1 Introduction

Landslides are categorized as destructive natural disasters and are heterogeneously distributed in the world (Petley 2012; Froude and Petley 2018). Although floods are the deadliest natural disaster (Jonkman and Kelman 2005), the number of deaths from landslides is greater than that from floods in some regions and countries such as East Uganda (Agrawal et al. 2013), Italy (Guzzetti et al. 2005), and Japan (Shinohara and Kume 2022). Landslide susceptibility is the likelihood of a landslide occurrence in an area based on local and environmental conditions (Reichenbach et al. 2018). The evaluation of causative factors affecting landslide susceptibility is an initial step for reducing landslides.

Land use is a factor affecting landslide susceptibility that may change considerably owing to human activity and natural disasters (Guns and Vanacker 2013), unlike other geo-environmental factors (e.g., morphology and geology). The presence of forests and trees contributes to slope stability owing to both hydrological and mechanical processes (Cislaghi et al. 2021). In most temperate regions, the latter contribution is much larger than the former (Stokes et al. 2009). Soil shear strength is determined by cohesion, friction angle, and total stress on the shear surface (Zhang et al. 2001). Root reinforcement, represented as a function of the empirical root-soil interaction factor, root-area ratio, and root tensile strength (Waldron 1977; Wu et al. 1979), increases soil shear strength, commonly by increasing cohesion (O’loughlin and Ziemer 1982). Therefore, landslide susceptibility in forested areas is generally lower than that in other land-use types (Shu et al. 2019; Lusiana and Shinohara 2022).

Landslide susceptibility may also differ depending on forest characteristics, such as species (Fattet et al. 2011; Ji et al. 2012), density (Douglas et al. 2013; Spiekermann et al. 2022), height (Evans et al. 2020), age (Genet et al. 2008), and forest management strategies (Cislaghi et al. 2021). Landslide susceptibility in mature forests is generally smaller than that in younger forests owing to thicker, deeper, and greater amounts of roots. This indicates that changes in root strength caused by changes in forest age after harvesting will affect landslide susceptibility (Brardinoni et al. 2002). Root reinforcement gradually reduces after harvesting owing to root decay (Cislaghi et al. 2021). The increase in landslide susceptibility occurs during the period of minimum root strength owing to root decay after harvesting and prior to substantial regeneration (Sidle and Ochiai 2006). The susceptibility gradually decreases over time as root strength increases during regrowth of planted and natural forests (Gorsevski et al. 2006; Turner et al. 2010).

Many studies have reported changes in landslide susceptibility owing to changes in forest age distribution on the local scale (Lee and Lee 2006; Imaizumi et al. 2008; Turner et al. 2010); some of them (e.g., Zhang et al. 2022) have analyzed multiple causative factors. On the national scale, Shinohara and Kume (2022) examined the trends in the number of landslide fatalities and underlying factors, including forest maturity, and found that increasing forest maturity reduced landslide numbers and fatalities in Japan, especially from 1960 to 1989. Sato and Shuin (2022) obtained a similar result using the data for forest growing stocks and flooded areas. However, these findings were obtained based on comparing trends, and no studies since have quantified the effect of increasing forest maturity on landslides on a national scale. The contribution of forest maturity to reducing landslide susceptibility on a national scale should be determined because this information is of fundamental importance in designing and implementing risk management policies (Malet et al. 2013), particularly in providing proper recommendations for future forest management-affected landslides over the whole of Japan.

Forest harvesting and rainfall are key factors affecting slope stability owing to the removal of root reinforcement and the increase in soil saturation, respectively (Malek et al. 2015; Barik et al. 2017). Because a change in rainfall regimes (i.e., an increase in rainfall) is expected in the future owing to the warmer climate, the rates of rainfall-triggered landslides are consequently predicted to increase (Gariano and Guzzetti 2016). To control rising landslide frequency, we hypothesized that appropriate forest management by adjusting age distribution may effectively mitigate the impact of climate change on landslide hazards. This study aimed to quantify the effects of an increase in forest maturity in different forest age distributions on rainfall-triggered landslides in Japan. In Japan, the forested area has been constant since from 1966 to 2017 and forests have become more mature. The authors first collected case studies reporting landslide susceptibility in several forest age classes and developed a general relation between landslide susceptibility and forest age. Then, we developed models to estimate the annual number of rainfall-triggered landslides in Japan using a rainfall index and forest age distributions. Finally, we simulated potential changes in the number of landslides owing to changes in forest age distributions and rainfall.

