temporal patterns (mean patch area, large patch index and aggregation index) of the persisting forest and deforestation per year and UMAFOR due to disturbance mapped by the VCT algorithm during the period 1986–2012 for the temperate forest of the SMO of Durango State. Fig. S3 in Supplementary Material shows the ecological stability of the persisting forest in each UMAFOR and its corresponding susceptibility according to the degree of spatial attributes of fragmentation. We found that the UMAFORs in which forest loss was highest (UMAFORs 1001, 1007 and 1011) present low values of mean patch area, large patch index and aggregation index, contrary to those in which forest loss was lowest (UMAFORs 1006 and 1008), yielding a Spearman’s coefficient rs = -0.95 for mean patch area, rs = −0.77 for large patch index and rs =−0.91 for aggregation index. High values of these fragmentation indices therefore indicate more stable forest masses that are less susceptible to disturbance. Figs. S4, S5 and S6 in Supplementary Material show the spatial attributes of fragmentation corresponding to the occurrence of disturbance per UMAFOR and year. In the same way as in the previous figure, forest loss is correlated with these attributes, in this case, rs = 0.61 for mean patch area, rs = 0.70 for large patch index and rs = 0.50 for aggregation index, indicating that the years with the

3.2. Analysis of USV and INEGI geospatial databases Table 9 shows the results obtained for the INEGI series considered. The results obtained by analysis of the USV databases reveal a total forest area of 4,880,818 ha in 1986 and a corresponding area of 3,482,490 ha in 2012, indicating that a total of 1,398,328 ha of temperate forest was affected by disturbance in the 11 UMAFORs in the SMO in the State of Durango. This represents a rate of forest area loss of 28.7%, equating to an annual rate of 1.02%, or 49,940 ha a year for the period 1986–2012. Regarding the UMAFORs, the absolute rate of loss of forest area was lowest in UMAFOR 1003 (334 ha/year) and highest in UMAFOR 1001 (13,520 ha/year). However, the percentage loss was lowest in UMAFOR 1006 (total loss of 6.8%, 0.24% annually) and highest in UMAFOR 1011 (total loss of 53.2%, 1.9% annually). 3.3. Comparison of forest cover with VCT results Table 10 shows the data estimated by the VCT algorithm and the INEGI USV data for rates of total forest loss during the period considered (1986–2012). As shown above, the rate of loss of forest area for the UMAFORs was estimated to 28.7%, based on the analysis of INEGI USV data, and 34.2%, based on application of the VCT algorithm. Although the mean rate of loss was not very different, some differences were observed in the data for each UMAFOR. The VCT estimates were higher than the USV-derived estimates, except for UMAFORs 1001 and 1011, for which the opposite was observed. The values obtained for forest cover also 237

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Fig. 5. Disturbance statistics for all UMAFORs considered as a single set.

V, the estimated rates of loss were very different depending on the method used, with VCT again indicating a higher rate of loss. Fig. 10 shows an example of the difference in accuracy of the zonification yielded by the two methods used in the study (USV and VCT).

4. Discussion The findings of the present study showed an annual rate of forest loss of 1.3% for the period 1986–2012, which is intermediate relative to other parts of Mexico. This includes temporary loss due to fire or felling or permanent loss due to land use change. A deforestation study for all of Mexico (Rosete-Vergés et al., 2014) determined an annual deforestation rate of 0.08% for the period 1976-2007. Previous studies of temperate forest in Mexico reported annual deforestation rates of 2–3% for 1982 and 1992 in southeastern Mexico (Lal, 2001), of 2.1% in Tingambato (Michoacán) for the period 1974–1991 (Rosete et al., 1977), of 3.9% (1993–2002) and 3.1% (2002–2007) in the Mexican part of the Mesoamerican Biological Corridor (Ramírez-Mejía et al., 2011), of 1.8% in the state of Michoacán during the period 1975–1993 (Bocco et al., 2001), and of 0.48% in the Nevado de Toluca National Park for the period 1972–2000 (Maass et al., 2006). In contrast to the most recent official rates, Rosete et al. observed that deforestation had not decreased in Mexico, but actually stabilized in the period 1993–2007. Furthermore, the surface area occupied by forest plantations and secondary vegetation had increased since 2000 (RoseteVergés et al., 2014). In the present study, we also observed such stability, except in the period 1997–1999, when high forest losses occurred due to forest pests and forest fires. We also observed an increase in the forest plantation rate from 2002, which represents stabilization of loss, which was then interrupted by the drought that began in 2010. In the present study, we observed that the UMAFORs characterized by the highest degree of aridity correspond to those in which loss of forest area was highest. Thus, we found that the aridity index is a powerful climatic variable for predicting forest disturbance in the SMO in Mexico. The harsher the climate conditions, the lower the density and dimension of trees that can maintain local conditions (Kalin-Arroyo et al., 1988; Currie et al., 2004). Therefore, for forest stands situated in areas with more arid conditions, a smaller reduction in tree density is required than for stands located in areas with more favourable climate conditions in order to pass the threshold value from the “forest” to “disturbance” class. The temporal evaluation of forest change associated with fragmentation analysis involves a valuable set of techniques for assessing

Fig. 6. Forest disturbance detected using the VCT algorithm for the forest area and aridity index in each UMAFOR. Note that aridity index values are higher for more humid conditions, and lower for more arid conditions.

