The general method of our analysis is organized in four parts: reclassification of all databases as two classes Forest/No Forest (F/NF), comparison of EO-based products with the MAATE reference maps, production of confusion matrix and associated metrics and validation of selected EO-based products through analysis of forest cover losses (Figure 2).

The metrics associated with the confusion matrix (Figure 3) allow us to analyse accuracy, F1-score and MCC, and to guide the analysis by selecting the best-performing databases. The databases with the highest accuracy are HILDA+ and TMF, with values between 0.75 and 0.82 and between 0.75 and 0.78 respectively. In 2018, HILDA + correctly classified 82% of instances. In contrast, the MODIS database displays the lowest accuracy values, correctly classifying between 40% and 42% of instances over the 2008–2020 period. Accuracy values of all database range between 0.4 and 0.65. The highest F1-score is displayed by HILDA+, varying between 0.61 and 0.64. In contrast, the lowest is displayed by MODIS, which ranges from 0.44 to 0.47. Analysing the MCC metric, we can see that the highest values displayed by HILDA + are between 0.45 and 0.5. TMF also has an MCC close to that of HILDA+, with values between 0.43 and 0.45. Those for MODIS vary between 0.18 and 0.19. The MODIS values show a strong difference between F1-score and MCC, with F1-score values twice as high as MCC values. If the MCC is much lower than the F1-score, this may indicate that the model has a problem with True Negatives or False Positives.

Following this initial spatial analysis with confusion matrices based on Forest cover and No-Forest cover, we selected the databases used for a more precise analysis of forest cover gains and losses. The chosen databases are HILDA+, which has a low spatial resolution but goes back to the 1960s, and TMF, which has a more precise spatial resolution but goes back to the 1990s.

To validate the two EO-based products selected, we carried out a second analysis that showing the loss of forest cover over the whole period for HILDA+ and TMF compared with MAATE (Figure 4). The gross losses recorded by MAATE are 13,589 km² between 1990 and 2018 on EPSC. The gross losses recorded by HILDA + are 7,905 km² corresponding to 58.2% of the gross losses recorded by MAATE. The gross losses recorded by TMF are 4,890 km² corresponding to 36% of the gross losses recorded by MAATE. Forest cover gains are 2,562 km² according to MAATE. HILDA + observed 27.4% of the gains recorded by MAATE, compared to 66% for TMF. MAATE reports a net loss of 11,028 km². HILDA + reports a net loss of 7,203 km², or 65.3% of the net losses reported by MAATE. TMF reports a net loss of 3,018 km², or 27.4% of the net losses reported by MAATE.

Following this analysis, we chose HILDA + to analyse deforestation. Confusion matrices were used to spatially validate HILDA+, which showed good agreement in terms of the spatial distribution of forest cover compared with MAATE. The quantitative comparison of forest cover losses shows that over the period 1990–2018, HILDA + managed to detect more than 65% of the losses reported by MAATE, over a recent period when changes are complex to identify. HILDA + has the advantage of going back to the 1960s. At that date, the uncertainties presented by HILDA + are much smaller than for the recent period. HILDA + can therefore be a good product for obtaining accurate estimates of forest cover and associated losses over the long term. The annual resolution makes it possible to analyse the evolution of the deforestation process and to define phases of deforestation that are more intense than others. From a historical point of view, 1960 marked the beginning of the agrarian reforms that led to the massive displacement of the population and the reorganization of the land, with consequences for the deforestation process.