2 Materials and methods

2.1 Developing the relation between the forest age and landslide susceptibility

Landslide susceptibility is usually represented as a frequency ratio (FR) (i.e., the ratio of the percentage landslide area for a class to the percentage landslide area for the site) or landslide density (the number of landslides / the total area for a class). Assuming the mean of a landslide area (i.e., the total landslide area / the number of landslides) does not change owing to the forest age, both FR and the landslide density can be treated as equal when examining the relation between the forest age and landslide susceptibility. In this study, FR and the landslide density are collectively referred to as the landslide susceptibility index (LSI).

Case studies reporting FR or landslide density in several forest age classes were collected by referring to the SCOPUS and ISI Web of Knowledge databases using landslide susceptibility, landslide density, and forest age as keyword searches. It is known that landslide susceptibility generally varies when the forest age is less than 20 years (Sidle and Wu 1999; Dhakal and Sidle 2003; Turner et al. 2010). Considering this and the data availability from previous studies, the authors set the maximum forest age as 60 years and established six age classes with 10-year intervals (i.e., 1–10, 11–20, 21–30, 31–40, 41–50, and 51–60 years). Therefore, only studies with an interval of 5 or 10 years were collected. For studies with an interval of 5 years, the LSI for each forest age class with an interval of 10 years was calculated by averaging LSI for two subject classes. In total, 22 datasets were collected. Some datasets targeted the same rainfall event and location as other studies (or another study). For these cases, the average LSI for each forest age category was used for analysis.

The magnitude of LSI was different because of site-specific causative factors. Therefore, the normalized LSI (NLSI) was calculated for each age class and the LSI of 21–30 years, available in all datasets, was set as 1. Using all data available for each forest age class, the average of the NLSI was calculated. The relations between the forest age and average NLSI were examined using four types of regression (i.e., power, logarithmic, exponential, and linear regression). When examining the regression models, the forest age classes of 1–10, 11–20, 21–30, 31–40, 41–50, and 51–60 years were set at 5, 15, 25, 35, 45, and 55 years, respectively. Using the regressions, NLSI was calculated for each year between 1 and 60 years. The NLSIs based on power, logarithmic, exponential, and linear regressions are referred to as NLSIpower, NLSIlog, NLSIexp, and NLSIlinear, respectively.

2.2 Developing models to estimate the annual number of rainfall-triggered landslides in Japan

Generalized linear models (GLMs) were developed to estimate variations in the annual number of rainfall-triggered landslides in Japan from 1966 to 2017. In Japan, 66% of the total land is covered by forests and the forested area remained almost constant during the analysis period. Geologically, the accretionary complexes, metamorphic rocks, plutonic and volcanic rocks, and surface sediments have an extremely complex distribution. The annual precipitation and average temperature are 1,650 mm and 15.5 °C, respectively. Detailed geological and meteorological conditions are available in Shinohara and Watanabe (2023). When developing the models, we assumed that the temporal change in the number of landslides was determined by forest age distributions and rainfall. The validity of this assumption is discussed in Sect. 4. Although the definition of forest age was not available for most datasets used in this study, the forest age for planted and natural forests were expected to be the number of years after planting and after clearcutting, respectively.

The area of forest age classes at 5-year intervals was available for planted forests in 12 years between 1966 and 2017 and for natural forests in 2012 and 2017. Using these data, the areas of planted and natural forests in 13 forest age classes (i.e., 1–5, 6–10, 11–15, 16–20, 21–25, 26–30, 31–35, 36–40, 41–45, 46–50, 51–55, 56–60, and > 60 years) were estimated for every year during the analysis period. The detailed procedures for the calculation and the estimated area for each age class are available in Supplementary Information Text S1 and Fig. S1, respectively.