differed depending on the method used. The analysis of the cumulative loss of forest cover for the USV series relative to the initial forest area (1986) (Fig. 8) reveals very similar trends for the procedures evaluated. For Series II (1993), the estimated rates of loss of forest area (relative to the initial surface area in 1986) were 6.9% (VCT) and 6.1% (USV data). For Series III (2002), the estimated rates of loss of forest area were 24.6% (VCT) and 23.7% (USV). For Series IV (2008), the estimated rates of forest loss were 27.8% (VCT) and 28.7% (USV). Finally, for Series V (2012), the estimated rates of forest loss were 34.1% (VCT) and 28.7% (USV). Although the rates are similar, for series IV, analysis of the USV data yielded a higher rate than that estimated by VCT, unlike for the other series. The differences in the estimates yielded by the two methods were more pronounced for Series V, with VCT again indicating a higher rate of loss of forest area than estimated by the USV method. Comparing the forest area loss rates of the USV and VCT algorithm series for the UMAFOR set (Fig. 9) also shows a uniform trend in the studied procedures. Thus, the rate of loss for Series I and II was 6.1% according to USV and 6.92% according to VCT. For series II and III, the estimated rates of loss were 18.8% (USV) and 19% (VCT). For series III and IV, the estimated rates of loss were 6.4% (USV) and 4.2% (VCT). Finally, for series IV and V, the estimated rates of loss rate were 0% (USV) and 8.8% (VCT). Once again, the rate of loss estimated by VCT was higher than that estimated by USV for series I and II, and for series II and III. However, for series III and IV, in addition to a greater difference, the USV indicated a higher rate of loss, while for series IV and 238

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Fig. 7. Aridity map for the temperate forest of the Sierra Madre Occidental of Durango State. Note that the more humid areas, shown in green, present higher aridity index values and the more arid areas, shown in red, present lower aridity index values (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article).

Table 9 Forest area loss determined by analysis of INEGI USV databases (1986–2012) relative to the total forest area in each UMAFOR. UMAFOR

Forest area in 1986 (ha)

Forest area in 2012 (ha)

Forest area loss (ha)

Forest area loss (%)

Annual forest area loss (%)

Forest area loss (ha/year)

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 TOTAL

806,910 447,664 134,846 356,376 619,213 460,798 387,821 501,648 192,003 418,046 555,492 4,880,817

428,339 348,869 125,481 317,992 424,753 429,468 253,619 372,948 162,075 358,845 260,101 3,482,491

378,571 98,796 9,365 38,384 194,461 31,330 134,202 128,700 29,928 59,201 295,391 1,398,328

46.92 22.07 6.94 10.77 31.40 6.80 34.60 25.66 15.59 14.16 53.18 28.65

1.68 0.79 0.25 0.38 1.12 0.24 1.24 0.92 0.56 0.51 1.90 1.02

13,520 3,528 334 1,371 6,945 1,119 4,793 4,596 1,069 2,114 10,550 49,940

structure and spatial arrangement of patches (Jose et al., 2011). The fragmentation analysis for this study was based on the spatial attributes of mean patch area, large patch index and aggregation index, revealing that high values of these indices for persisting forest indicate more

the degree of threat to ecosystems (Armenteras et al., 2003; Franklin et al., 2002). Habit fragmentation is a problem for many species and may result in the loss of regional and global biodiversity (Harris, 1984). Furthermore, at the landscape level, disturbance is related to the

Table 10 The data estimated by the VCT algorithm and the USV data for total loss of forest area. UMAFOR

1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 TOTAL

Forest area in 1986 (ha)

Forest area in 2012 (ha)

Total forest area loss rates (%)

VCT

USV

VCT

USV

VCT

USV

345,136 295,156 89,956 193,077 299,906 339,314 120,458 332,784 107,746 239,451 230,905 2,593,890

806,910 447,664 134,846 356,376 619,213 460,798 387,821 501,648 192,003 418,046 555,492 4,880,817

211,274 185,562 60,295 128,178 191,580 251,930 57,133 238,523 74,057 170,698 137,980 1,707,211

428,339 348,869 125,481 317,992 424,753 429,468 253,619 372,948 162,075 358,845 260,101 3,482,491

38.79% 37.13% 32.97% 33.61% 36.12% 25.75% 52.57% 28.33% 31.27% 28.71% 40.24% 34.18%

46.92% 22.07% 6.94% 10.77% 31.40% 6.80% 34.60% 25.66% 15.59% 14.16% 53.18% 28.65%

239

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

stable UMAFORs that are less susceptible to disturbance. Moreover, human induced disturbance is different from natural disturbance, especially in relation to the extent, severity and frequency (Jose et al., 2011). Thus, forest pests and forest fires are events characterized by affecting large areas, and we also found that the years during which disturbance was greatest correspond to the years in which large disturbance events occurred, such as forest pests and forest fires. The results reveal that forest pests and forest fires are the principal disturbance events in the SMO of Durango. Forest fires were estimated to affect about 23,330 ha/year in tropical, subtropical and temperate forests and scrublands in state of Durango during the period 1991–2012 (SEMARNAP, 1997, 1998, 1999, 1998-2000; SRNyMA-CONAFOR, 2007; CONAFOR, 2013). However, the annual area affected by this type of disturbance is very variable and is related to annual weather patterns and variations in temperature and precipitation (Delong and Tanner, 1996), which are closely related to the aridity index. The cool season precipitation over northern Mexico is significantly correlated with indices of El Niño Southern Oscillation (ENSO) (Stahle et al., 2012), and this phenomenon thus has a strong influence on the activity of forest fires in the state of Durango, with the consequent changes in land use (Pompa-García and Sensibaugh, 2014). The cold phase (i.e. La Niña) of the ENSO caused a decrease in precipitation in 1983, 1992, 1997, 2003 and 2010; likewise, the warm phase (El Niño) of the ENSO caused important increases in precipitation in 1987, 1989, 1998, 2002 and 2008 (Pompa-García and Jurado, 2014). These changes are closely correlated with periods of greater and lower loss of forest cover that we found in the present study, thus explaining the peaks in disturbance observed in 1992, 1997 and 2010–2011. Drought and fire also occurred in 2003, although to a lesser extent than in the aforementioned years. The disturbances detected in 1999 were due to a plague of bark boring insects, the presence of which was favoured by the drought that occurred in previous years (Quiñones, 2007). Insect

Fig. 8. Comparison of rates of forest area loss estimated by USV analysis and the VCT algorithm.

Fig. 9. Comparison of rate of loss of forest area estimated by the USV and VCT methods applied to the UMAFOR set.

