Mass movements, particularly soil landslides, are considered frequent natural disasters that result in loss of human lives and substantial economic losses [1] . In Brazil, the most affected regions are the densely urbanized slopes, primarily located in the major urban centers of the South, Southeast, and Northeast. In the South and Southeast regions, the mass movement majority occurrences happen in the occupied slopes of the mountainous coastal areas of Brazil [2] [3] . In the Northeast region, according to the Brazilian Atlas of Natural Disasters (2013), the occurrences are concentrated in Bahia and Pernambuco, which have characteristics of barrier formation [4] .

The primary preparatory and triggering environmental variable for landslide processes is rainfall. Once in Brazil there are several climate conditions [5] , certain rainfall indices, such as accumulated rainfall over 24 hours, 48 hours, or 72 hours obtained through empirical correlation methods [6] [7] [8] are typically employed in Preventive Civil Defense Plans as critical operational thresholds in alert systems to predict landslides in urbanized slopes. However, relying solely on rainfall thresholds does not allow for precise and consistent issuance of alerts, as the landslide initiation process involves other equally important geo-environmental factors or variables.

In this context, geo-environmental or geotechnical variables, namely matric suction, and soil moisture [9] [10] are prominent. Understanding the behavior of these variables in unsaturated soils is crucial for better comprehension of the landslide rupture mechanism on urban slopes and consequently defining more precise and robust critical operational thresholds [11] [12] . These thresholds can be used in conjunction with rainfall thresholds (accumulated and intensity) to predict the occurrence of this type of process.

Soil moisture is a geotechnical variable that can be obtained through direct in-situ measurements using real-time monitoring sensors/equipment [13] or estimated through various empirical techniques or methods [14] - [22] . This estimation may also involve utilizing monitoring data for calibration or adjustment of prediction models [23] - [28] .

Soil moisture estimates based on models can cover a wide range of spatial and temporal scales, as well as their complexity. A widely used approach to infer soil moisture is based on the Antecedent Precipitation Index (API), which is calculated by summing up the antecedent daily rainfall, weighted by a specific factor to be determined. This factor may encompass the entire complexity of soil characteristics and processes due to a rainfall event.

Therefore, this study aims to estimate soil moisture and establish critical operational thresholds based on past landslide occurrences, using an empirical modeling of the Antecedent Precipitation Index (API). This will be achieved by utilizing field measurements of soil moisture obtained through sensors and equipment utilizing multisensor capacitance, installed in a high-risk area located in the mountainous region of Campos do Jordão municipality, São Paulo, Brazil.

The municipality of Campos do Jordão is situated in the state of São Paulo, bordering Minas Gerais, precisely in the mountainous region known as Serra da Mantiqueira, which is part of the Atlantic Plateau. The area features altitudes ranging from 922 meters in the Rio Sapucaí valley near the border with the Municipality of Piranguçu (Minas Gerais) to 2008 meters in the region of Fazenda da Lavrinha in Guaratinguetá (SP). Figure 1 illustrates the location of Campos do Jordão, the digital elevation model of the terrain, and the specific area referred to as the “monitoring region” in this study.

The urban region of Campos do Jordão is situated in the plateau compartment known as Planalto de Campos do Jordão, characterized by hilly terrain with restricted ridges. The area features rounded hilltops, slopes with rectilinear and sometimes steep profiles, high-density drainage, dendritic to pinnate patterns, closed valleys with restricted interior plains [29] .

The terrain is supported by metamorphic rocks with granitic intrusions, following a regional NE direction structuring by metamorphic foliation and shear zones in the Precambrian rocks. Notably, the Jundiuvira Fault is located at the edge of the Campos do Jordão Plateau, adjacent to the escarpment of Serra da Mantiqueira. The predominant rocks consist of gneisses and migmatites from the Paraíba do Sul Complex [30] . Unconsolidated materials are found in the form of sedimentary deposits and residual soils.