Using the areas of planted and natural forests (PFAi and NFAi, respectively) and the NLSI (NLSIi) for each age class i (with an interval of 5 years), we calculated the national-scale NLSI (NLSIJpn) for every year from 1966 to 2017.

$${\text{N}\text{L}\text{S}\text{I}}_{\text{J}\text{p}\text{n}}=\sum _{i=1}^{n}\left\{{\text{N}\text{L}\text{S}\text{I}}_{i}\times \left({\text{P}\text{F}\text{A}}_{i}+{\text{N}\text{F}\text{A}}_{i}\right)+\dots +{\text{N}\text{L}\text{S}\text{I}}_{n}\times \left({\text{P}\text{F}\text{A}}_{n}+{\text{N}\text{F}\text{A}}_{n}\right)\right\}$$
(1)

where n is the number of the forest age classes (= 13). NLSIi, except NLSI13, is set as the average NLSI over the 5 years including in the interval. NLSI13 (i.e., NLSI for the age of > 60 years) is set as NLSI at the age of 60 years.

A rainfall index proposed by Shinohara and Komatsu (2016)—the total rainfall from May to October in northern, eastern, and western Japan—was used as the other explanatory variable. The rainfall index has a good correlation with the annual number of rainfall-triggered landslides in Japan (Shinohara and Komatsu 2016; Shinohara and Kume 2022). For the response variable, the number of rainfall-triggered landslides, reported by the Sabo and landslide Technical Center (STC), was used. The dataset includes landslides that affected houses or public facilities and is the most accurate among the four landslide datasets in Japan (Shinohara and Kume 2022). The dataset is available from 1982 to 2017 only. The number of landslides from 1966 to 1981 was estimated using the dataset of the National Police Agency (NPA) and the relation between the number by STC and that by NPA during 1982–2017 (r = 0.76, P < 0.05). The sources of datasets for forests, rainfall, and the number of landslides can be found in Table S1.

In the GLMs, a gamma distribution was assumed, and the logarithm link function was selected (Atkinson and Massari 1998). The explanatory variables (i.e., NLSIJpn and rainfall index) were selected using a stepwise method based on the Akaike Information Criterion (AIC). The models were developed using the four types of NLSI (see Sect. 2.1). The AIC, correlation coefficient (r), and root mean square error (RMSE) were used to evaluate the models. All analysis was calculated using the R program version 4.2.1 (R Core Team 2023).

2.3 Simulation of landslide frequency with different forest age distributions and rainfall amount scenarios

To quantify potential changes in the landslide frequency owing to changes in the forest age distribution, three scenarios of forest age distributions were prepared as inputs for the GLMs. In the first scenario, young-age-dominated forests were assumed using the forest age distribution in 1966 (Fyoung). In the second scenario, middle-age-dominated forests were assumed using the forest age distribution in 2002 (Fmiddle). In the third scenario, mature-age-dominated forests were assumed using the forest age distribution in 2017 (Fmature). Here, young-, middle-, and mature-ages corresponded to 1–20, 21–40, and > 40 years, respectively. Assuming a normal distribution with the mean and standard deviation (1046.2 ± 157.7 mm) for the rainfall index during 1966–2017 (Rcurrent), a datum was extracted using random numbers. The extraction was repeated 1,000 times and used as the input for the GLMs. Thus, the average and standard deviation of the number of landslides were obtained for the three forest distribution scenarios. In addition, four rainfall scenarios were prepared with an increase in the mean of 10% (R+ 10%) and 20% (R+ 20%) and a decrease in the mean of 10% (R–10%) and 20% (R–20%). For all scenarios, the standard deviation was assumed to be constant, and the extracted datum was repeated 1,000 times. The number of landslides was calculated for 15 scenarios with combinations of three forest age distribution (i.e., Fyoung, Fmiddle, and Fmature) and five rainfall (i.e., Rcurrent, R− 20%, R− 10%, R+ 10%, and R+ 20%) scenarios. The above simulation was conducted using the four types of NLSI.