Fig. 10. Differences in zonification yielded by the VCT and USV methods. 240

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

change in the primary vegetation. The official data regarding loss of forest cover therefore underestimate the values as they discount 100% of the area occupied by secondary vegetation, as clearly detected from series IV onwards (Rosete-Vergés et al., 2014). Regarding the rates of loss for each UMAFOR, those obtained for UMAFOR 1001 and 1011, according to the USV series, were much higher than those estimated by the VCT method. This is because these UMAFORs include a higher proportion of secondary vegetation than the others, thus leading to a large decrease in the surface area detected as forest, relative to the previous series. By contrast, UMAFORs 1003, 1004 and 1006 contain the lowest proportions of secondary vegetation and therefore the rates of loss were lower than in the other plots.

plagues were reported to affect a total area of 155,000 ha in Durango forests after extreme drought in 2011 (CONAFOR, 2012). However, the forest cover was not re-established during the more humid intervals of the study period. Therefore, higher-ranking factors adversely affected the forest cover, possibly including a mixture of drier climate (Conagua, 2013; Pompa-García and Jurado, 2014; PompaGarcía et al., 2015; Díaz-Ramírez et al.,2016) along with human activities, e.g. cattle grazing and tree logging (SRNyMA-CONAFOR, 2007). Furthermore, forecasts of future climate scenarios for Mexico predict an increase in aridity of about 26% by 2090 (Sáenz-Romero et al., 2010). Under this scenario, the diversity of tree species and forest stand density in the state of Durango will be drastically reduced because of poorer climate conditions (Silva-Flores et al., 2014) and, thus, forest loss will increase. As in many countries, the data production process in cartographic practices is not subjected to rigorous accuracy assessment in Mexico, leading to some doubts regarding the quality of the information derived from these database (such as total surfaces and rates of change) (Mas et al., 2009). Global land cover maps generated from coarse spatial resolution data are typically characterised by low local accuracy (Frey and Smith, 2007; Fritz et al., 2010), particularly in regions with heterogeneous land cover (Herold et al., 2008). Moreover, owing to disparities in classification schemes, spatial resolution, thematic detail and estimated accuracy, comparison of maps describing change over time is problematic (Bai et al., 2014; Perez-Hoyos et al., 2012) and generally discouraged (Gebhardt et al., 2014; Homer et al., 2007), as inaccuracies in thematic classes are further compounded when change is considered (Fuller et al., 2003). Although we also observed differences on comparing the results obtained using the VCT algorithm and analysis of the INEGI series (USV), the trends in loss of cover were very similar (except for the losses from Series IV and V), and the differences between the two methods were generally related to the values of forest surface area and loss of forest cover. The main reason for the differences is the different scales used: the VCT algorithm uses a pixel scale (30 x 30 m) and the INEGI series used a minimum mapping unit of 50 ha. The INEGI series-derived calculations will overestimate the surface area as the scale used is not sufficiently accurate to enable differentiation of different types of land use within land forest stands. The second reason for the differences in results obtained by the two methods is the identification of forest cover samples used to calculate DI and initiate the VCT algorithm (by thresholding NDVI and MODIS VCF). Identification of dense forest classes (pure forest pixels) results in major differences in the absolute values estimated, as the study area is characterised by a large number of forest stands and therefore a small canopy cover fraction (of between 10 and 40%). As the algorithm is sensitive to tree density, it will not identify some open forest stands. Analysis of the results obtained on comparing the rates of forest loss determined using the USV series and VCT algorithm for UMAFOR set showed very similar trends in series II and III and differences between these series, with VCT indicating higher rates of loss than USV. By contrast, the USV-derived estimates were higher for series IV and for the difference between series III and IV, although the values were generally similar. The differences are due to the fact that secondary vegetation (defined as vegetation established naturally after a dramatic disturbance where secondary species account for more than 70% of the vegetation cover (INEGI, 2014) was not identified in series I, II and III and was included as forest. However, secondary vegetation was differentiated in series IV and therefore the surface area identified as forest was significantly lower than in the previous series. For series V and the change between series IV and V, which did not consider secondary vegetation as forest loss, because this type of cover was considered a separate class, forest loss was minimal and forest cover was maintained relative to the previous series. Rosete et al. estimated that 13% of cover by secondary vegetation is the product of vegetation recovery processes, whereas 50% is maintained as such and 35% is the product of

5. Conclusions The study findings show that, according to the estimates produced by the VCT algorithm, the SMO in the State of Durango experienced a total loss of 886,679 ha of forest cover during the period 1986–2012, with an annual forest area loss rate of 1.3%, which is intermediate relative to the rates determined for other parts of Mexico. The extent of forest area disturbance, determined by both the VCT algorithm and USV databases, indicates that the losses represent about 30% of forest loss in the UMAFOR set. UMAFOR 1003 experienced the lowest rate of forest loss, and 1001 experienced the highest level of forest disturbance. Considering the percentage loss, the lowest relative rate of forest loss occurred in UMAFOR 1006, while the highest rate of loss occurred in UMAFOR 1007. Similarly, the VCT algorithm revealed four marked periods of increased loss of forest cover (in 1992, 1997, 1999, and 2010–2011), corresponding to the occurrence of fires and forest pests largely as a result of the impact of the ENSO phenomenon. Stabilization of the losses was generally observed from 2000, except in the period 2010–2011. Moreover, the aridity index analysis indicated that the UMAFORs in which the rate of forest loss was highest were those in which the level of aridity was highest. Furthermore, the fragmentation analysis revealed that high values of mean patch area, large patch index and aggregation index for persisting forest indicate more stable forest masses that are less susceptible to disturbance, and that the years in which the largest area of disturbances accumulate correspond to the years in which large disturbance events (such as forest pests and forest fires) occurred. Therefore, the results showed that forest pests and forest fires are the principal disturbance events in the SMO of Durango, and that the climate greatly influenced the incidence of disturbances. Although comparison of the forest cover estimated using VCT algorithm and the INEGI (USV) series showed similar trends in the loss of cover (except in series IV and V due to the inclusion of secondary vegetation in series IV), the different scales used by each method and the identification of forest cover samples used to calculate DI and initiate the VCT algorithm resulted in major differences in the estimated absolute values, for values of forest cover and rates of loss of cover. On the basis of the study findings, we conclude that the land disturbance determined using the LTSS-VCT approach was spatially explicit and included many more temporal details than provided by the conventional analysis of changes in INEGI (USV) series. Thus, we can conclude that the study findings indicate that any likely increase in aridity level as a result of climatic variation or change will have serious consequences for the forest cover and, thus, the ecological functions of temperate forest of the SMO in the State of Durango (Mexico). Given the importance of this study area in terms of biodiversity, long-term monitoring of changes in vegetation in relation to climate variation or change will be required, in addition to study of the real consequences of climate change. Moreover, future research should examine the potential use of land disturbance products in carbon modelling studies to quantify the carbon fluxes from the mapped disturbance events and their impacts on future carbon capture potential in the SMO in the State of Durango. 241