We used HILDA + to monitor the evolution of forest cover on the EPSC over 1960–2019 (Figure 5). On the EPSC, between 1960 and 2019, 4 phases were identified in the deforestation process. The first phase between 1960 and 1968 showed a rapid decrease in forest cover from 34,487 km² to 31,714 km², with a net loss of 2,773 km² or 8.04% over 8 years. During the second phase, between 1968 and 1982, forest cover continued to decrease much less rapidly until 1974, when it finally stabilized at about 31,100 km². At the end of the period, in 1982, forest cover had decreased by 9.82% since the 1960s. The third phase between 1982 and 2000, recorded a continuous large loss of forest cover from 31,100 km² to 22,778 km², representing a loss of 8,322 km². At the end of this period, in 2000, forest cover had decreased by 33.95% since the 1960s. The last phase, from 2000 to 2019, exhibited a continuous but less regular and weaker decrease in forest cover than during the previous period, reaching a forest cover of 19,354 km², representing a loss of 3,424 km². Over the last 2 decades, while the trend is still towards a decrease in forest cover, this decrease is 2.4 times less important than over the 18-year period from 1982 to 2000. Over the entire period from 1960 to 2019, forest cover was reduced by 43.88%.

Analysis of changes in forest cover at the scale of the EPSC has revealed 4 phases of deforestation. We can then carry out an analysis on the scale of bioclimatic zones to better distinguish the spatio-temporal dynamics of the deforestation process according to forest cover types. The analysis by bioclimatic zones with the HILDA + database (Figure 6) shows that over the entire period between 1960 and 2019, tropical rainforests lost 32% of their area, compared to 54.6% for transition forests and 50.5% for dry tropical forests (Figure 6D). However, while the tropical rain forest and seasonal forests have a similar area, 16,122 km² and 15,938 km² respectively in 1960, dry tropical forests initially have an area almost 6.6 times smaller, with a coverage of 2,413 km² in 1960 (Figure 6D). The dynamics of deforestation differ according to the type of forest considered. There was a net loss of forest cover twice as large in the period 1960–1970 for tropical transition forests with a loss of 2,000 km² at the end of the 1970s against 1,000 km² for tropical rainforests (Figures 6A,B). Deforestation rates remained stable during the 1970s and 1980s and there was almost no deforestation until the early 1980s for dry tropical forests. The period 1980–1990 marks the beginning of a strong deforestation of dry tropical forests with net losses multiplied by three and a loss of 14.3% of forest cover in 10 years (Figure 6C). In this period, the net losses recorded for tropical transition forests are 2.5 times higher than in the previous period and 2 times higher for tropical rain forests. At the end of the 1990s, transition forests lost 30.6% of their area compared to 1960, while tropical rainforests lost only 11.6% of their original area in 1960 (Figures 6A,B). The largest losses for the period 1990–2000 were recorded for tropical rainforests with a net loss multiplied by 2, reaching a loss of 24.6% since the 1960s (Figure 6A). For the last two periods, net loss of forest cover for all bioclimatic zones continues to increase steadily but less rapidly than in the other two periods (Figures 6A–C).

Between 1960 and 2019, the EPSC experienced four periods of deforestation (Figure 5). These four phases can be explained by anthropogenic and climatic factors.

The first phase of deforestation highlighted by HILDA + took place between 1960 and 1968. During this period, the forest cover has been steadily and rapidly decreasing from 34,487 km² to 31,714 km² (Figure 5). Over the whole period, a net loss of 2,773 km² was recorded. The first phase is the consequence of a structural transformation of Ecuadorian society and economy that began in the 1950s, with a shift from a rural and agrarian economy to an urban society and a commercial economy. The banana economy replaces the predominant cocoa crop after the end of World War II (Cabarle et al., 1989) and marks the culmination of the agro-exporter model. This agricultural change, with export agriculture, created a need for new land having land or river access to the ports (Delavaud, 1980) to be directly linked to world markets (Fauroux, 1981).