3 Results

3.1 The relation between forest age and landslide susceptibility

We obtained 21 studies covering 11 sites that provided LSI in multiple forest age classes (Table 1). Among the 21 studies, 19 studies were conducted in South Korea, and the other two were in the United States and Japan. The 19 studies in South Korea reported FR and the two studies in the United States and Japan described the landslide density. Although not all studies indicated the types or species of the forests, these studies were conducted in both planted and natural forests and in both coniferous and broadleaved forests.

The LSI in all age classes for the 11 sites ranged from 0 to 18.00 (Fig. 1a), while the NLSI ranged from 0.13 to 4.93 (Fig. 1b). The NLSI for younger age classes tended to be larger than that for older age classes. For six sites, the highest NLSI was observed in the age class of 1–10 years. In 10 out of the 11 sites, the highest NLSI was observed in the age classes < 30 years. The relation between the forest age and NLSI was examined for the four types of regression (Fig. 2). The linear regression had the highest coefficient of determination (0.91), followed by the exponential (0.89), logarithmic (0.85), and power (0.74) regressions.

Table 1 List of studies reporting a landslide susceptibility index. The location of each site is shown in Fig. S2
Full size table
Fig. 1
figure 1

(a) Landslide susceptibility index (LSI) and (b) normalized LSI (NLSI) for the six forest age classes at the 11 sites. The bars for each forest age class align with site no.1 to site no.11 in Table 1. The asterisks and downward arrows in Fig. 1a indicate the forest age class with the highest LSI in each site and the sites with the LSI of 0, respectively. The blue and purple bars indicate data reported using the frequency ratio and landslide density, respectively. In Fig. 1b, LSI was normalized by setting the LSI as one for the forest age class of 21‒30 years (this age class was available in all datasets)

Full size image
Fig. 2
figure 2

Relations between the forest age and normalized landslide susceptibility index (NLSI). The blue squares and error bars indicate the average and standard deviation of NLSI. The power, logarithmic, exponential, and linear regressions are shown by red, orange, blue, and brown lines, respectively

Full size image

3.2 Development of models for estimating the annual number of rainfall-triggered landslides

The ratio of younger or older forests to the total changed substantially during the 51 years (Fig. 3a). The ratio of young forests decreased from 51% in 1966 to 3% in 2017. The ratio of mature forests increased from 29% in 1966 to 85% in 2017. Consequently, in all four types of NLSI, NLSIJpn decreased from 1966 to 2017. In 1966, NLSIJpn ranged from 2743.9 to 3209.7. In 2017, NLSIJpn ranged from 1288.1 to 1674.4. The maximum NLSIJpn was observed between 1966 and 1967. The average rainfall index during the analyzed period was 1046.2 mm with large year-to-year variations (Fig. 3b), whereas the trend of landslide numbers declined within the 51 years with an average of 1059 landslides per year. The average landslides were 1411 and 903 per year within 1966–1981 and 1982–2017, respectively (Fig. 3c). The number of landslides in 1972 was much larger than those in the years around 1972, probably because of the huge numbers of landslides over wide areas of Japan in that rainy season (Tani 1973), although the impact did not appear in the rainfall index.

The results from the GLMs using the four types of NLSIJpn are shown in Table 2. For all cases, both rainfall and NLSIJpn were selected as significant parameters (P < 0.01). The model using NLSIpower had the lowest AIC and RMSE and highest r values among the four cases.

Fig. 3
figure 3

Trends in (a) the area of the 13 forest age classes and national-scale normalized landslide index (NLSIJpn), (b) the rainfall index, and (c) the number of landslides during 1966–2017. In Fig. 3a, the area and NLSIJpn are shown in bars and lines, respectively. NLSIJpn was calculated using the four types of regressions; the colors correspond to the colors in Fig. 2. In Fig. 3b, the average values in 1990s, 2000s, and 2010s are shown in red lines. In Fig. 3c, data estimated from the National Police Agency (NPA) are shown in blue; data from the Sabo and landslide Technical Center (STC) are shown in red