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

In the light of the above, the study findings may be helpful in the development of forest management strategies, including reforestation programmes and the implementation of policies that contribute to mitigating the effects of climate change on the Durango forests.

Presidio-San Pedro. Madera y bosques 22 (1), 111–123. FAO-CONAFOR, 2009. Revisión del Programa Estratégico Forestal 2025 (Pef 2025) y del Programa Nacional Forestal 2001‐2006 (Pnf 2001‐2006), en el marco del Proyecto Utf/Mex/072. Franklin, J., Simons, D.K., Beardsley, D., Rogan, J.M., Gordon, H., 2002. Evaluating errors in a digital vegetation map with forest inventory data and accuracy assessment using fuzzy sets. Transactions GIS 5 (4), 285–304. Frey, K.E., Smith, L.C., 2007. How Well Do We know Northern Land cover? Comparison of Four global vegetation and wetland products with a New Ground-truth database for West Siberia. Global Biogeochem. Cycles 21. Fritz, S., See, L., Rembold, F., 2010. Comparison of global and regional Land cover maps with statistical information for the agricultural domain in Africa. Int. J. Remote Sens. 31, 2237–2256. Fuller, R., Smith, G., Devereux, B., 2003. The characterisation and measurement of Land cover change through remote sensing: problems in operational applications? Int. J. Appl. Earth Obs. 4, 243–253. Gebhardt, S., Wehrmann, T., Ruiz, M.A.M., Maeda, P., Bishop, J., Schramm, M., Kopeinig, R., Cartus, O., Kellndorfer, J., Ressl, R., Santos, L.A., Schmidt, M., 2014. Mad-Mex: automatic wall-to-wall land cover monitoring for the Mexican Redd-Mrv program using all landsat data. Remote Sens. 6, 3923–3943. Gebhardt, S., Maeda, P., Wehrmann, T., Espinoza, J.A., Schmidt, M., 2015. A proper Land cover and Forest type classification Scheme for Mexico. Int. Arch. Photogrammetry, Remote Sens. Spatial Inf. Sci. 40, 383. Goodchild, M.F., 1994. Integrating GIS and remote sensing for vegetation analysis and modeling: methodological issues. J. Veg. Sci. 5 (5), 615–626. http://dx.doi.org/10. 2307/3235878. Gomez, C., White, J.C., Wulder, M.A., 2011. Characterizing the State and processes of change in a dynamic Forest environment using hierarchical spatio-temporal segmentation. Remote Sens. Environ. 115, 1665–1679. Gomez, C., Wulder, M.A., White, J.C., Montes, F., Delgado, J.A., 2012. Characterizing 25 years of change in the Area, distribution, and carbon stock of Mediterranean pines in Central Spain. Int. J. Remote Sens. 33, 5546–5573. Gomez, C., White, J.C., Wulder, M.A., Alejandro, P., 2014. Historical Forest biomass dynamics modelled with landsat spectral trajectories. ISPRS-J. Photogramm. Remote Sens. 93, 14–28. Gomez, C., White, J.C., Wulder, M.A., 2016. Optical remotely sensed time series data for Land cover classification: a review. ISPRS-J. Photogramm. Remote Sens. 116, 55–72. González-Elizondo, M.S., González-Elizondo, M., Tena-Flores, J., Ruacho-González, L., López-Enríquez, I.L., 2012. Vegetación de la Sierra Madre Occidental, México: Una síntesis. Acta botánica mexicana 351–403. Hamunyela, E., Verbesselt, J., de Bruin, S., Herold, M., 2016. Monitoring deforestation at sub-annual scales as extreme events in landsat data cubes. Remote Sens. 8. Hansen, M.C., Defries, R.S., Townshend, J.R.G., Sohlberg, R., Dimiceli, C., Carroll, M., 2002. Towards and operational MODIS continuous field of percent tree cover algorithm: examples using avhrr and MODIS data. Remote Sens. Environ. 83, 303–319. Hansen, M.C., Krylov, A., Tyukavina, A., Potapov, P.V., Turubanova, S., Zutta, B., Ifo, S., Margono, B., Stolle, F., Moore, R., 2016. Humid tropical Forest disturbance alerts using landsat data. Environ. Res. Lett. 11. Harris, L.D., 1984. The Fragmented Forest: Island Biogeographic Theory and the Preservation of Biotic Diversity. The University of Chicago Press 211 p. Healey, S.P., Cohen, W.B., Yang, Z.Q., Krankina, O.N., 2005. Comparison of tasseled capbased landsat data structures for use in forest disturbance detection. Remote Sens. Environ. 97, 301–310. Hermosilla, T., Wulder, M.A., White, J.C., Coops, N.C., Hobart, G.W., 2015. Regional detection, characterization, and attribution of annual forest change from 1984 to 2012 using landsat-derived time-series metrics. Remote Sens. Environ. 170, 121–132. Herold, M., Mayaux, P., Woodcock, C.E., Baccini, A., Schmullius, C., 2008. Some challenges in global Land cover mapping: an assessment of agreement and accuracy in existing 1 km datasets. Remote Sens. Environ. 112, 2538–2556. Hilker, T., Wulder, M.A., Coops, N.C., Linke, J., McDermid, G., Masek, J.G., Gao, F., White, J.C., 2009. A New data fusion model for High spatial- and temporalResolution mapping of Forest disturbance based on landsat and MODIS. Remote Sens. Environ. 113, 1613–1627. Hirsch, A.I., Little, W.S., Houghton, R.A., Scott, N.A., White, J.D., 2004. The net carbon flux due to deforestation and Forest Re-growth in the Brazilian amazon: analysis using a process-based model. Glob. Change Biol. 10, 908–924. Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A., VanDriel, J.N., Wickham, J., 2007. Completion of the 2001 national Land cover database for the conterminous United States. Photogramm. Eng. Remote Sens. 73, 337–341. Huang, C.Q., Townshend, J.R.G., Liang, S.L., Kalluri, S.N.V., DeFries, R.S., 2002. Impact of sensor’s point spread function on land cover characterization: assessment and deconvolution. Remote Sens. Environ. 80 (2), 203–212. Huang, C.Q., Coward, S.N., Masek, J.G., Thomas, N., Zhu, Z.L., Vogelmann, J.E., 2010. An automated approach for reconstructing recent Forest disturbance history using dense landsat time series stacks. Remote Sens. Environ. 114, 183–198. Huang, C.Q., Ling, P.Y., Zhu, Z.L., 2015. North Carolina’s Forest disturbance and timber production assessed using time series landsat observations. Int. J. Digital Earth 8, 947–969. Hussain, M., Chen, D.M., Cheng, A., Wei, H., Stanley, D., 2013. Change detection from remotely sensed images: from pixel-based to object-based approaches. ISPRS-J. Photogramm. Remote Sens. 80, 91–106. INEGI, 1989. Conjunto Nacional de Uso de Suelo y Vegetación a Escala 1:250,000, Serie I. DGG-INEGI, M., Ed. INEGI, 1996. Conjunto Nacional de Uso de Suelo y Vegetación a Escala 1:250,000, Serie II. DGG-INEGI, M., Ed. INEGI, 2005. Conjunto Nacional de Uso de Suelo y Vegetación a Escala 1:250,000, Serie III. DGG-INEGI, M., Ed. INEGI, 2010. Conjunto Nacional de Uso de Suelo y Vegetación a Escala 1:250,000, Serie IV. DGG-INEGI, M., Ed.

