Predicting Drought and Subsidence Risks in France
- 1UQAM, Université du Québec à Montréal (UQAM), Montréal (Québec), Canada
- 2EURo Institut d’Actuariat (EURIA), Université de Brest, France
- 3Willis Re, Paris, France
- 1UQAM, Université du Québec à Montréal (UQAM), Montréal (Québec), Canada
- 2EURo Institut d’Actuariat (EURIA), Université de Brest, France
- 3Willis Re, Paris, France
Abstract. The economic consequences of drought episodes are increasingly important, although they are often difficult to apprehend in part because of the complexity of the underlying mechanisms. In this article, we will study one of the consequences of drought, namely the risk of subsidence (or more specifically clay shrinkage induced subsidence), for which insurance has been mandatory in France for several decades. Using data obtained from several insurers, representing about a quarter of the household insurance market, over the past twenty years, we propose some statistical models to predict the frequency but also the intensity of these droughts, for insurers, showing that climate change will have probably major economic consequences on this risk. But even if we use more advanced models than standard regression-type models (here random forests to capture non linearity and cross effects), it is still difficult to predict the economic cost of subsidence claims, even if all geophysical and climatic information is available.
Arthur Charpentier et al.
Status: final response (author comments only)
-
RC1: 'Comment on nhess-2021-214', Sien Kok, 24 Sep 2021
The content of the article is very novel and addresses a concrete societal problem: the difficulties in predicting building damage from subsidence, an increasingly costly climate risk for insurers. In some countries, such as France, this risk is insured via household policies. The methodology and results will be useful for the insurance industry in France, and may also be useful in other countries with similar established insurance products, as well as countries where insurance industry and policy makers investigate the possibly for developing insurance products for this risk. Overall the paper is well-structured. It compares various statistical models using a range of indicators to predict subsidence claims, calibrated against a historical dataset from insurance companies.
For detailed comments and suggestions, please see attached word file in zip folder.
- AC2: 'Reply on RC1', Arthur Charpentier, 01 Feb 2022
-
RC2: 'Comment on nhess-2021-214', Anonymous Referee #2, 20 Dec 2021
The article relates subsidence to shrinkage of clay happening from drought and long term precipitation deficit. While that is true, the clay shrinkage process relates back to groundwater withdrawal for fulfilling agricultural and urban demands during drought period. Even areas with high precipitation can subside if amount of precipitation is not sufficient to meet water need in an area. The article should clearly discuss how drought conditions increase groundwater demand and how groundwater withdrawal is affects clay layer causing subsidence.
The article should mention how water balance is represented in this research using combination of variable representative precipitation and soil moisture indices. The mechanism of land subsidence is related to the imbalance in hydrologic cycle. The predictor datasets used in the models are indices based mostly on precipitation and soil moisture. Evapotranspiration (ET) is an important component of the hydrologic cycle which has not been incorporated in these variables. ET strongly represents the water demand of an area. Therefore, an explanation on why ET was not added to the variables or how the used indices might fill the gap of ET should be mentioned in the article.
In tree-based method for number of claims prediction, no error metrices have been presented to represent each model performance compared to original observations. Same goes for cost predictions where no error metrices have been mentioned. In addition, a discussion can be added about among regression or tree-based models which one is better suited for such prediction. Future directions on how to make mnodel outputs more homogeneous to each other can be discussed. In colclusion, the methodology, result analaysis and uncertainty discussion are coherent. The paper shows some promising outcome and employs some useful techniques for subsidence risk modeling based on open source datasets.
-
AC1: 'Reply on RC2', Arthur Charpentier, 01 Feb 2022
1) "The article should clearly discuss how drought conditions increase groundwater demand and how groundwater withdrawal is affects clay layer causing subsidence"Those are indeed important in the US (https://www.usgs.gov/special-topics/water-science-school/science/land-subsidence) or in Asia, such as in Vietnam (https://hal.archives-ouvertes.fr/hal-01888487/document) or Jakarta (https://doi.org/10.1016/j.ocecoaman.2021.105775), but such a phenonema is not reported in France.2) "Evapotranspiration (ET) is an important component of the hydrologic cycle which has not been incorporated in these variables"Indeed, our variables do not take into account ET explicitly (we only consider soil moisture, soil temperature and precipitation), even if a correlation undoubtedly exists between our variables and ET because we consider soil heat, moisture and precipitation.
Other interesting indicators can indeed be constructed by adding ET (e.g. SPEI, more powerful than SPI), however this indicator is sensitive to the method of calculation of "potential evapotranspiration". And there were granularity issues with the data, that were not on the same scale as other variables. This is why we did not incorporate that component explicitely.3) "no error metrices have been presented to represent each model performance compared to original observations"Here a a summary of various statistics,TPR (%) Gini (%) RMSE (%) AIC BIC Binomial 18.5% 84.0% 0.0080 115,051 115,113 Poisson 18.5% 92.7% 0.0081 114,189 114,252 Quasi-poisson 17.8% 92.7% 0.0081 Negative-Binomial 20.9% 94.1% 0.0142 100,491 100,564 ZI Poisson 14.9% 93.2% 0.0079 71,154 71,259 ZI Negative-Binomial 18.1% 93.4% 0.0080 54,375 54,510 RFF 16.7% 91.5% 0.0083 RF poisson 15.2% 91.5% 0.0079 Such a table was not incorporated since we are not really confortable with those measures on counting variables (related to some Poisson loss). There are standard measures, with pros and cons, for continuous variables (RMSE) or binary ones (TPR, Gini), but no real consensus on counting variables. We can add that table if necessary
-
AC1: 'Reply on RC2', Arthur Charpentier, 01 Feb 2022
Arthur Charpentier et al.
Arthur Charpentier et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
935 | 361 | 22 | 1,318 | 19 | 13 |
- HTML: 935
- PDF: 361
- XML: 22
- Total: 1,318
- BibTeX: 19
- EndNote: 13
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1