Significant packages of organic clay occur in amphitheater erosive floodplains. Residual soils overlaid on crystalline rock slopes correspond to cambisols [31] . The textural variation of these soils results from the heterogeneity of the gneiss-migmatite rock mass, generally composed mainly of silt and fine sand, with variable occurrences of clay depending on the degree of profile alteration. The saprolitic material formed beneath the residual horizons, resulting from the decomposition of the gneiss-migmatite rock, exhibits a predominantly sandy-clayey texture, with the quantity of silt and clay varying with depth.

The annual mean temperatures vary around 14˚C, with minimums below 0˚C and maximums potentially exceeding 30˚C. Precipitation varies greatly, with annual totals ranging from 800 to 2800 mm. The distribution of rainfall throughout the year shows a concentration of over 80% between the months of October to March, with December, January, and February being the wettest months [32] . This rainy period is part of the large-scale phenomenon known as the South Atlantic Convergence Zone, SACZ, one of the most significant phenomena in the summer regional scale in South America. Many flood, inundation, and mass movement events are associated with extreme rainfall caused by SACZ [33] .

Due to its steep terrain with valleys, amphitheaters, and rugged scarps, Campos do Jordão experiences a high occurrence of mass movement events and flooding [34] [35] [36] . In January 2000, one of the worst landslides was recorded, resulting in the loss of 10 lives, over 100 injured individuals, and severe damage to 423 houses [37] [38] . The neighborhoods most affected by mass movements in the municipality are Vila Albertina, Santo Antônio, and Britador, located within the Piracuama River basin. Due to the recurrent events, this region of the municipality has been equipped with real-time monitoring devices and sensors, including soil moisture sensors, rain gauges, and river level sensors [39] . The location of these sensors, mapping of landslide risk, drainage, and delineation of the basin are shown in Figure 2. The locations of recorded mass movement occurrences from 2009 to 2020 are also depicted.

The moisture sensors shown in Figure 2 are EnviroScan^TM type sensors installed in access tubes manufactured by Sentek Pty. Ltd, containing six capacitive sensors (spaced every 50 cm) distributed along three meters inserted into the soil. The sensors provide outputs in volumetric water content in the soil (millimeters of water per 100 mm of soil). These values are converted from a frequency to a scale using a standard calibration equation, based on data obtained from various scientific studies across a variety of soil textures [37] . The available data from the moisture sensors is limited to the period from 12/04/2019 to the present moment, but the data for December 2019 had to be discarded due to calibration errors. Therefore, the period from 01/01/2020 to 12/31/2021 was defined as the database for analysis in this study.

Figure 3 presents the variations in daily temperature (maximum, minimum, and average daily), the climatological daily average temperature (calculated based on the historical series of the INMET meteorological station, data from 2002 until 2021), daily precipitation, and soil moisture at different depths (50 cm to 300 cm) for one of the sampling points shown in Figure 2. A clear distinction between the dry period, starting in April, and the wet period, starting in October, can be observed, with maximum temperatures well above the climatological average. It is also evident that soil moisture at a depth of 50 cm responds to rainfall much more quickly. Regarding temperature, during the dry period, it can be observed that temperatures reach their minimum values, both for daily maximums and minimums.

The Antecedent Precipitation Index (API) is typically used to estimate surface runoff based on recorded rainfall events using rain gauges or other estimation techniques in unmonitored watersheds or those lacking direct calculation information, such as discharge data [40] . Antecedent precipitation is calculated based on the recorded rainfall in the days leading up to the event day it refers to. It can also serve as a measure of soil moisture. The initial condition of the soil in terms of its water content is of crucial importance to have a quantitative understanding, as dry soil responds differently than moist or saturated soil. The physical process can be quite complex, but conceptually, it can be said that the response differs due to the reduction in infiltration capacity [41] .

Reference [42] presented an equation for the calculation of API as follows:

API = ∑ t = − 1 − i P t k − t . (1)

where i represents the number of antecedent days, k is a decay constant, also known as the recession coefficient, and P_t is the precipitation during day t. This model can also be considered as “retained rainfall” [43] .