Full size image
Table 2 Results from the generalized linear models using normalized landslide susceptibility indexes (NLSIJpn), using power, logarithmic, exponential, and linear regressions
Full size table

3.3 Effect of differences in forest age distributions and rainfall on variability in landslide frequency

The areas of each forest age class for the three scenarios are shown in Fig. 4. Note that the area of each planted and natural forest is available in Fig. S1. In the current rainfall and forest age distribution scenarios (Rcurrent and Fmature), the number was not significantly different among the four types of NLSI (ANOVA, P = 0.54), as shown in Fig. S3. Therefore, the GLM using NLSI power, with the lowest AIC and RMSE and highest r values, was used in the following analysis. Results from the GLMs using NLSIlog, NLSIexp, and NLSIlinear, are shown in Table S2. Table 3; Fig. 5 indicate the number of landslides estimated in the combinations of five rainfall and three forest age distribution scenarios. The number of landslides based on the scenarios Fyoung and Fmiddle was estimated as ca. 2.4 and ca. 1.1 times that of the scenario Fmature. Under all forest age distribution scenarios, the number of landslides based on the scenarios R+ 20% and R+ 10% was estimated as ca. 1.7 and 1.3 times that of the scenario to Rcurrent. The ratio of Fmiddle to Fmature (ca. 1.1) was slightly smaller than that of Rcurrent to R+ 10% (ca. 1.3). The ratio of Fyoung to Fmature (ca. 2.4) was considerably larger than that of Rcurrent to R+ 20% (ca. 1.7).

Fig. 4
figure 4

The area of the 13 forest age classes for the three scenarios of (a) Fyoung, (b) Fmiddle, and (c) Fmature

Full size image
Table 3 The number of landslides calculated for the combination of three forest age distributions and five rainfall scenarios. The average and standard deviation are shown
Full size table
Fig. 5
figure 5

Bubble plots of the number of landslides under scenarios of forest age distributions and rainfall

Full size image

4 Discussion

The relation was examined between the forest age and NLSI based on 21 previous studies in 11 sites. Using the relation and the area of each age class, we estimated changes in NLSIJpn between 1966 and 2017. The GLMs for estimating the number of landslides were developed from NLSIJpn and a rainfall index. On the basis of the GLMs, the authors found that the age distribution of forests potentially had a strong effect on landslide frequency. The number with Fmature, corresponding to the condition in 2017, was simulated to be 41% of that with Fyoung, corresponding to the condition in 1966. Thus, we succeeded in quantifying the reduction in landslide frequency through an increase in the maturity of forests at a national scale.

Shinohara and Kume (2022) speculated that the increase in forest maturity contributed to the decrease in the number of landslides in Japan between 1960 and 1989. Sato and Shuin (2022) concluded that there was a contribution from increasing forest carbon stocks to the reduction of landslides for the period before 1990. In our simulation (Table 3; Fig. 5), the number of landslides with Fyoung corresponding to the condition in 1966 was estimated at 2.2 times that of Fmiddle, corresponding to the condition in 2002. The result showed that the increase in forest maturity did contribute to the reduction in landslide frequency from 1966 to 2002, similar to the results in the two studies mentioned above. In addition, the number of landslides with Fmiddle was simulated to be 1.1 times that with Fmature, corresponding to the condition in 2017. The result suggests that increasing the maturity of forests potentially reduced landslide frequency even after 1990. The average value of rainfall index in the 2010s (= 1141.8 mm) was higher than that in the 1990s (= 1055.3 mm) and 2000s (= 1016.2 mm) (Fig. 3b). The higher amounts of rainfall might have offset the reduction in landslide numbers owing to increasing the maturity of forests after 1990.

When developing the relation between the forest age and NLSI, we collected data suitable for our criteria, that is, reporting the landslide density or FR with intervals of 5 or 10 years. Other studies that did not meet our criteria also reported decreases in landslide susceptibility owing to increasing forest age. For example, Gorsevski et al. (2006) examined landslide susceptibility for each class in north-central Idaho, USA, and found that the landslide susceptible area for the 0–5 and 6–10 year stand age categories was 2–3 times larger than that in older categories under rainstorm conditions. In addition, a study by the Oregon Board of Forestry (2001) showed that the landslide density in 0–9 years stand age forest was 2.5 times larger compared with a stand age of 31‒100 years. Thus, the relation we found may be similar in forests around the world.