This evolution of the economic model was accompanied by a demographic boom. The increase in the population of cities during this period caused a need to increase agricultural supply (Salvador Lara, 1995). The increase in population in the Sierra generated pressure on the existing production area. These changes led the state to intervene in the reorganization of production systems to meet the need to increase agricultural supply and encourage the population of the Sierra to colonize the regions of the Costa. The State established a project for restructuring Ecuadorian agriculture that resulted in the agrarian reforms of July 1964 and October 1973. These reforms abolished old colonial inherited production relationships where a few owned most of lands and aimed at a more equitable redistribution allowing landless farmers to own small plots (Fauroux, 1981; Deler, 1987). The objective of the State was to increase productivity and expand the supply of manufactured products (Fauroux, 1981). The migration of landless peasants, especially from the uplands, was an essential factor in the expansion of agricultural borders into the forested areas of the coast (Commander and Peek, 1986). This economic change is encouraged by the Ecuadorian state, which promotes the expansion of agricultural land by granting loans and land to Ecuadorian and foreign entrepreneurs and peasants.

Nevertheless, we found with HILDA + data that in the period 1960–1970, seasonal forests on the central coast are twice as affected as tropical rainforests in the north of the country (Figures 6A,B). This spatial distinction is confirmed by Sierra and Stallings (1998) which explains that this first period of deforestation is concentrated on the central and northern coast and is explained by the fact that the process of deforestation remains limited by an emerging transport network outside the traditional axes: Quito-Guayaquil-Cuenca. At that time, the north of the province of Esmeraldas was still inaccessible (Figure 6A) (Fauroux, 1981; Sierra and Stallings, 1998).

The second phase of deforestation identified with HILDA + corresponds to a period of stabilization of forest cover between 1968 and 1982 over the entire EPSC (Figure 5). The analysis of the losses of forest cover in the three bioclimatic zones shows a stabilization of the forest cover loss over the whole period (Figures 6A–C). This stabilization is explained by several factors. The failure of the State in the restructuring of agriculture has social and economic consequences that affect forest cover, with a slowdown in deforestation and a gradual stabilization of forest cover over the period. This stabilization of the forest cover on the EPSC can also be related to the beginning of the deforestation period in the Amazon basin with the exploitation of oil resources (Southgate and Whitaker, 1992).

The failure of agrarian reforms caused a multidimensional crisis on the coast, with a peasantry revolt that materialized by a phenomenon of invasion of the land which accelerated in 1975, giving rise to clashes between peasants and the police and tensions between employers and workers (Fauroux, 1981). This period of tension slowed the clearing of land on the Pacific coast of Ecuador. This social crisis was compounded by the problem of overproduction of bananas in 1968, which also led to a crisis and massacres in the cities of Guayaquil. These socio-economic phenomena caused deforestation to slow down. The impoverishment of farmers and the failure of agricultural products markets were slowed down by inefficient storage methods and the still little developed ENSO communication links between the different production areas lead to shortages in some places and overproduction of products in others.

Another phenomenon explained a slowdown in deforestation on the coast. During this period, the Oriente began to be coveted for its oil resources, which led to the construction of roads to serve oil production, especially around Nueva-Loja that began in the 1970s (Deler, 1987; Lopez, 2021). Infrastructure development was supporting rapid settlement and agricultural expansion in the Oriente. Between the 1974 and 1982 censuses, 92,700 people settled in the four eastern provinces of Ecuador (Southgate et al., 1991). The 1982 census showed that a relatively small proportion of the labor force found employment in the oil industry and 60% of the economically active population of the region worked in agriculture. Thus, the oil boom of the 1970s caused the share of agricultural products in total exports to fall: in 1964 all agricultural products (bananas, coffee, cocoa) represent 89.2% of the country’s exports, compared with 36.2% in 1976, while oil exports account for 47.8% (Fauroux, 1981).