Author contributions Alís Novo-Fernández and Carlos A. López-Sánchez conceived, designed and performed the experiments and wrote the manuscript. Shannon Franks assisted with analysis of the results and revised the manuscript. Christian Wehenkel, Pablito M. López-Serrano and Matthieu Molinier revised the manuscript. Conflicts of interest The authors declare no conflict of interest. Acknowledgements This research was derived from the Workshop "Training for Landsat Time Series Analysis – Vegetation Change Tracker (VCT)" supported by Fondo Mexicano para la Conservación de la Naturaleza, A.C. US Forest Service and United States Agency for International Development (USAID). The authors acknowledge Rafael Flores (Coordinador de Programa para México, US Forest Service, International Programs) for support provided, as well as Alejandra Saly Franks for help with translating an earlier version of the manuscript into English. We are grateful to the journal editors and reviewers for their helpful comments. Appendix A. Supplementary data Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.jag.2018.06.015. References Armenteras, D., Gast, F., Villareal, H., 2003. Andean forest fragmentation and the representativeness of protected natural areas in the eastern Andes, Colombia. Biol. Conserv. 113, 245–256. Bai, Y., Feng, M., Jiang, H., Wang, J.L., Zhu, Y.Q., Liu, Y.Z., 2014. Assessing consistency of Five global Land cover data sets in China. Remote Sens. 6, 8739–8759. Bocco, G., Mendoza, M., Masera, O.R., 2001. La dinámica del cambio del uso del suelo en Michoacán: Una propuesta metodológica para el estudio de los procesos de deforestación. Investigaciones geográficas 18–36. CONAFOR, 2012. Plaga del insecto descortezador en Durango. Retrieved from:. . México. http://www.conafor.gob.mx/portal/index.php/component/content/ article/443. CONAFOR, 2013. Gerencia de Incendios Forestales, Secretaría de Medio Ambiente y Recursos Naturales. México. . CONAFOR, 2017. Estrategia Nacional de Manejo Forestal Sustentable para el incremento de la Producción y Productividad 2013-2018 (Enaipros). Comisión Nacional Forestal, Mexico. Conagua, 2013. Reporte del Clima en México. Reporte Anual 2013. México. . Conrad, O., Bechtel, B., Bock, M., Dietrich, H., Fischer, E., Gerlitz, L., Wehberg, J., Wichmann, V., Bohner, J., 2015. System for automated geoscientific analyses (Saga) V. 2.1.4. Geosci Model. Dev. 8, 1991–2007. Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., Lambin, E., 2004. Digital change detection methods in ecosystem monitoring: a review. Int. J. Remote Sens. 25, 1565–1596. Crist, E.P., Cicone, R.C.A., 1984. Physically-based transformation of thematic mapper data – the Tm tasseled cap. IEEE Trans. Geosci. Remote Sens. 22, 256–263. Currie, D.J., Mittelbach, G.G., Cornell, H.V., Field, R., Guégan, J.-F., et al., 2004. Predictions and tests of climate-based hypotheses of broad-scale variation in taxonomic richness. Ecol. Lett. 7, 1121–1134. Delong, S.C., Tanner, D., 1996. Managing the pattern of forest harvest: lessons from wildfire. Biodivers. Conserv. 5, 1191–1205. http://dx.doi.org/10.1007/BF00051571. Deo, R.K., Russell, M.B., Domke, G.M., Andersen, H.E., Cohen, W.B., Woodall, C.W., 2017a. Evaluating site-specific and generic spatial models of aboveground Forest biomass based on landsat time-series and lidar strip samples in the Eastern USA. Remote Sens. 9. Deo, R.K., Russell, M.B., Domke, G., Woodall, C.W., Falkowski, M.J., Cohen, W.B., 2017b. Using landsat time-series and lidar to inform aboveground forest biomass baselines in Northern Minnesota, USA. Can. J. Remote Sens. 43, 28–47. Diario Oficial de la Federación. México. 2017. Ley nº CD-LXIII -II -2P-177. Ley General de Desarrollo Forestal Sustentable. 24 de enero de 2017. Díaz-Ramírez, B., Villanueva-Díaz, J., Cerano-Paredes, J., 2016. Reconstrucción de la precipitación estacional con anillos de crecimiento para la región hidrológica