The current study assumed a monotonic decrease in NLSI owing to the increase in forest age from 0 to 60 years (Fig. 2) and NLSI at ages of > 60 years was assumed to be the same as NLSI at an age of 60 years. However, because the forest development stage following “overmatured” is “decay” (Jacob et al. 2013), it is possible that NLSI at ages of > 60 years might have different trends. Furthermore, Cislaghi et al. (2021) suggested that root reinforcements could vary owing to the treatment of forests even at the same age. These factors were not considered in our study because of data limitations. This study also did not consider changes in root strength after harvesting and natural hazards (e.g., windstorms). When mature forests are harvested, the root reinforcement initially decreases owing to the root decay of the harvested trees and then increases owing to the root growth of the planted (or regenerated) trees. The rate of root reinforcement decay would be different depending on species, environmental conditions, and/or methodology for measuring root reinforcement (O’loughlin and Watson 1979; Cislaghi et al. 2021). In Japan, Kitamura and Namba (1981) examined the root pullout strength for living trees of various ages and stumps after various times in three coniferous and two broadleaved forest types and demonstrated that the total root reinforcement was at its minimum 5‒10 years after harvesting. Okada et al. (2023) reported that the minimum root reinforcement was approximately 9 years after harvesting in Japanese cedar trees. Imaizumi et al. (2008) described the landslide frequency in forests aged 26‒40 years as equal to control sites. This indicates that more than 20 years after forest harvesting, the root strength had almost completely returned to 0 years or the condition before harvesting (Kitamura and Namba 1981; Sidle and Ochiai 2006). In the current study, we calculated NLSIJpn simply considering the process of root regrowth and decay. In this calculation, we assumed that (1) NLSIs at 0 and 10 years were the same as NLSIlinear at 25 and 10 years, respectively; (2) NLSI increased linearly from 0 to 10 years; and (3) NLSI more than 10 years followed NLSIlinear. Although the recalculated NLSIJpn was different from the originals in the initial period of our analysis (Fig. S4), the estimated parameters of the GLMs (Table S3) and the simulated number of landslides (Table S4) were not altered. Thus, we verified that our conclusion remained the same, even if we used NLSIJpn on the assumption of root regrowth and decay.

In our simulation (Table 3; Fig. 5), the increase in the number of landslides from Fmature to Fmiddle under any rainfall scenario was slightly smaller than that from Rcurrent to R+ 10% (or from R–10% to Rcurrent) under any forest age distribution scenario. In contrast, the increase in the number of landslides from Fmature to Fyoung under any rainfall scenario was considerably larger than that from scenario Rcurrent to R+ 20% (or from R–20% to Rcurrent) under any forest age distributions. These results suggest that the change in maturity of forests potentially changes landslide frequency more than a change in rainfall. Furthermore, Sato et al. (2023) compared two landslide events that occurred in plantation forests of different ages in Japan and reported a 3.0-fold higher return period for the estimated groundwater level in mature forests than in young forests. Thus, the protection of mature forests is essential to avoid unnecessarily increasing landslide frequency in a changing climate. Japan has 10.3 million ha of planted forests; the ratio of planted forests to the total forested area is 42%, the second largest in the world (Yamaura et al. 2012). The increase in forest maturity results from low forestry activity because of increasing employer costs and decreasing timber prices (Komatsu et al. 2010). Recently, the loss of forestry workers has slowed, and the ratio of young workers has increased (Ministry of Agriculture Forestry and Fisheries 2021). In addition, large uncertainties exist in future rainfall prediction (Nikolopoulos et al. 2014). If both rainfall and the area of young forests increase in the future, the risk of landslides may substantially increase in Japan.