The third deforestation period runs from 1982 to 2000. There is a continuous large loss of forest cover from 31,100 km² in 1982 to 22,778 km² in 2,000 (Figure 5). Deforestation now affects all bioclimatic zones (Figures 6A–C). By establishing a spatio-temporal differentiation, between 1980 and 1990, we observed greater losses of dry tropical forests and transitions and from the 1990s, the deforestation process moved towards the north of the country, and since 1990, tropical rainforests were the most affected biome by deforestation. Sierra and Stallings (1998) also point to a resumption of deforestation on the Costa. The period of deforestation is due to the development of the road network mainly in the north of the country. On the Costa, construction of new roads began in the 1970s, such as those to access La Tola, at the extreme northwest of the country and later, Borbón and the lower Santiago River. The consequences of these constructions on deforestation are visible only from the 1980s with the extension of the agricultural border to the forests of the north of Esmeraldas. In 1982, almost half of the population of the parish of Borbón was made up of immigrants, compared to 20% in 1978. The number of immigrants in Malimpia Parish tripled between 1978 and 1982 and quadrupled between 1982 and 1990 (Sierra and Stallings, 1998). Then, in the 1980s, new roads were built from the coast inland, following rivers and forest paths (for example, in the parishes of Chontaduro and Chumunde). By the late 1980s, many areas north of Esmeraldas, particularly near the coast, were dominated by settlers. The 1990s marked the beginning of a process of rapid expansion of agriculture on the coast, and mainly in the north, where Esmeraldas became a province that attracted people through the availability of land to produce agricultural products for export (mainly bananas, coffee and cocoa).

The oil economy also had important multiplier effects. Subsequently, oil extraction has created urban and rural jobs with relatively high wages, generating demand for services, food and other commodities produced in the deforested areas. The consequences of the oil economy are the development of regional markets, stimulating the development of a commercial agricultural sector for domestic markets. During this period, there is an expansion of agricultural land on the Costa.

Sierra and Stallings (1998) have highlighted the role of selective timber extraction in north-western Ecuador [for example, heavy woods such as chanul (Humiriastrum sp.) and guayacán (Tabebuia sp.)]. This aspect is also widely discussed by Amelung and Diehl (1992) and Guppy (1984). Nearly 70% of the deforestation that occurred in the forests of the north-west of the EPSC between the 1980s and 1990s can be explained solely by timber extraction (Sierra and Stallings, 1998). In addition to the anthropogenic disturbances described, there were also climatic disturbances during this period, with two major ENSO episodes, in 1982–1983 and 1997–1998, responsible for extremes inundations (França et al., 2020; Quiroz, 1983).

The fourth period of deforestation between 2000 and 2019 is characterized by a slowdown in deforestation compared to the previous period (Figure 5). With HILDA + there was a decrease in forest cover which is 2.5 times less rapid compared to the period 1982–2000. This period of deforestation, which corresponds to the last 20 years, is characterized by a contraction of the agricultural area and a tendency to a decrease in deforestation compared with the previous period, with a significant slowdown in the expansion of agricultural land, a decrease in demand for new agricultural land and an intensification of rural production systems. This is due to much lower population growth rates than in the past and the concentration of the country’s population in dense urban and rural areas. This population concentration was responsible for about 60% of the decrease in deforestation in the Ecuadorian provinces (MAATE, 2017). When the annual growth of urban areas was high (between 2000 and 2008) compared to the 1990s, deforestation was lower. Beyond these social factors, which contributed to the decrease in deforestation, when the country’s economy entered crisis in 1999, deforestation decreased. Thus, the recurrence of El Niño, the fall in oil prices and the financial crisis of 1998–1999 have had a major impact on the national economy and household purchasing power (Ramírez and Ramírez, 2005). The 1998 EL Niño event caused losses equivalent to 14.5% of GDP and the 1998/1999 crisis cost between 25% and 22% of GDP. The fall in oil prices in 1998 aggravated the fiscal crisis, forcing the state to cut public services and transfers. Public social expenditure, which had increased between 1992 and 1996, fell by 37% between 1996 and 1999. This has resulted in an increase in urban poverty, from 19% of households in 1995 to 42% in 1999. In rural areas, it rose from 56% to 77% (Ramírez and Ramírez, 2005). The crisis has been a trigger for a general decline in demand for agricultural products and deforestation. From 2005, once the impact of the crisis and inflation linked to dollarization are under control, we observe a reactivation of deforestation in some regions such as north of Esmeraldas (Figure 6A).