Equation (1) presents the recession coefficient as constant, which disagrees with the physical processes that vary at different temporal scales. The recession coefficient should consider water losses due to evapotranspiration or drainage. Since evapotranspiration depends on various factors (e.g., temperature, solar radiation, wind speed, among others) that vary throughout the day, the recession coefficient should be calculated to account for this variability. Reference [27] proposed a reformulation in the calculation of the recession coefficient considering only the daily temperature variation, expressed by Equation (2):

k = 0.84 + δ ( 20 − T a v g ) . (2)

where T_avg is the daily average temperature (˚C), and δ is a sensitivity parameter (˚C⁻¹). The value 0.84 used in the equation is derived from the recommendation of [44] , and the temperature of 20˚C is the temperature used when the value 0.84 is considered.

Building upon Equation (2), this work proposes modifications that consider variability in the two constant values used in [27] , with the aim of capturing local dynamics, expressed by Equation (3):

k = k o p t + δ ( T c l t − T a v g ) . (3)

where k_opt represents the optimized value, for each point where the API is calculated, of the recession coefficient to be determined; T_clt is the calculated daily climatological mean temperature based on the recorded historical series. With the modifications, it is expected that the correlation between API and soil moisture will be improved since the parameters will be optimized taking into account the daily temperature variation (difference from the climatological mean) as well as the optimized recession coefficient.

The strategy used to find the optimal values of k_opt and δ involved constructing a function, k ( t , k o p t , δ , T c l t , T a v g ) , with daily variation (t) dependent on the standard recession coefficient (k_opt), the sensitivity parameter (δ), and temperatures (T_clt, T_avg). This function is used to calculate the API (Equation (1)), where the API ( P t , k , p s , n a p i ) depends on precipitation (P_t), the recession coefficient (k), calculated for each depth (ps) recorded in the soil moisture sensors (50 cm a 300 cm), and for different numbers of antecedent days (napi = 3, 7, 14, 21, and 28 days).

To achieve the optimal values of the parameters k_opt and δ, the resulting API was compared with soil moisture values for different depths, and the Pearson correlation coefficient was calculated. Different combinations of k_opt and δ were used to calculate the recession coefficient, aiming to adjust k to better represent the rate of water infiltration into the soil in the studied region. Figure 4 presents the ranges of values for k_opt and δ for different soil depths, considering that 28 days shows the best performance to Pearson correlation showed in Figure 5, this number of API days was taken to calculate the API, as well as the region where the correlation coefficients achieve the best results. It can be observed that the depths of 50 cm and 100 cm appear to be the best candidates to establish a relationship between API and soil moisture.

To determine the best correlation coefficients, a search was conducted for all combinations of the parameters k_opt and δ (within the intervals according to Figure 4), for different soil depths, as well as for different numbers of antecedent API days. Figure 5 summarizes the search for the best correlation values between soil moisture and API. It can be observed that for API with 3, 7, and 14 days of antecedents (represented by the numbers inside the circles in the figure), in almost all soil depths (except 50 cm), the correlations were below 0.70. For 21 and 28 days, the correlations achieved the best results, with values above 0.7 for depths of 50 cm and 100 cm, and the worst results were for depths of 200 cm and 300 cm. Therefore, to achieve the best correlation, the 28-day API and a depth of

50 cm should be used. It should be noted that the correlations were obtained with a limited two-year historical series, which may result in low correlation values and also limit correlations for other depths. Ideally, a good correlation should be achieved with deeper soils, especially at 250 cm depth, as this depth is more characteristic of mass movement events. The 0 cm value in Figure 5 refers to the average soil moisture calculated based on different depths, which did not show a good correlation.

In [27] , the parameter k_opt had a fixed value of 0.84, and the value of δ was explored between 0 and 0.03, obtaining the optimal value at 0.012. This optimization was valid for soil moisture up to a depth of 10 cm and API calculated for 3 days in advance, using an 11-year historical series. The limitation on the value of API_max was also explored to express the maximum water retention capacity in the

soil, which can overestimate soil moisture. This limitation is imposed in the API calculation when it reaches a certain threshold. In [27] , the threshold value was set at 35 mm, which considerably improved the correlation between API and soil moisture. This procedure was also used in the proposed API in this work, where the correlation coefficient reached a value of 0.89 at a depth of 50 cm, with API max = 120   mm .