242

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

INEGI, 2014. Guía para la interpretación de cartografía: Uso del Suelo y Vegetación: Escala 1:250, 000: Serie V. INEGI, I.N.d.E.y.G.-.-M., Ed. INEGI, 2015. Conjunto Nacional de Uso de Suelo y Vegetación a Escala 1:250,000, Serie V. DGG-INEGI, M., Ed. Jose, S.K., Alex, C.J., Kumar, S., Varghese, A., Madhu, G., 2011. Landscape metric modeling - a technique for Forest disturbance assessment in shendurney wildlife sanctuary. Environ. Res., Eng. Manage. 4 (58), 34–41. Kalin-Arroyo, M.T., Squeo, F.A., Armesto, J.J., Villagran, C., 1988. Effects of aridity on plant diversity in the Northern Chilean Andes: results of a natural experiment. Ann. Mo. Bot. Gard. 75, 55–78. Kennedy, R.E., Yang, Z.G., Cohen, W.B., 2010. Detecting trends in Forest disturbance and recovery using yearly landsat time series: 1. Landtrendr - temporal segmentation algorithms. Remote Sens. Environ. 114, 2897–2910. Kotchenova, S.Y., Vermote, E.F., Matarrese, R., Klemm Jr, F.J., 2006. Validation of a vector version of the 6s radiative transfer code for atmospheric correction of satellite data. Part I: path radiance. Appl. Opt. 45, 6762–6774. Soil Science Society of America, 2001. Meeting (90th : 1998 : baltimore Md.). In: Lal, R. (Ed.), Soil Carbon Sequestration and the Greenhouse Effect : Proceedings of a Symposium Sponsored by Divisions S-3, S-5, and S-7 of the Soil Science Society of America at the 90th Annual Meeting in Baltimore, Md, 18-22 October 1998. Soil Science Society of America, Madison, WI p xvii, 236 p. Law, B.E., Turner, D., Campbell, J., Sun, O.J., Van Tuyl, S., Ritts, W.D., Cohen, W.B., 2004. Disturbance and climate effects on carbon stocks and fluxes across Western Oregon USA. Glob. Change Biol. 10, 1429–1444. Ley General de Cambio Climático, 2012. Diario Oficial de la Federación 06/06/2012, Secretaría de Gobernación (SEGOB). Obtenido de. http://www.dof.gob.mx/nota_ detalle.php?codigo=5249899&fecha=06/06/2012. Lopez-Sanchez, C.A., Garcia-Ramirez, P., Resl, R., Hernandez-Diaz, J.C., Lopez-Serrano, P.M., Wehenkel, C., 2017. Modelling dasometric attributes of mixed and unevenaged forests using landsat-8 oli spectral data in the Sierra Madre Occidental, Mexico. Iforest 10, 288–295. Lopez-Serrano, P.M., Corral-Rivas, J.J., Diaz-Varela, R.A., Alvarez-Gonzalez, J.G., LopezSanchez, C.A., 2016a. Evaluation of radiometric and atmospheric correction algorithms for aboveground Forest biomass estimation using landsat 5 Tm data. Remote Sens. 8. Lopez-Serrano, P.M., Lopez Sanchez, C.A., Solis-Moreno, R., Corral-Rivas, J.J., 2016b. Geospatial estimation of above Ground Forest biomass in the Sierra Madre Occidental in the State of Durango, Mexico. Forests 7. Lopez-Serrano, P.M., Lopez-Sanchez, C.A., Alvarez-Gonzalez, J.G., Garcia-Gutierrez, J., 2016c. A comparison of machine learning techniques applied to landsat-5 Tm spectral data for biomass estimation. Cana. J. Remote Sens. 42, 690–705. Lopez-Serrano, P.M., Lopez-Sanchez, C.A., Diaz-Varela, R.A., Corral-Rivas, J.J., SolisMoreno, R., Vargas-Larreta, B., Alvarez-Gonzalez, J.G., 2016d. Estimating biomass of mixed and uneven-aged forests using spectral data and a hybrid model combining regression trees and linear models. Iforest 9, 226–234. Lu, D., Mausel, P., Brondizio, E., Moran, E., 2004. Change detection techniques. Int. J. Remote Sens. 25, 2365–2407. Maass, S.F., Regil Garcia, H., Benjamín Ordoñez, J., 2006. Dinámica de perturbaciónrecuperación de las zonas forestales en el Parque Nacional Nevado de Toluca. Madera Bosques 12, 17–28. Main-Knorn, M., Cohen, W.B., Kennedy, R.E., Grodzki, W., Pflugmacher, D., Griffiths, P., Hostert, P., 2013. Monitoring coniferous Forest biomass change using a landsat trajectory-based approach. Remote Sens. Environ. 139, 277–290. Mas, J.-F., Velázquez, A., Couturier, S., 2009. La Evaluación de los Cambios de Cobertura/ Uso del Suelo en la República Mexicana. Investigación ambiental Ciencia y política pública, 1. Mascorro, V.S., Coops, N.C., Kurz, W.A., Olguín, M., 2015. Choice of satellite imagery and attribution of changes to disturbance type strongly affects Forest carbon balance estimates. Carbon Balance Manage. 10, 30. Masek, J.G., Huang, C.Q., Wolfe, R., Cohen, W., Hall, F., Kutler, J., Nelson, P., 2008. North American Forest disturbance mapped from a decadal landsat record. Remote Sens. Environ. 112, 2914–2926. Masek, J.G., Goward, S.N., Kennedy, R.E., Cohen, W.B., Moisen, G.G., Schleeweis, K., Huang, C.Q., 2013. United States Forest disturbance trends observed using landsat time series. Ecosystems 16, 1087–1104. McGarigal, K., Cushman, S.A., Neel, M.C., Ene, E., 2002. FRAGSTATS v3: Spatial Pattern Analysis Program for Categorical Maps. Computer Software Program Produced by the Authors at the University of Massachusett, Amherst. http://www.umass.edu/ landeco/research/fragstats/fragstats.html. Molinier, M., Lopez-Sanchez, C.A., Toivanen, T., Korpela, I., Corral-Rivas, J.J., Tergujeff, R., Hame, T., 2016. Relasphone-Mobile and participative in situ Forest biomass measurements supporting satellite image mapping. Remote Sens. 8. NASA, 2017. Landsat, N. Science Data Users Handbook. Olofsson, O., Foody, G.M., Stehman, S.V., Woodcock, C.E., 2013. Making better use of accuracy data in land change studies: estimating accuracy and area and quantifying uncertainty using stratified estimation. Remote Sens. Environ. 129, 122–131. http:// dx.doi.org/10.1016/j.rse.2012.10.031. Olthof, I., Fraser, R.H., 2014. Detecting landscape changes in High latitude environments using landsat trend analysis: 2. Classification. Remote Sens. 6, 11558–11578. Perez-Hoyos, A., Garcia-Haro, F.J., San-Miguel-Ayanz, J., 2012. Conventional and fuzzy comparisons of large scale Land cover products: application to corine, Glc2000, MODIS and globcover in Europe. ISPRS-J. Photogramm. Remote Sens. 74, 185–201. Peterken, G.F., 2001. Structural dynamics of Forest stands and natural processes. In: Evans, J.W. (Ed.), Procedures for Developing Allowable Properties for a Single Species Under Astm D1990 and Computer Programs Useful for the Calculations. Forest Products Laboratory (U.S.), U.S. Dept. of Agriculture, Forest Service, Forest Products Laboratory, Madison, WI p 42. Pickell, P.D., Hermosilla, T., Coops, N.C., Masek, J.G., Franks, S., Huang, C.Q., 2014. Monitoring anthropogenic disturbance trends in an industrialized boreal Forest with landsat time series. Remote Sens. Lett. 5, 783–792.

