The use of IoT technologies in various areas of our life, including environmental monitoring of green spaces, is increasing every year. One such solution is the TreeTalker sensor-based monitoring system, which collects data on various parameters of trees. One of the most important parameters is the rate of tree sap flow. Predicting the density of sap flow and studying the relationship between the parameters of trees and the environment is an urgent task. In this work, a statistical analysis of the data collected using the TreeTalker monitoring system was carried out. The data was pre-processed: outliers in the data were removed using mean value replacement, z-score replacement and cumulative moving average replacement. Groups of trees that were homogeneous in time were identified, and regression models were built to predict the sap flow parameter using auto-regressive moving average and linear modeling. The results obtained can be used for further studies of the dependence of the state of the tree on external factors.
Monitoring of the health of the trees helps to achieve a comprehensive view of ecosystems.
Nowadays environment is stressed by human activities. Providing a monitoring of trees health
can answer a lot of questions about the efectiveness of the measures to maintain ecosystem’s
health. TreeTalker(TT) is an IoT device that collects information about the health state of
the trees based on various internal and external factors. The main factor of the tree which is
considered the most important is sap flow [
This work has the following structure: the section 2 is devoted to a primary statistical analysis, work with the outliers in data with three methods: Mean Replacement, Z-score replacement and Cumulative Mean Average. In section 3 we perform prediction of the sap flow using linear models: auto-regressive moving average and linear regression.
CEUR Workshop Proceedings Workshop on information technology and scientific computing in the framework of the XI International Conference
TreeTalker sensors were installed on 195 trees in seven territories located in the center of
Moscow, on the RUDN University campus and in parks in the Moscow region. During 2019,
measurements were made of the parameters of 22 diferent tree species, diferent in age, as well
as in diferent states of ”health”. Every hour, data was collected on eight parameters – sap flow,
air temperature and humidity, negative pressure of water vapor in the leaves of a tree, angles of
tree deviations from the axis, wood moisture, temperature inside the trunk, and the normalized
relative vector of vegetation (NDVI) (Tab.1 and 2). Age group and VTA score are constants. For
the following work the data from the Troitsk Territory was selected.
time due to damage to the electronics after heavy rain (Fig. 1) and abnormally high-values (Fig.2).
The presence of gaps in measurements leads to false statistical analysis, as well as incorrect
modeling of dependencies. Therefore, the next task was to identify time-homogeneous groups
of data [
The presence of time gaps, that Fig.1 showed, makes it impossible to build models. The set
of time values = { 1, 2, ..., } consists of time-homogeneous subgroups = { ,
selected according to the algorithm [
Data: = { 1, 2, ..., } - set of time values Result: = { ,
for = 1, 2, ... do
+ } - time-homogeneous subgroups else end end if [ + 1] − [] > 1
then put [ + 1] in the new group;
Mean Replacement Method. The first way to work with outliers in your data is to replace outliers with mean values. – source row of one of eight parameters. – row after processing from outliers after applying the following algorithm:
1
The results of the replacement of outliers with mean values can be seen on Fig.5. From
(1)
(2)
the graph it is clear that the values are no more than 5 points of the sup tree flux units of
measurement.
analysis of values using – estimation and subsequent processing of abnormally high values
of the parameter [
1 √ − 1 =1
∑( − ) ̄ 2 =
, if ≤ if > curves is diferent. period:
The results of the replacement of outliers with mean values can be seen on Fig.6. From the graph it is clear that the values are no more than 5 points of the sup tree flux units of measurement as it was with the mean replacement method. However, the structure of the
Cumulative Moving Average Method. The third method used to deal with outliers in this
work is the cumulative moving average method. It is used for smoothing time series [
The methods of working with outliers are presented in the form of the algorithm (Alg.2). 3. Sap Flow Density Prediction
The construction of a mathematical model of sap flow and prediction of sap flow will help
to find out whether there is really a direct relationship between the sap flow and the air
temperature, whether other factors afect the sap flow parameter. To analyze the obtained
dependencies, 4 parameters for assessing the quality of the models were investigated: 2, –
statistics, root-mean-square error (RMSE) and mean absolute error (MAE) [
The ARMA model is an auto-regressive moving average model. The formula is as follows: = 1.4220 + 0.7150 + 1.39 + 0.134, (7) where = 0.7150 is the parameter of the model, is the parameter of the regression model, = 1.39 is the coeficient of the moving average, is the parameter of the moving average, = 1.4240 is a constant. The graph of the sup flow prognostication shows deceleration of the lfow (Fig. 8). Data: - original series, ̄ - mean value of original series, - z-score of original series Result: - processed series Case 1: Mean-value Replacement for = 1, 2, ... do if [] > ̄ then [] = ;̄ [] = [] else end else end if [] >
then [] = ; [] = [] end Case 2: Z-score Replacement for = 1, 2, ... do end Case 3: Cumulative Moving Average for = 1, 2, ... do
end end for = 1, 2, ... do [] =
∑=1 []
We will forecast 25 observations ahead. We will draw the plot of the result, which turned
out as a result of applying the linear regression model (Fig.9) [
It was found in the work that the negative pressure of water vapor in the leaves is significantly correlated with the parameter of tree sap flow. After analyzing the data, it was found that the sap flow of trees depends on 7 factors. The results obtained during the work showed which parameters should be taken into account when analyzing the state of the tree and predicting time-dependent factors. This study will provide a starting point for more sophisticated modeling approaches. For example, predicting a model using Fourier series can provide more accurate parameter estimates. In addition, assessing the flow density of sap flow is the main goal of researching the health of green spaces to predict changes in their health status.
The authors grateful to Dr. Alexey Yaroslavtsev (RUDN University) for providing the dataset from TreeTalker system.
The work was supported by the Russian Science Foundation, project 19-77-30012 (recipient Irina Kochetkova). This paper has been supported by the RUDN University Strategic Academic