With the optimized parameters, k_opt and δ, it was possible to achieve the best API, and from this, adjusted curves were constructed to represent the relationship between API and soil moisture (θ). Figure 6 presents the scatter plot of θ × API_c (top), with the best-fitted curve exhibiting an exponential behavior and a Pearson coefficient with a value of 0.82. In the lower part, there is the scatter plot of θ × API_max, with the fitted curve displaying a linear behavior and a Pearson coefficient of 0.89. It can be observed that by limiting the API calculation, the relationship with soil moisture tends to be linear, even using a short two-year historical series.

Since the aim is to establish the API to estimate soil moisture at points where there are no soil moisture sensors and to define thresholds that trigger mass movements, a survey of landslide occurrences in the study area was conducted. The summarized landslide records for the sensor data period (01/01/2020-31/12/2021) are presented in Table 1. It is evident that the occurrences were recorded during the rainy period of the municipality, spanning from October to April. Figure 7 displays the location of the registered landslide events, showing only those that occurred near the soil moisture sensor. This information is crucial for verifying the efficiency of using the API as a soil moisture estimator and consequently establishing thresholds.

The performance verification of the API as an index to predict mass movements can be initially observed through Figure 8, where the occurrence of landslides and the values for API_c, API_max (Figure 8(a)), observed soil moisture (θ), and soil moisture calculated through API_max (Figure 8(b)) are depicted. In Figure 8(a), it can be observed that in the first landslide event (07/02/2020), API_c reached its maximum before the landslide, and in the second event (09/12/2020), the maximum value of API_c coincided. Thus, API_c could not be considered a good predictor since it did not exhibit a consistent behavior during the events. For API_max, in both events, the behavior was consistent, where the events occurred above a certain maximum value, indicating good performance in predicting landslide events. Conducting a similar analysis for observed soil moisture and calculated

soil moisture using API_max, ( θ API ) (Figure 8(b)), it can be observed that the maximum value was reached before the landslides occurred, also indicating it to be a good predictor of events. It is worth noting that the flattened behavior for the API_max, observed in the figures, is due to the cutoff value in accumulated rainfall, as explained previously.

A more detailed analysis of Figure 8, highlighting the landslide dates, reveals that θ API reaches a maximum value of 0.29 a few days before the landslides occur on the two distinct dates (Figure 9). On the other hand, observed soil moisture shows variations concerning the occurrences, not indicating a clear threshold for landslides. Thus, an initial threshold can be established for the observed point, set at θ API = 0.29 . It is important to note that the historical series, both for soil moisture records and landslide events, is quite short, which could limit the calculated correlations as well as the performance analysis of the API as a potential predictor of mass movements.

The objective of this study was to establish a model for calculating the antecedent precipitation index and using it to obtain soil moisture to estimate a critical threshold for mass movements. Despite the short historical series of soil moisture and landslide events, the results have shown promise. Applying the fitted curve to calculate the soil moisture, θ API reached its maximum value ( θ API = 0.29 ) before the landslides occurred in the verified period, indicating that it can be used as a threshold for the vicinity of the soil moisture sensor (Figure 7). The next step of this work will be to expand this model to other regions within the municipality as well as to different historical periods. This expansion will help establish more precise and reliable thresholds. Consequently, this model can be used in a forecasting and mass movement alert system.

The authors thank FINEP (Financiadora de Estudos e Projetos) for financing the REDEGEO project (Carta Convite MCTI/FINEP/FNDCT 01/2016), responsible for the PCD Geo network installation and maintenance.

The authors declare no conflicts of interest regarding the publication of this paper.

De Moraes, M.A.E., Filho, W.M.M., Mendes, R.M., Bortolozo, C.A., Metodiev, D., De Andrade, M.R.M., Egas, H.M., Mendes, T.S.G. and Pampuch, L.A. (2024) Antecedent Precipitation Index to Estimate Soil Moisture and Correlate as a Triggering Process in the Occurrence of Landslides. International Journal of Geosciences, 15, 70-86. https://doi.org/10.4236/ijg.2024.151006