Pompa-García, M., Jurado, E., 2014. Seasonal precipitation reconstruction and teleconnections with enso based on tree ring analysis of Pinus cooperi. Theoretical Appl. Climatology 117, 495–500. Pompa-García, M., Sensibaugh, M., 2014. Ocurrencia de incendios forestales y su teleconexión con fenómenos enso. CienciaUAT 8, 6–10. Pompa-García, M., Miranda-Aragón, L., Aguirre-Salado, C.A., 2015. Tree growth response to ENSO in Durango, Mexico. Int. J. Biometeorol. 59, 89–97. http://dx.doi.org/10. 1007/s00484-014-0828-2. Quiñones, B.S., 2007. Plagas y Enfermedades. Región III “Pacífico-Norte” Conafor. Durango, Mexico. . R Core team, 2015. R Core Team. R A Lang. Environ. Stat. Comput. R Found. Stat. Comput. Vienna, Austria. pp. 275–286 ISBN 3-900051-07-0. URL < http// www.Rproject.org/.55 > . Ramírez-Mejía, D., G. Cuevas, Mendoza, E., 2011. Escenarios de cambio de cobertura y uso del suelo en el Corredor Biológico Mesoamericano-méxico. Memorias de la XIX reunión SELPER-México 410. Reiche, J., de Bruin, S., Hoekman, D., Verbesselt, J., Herold, M.A., 2015. Bayesian approach to combine landsat and alos palsar time series for near real-time deforestation detection. Remote Sens. 7, 4973–4996. Rosete, F., Ordóñez, J., Masera, O., 1977. Dinámica del Cambio de Uso del Suelo y Emisiones de Carbono en La Meseta Purépecha. Reporte interno. Instituto de Ecología. UNAM, México. Rosete-Vergés, F.A., Pérez-Damián, J.L., Villalobos-Delgado, M., Navarro-Salas, E.N., Salinas-Chávez, E., Remond-Noa, R., 2014. El avance de la deforestación en México 1976-2007. Madera Bosques 20, 21–35. Rzedowski, J., 1978. Vegetación de México. Ed. Limusa. 4 A. Reimpresión. México, DF México. . Sáenz-Romero, C., Rehfeldt, G.E., Crookston, N.L., Duval, P., St-Amant, R., Beaulieu, J., et al., 2010. Spline models of contemporary, 2030, 2060 and 2090 climates for Mexico and their use in understanding climate-change impacts on the vegetation. Clim. Change 102, 595–623. Sarh, S.F., 1994. Inventario Nacional Forestal Periódico 1992-1994: Aprovechamiento y conservación de los Bosques. 9688003794, México. . Sarukhán, J., Carabias, J., Koleff, P., Urquiza-Haas, T., 2012. Capital natural De México: Acciones Estratégicas para su valoración, preservación y recuperación. Comisión Nacional para el Conocimiento y Uso de la Biodiversidad. Schroeder, T.A., Healey, S.P., Moisen, G.G., Frescino, T.S., Cohen, W.B., Huang, C.Q., Kennedy, R.E., Yang, Z.Q., 2014. Improving Estimates of Forest disturbance by combining observations from landsat time series with Us Forest service Forest inventory and analysis data. Remote Sens. Environ. 154, 61–73. SEMARNAP, 1997, 1998, 1999, 1998-2000. Anuario Estadístico de la Producción Forestal, Secretaría de Medio Ambiente, Recursos Naturales y Pesca, México. SEMARNAT, 2013. Anuario Estadístico de la Producción Forestal 2013, Mexico. p 236. . SEMARNAT-CONAFOR, 2001a. Programa Estratégico Forestal Para México 2025. Semarnat, México, D. F. México p 136. SEMARNAT-CONAFOR, 2001b. Programa Nacional Forestal 2001-2006. Guadalajara, Jalisco, Mexico, p 143. . SEMARNAT-SNIARN, 2017. Base de datos. Consulta temática. Plantaciones forestales comerciales. http://dgeiawf.semarnat.gob.mx:8080/ibi_apps/WFServlet?IBIF_ex= D3_RFORESTA09_02&IBIC_user=dgeia_mce&IBIC_pass=dgeia_mce& NOMBREENTIDAD=*&NOMBREANIO=*. Shimabukuro, Y.E., Beuchle, R., Grecchi, R.C., Achard, F., 2014. Assessment of Forest degradation in Brazilian amazon due to selective logging and fires using time series of fraction images derived from landsat ETM+ images. Remote Sens. Lett. 5, 773–782. Silva-Flores, R., Pérez-Verdín, G., Wehenkel, C., 2014. Patterns of tree species diversity in relation to climatic factors on the Sierra Madre Occidental, Mexico. PLoS One 9 (8). http://dx.doi.org/10.1371/journal.pone.0105034. e105034. Singh, A., 1989. Digital change detection techniques using remotely-sensed data. Int. J. Remote Sens. 10, 989–1003. SRNyMA, S.d.R.N.y.M.A. 2002. Programa Estratégico Forestal 2030, Secretaría de Recursos Naturales y Medio Ambiente del Estado de Durango., Durango, Dgo., p p 242. SRNyMA-CONAFOR, 2007. Plan Estrátegico Forestal 2030 (2030 Forest Strategic Plan). Durango State Government, Durango, México. Stahle, D., Burnette, D., Díaz, J.V., Heim, R., Fye, F., Paredes, J.C., Soto, R.A., Cleaveland, M., 2012. Pacific and Atlantic influences on mesoamerican climate over the past millennium. Clim. Dyn. 39, 1431–1446. Stibig, H.J., Achard, F., Carboni, S., Rasi, R., Miettinen, J., 2014. Change in tropical Forest cover of Southeast Asia from 1990 to 2010. Biogeosciences 11, 247–258. Tewkesbury, A.P., Comber, A.J., Tate, N.J., Lamb, A., Fisher, P.F., 2015. A critical synthesis of remotely sensed optical image change detection techniques. Remote Sens. Environ. 160, 1–14. Townshend, J.R.G., Justice, C.O., 1988. Selecting the spatial-resolution of satellite sensors required for global monitoring of land transformations. International J. Remote Sens. 9, 187–236. Tulbure, M.G., Broich, M., 2013. Spatiotemporal dynamic of surface water bodies using landsat time-series data from 1999 to 2011. ISPRS-J. Photogramm. Remote Sens. 79, 44–52. UNEP (United Nations Environment Programme), 1997. World Atlas of Desertification 2ED. UNEP, London. USGS, 2017. Earth Explorer. (November 2). https://earthexplorer.usgs.gov/. Vargas-Larreta, B., López-Sánchez, C.A., Corral-Rivas, J.J., López-Martínez, J.O., AguirreCalderón, C.G., Álvarez-González, J.G., 2017. Allometric equations for estimation of biomass and carbon stocks in temperate forests of North-Western Mexico. Forest 8, 269. Verbesselt, J., Hyndman, R., Newnham, G., Culvenor, D., 2010. Detecting trend and seasonal changes in satellite image time series. Remote Sens. Environ. 114, 106–115. Verbesselt, J., Zeileis, A., Herold, M., 2012. Near real-time disturbance detection using satellite image time series. Remote Sens. Environ. 123, 98–108. Vogelmann, J.E., Kost, J.R., Tolk, B., Howard, S., Short, K., Chen, X.X., Huang, C.Q.,