This study had several limitations. First, as we previously mentioned, the relation between the forest age and NLSI mostly came from South Korea. Second, the number of landslides was collected only for landslides that affected houses or public facilities. Shinohara and Kume (2022) expected that the number of people living in landslide-prone areas would increase during the analyzed period of this study. However, the construction of retaining walls, crib works, and ground anchors (Japan Landslide Society 2002; Junichi and Naoki 2020) will reduce the possibility of landslides on some slopes. These things might affect the number of landslides that we used. Third, we assumed that the relation between the total rainfall and rainfall intensity was constant for the simulation. The rainfall index that we used was highly correlated with the temporal changes in the number of landslides in recent years (Shinohara and Komatsu 2016; Shinohara and Kume 2022). The occurrence of landslides is affected by both the total (or long-term) rainfall and rainfall intensity (i.e., short-term rainfall) (Berti et al. 2012; Chen et al. 2015; Zhang et al. 2019). In Japan, a clear relation was observed between rainfall intensity and total rainfall (Shinohara et al. 2010), which is the probable reason why the simple index was well correlated with the number of landslides. Shinohara et al. (2010) reported that the relation between rainfall intensity and total rainfall had not changed from 1976 to 2007 in most stations of Japanese mountain regions. However, predictions of future trends in rainfall intensity are often different from that of total rainfall (Barbero et al. 2019; Lei et al. 2022). This may change the relation between the number of landslides and the rainfall index.

There are many types of landslides (Varnes 1978; Hungr et al. 2014). In Japan, landslides are classified into steep-slope failure, deep-seated landslide, and debris flow (Osanai et al. 2010). Shinohara and Watanabe (2023) indicated that the effect of forests on the number of rainfall-triggered landslides was clearly observed only for steep-slope failures among the three types of landslides. Some studies (e.g., Michelini et al. 2017; Wang et al. 2017) pointed out the contribution of forest roots to reducing the frequency of debris flows. In contrast, the slip surface of deep-seated landslides is often deeper than the depth of forest roots, suggesting that there would be no (or a weak) contribution of forests to reducing deep-seated landslide frequency. If we had used only the data for steep-slope failure (and debris flow), clearer effects might be observed. Furthermore, the authors combined planted and natural forests and did not consider the landslide susceptibility in each forest type (or species). The authors believe that this assumption is reasonable at the current stage. Nishioka et al. (2023) reported clear differences in landslide susceptibility between young and mature forests, but no differences in mature coniferous forests and broadleaved forests in Asakura, Japan. Note that, in Japan, most planted and natural forests are coniferous and broadleaved forests, respectively. If the relation between landslide susceptibility and the age is obtained for various types (or species) of forests, more detailed forest management (i.e., change in species) may be able to reduce landslide susceptibility. Furthermore, when rainfall exceeds the landslide resistance of mature forests, huge amounts of driftwood may cause substantial damage to residential areas (Sato et al. 2023). Thus, it may be important to consider both the positive and negative effects of mature forests in further studies.

5 Conclusion

The contribution of increased maturity of forests to landslide susceptibility was evaluated in Japan. The relationship between forest age and NLSI was developed based on 11 sites from three countries. The GLMs were created to estimate the annual number of rainfall-triggered landslides from a rainfall index and NLSIJpn, which was used to estimate the above relation and the area for each age class. Although NLSI was assumed to be a monotonic decreasing trend, the GLMs accurately represented the annual number of rainfall-triggered landslides. Using the GLMs and 15 scenarios with different rainfall amounts and forest age distributions, potential changes in the landslide frequency were simulated. The number of landslides with mature-age-dominated forests compared with middle- and young-age-dominated forests was estimated to increase by ca. 12% and 144%, respectively. Increases of 10% and 20% from current rainfall were estimated to increase the number of landslides by ca. 31% and 70%, respectively. Thus, the maturity of forests had potentially larger impacts on landslide frequency than changing rainfall. The authors conclude that the protection of forests is required to avoid unnecessarily increasing landslide frequency during climate change. In this study, the effects of forest characteristics except for age were not considered. If clear relations are obtained between landslide susceptibility and other forest characteristics, more detailed suggestions for forest management could possibly reduce landslide susceptibility.