243

Int J Appl  Earth Obs Geoinformation 73 (2018) 230–244

A. Novo-Fernández et al.

Pabst, K., Rollins, M.G., 2011. Monitoring landscape change for landfire using multitemporal satellite imagery and ancillary data. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 4, 252–264. Wehenkel, C., Corral-Rivas, J.J., Hernandez-Diaz, J.C., von Gadow, K., 2011. Estimating balanced structure areas in multi-species forests on the Sierra madre Occidental, Mexico. Ann. For. Sci. 68, 385–394. White, J.C., Wulder, M.A., Hermosilla, T., Coops, N.C., Hobart, G.W., 2017. A nationwide annual characterization of 25 years of Forest disturbance and recovery for Canada using landsat time series. Remote Sens. Environ. 194, 303–321. Williams-Linera, G., Alvarez-Aquino, C., 2010. Tropical dry Forest landscape restoration in Central Veracruz, Mexico. Ecol. Restor. 28, 259–261. Xin, Q.C., Olofsson, P., Zhu, Z., Tan, B., Woodcock, C.E., 2013. Toward near Real-time monitoring of Forest disturbance by fusion of MODIS and landsat data. Remote Sens. Environ. 135, 234–247.

Zhao, X., Corral-Rivas, J., Zhang, C., Temesgen, H.v., Gadow, K., 2014. Forest observational studies-an essential infrastructure for sustainable use of natural resources. For. Ecosyst. 1, 1–10. Zhao, F., Huang, C.Q., Zhu, Z.L., 2015. Use of vegetation change tracker and support vector machine to map disturbance types in Greater yellowstone ecosystems in a 1984-2010 landsat time series. IEEE Geosci. Remote Sens. Lett. 12, 1650–1654. Zhu, Z., 2017. Change detection using landsat time series: a review of frequencies, preprocessing, algorithms, and applications. ISPRS J. Photogramm. Remote Sens. 130, 370–384. http://dx.doi.org/10.1016/j.isprsjprs.2017.06.013. Zhu, Z., Woodcock, C.E., 2014. Continuous change detection and classification of Land cover using all available landsat data. Remote Sens. Environ. 144, 152–171. Zhu, Z., Woodcock, C.E., Olofsson, P., 2012. Continuous monitoring of Forest disturbance using all available landsat imagery. Remote Sens. Environ. 122, 75–91.

244